2bc56e2a35262141859f8da21d09a54dec852e52
[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 rcu_read_lock();
182 if (rdev == NULL)
183 /* start at the beginning */
184 rdev = list_entry_rcu(&mddev->disks, struct md_rdev, same_set);
185 else {
186 /* release the previous rdev and start from there. */
187 rdev_dec_pending(rdev, mddev);
188 }
189 list_for_each_entry_continue_rcu(rdev, &mddev->disks, same_set) {
190 if (rdev->raid_disk >= 0 &&
191 !test_bit(Faulty, &rdev->flags)) {
192 /* this is a usable devices */
193 atomic_inc(&rdev->nr_pending);
194 rcu_read_unlock();
195 return rdev;
196 }
197 }
198 rcu_read_unlock();
199 return NULL;
200 }
201
202 static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
203 {
204 struct md_rdev *rdev = NULL;
205 struct block_device *bdev;
206 struct mddev *mddev = bitmap->mddev;
207 struct bitmap_storage *store = &bitmap->storage;
208 int node_offset = 0;
209
210 if (mddev_is_clustered(bitmap->mddev))
211 node_offset = bitmap->cluster_slot * store->file_pages;
212
213 while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
214 int size = PAGE_SIZE;
215 loff_t offset = mddev->bitmap_info.offset;
216
217 bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev;
218
219 if (page->index == store->file_pages-1) {
220 int last_page_size = store->bytes & (PAGE_SIZE-1);
221 if (last_page_size == 0)
222 last_page_size = PAGE_SIZE;
223 size = roundup(last_page_size,
224 bdev_logical_block_size(bdev));
225 }
226 /* Just make sure we aren't corrupting data or
227 * metadata
228 */
229 if (mddev->external) {
230 /* Bitmap could be anywhere. */
231 if (rdev->sb_start + offset + (page->index
232 * (PAGE_SIZE/512))
233 > rdev->data_offset
234 &&
235 rdev->sb_start + offset
236 < (rdev->data_offset + mddev->dev_sectors
237 + (PAGE_SIZE/512)))
238 goto bad_alignment;
239 } else if (offset < 0) {
240 /* DATA BITMAP METADATA */
241 if (offset
242 + (long)(page->index * (PAGE_SIZE/512))
243 + size/512 > 0)
244 /* bitmap runs in to metadata */
245 goto bad_alignment;
246 if (rdev->data_offset + mddev->dev_sectors
247 > rdev->sb_start + offset)
248 /* data runs in to bitmap */
249 goto bad_alignment;
250 } else if (rdev->sb_start < rdev->data_offset) {
251 /* METADATA BITMAP DATA */
252 if (rdev->sb_start
253 + offset
254 + page->index*(PAGE_SIZE/512) + size/512
255 > rdev->data_offset)
256 /* bitmap runs in to data */
257 goto bad_alignment;
258 } else {
259 /* DATA METADATA BITMAP - no problems */
260 }
261 md_super_write(mddev, rdev,
262 rdev->sb_start + offset
263 + page->index * (PAGE_SIZE/512),
264 size,
265 page);
266 }
267
268 if (wait)
269 md_super_wait(mddev);
270 return 0;
271
272 bad_alignment:
273 return -EINVAL;
274 }
275
276 static void bitmap_file_kick(struct bitmap *bitmap);
277 /*
278 * write out a page to a file
279 */
280 static void write_page(struct bitmap *bitmap, struct page *page, int wait)
281 {
282 struct buffer_head *bh;
283
284 if (bitmap->storage.file == NULL) {
285 switch (write_sb_page(bitmap, page, wait)) {
286 case -EINVAL:
287 set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
288 }
289 } else {
290
291 bh = page_buffers(page);
292
293 while (bh && bh->b_blocknr) {
294 atomic_inc(&bitmap->pending_writes);
295 set_buffer_locked(bh);
296 set_buffer_mapped(bh);
297 submit_bh(WRITE | REQ_SYNC, bh);
298 bh = bh->b_this_page;
299 }
300
301 if (wait)
302 wait_event(bitmap->write_wait,
303 atomic_read(&bitmap->pending_writes)==0);
304 }
305 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
306 bitmap_file_kick(bitmap);
307 }
308
309 static void end_bitmap_write(struct buffer_head *bh, int uptodate)
310 {
311 struct bitmap *bitmap = bh->b_private;
312
313 if (!uptodate)
314 set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
315 if (atomic_dec_and_test(&bitmap->pending_writes))
316 wake_up(&bitmap->write_wait);
317 }
318
319 /* copied from buffer.c */
320 static void
321 __clear_page_buffers(struct page *page)
322 {
323 ClearPagePrivate(page);
324 set_page_private(page, 0);
325 page_cache_release(page);
326 }
327 static void free_buffers(struct page *page)
328 {
329 struct buffer_head *bh;
330
331 if (!PagePrivate(page))
332 return;
333
334 bh = page_buffers(page);
335 while (bh) {
336 struct buffer_head *next = bh->b_this_page;
337 free_buffer_head(bh);
338 bh = next;
339 }
340 __clear_page_buffers(page);
341 put_page(page);
342 }
343
344 /* read a page from a file.
345 * We both read the page, and attach buffers to the page to record the
346 * address of each block (using bmap). These addresses will be used
347 * to write the block later, completely bypassing the filesystem.
348 * This usage is similar to how swap files are handled, and allows us
349 * to write to a file with no concerns of memory allocation failing.
350 */
351 static int read_page(struct file *file, unsigned long index,
352 struct bitmap *bitmap,
353 unsigned long count,
354 struct page *page)
355 {
356 int ret = 0;
357 struct inode *inode = file_inode(file);
358 struct buffer_head *bh;
359 sector_t block;
360
361 pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE,
362 (unsigned long long)index << PAGE_SHIFT);
363
364 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
365 if (!bh) {
366 ret = -ENOMEM;
367 goto out;
368 }
369 attach_page_buffers(page, bh);
370 block = index << (PAGE_SHIFT - inode->i_blkbits);
371 while (bh) {
372 if (count == 0)
373 bh->b_blocknr = 0;
374 else {
375 bh->b_blocknr = bmap(inode, block);
376 if (bh->b_blocknr == 0) {
377 /* Cannot use this file! */
378 ret = -EINVAL;
379 goto out;
380 }
381 bh->b_bdev = inode->i_sb->s_bdev;
382 if (count < (1<<inode->i_blkbits))
383 count = 0;
384 else
385 count -= (1<<inode->i_blkbits);
386
387 bh->b_end_io = end_bitmap_write;
388 bh->b_private = bitmap;
389 atomic_inc(&bitmap->pending_writes);
390 set_buffer_locked(bh);
391 set_buffer_mapped(bh);
392 submit_bh(READ, bh);
393 }
394 block++;
395 bh = bh->b_this_page;
396 }
397 page->index = index;
398
399 wait_event(bitmap->write_wait,
400 atomic_read(&bitmap->pending_writes)==0);
401 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
402 ret = -EIO;
403 out:
404 if (ret)
405 printk(KERN_ALERT "md: bitmap read error: (%dB @ %llu): %d\n",
406 (int)PAGE_SIZE,
407 (unsigned long long)index << PAGE_SHIFT,
408 ret);
409 return ret;
410 }
411
412 /*
413 * bitmap file superblock operations
414 */
415
416 /* update the event counter and sync the superblock to disk */
417 void bitmap_update_sb(struct bitmap *bitmap)
418 {
419 bitmap_super_t *sb;
420
421 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
422 return;
423 if (bitmap->mddev->bitmap_info.external)
424 return;
425 if (!bitmap->storage.sb_page) /* no superblock */
426 return;
427 sb = kmap_atomic(bitmap->storage.sb_page);
428 sb->events = cpu_to_le64(bitmap->mddev->events);
429 if (bitmap->mddev->events < bitmap->events_cleared)
430 /* rocking back to read-only */
431 bitmap->events_cleared = bitmap->mddev->events;
432 sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
433 sb->state = cpu_to_le32(bitmap->flags);
434 /* Just in case these have been changed via sysfs: */
435 sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ);
436 sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind);
437 /* This might have been changed by a reshape */
438 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
439 sb->chunksize = cpu_to_le32(bitmap->mddev->bitmap_info.chunksize);
440 sb->nodes = cpu_to_le32(bitmap->mddev->bitmap_info.nodes);
441 sb->sectors_reserved = cpu_to_le32(bitmap->mddev->
442 bitmap_info.space);
443 kunmap_atomic(sb);
444 write_page(bitmap, bitmap->storage.sb_page, 1);
445 }
446
447 /* print out the bitmap file superblock */
448 void bitmap_print_sb(struct bitmap *bitmap)
449 {
450 bitmap_super_t *sb;
451
452 if (!bitmap || !bitmap->storage.sb_page)
453 return;
454 sb = kmap_atomic(bitmap->storage.sb_page);
455 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
456 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
457 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
458 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
459 *(__u32 *)(sb->uuid+0),
460 *(__u32 *)(sb->uuid+4),
461 *(__u32 *)(sb->uuid+8),
462 *(__u32 *)(sb->uuid+12));
463 printk(KERN_DEBUG " events: %llu\n",
464 (unsigned long long) le64_to_cpu(sb->events));
465 printk(KERN_DEBUG "events cleared: %llu\n",
466 (unsigned long long) le64_to_cpu(sb->events_cleared));
467 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
468 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
469 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
470 printk(KERN_DEBUG " sync size: %llu KB\n",
471 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
472 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
473 kunmap_atomic(sb);
474 }
475
476 /*
477 * bitmap_new_disk_sb
478 * @bitmap
479 *
480 * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb
481 * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
482 * This function verifies 'bitmap_info' and populates the on-disk bitmap
483 * structure, which is to be written to disk.
484 *
485 * Returns: 0 on success, -Exxx on error
486 */
487 static int bitmap_new_disk_sb(struct bitmap *bitmap)
488 {
489 bitmap_super_t *sb;
490 unsigned long chunksize, daemon_sleep, write_behind;
491
492 bitmap->storage.sb_page = alloc_page(GFP_KERNEL);
493 if (bitmap->storage.sb_page == NULL)
494 return -ENOMEM;
495 bitmap->storage.sb_page->index = 0;
496
497 sb = kmap_atomic(bitmap->storage.sb_page);
498
499 sb->magic = cpu_to_le32(BITMAP_MAGIC);
500 sb->version = cpu_to_le32(BITMAP_MAJOR_HI);
501
502 chunksize = bitmap->mddev->bitmap_info.chunksize;
503 BUG_ON(!chunksize);
504 if (!is_power_of_2(chunksize)) {
505 kunmap_atomic(sb);
506 printk(KERN_ERR "bitmap chunksize not a power of 2\n");
507 return -EINVAL;
508 }
509 sb->chunksize = cpu_to_le32(chunksize);
510
511 daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep;
512 if (!daemon_sleep ||
513 (daemon_sleep < 1) || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) {
514 printk(KERN_INFO "Choosing daemon_sleep default (5 sec)\n");
515 daemon_sleep = 5 * HZ;
516 }
517 sb->daemon_sleep = cpu_to_le32(daemon_sleep);
518 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
519
520 /*
521 * FIXME: write_behind for RAID1. If not specified, what
522 * is a good choice? We choose COUNTER_MAX / 2 arbitrarily.
523 */
524 write_behind = bitmap->mddev->bitmap_info.max_write_behind;
525 if (write_behind > COUNTER_MAX)
526 write_behind = COUNTER_MAX / 2;
527 sb->write_behind = cpu_to_le32(write_behind);
528 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
529
530 /* keep the array size field of the bitmap superblock up to date */
531 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
532
533 memcpy(sb->uuid, bitmap->mddev->uuid, 16);
534
535 set_bit(BITMAP_STALE, &bitmap->flags);
536 sb->state = cpu_to_le32(bitmap->flags);
537 bitmap->events_cleared = bitmap->mddev->events;
538 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
539
540 kunmap_atomic(sb);
541
542 return 0;
543 }
544
545 /* read the superblock from the bitmap file and initialize some bitmap fields */
546 static int bitmap_read_sb(struct bitmap *bitmap)
547 {
548 char *reason = NULL;
549 bitmap_super_t *sb;
550 unsigned long chunksize, daemon_sleep, write_behind;
551 unsigned long long events;
552 int nodes = 0;
553 unsigned long sectors_reserved = 0;
554 int err = -EINVAL;
555 struct page *sb_page;
556
557 if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) {
558 chunksize = 128 * 1024 * 1024;
559 daemon_sleep = 5 * HZ;
560 write_behind = 0;
561 set_bit(BITMAP_STALE, &bitmap->flags);
562 err = 0;
563 goto out_no_sb;
564 }
565 /* page 0 is the superblock, read it... */
566 sb_page = alloc_page(GFP_KERNEL);
567 if (!sb_page)
568 return -ENOMEM;
569 bitmap->storage.sb_page = sb_page;
570
571 re_read:
572 /* If cluster_slot is set, the cluster is setup */
573 if (bitmap->cluster_slot >= 0) {
574 sector_t bm_blocks = bitmap->mddev->resync_max_sectors;
575
576 sector_div(bm_blocks,
577 bitmap->mddev->bitmap_info.chunksize >> 9);
578 /* bits to bytes */
579 bm_blocks = ((bm_blocks+7) >> 3) + sizeof(bitmap_super_t);
580 /* to 4k blocks */
581 bm_blocks = DIV_ROUND_UP_SECTOR_T(bm_blocks, 4096);
582 bitmap->mddev->bitmap_info.offset += bitmap->cluster_slot * (bm_blocks << 3);
583 pr_info("%s:%d bm slot: %d offset: %llu\n", __func__, __LINE__,
584 bitmap->cluster_slot, (unsigned long long)bitmap->mddev->bitmap_info.offset);
585 }
586
587 if (bitmap->storage.file) {
588 loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host);
589 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
590
591 err = read_page(bitmap->storage.file, 0,
592 bitmap, bytes, sb_page);
593 } else {
594 err = read_sb_page(bitmap->mddev,
595 bitmap->mddev->bitmap_info.offset,
596 sb_page,
597 0, sizeof(bitmap_super_t));
598 }
599 if (err)
600 return err;
601
602 err = -EINVAL;
603 sb = kmap_atomic(sb_page);
604
605 chunksize = le32_to_cpu(sb->chunksize);
606 daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ;
607 write_behind = le32_to_cpu(sb->write_behind);
608 sectors_reserved = le32_to_cpu(sb->sectors_reserved);
609 nodes = le32_to_cpu(sb->nodes);
610 strlcpy(bitmap->mddev->bitmap_info.cluster_name, sb->cluster_name, 64);
611
612 /* verify that the bitmap-specific fields are valid */
613 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
614 reason = "bad magic";
615 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
616 le32_to_cpu(sb->version) > BITMAP_MAJOR_HI)
617 reason = "unrecognized superblock version";
618 else if (chunksize < 512)
619 reason = "bitmap chunksize too small";
620 else if (!is_power_of_2(chunksize))
621 reason = "bitmap chunksize not a power of 2";
622 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT)
623 reason = "daemon sleep period out of range";
624 else if (write_behind > COUNTER_MAX)
625 reason = "write-behind limit out of range (0 - 16383)";
626 if (reason) {
627 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
628 bmname(bitmap), reason);
629 goto out;
630 }
631
632 /* keep the array size field of the bitmap superblock up to date */
633 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
634
635 if (bitmap->mddev->persistent) {
636 /*
637 * We have a persistent array superblock, so compare the
638 * bitmap's UUID and event counter to the mddev's
639 */
640 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
641 printk(KERN_INFO
642 "%s: bitmap superblock UUID mismatch\n",
643 bmname(bitmap));
644 goto out;
645 }
646 events = le64_to_cpu(sb->events);
647 if (!nodes && (events < bitmap->mddev->events)) {
648 printk(KERN_INFO
649 "%s: bitmap file is out of date (%llu < %llu) "
650 "-- forcing full recovery\n",
651 bmname(bitmap), events,
652 (unsigned long long) bitmap->mddev->events);
653 set_bit(BITMAP_STALE, &bitmap->flags);
654 }
655 }
656
657 /* assign fields using values from superblock */
658 bitmap->flags |= le32_to_cpu(sb->state);
659 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
660 set_bit(BITMAP_HOSTENDIAN, &bitmap->flags);
661 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
662 strlcpy(bitmap->mddev->bitmap_info.cluster_name, sb->cluster_name, 64);
663 err = 0;
664
665 out:
666 kunmap_atomic(sb);
667 /* Assiging chunksize is required for "re_read" */
668 bitmap->mddev->bitmap_info.chunksize = chunksize;
669 if (nodes && (bitmap->cluster_slot < 0)) {
670 err = md_setup_cluster(bitmap->mddev, nodes);
671 if (err) {
672 pr_err("%s: Could not setup cluster service (%d)\n",
673 bmname(bitmap), err);
674 goto out_no_sb;
675 }
676 bitmap->cluster_slot = md_cluster_ops->slot_number(bitmap->mddev);
677 goto re_read;
678 }
679
680
681 out_no_sb:
682 if (test_bit(BITMAP_STALE, &bitmap->flags))
683 bitmap->events_cleared = bitmap->mddev->events;
684 bitmap->mddev->bitmap_info.chunksize = chunksize;
685 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
686 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
687 bitmap->mddev->bitmap_info.nodes = nodes;
688 if (bitmap->mddev->bitmap_info.space == 0 ||
689 bitmap->mddev->bitmap_info.space > sectors_reserved)
690 bitmap->mddev->bitmap_info.space = sectors_reserved;
691 if (err) {
692 bitmap_print_sb(bitmap);
693 if (bitmap->cluster_slot < 0)
694 md_cluster_stop(bitmap->mddev);
695 }
696 return err;
697 }
698
699 /*
700 * general bitmap file operations
701 */
702
703 /*
704 * on-disk bitmap:
705 *
706 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
707 * file a page at a time. There's a superblock at the start of the file.
708 */
709 /* calculate the index of the page that contains this bit */
710 static inline unsigned long file_page_index(struct bitmap_storage *store,
711 unsigned long chunk)
712 {
713 if (store->sb_page)
714 chunk += sizeof(bitmap_super_t) << 3;
715 return chunk >> PAGE_BIT_SHIFT;
716 }
717
718 /* calculate the (bit) offset of this bit within a page */
719 static inline unsigned long file_page_offset(struct bitmap_storage *store,
720 unsigned long chunk)
721 {
722 if (store->sb_page)
723 chunk += sizeof(bitmap_super_t) << 3;
724 return chunk & (PAGE_BITS - 1);
725 }
726
727 /*
728 * return a pointer to the page in the filemap that contains the given bit
729 *
730 */
731 static inline struct page *filemap_get_page(struct bitmap_storage *store,
732 unsigned long chunk)
733 {
734 if (file_page_index(store, chunk) >= store->file_pages)
735 return NULL;
736 return store->filemap[file_page_index(store, chunk)];
737 }
738
739 static int bitmap_storage_alloc(struct bitmap_storage *store,
740 unsigned long chunks, int with_super,
741 int slot_number)
742 {
743 int pnum, offset = 0;
744 unsigned long num_pages;
745 unsigned long bytes;
746
747 bytes = DIV_ROUND_UP(chunks, 8);
748 if (with_super)
749 bytes += sizeof(bitmap_super_t);
750
751 num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE);
752 offset = slot_number * (num_pages - 1);
753
754 store->filemap = kmalloc(sizeof(struct page *)
755 * num_pages, GFP_KERNEL);
756 if (!store->filemap)
757 return -ENOMEM;
758
759 if (with_super && !store->sb_page) {
760 store->sb_page = alloc_page(GFP_KERNEL|__GFP_ZERO);
761 if (store->sb_page == NULL)
762 return -ENOMEM;
763 }
764
765 pnum = 0;
766 if (store->sb_page) {
767 store->filemap[0] = store->sb_page;
768 pnum = 1;
769 store->sb_page->index = offset;
770 }
771
772 for ( ; pnum < num_pages; pnum++) {
773 store->filemap[pnum] = alloc_page(GFP_KERNEL|__GFP_ZERO);
774 if (!store->filemap[pnum]) {
775 store->file_pages = pnum;
776 return -ENOMEM;
777 }
778 store->filemap[pnum]->index = pnum + offset;
779 }
780 store->file_pages = pnum;
781
782 /* We need 4 bits per page, rounded up to a multiple
783 * of sizeof(unsigned long) */
784 store->filemap_attr = kzalloc(
785 roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
786 GFP_KERNEL);
787 if (!store->filemap_attr)
788 return -ENOMEM;
789
790 store->bytes = bytes;
791
792 return 0;
793 }
794
795 static void bitmap_file_unmap(struct bitmap_storage *store)
796 {
797 struct page **map, *sb_page;
798 int pages;
799 struct file *file;
800
801 file = store->file;
802 map = store->filemap;
803 pages = store->file_pages;
804 sb_page = store->sb_page;
805
806 while (pages--)
807 if (map[pages] != sb_page) /* 0 is sb_page, release it below */
808 free_buffers(map[pages]);
809 kfree(map);
810 kfree(store->filemap_attr);
811
812 if (sb_page)
813 free_buffers(sb_page);
814
815 if (file) {
816 struct inode *inode = file_inode(file);
817 invalidate_mapping_pages(inode->i_mapping, 0, -1);
818 fput(file);
819 }
820 }
821
822 /*
823 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
824 * then it is no longer reliable, so we stop using it and we mark the file
825 * as failed in the superblock
826 */
827 static void bitmap_file_kick(struct bitmap *bitmap)
828 {
829 char *path, *ptr = NULL;
830
831 if (!test_and_set_bit(BITMAP_STALE, &bitmap->flags)) {
832 bitmap_update_sb(bitmap);
833
834 if (bitmap->storage.file) {
835 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
836 if (path)
837 ptr = d_path(&bitmap->storage.file->f_path,
838 path, PAGE_SIZE);
839
840 printk(KERN_ALERT
841 "%s: kicking failed bitmap file %s from array!\n",
842 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
843
844 kfree(path);
845 } else
846 printk(KERN_ALERT
847 "%s: disabling internal bitmap due to errors\n",
848 bmname(bitmap));
849 }
850 }
851
852 enum bitmap_page_attr {
853 BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */
854 BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned.
855 * i.e. counter is 1 or 2. */
856 BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */
857 };
858
859 static inline void set_page_attr(struct bitmap *bitmap, int pnum,
860 enum bitmap_page_attr attr)
861 {
862 set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
863 }
864
865 static inline void clear_page_attr(struct bitmap *bitmap, int pnum,
866 enum bitmap_page_attr attr)
867 {
868 clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
869 }
870
871 static inline int test_page_attr(struct bitmap *bitmap, int pnum,
872 enum bitmap_page_attr attr)
873 {
874 return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
875 }
876
877 static inline int test_and_clear_page_attr(struct bitmap *bitmap, int pnum,
878 enum bitmap_page_attr attr)
879 {
880 return test_and_clear_bit((pnum<<2) + attr,
881 bitmap->storage.filemap_attr);
882 }
883 /*
884 * bitmap_file_set_bit -- called before performing a write to the md device
885 * to set (and eventually sync) a particular bit in the bitmap file
886 *
887 * we set the bit immediately, then we record the page number so that
888 * when an unplug occurs, we can flush the dirty pages out to disk
889 */
890 static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
891 {
892 unsigned long bit;
893 struct page *page;
894 void *kaddr;
895 unsigned long chunk = block >> bitmap->counts.chunkshift;
896
897 page = filemap_get_page(&bitmap->storage, chunk);
898 if (!page)
899 return;
900 bit = file_page_offset(&bitmap->storage, chunk);
901
902 /* set the bit */
903 kaddr = kmap_atomic(page);
904 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
905 set_bit(bit, kaddr);
906 else
907 set_bit_le(bit, kaddr);
908 kunmap_atomic(kaddr);
909 pr_debug("set file bit %lu page %lu\n", bit, page->index);
910 /* record page number so it gets flushed to disk when unplug occurs */
911 set_page_attr(bitmap, page->index, BITMAP_PAGE_DIRTY);
912 }
913
914 static void bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block)
915 {
916 unsigned long bit;
917 struct page *page;
918 void *paddr;
919 unsigned long chunk = block >> bitmap->counts.chunkshift;
920
921 page = filemap_get_page(&bitmap->storage, chunk);
922 if (!page)
923 return;
924 bit = file_page_offset(&bitmap->storage, chunk);
925 paddr = kmap_atomic(page);
926 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
927 clear_bit(bit, paddr);
928 else
929 clear_bit_le(bit, paddr);
930 kunmap_atomic(paddr);
931 if (!test_page_attr(bitmap, page->index, BITMAP_PAGE_NEEDWRITE)) {
932 set_page_attr(bitmap, page->index, BITMAP_PAGE_PENDING);
933 bitmap->allclean = 0;
934 }
935 }
936
937 static int bitmap_file_test_bit(struct bitmap *bitmap, sector_t block)
938 {
939 unsigned long bit;
940 struct page *page;
941 void *paddr;
942 unsigned long chunk = block >> bitmap->counts.chunkshift;
943 int set = 0;
944
945 page = filemap_get_page(&bitmap->storage, chunk);
946 if (!page)
947 return -EINVAL;
948 bit = file_page_offset(&bitmap->storage, chunk);
949 paddr = kmap_atomic(page);
950 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
951 set = test_bit(bit, paddr);
952 else
953 set = test_bit_le(bit, paddr);
954 kunmap_atomic(paddr);
955 return set;
956 }
957
958
959 /* this gets called when the md device is ready to unplug its underlying
960 * (slave) device queues -- before we let any writes go down, we need to
961 * sync the dirty pages of the bitmap file to disk */
962 void bitmap_unplug(struct bitmap *bitmap)
963 {
964 unsigned long i;
965 int dirty, need_write;
966
967 if (!bitmap || !bitmap->storage.filemap ||
968 test_bit(BITMAP_STALE, &bitmap->flags))
969 return;
970
971 /* look at each page to see if there are any set bits that need to be
972 * flushed out to disk */
973 for (i = 0; i < bitmap->storage.file_pages; i++) {
974 if (!bitmap->storage.filemap)
975 return;
976 dirty = test_and_clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
977 need_write = test_and_clear_page_attr(bitmap, i,
978 BITMAP_PAGE_NEEDWRITE);
979 if (dirty || need_write) {
980 clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING);
981 write_page(bitmap, bitmap->storage.filemap[i], 0);
982 }
983 }
984 if (bitmap->storage.file)
985 wait_event(bitmap->write_wait,
986 atomic_read(&bitmap->pending_writes)==0);
987 else
988 md_super_wait(bitmap->mddev);
989
990 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
991 bitmap_file_kick(bitmap);
992 }
993 EXPORT_SYMBOL(bitmap_unplug);
994
995 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
996 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
997 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
998 * memory mapping of the bitmap file
999 * Special cases:
1000 * if there's no bitmap file, or if the bitmap file had been
1001 * previously kicked from the array, we mark all the bits as
1002 * 1's in order to cause a full resync.
1003 *
1004 * We ignore all bits for sectors that end earlier than 'start'.
1005 * This is used when reading an out-of-date bitmap...
1006 */
1007 static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
1008 {
1009 unsigned long i, chunks, index, oldindex, bit, node_offset = 0;
1010 struct page *page = NULL;
1011 unsigned long bit_cnt = 0;
1012 struct file *file;
1013 unsigned long offset;
1014 int outofdate;
1015 int ret = -ENOSPC;
1016 void *paddr;
1017 struct bitmap_storage *store = &bitmap->storage;
1018
1019 chunks = bitmap->counts.chunks;
1020 file = store->file;
1021
1022 if (!file && !bitmap->mddev->bitmap_info.offset) {
1023 /* No permanent bitmap - fill with '1s'. */
1024 store->filemap = NULL;
1025 store->file_pages = 0;
1026 for (i = 0; i < chunks ; i++) {
1027 /* if the disk bit is set, set the memory bit */
1028 int needed = ((sector_t)(i+1) << (bitmap->counts.chunkshift)
1029 >= start);
1030 bitmap_set_memory_bits(bitmap,
1031 (sector_t)i << bitmap->counts.chunkshift,
1032 needed);
1033 }
1034 return 0;
1035 }
1036
1037 outofdate = test_bit(BITMAP_STALE, &bitmap->flags);
1038 if (outofdate)
1039 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
1040 "recovery\n", bmname(bitmap));
1041
1042 if (file && i_size_read(file->f_mapping->host) < store->bytes) {
1043 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
1044 bmname(bitmap),
1045 (unsigned long) i_size_read(file->f_mapping->host),
1046 store->bytes);
1047 goto err;
1048 }
1049
1050 oldindex = ~0L;
1051 offset = 0;
1052 if (!bitmap->mddev->bitmap_info.external)
1053 offset = sizeof(bitmap_super_t);
1054
1055 if (mddev_is_clustered(bitmap->mddev))
1056 node_offset = bitmap->cluster_slot * (DIV_ROUND_UP(store->bytes, PAGE_SIZE));
1057
1058 for (i = 0; i < chunks; i++) {
1059 int b;
1060 index = file_page_index(&bitmap->storage, i);
1061 bit = file_page_offset(&bitmap->storage, i);
1062 if (index != oldindex) { /* this is a new page, read it in */
1063 int count;
1064 /* unmap the old page, we're done with it */
1065 if (index == store->file_pages-1)
1066 count = store->bytes - index * PAGE_SIZE;
1067 else
1068 count = PAGE_SIZE;
1069 page = store->filemap[index];
1070 if (file)
1071 ret = read_page(file, index, bitmap,
1072 count, page);
1073 else
1074 ret = read_sb_page(
1075 bitmap->mddev,
1076 bitmap->mddev->bitmap_info.offset,
1077 page,
1078 index + node_offset, count);
1079
1080 if (ret)
1081 goto err;
1082
1083 oldindex = index;
1084
1085 if (outofdate) {
1086 /*
1087 * if bitmap is out of date, dirty the
1088 * whole page and write it out
1089 */
1090 paddr = kmap_atomic(page);
1091 memset(paddr + offset, 0xff,
1092 PAGE_SIZE - offset);
1093 kunmap_atomic(paddr);
1094 write_page(bitmap, page, 1);
1095
1096 ret = -EIO;
1097 if (test_bit(BITMAP_WRITE_ERROR,
1098 &bitmap->flags))
1099 goto err;
1100 }
1101 }
1102 paddr = kmap_atomic(page);
1103 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
1104 b = test_bit(bit, paddr);
1105 else
1106 b = test_bit_le(bit, paddr);
1107 kunmap_atomic(paddr);
1108 if (b) {
1109 /* if the disk bit is set, set the memory bit */
1110 int needed = ((sector_t)(i+1) << bitmap->counts.chunkshift
1111 >= start);
1112 bitmap_set_memory_bits(bitmap,
1113 (sector_t)i << bitmap->counts.chunkshift,
1114 needed);
1115 bit_cnt++;
1116 }
1117 offset = 0;
1118 }
1119
1120 printk(KERN_INFO "%s: bitmap initialized from disk: "
1121 "read %lu pages, set %lu of %lu bits\n",
1122 bmname(bitmap), store->file_pages,
1123 bit_cnt, chunks);
1124
1125 return 0;
1126
1127 err:
1128 printk(KERN_INFO "%s: bitmap initialisation failed: %d\n",
1129 bmname(bitmap), ret);
1130 return ret;
1131 }
1132
1133 void bitmap_write_all(struct bitmap *bitmap)
1134 {
1135 /* We don't actually write all bitmap blocks here,
1136 * just flag them as needing to be written
1137 */
1138 int i;
1139
1140 if (!bitmap || !bitmap->storage.filemap)
1141 return;
1142 if (bitmap->storage.file)
1143 /* Only one copy, so nothing needed */
1144 return;
1145
1146 for (i = 0; i < bitmap->storage.file_pages; i++)
1147 set_page_attr(bitmap, i,
1148 BITMAP_PAGE_NEEDWRITE);
1149 bitmap->allclean = 0;
1150 }
1151
1152 static void bitmap_count_page(struct bitmap_counts *bitmap,
1153 sector_t offset, int inc)
1154 {
1155 sector_t chunk = offset >> bitmap->chunkshift;
1156 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1157 bitmap->bp[page].count += inc;
1158 bitmap_checkfree(bitmap, page);
1159 }
1160
1161 static void bitmap_set_pending(struct bitmap_counts *bitmap, sector_t offset)
1162 {
1163 sector_t chunk = offset >> bitmap->chunkshift;
1164 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1165 struct bitmap_page *bp = &bitmap->bp[page];
1166
1167 if (!bp->pending)
1168 bp->pending = 1;
1169 }
1170
1171 static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap,
1172 sector_t offset, sector_t *blocks,
1173 int create);
1174
1175 /*
1176 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1177 * out to disk
1178 */
1179
1180 void bitmap_daemon_work(struct mddev *mddev)
1181 {
1182 struct bitmap *bitmap;
1183 unsigned long j;
1184 unsigned long nextpage;
1185 sector_t blocks;
1186 struct bitmap_counts *counts;
1187
1188 /* Use a mutex to guard daemon_work against
1189 * bitmap_destroy.
1190 */
1191 mutex_lock(&mddev->bitmap_info.mutex);
1192 bitmap = mddev->bitmap;
1193 if (bitmap == NULL) {
1194 mutex_unlock(&mddev->bitmap_info.mutex);
1195 return;
1196 }
1197 if (time_before(jiffies, bitmap->daemon_lastrun
1198 + mddev->bitmap_info.daemon_sleep))
1199 goto done;
1200
1201 bitmap->daemon_lastrun = jiffies;
1202 if (bitmap->allclean) {
1203 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1204 goto done;
1205 }
1206 bitmap->allclean = 1;
1207
1208 /* Any file-page which is PENDING now needs to be written.
1209 * So set NEEDWRITE now, then after we make any last-minute changes
1210 * we will write it.
1211 */
1212 for (j = 0; j < bitmap->storage.file_pages; j++)
1213 if (test_and_clear_page_attr(bitmap, j,
1214 BITMAP_PAGE_PENDING))
1215 set_page_attr(bitmap, j,
1216 BITMAP_PAGE_NEEDWRITE);
1217
1218 if (bitmap->need_sync &&
1219 mddev->bitmap_info.external == 0) {
1220 /* Arrange for superblock update as well as
1221 * other changes */
1222 bitmap_super_t *sb;
1223 bitmap->need_sync = 0;
1224 if (bitmap->storage.filemap) {
1225 sb = kmap_atomic(bitmap->storage.sb_page);
1226 sb->events_cleared =
1227 cpu_to_le64(bitmap->events_cleared);
1228 kunmap_atomic(sb);
1229 set_page_attr(bitmap, 0,
1230 BITMAP_PAGE_NEEDWRITE);
1231 }
1232 }
1233 /* Now look at the bitmap counters and if any are '2' or '1',
1234 * decrement and handle accordingly.
1235 */
1236 counts = &bitmap->counts;
1237 spin_lock_irq(&counts->lock);
1238 nextpage = 0;
1239 for (j = 0; j < counts->chunks; j++) {
1240 bitmap_counter_t *bmc;
1241 sector_t block = (sector_t)j << counts->chunkshift;
1242
1243 if (j == nextpage) {
1244 nextpage += PAGE_COUNTER_RATIO;
1245 if (!counts->bp[j >> PAGE_COUNTER_SHIFT].pending) {
1246 j |= PAGE_COUNTER_MASK;
1247 continue;
1248 }
1249 counts->bp[j >> PAGE_COUNTER_SHIFT].pending = 0;
1250 }
1251 bmc = bitmap_get_counter(counts,
1252 block,
1253 &blocks, 0);
1254
1255 if (!bmc) {
1256 j |= PAGE_COUNTER_MASK;
1257 continue;
1258 }
1259 if (*bmc == 1 && !bitmap->need_sync) {
1260 /* We can clear the bit */
1261 *bmc = 0;
1262 bitmap_count_page(counts, block, -1);
1263 bitmap_file_clear_bit(bitmap, block);
1264 } else if (*bmc && *bmc <= 2) {
1265 *bmc = 1;
1266 bitmap_set_pending(counts, block);
1267 bitmap->allclean = 0;
1268 }
1269 }
1270 spin_unlock_irq(&counts->lock);
1271
1272 /* Now start writeout on any page in NEEDWRITE that isn't DIRTY.
1273 * DIRTY pages need to be written by bitmap_unplug so it can wait
1274 * for them.
1275 * If we find any DIRTY page we stop there and let bitmap_unplug
1276 * handle all the rest. This is important in the case where
1277 * the first blocking holds the superblock and it has been updated.
1278 * We mustn't write any other blocks before the superblock.
1279 */
1280 for (j = 0;
1281 j < bitmap->storage.file_pages
1282 && !test_bit(BITMAP_STALE, &bitmap->flags);
1283 j++) {
1284 if (test_page_attr(bitmap, j,
1285 BITMAP_PAGE_DIRTY))
1286 /* bitmap_unplug will handle the rest */
1287 break;
1288 if (test_and_clear_page_attr(bitmap, j,
1289 BITMAP_PAGE_NEEDWRITE)) {
1290 write_page(bitmap, bitmap->storage.filemap[j], 0);
1291 }
1292 }
1293
1294 done:
1295 if (bitmap->allclean == 0)
1296 mddev->thread->timeout =
1297 mddev->bitmap_info.daemon_sleep;
1298 mutex_unlock(&mddev->bitmap_info.mutex);
1299 }
1300
1301 static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap,
1302 sector_t offset, sector_t *blocks,
1303 int create)
1304 __releases(bitmap->lock)
1305 __acquires(bitmap->lock)
1306 {
1307 /* If 'create', we might release the lock and reclaim it.
1308 * The lock must have been taken with interrupts enabled.
1309 * If !create, we don't release the lock.
1310 */
1311 sector_t chunk = offset >> bitmap->chunkshift;
1312 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1313 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1314 sector_t csize;
1315 int err;
1316
1317 err = bitmap_checkpage(bitmap, page, create);
1318
1319 if (bitmap->bp[page].hijacked ||
1320 bitmap->bp[page].map == NULL)
1321 csize = ((sector_t)1) << (bitmap->chunkshift +
1322 PAGE_COUNTER_SHIFT - 1);
1323 else
1324 csize = ((sector_t)1) << bitmap->chunkshift;
1325 *blocks = csize - (offset & (csize - 1));
1326
1327 if (err < 0)
1328 return NULL;
1329
1330 /* now locked ... */
1331
1332 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1333 /* should we use the first or second counter field
1334 * of the hijacked pointer? */
1335 int hi = (pageoff > PAGE_COUNTER_MASK);
1336 return &((bitmap_counter_t *)
1337 &bitmap->bp[page].map)[hi];
1338 } else /* page is allocated */
1339 return (bitmap_counter_t *)
1340 &(bitmap->bp[page].map[pageoff]);
1341 }
1342
1343 int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1344 {
1345 if (!bitmap)
1346 return 0;
1347
1348 if (behind) {
1349 int bw;
1350 atomic_inc(&bitmap->behind_writes);
1351 bw = atomic_read(&bitmap->behind_writes);
1352 if (bw > bitmap->behind_writes_used)
1353 bitmap->behind_writes_used = bw;
1354
1355 pr_debug("inc write-behind count %d/%lu\n",
1356 bw, bitmap->mddev->bitmap_info.max_write_behind);
1357 }
1358
1359 while (sectors) {
1360 sector_t blocks;
1361 bitmap_counter_t *bmc;
1362
1363 spin_lock_irq(&bitmap->counts.lock);
1364 bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 1);
1365 if (!bmc) {
1366 spin_unlock_irq(&bitmap->counts.lock);
1367 return 0;
1368 }
1369
1370 if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) {
1371 DEFINE_WAIT(__wait);
1372 /* note that it is safe to do the prepare_to_wait
1373 * after the test as long as we do it before dropping
1374 * the spinlock.
1375 */
1376 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1377 TASK_UNINTERRUPTIBLE);
1378 spin_unlock_irq(&bitmap->counts.lock);
1379 schedule();
1380 finish_wait(&bitmap->overflow_wait, &__wait);
1381 continue;
1382 }
1383
1384 switch (*bmc) {
1385 case 0:
1386 bitmap_file_set_bit(bitmap, offset);
1387 bitmap_count_page(&bitmap->counts, offset, 1);
1388 /* fall through */
1389 case 1:
1390 *bmc = 2;
1391 }
1392
1393 (*bmc)++;
1394
1395 spin_unlock_irq(&bitmap->counts.lock);
1396
1397 offset += blocks;
1398 if (sectors > blocks)
1399 sectors -= blocks;
1400 else
1401 sectors = 0;
1402 }
1403 return 0;
1404 }
1405 EXPORT_SYMBOL(bitmap_startwrite);
1406
1407 void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
1408 int success, int behind)
1409 {
1410 if (!bitmap)
1411 return;
1412 if (behind) {
1413 if (atomic_dec_and_test(&bitmap->behind_writes))
1414 wake_up(&bitmap->behind_wait);
1415 pr_debug("dec write-behind count %d/%lu\n",
1416 atomic_read(&bitmap->behind_writes),
1417 bitmap->mddev->bitmap_info.max_write_behind);
1418 }
1419
1420 while (sectors) {
1421 sector_t blocks;
1422 unsigned long flags;
1423 bitmap_counter_t *bmc;
1424
1425 spin_lock_irqsave(&bitmap->counts.lock, flags);
1426 bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 0);
1427 if (!bmc) {
1428 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1429 return;
1430 }
1431
1432 if (success && !bitmap->mddev->degraded &&
1433 bitmap->events_cleared < bitmap->mddev->events) {
1434 bitmap->events_cleared = bitmap->mddev->events;
1435 bitmap->need_sync = 1;
1436 sysfs_notify_dirent_safe(bitmap->sysfs_can_clear);
1437 }
1438
1439 if (!success && !NEEDED(*bmc))
1440 *bmc |= NEEDED_MASK;
1441
1442 if (COUNTER(*bmc) == COUNTER_MAX)
1443 wake_up(&bitmap->overflow_wait);
1444
1445 (*bmc)--;
1446 if (*bmc <= 2) {
1447 bitmap_set_pending(&bitmap->counts, offset);
1448 bitmap->allclean = 0;
1449 }
1450 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1451 offset += blocks;
1452 if (sectors > blocks)
1453 sectors -= blocks;
1454 else
1455 sectors = 0;
1456 }
1457 }
1458 EXPORT_SYMBOL(bitmap_endwrite);
1459
1460 static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1461 int degraded)
1462 {
1463 bitmap_counter_t *bmc;
1464 int rv;
1465 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1466 *blocks = 1024;
1467 return 1; /* always resync if no bitmap */
1468 }
1469 spin_lock_irq(&bitmap->counts.lock);
1470 bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1471 rv = 0;
1472 if (bmc) {
1473 /* locked */
1474 if (RESYNC(*bmc))
1475 rv = 1;
1476 else if (NEEDED(*bmc)) {
1477 rv = 1;
1478 if (!degraded) { /* don't set/clear bits if degraded */
1479 *bmc |= RESYNC_MASK;
1480 *bmc &= ~NEEDED_MASK;
1481 }
1482 }
1483 }
1484 spin_unlock_irq(&bitmap->counts.lock);
1485 return rv;
1486 }
1487
1488 int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1489 int degraded)
1490 {
1491 /* bitmap_start_sync must always report on multiples of whole
1492 * pages, otherwise resync (which is very PAGE_SIZE based) will
1493 * get confused.
1494 * So call __bitmap_start_sync repeatedly (if needed) until
1495 * At least PAGE_SIZE>>9 blocks are covered.
1496 * Return the 'or' of the result.
1497 */
1498 int rv = 0;
1499 sector_t blocks1;
1500
1501 *blocks = 0;
1502 while (*blocks < (PAGE_SIZE>>9)) {
1503 rv |= __bitmap_start_sync(bitmap, offset,
1504 &blocks1, degraded);
1505 offset += blocks1;
1506 *blocks += blocks1;
1507 }
1508 return rv;
1509 }
1510 EXPORT_SYMBOL(bitmap_start_sync);
1511
1512 void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted)
1513 {
1514 bitmap_counter_t *bmc;
1515 unsigned long flags;
1516
1517 if (bitmap == NULL) {
1518 *blocks = 1024;
1519 return;
1520 }
1521 spin_lock_irqsave(&bitmap->counts.lock, flags);
1522 bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1523 if (bmc == NULL)
1524 goto unlock;
1525 /* locked */
1526 if (RESYNC(*bmc)) {
1527 *bmc &= ~RESYNC_MASK;
1528
1529 if (!NEEDED(*bmc) && aborted)
1530 *bmc |= NEEDED_MASK;
1531 else {
1532 if (*bmc <= 2) {
1533 bitmap_set_pending(&bitmap->counts, offset);
1534 bitmap->allclean = 0;
1535 }
1536 }
1537 }
1538 unlock:
1539 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1540 }
1541 EXPORT_SYMBOL(bitmap_end_sync);
1542
1543 void bitmap_close_sync(struct bitmap *bitmap)
1544 {
1545 /* Sync has finished, and any bitmap chunks that weren't synced
1546 * properly have been aborted. It remains to us to clear the
1547 * RESYNC bit wherever it is still on
1548 */
1549 sector_t sector = 0;
1550 sector_t blocks;
1551 if (!bitmap)
1552 return;
1553 while (sector < bitmap->mddev->resync_max_sectors) {
1554 bitmap_end_sync(bitmap, sector, &blocks, 0);
1555 sector += blocks;
1556 }
1557 }
1558 EXPORT_SYMBOL(bitmap_close_sync);
1559
1560 void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector)
1561 {
1562 sector_t s = 0;
1563 sector_t blocks;
1564
1565 if (!bitmap)
1566 return;
1567 if (sector == 0) {
1568 bitmap->last_end_sync = jiffies;
1569 return;
1570 }
1571 if (time_before(jiffies, (bitmap->last_end_sync
1572 + bitmap->mddev->bitmap_info.daemon_sleep)))
1573 return;
1574 wait_event(bitmap->mddev->recovery_wait,
1575 atomic_read(&bitmap->mddev->recovery_active) == 0);
1576
1577 bitmap->mddev->curr_resync_completed = sector;
1578 set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags);
1579 sector &= ~((1ULL << bitmap->counts.chunkshift) - 1);
1580 s = 0;
1581 while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1582 bitmap_end_sync(bitmap, s, &blocks, 0);
1583 s += blocks;
1584 }
1585 bitmap->last_end_sync = jiffies;
1586 sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed");
1587 }
1588 EXPORT_SYMBOL(bitmap_cond_end_sync);
1589
1590 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1591 {
1592 /* For each chunk covered by any of these sectors, set the
1593 * counter to 2 and possibly set resync_needed. They should all
1594 * be 0 at this point
1595 */
1596
1597 sector_t secs;
1598 bitmap_counter_t *bmc;
1599 spin_lock_irq(&bitmap->counts.lock);
1600 bmc = bitmap_get_counter(&bitmap->counts, offset, &secs, 1);
1601 if (!bmc) {
1602 spin_unlock_irq(&bitmap->counts.lock);
1603 return;
1604 }
1605 if (!*bmc) {
1606 *bmc = 2;
1607 bitmap_count_page(&bitmap->counts, offset, 1);
1608 bitmap_set_pending(&bitmap->counts, offset);
1609 bitmap->allclean = 0;
1610 }
1611 if (needed)
1612 *bmc |= NEEDED_MASK;
1613 spin_unlock_irq(&bitmap->counts.lock);
1614 }
1615
1616 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1617 void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1618 {
1619 unsigned long chunk;
1620
1621 for (chunk = s; chunk <= e; chunk++) {
1622 sector_t sec = (sector_t)chunk << bitmap->counts.chunkshift;
1623 bitmap_set_memory_bits(bitmap, sec, 1);
1624 bitmap_file_set_bit(bitmap, sec);
1625 if (sec < bitmap->mddev->recovery_cp)
1626 /* We are asserting that the array is dirty,
1627 * so move the recovery_cp address back so
1628 * that it is obvious that it is dirty
1629 */
1630 bitmap->mddev->recovery_cp = sec;
1631 }
1632 }
1633
1634 /*
1635 * flush out any pending updates
1636 */
1637 void bitmap_flush(struct mddev *mddev)
1638 {
1639 struct bitmap *bitmap = mddev->bitmap;
1640 long sleep;
1641
1642 if (!bitmap) /* there was no bitmap */
1643 return;
1644
1645 /* run the daemon_work three time to ensure everything is flushed
1646 * that can be
1647 */
1648 sleep = mddev->bitmap_info.daemon_sleep * 2;
1649 bitmap->daemon_lastrun -= sleep;
1650 bitmap_daemon_work(mddev);
1651 bitmap->daemon_lastrun -= sleep;
1652 bitmap_daemon_work(mddev);
1653 bitmap->daemon_lastrun -= sleep;
1654 bitmap_daemon_work(mddev);
1655 bitmap_update_sb(bitmap);
1656 }
1657
1658 /*
1659 * free memory that was allocated
1660 */
1661 static void bitmap_free(struct bitmap *bitmap)
1662 {
1663 unsigned long k, pages;
1664 struct bitmap_page *bp;
1665
1666 if (!bitmap) /* there was no bitmap */
1667 return;
1668
1669 if (mddev_is_clustered(bitmap->mddev) && bitmap->mddev->cluster_info &&
1670 bitmap->cluster_slot == md_cluster_ops->slot_number(bitmap->mddev))
1671 md_cluster_stop(bitmap->mddev);
1672
1673 /* Shouldn't be needed - but just in case.... */
1674 wait_event(bitmap->write_wait,
1675 atomic_read(&bitmap->pending_writes) == 0);
1676
1677 /* release the bitmap file */
1678 bitmap_file_unmap(&bitmap->storage);
1679
1680 bp = bitmap->counts.bp;
1681 pages = bitmap->counts.pages;
1682
1683 /* free all allocated memory */
1684
1685 if (bp) /* deallocate the page memory */
1686 for (k = 0; k < pages; k++)
1687 if (bp[k].map && !bp[k].hijacked)
1688 kfree(bp[k].map);
1689 kfree(bp);
1690 kfree(bitmap);
1691 }
1692
1693 void bitmap_destroy(struct mddev *mddev)
1694 {
1695 struct bitmap *bitmap = mddev->bitmap;
1696
1697 if (!bitmap) /* there was no bitmap */
1698 return;
1699
1700 mutex_lock(&mddev->bitmap_info.mutex);
1701 spin_lock(&mddev->lock);
1702 mddev->bitmap = NULL; /* disconnect from the md device */
1703 spin_unlock(&mddev->lock);
1704 mutex_unlock(&mddev->bitmap_info.mutex);
1705 if (mddev->thread)
1706 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1707
1708 if (bitmap->sysfs_can_clear)
1709 sysfs_put(bitmap->sysfs_can_clear);
1710
1711 bitmap_free(bitmap);
1712 }
1713
1714 /*
1715 * initialize the bitmap structure
1716 * if this returns an error, bitmap_destroy must be called to do clean up
1717 */
1718 struct bitmap *bitmap_create(struct mddev *mddev, int slot)
1719 {
1720 struct bitmap *bitmap;
1721 sector_t blocks = mddev->resync_max_sectors;
1722 struct file *file = mddev->bitmap_info.file;
1723 int err;
1724 struct kernfs_node *bm = NULL;
1725
1726 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
1727
1728 BUG_ON(file && mddev->bitmap_info.offset);
1729
1730 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1731 if (!bitmap)
1732 return ERR_PTR(-ENOMEM);
1733
1734 spin_lock_init(&bitmap->counts.lock);
1735 atomic_set(&bitmap->pending_writes, 0);
1736 init_waitqueue_head(&bitmap->write_wait);
1737 init_waitqueue_head(&bitmap->overflow_wait);
1738 init_waitqueue_head(&bitmap->behind_wait);
1739
1740 bitmap->mddev = mddev;
1741 bitmap->cluster_slot = slot;
1742
1743 if (mddev->kobj.sd)
1744 bm = sysfs_get_dirent(mddev->kobj.sd, "bitmap");
1745 if (bm) {
1746 bitmap->sysfs_can_clear = sysfs_get_dirent(bm, "can_clear");
1747 sysfs_put(bm);
1748 } else
1749 bitmap->sysfs_can_clear = NULL;
1750
1751 bitmap->storage.file = file;
1752 if (file) {
1753 get_file(file);
1754 /* As future accesses to this file will use bmap,
1755 * and bypass the page cache, we must sync the file
1756 * first.
1757 */
1758 vfs_fsync(file, 1);
1759 }
1760 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1761 if (!mddev->bitmap_info.external) {
1762 /*
1763 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1764 * instructing us to create a new on-disk bitmap instance.
1765 */
1766 if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags))
1767 err = bitmap_new_disk_sb(bitmap);
1768 else
1769 err = bitmap_read_sb(bitmap);
1770 } else {
1771 err = 0;
1772 if (mddev->bitmap_info.chunksize == 0 ||
1773 mddev->bitmap_info.daemon_sleep == 0)
1774 /* chunksize and time_base need to be
1775 * set first. */
1776 err = -EINVAL;
1777 }
1778 if (err)
1779 goto error;
1780
1781 bitmap->daemon_lastrun = jiffies;
1782 err = bitmap_resize(bitmap, blocks, mddev->bitmap_info.chunksize, 1);
1783 if (err)
1784 goto error;
1785
1786 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1787 bitmap->counts.pages, bmname(bitmap));
1788
1789 err = test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0;
1790 if (err)
1791 goto error;
1792
1793 return bitmap;
1794 error:
1795 bitmap_free(bitmap);
1796 return ERR_PTR(err);
1797 }
1798
1799 int bitmap_load(struct mddev *mddev)
1800 {
1801 int err = 0;
1802 sector_t start = 0;
1803 sector_t sector = 0;
1804 struct bitmap *bitmap = mddev->bitmap;
1805
1806 if (!bitmap)
1807 goto out;
1808
1809 /* Clear out old bitmap info first: Either there is none, or we
1810 * are resuming after someone else has possibly changed things,
1811 * so we should forget old cached info.
1812 * All chunks should be clean, but some might need_sync.
1813 */
1814 while (sector < mddev->resync_max_sectors) {
1815 sector_t blocks;
1816 bitmap_start_sync(bitmap, sector, &blocks, 0);
1817 sector += blocks;
1818 }
1819 bitmap_close_sync(bitmap);
1820
1821 if (mddev->degraded == 0
1822 || bitmap->events_cleared == mddev->events)
1823 /* no need to keep dirty bits to optimise a
1824 * re-add of a missing device */
1825 start = mddev->recovery_cp;
1826
1827 mutex_lock(&mddev->bitmap_info.mutex);
1828 err = bitmap_init_from_disk(bitmap, start);
1829 mutex_unlock(&mddev->bitmap_info.mutex);
1830
1831 if (err)
1832 goto out;
1833 clear_bit(BITMAP_STALE, &bitmap->flags);
1834
1835 /* Kick recovery in case any bits were set */
1836 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
1837
1838 mddev->thread->timeout = mddev->bitmap_info.daemon_sleep;
1839 md_wakeup_thread(mddev->thread);
1840
1841 bitmap_update_sb(bitmap);
1842
1843 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
1844 err = -EIO;
1845 out:
1846 return err;
1847 }
1848 EXPORT_SYMBOL_GPL(bitmap_load);
1849
1850 /* Loads the bitmap associated with slot and copies the resync information
1851 * to our bitmap
1852 */
1853 int bitmap_copy_from_slot(struct mddev *mddev, int slot,
1854 sector_t *low, sector_t *high, bool clear_bits)
1855 {
1856 int rv = 0, i, j;
1857 sector_t block, lo = 0, hi = 0;
1858 struct bitmap_counts *counts;
1859 struct bitmap *bitmap = bitmap_create(mddev, slot);
1860
1861 if (IS_ERR(bitmap))
1862 return PTR_ERR(bitmap);
1863
1864 rv = bitmap_read_sb(bitmap);
1865 if (rv)
1866 goto err;
1867
1868 rv = bitmap_init_from_disk(bitmap, 0);
1869 if (rv)
1870 goto err;
1871
1872 counts = &bitmap->counts;
1873 for (j = 0; j < counts->chunks; j++) {
1874 block = (sector_t)j << counts->chunkshift;
1875 if (bitmap_file_test_bit(bitmap, block)) {
1876 if (!lo)
1877 lo = block;
1878 hi = block;
1879 bitmap_file_clear_bit(bitmap, block);
1880 bitmap_set_memory_bits(mddev->bitmap, block, 1);
1881 bitmap_file_set_bit(mddev->bitmap, block);
1882 }
1883 }
1884
1885 if (clear_bits) {
1886 bitmap_update_sb(bitmap);
1887 /* Setting this for the ev_page should be enough.
1888 * And we do not require both write_all and PAGE_DIRT either
1889 */
1890 for (i = 0; i < bitmap->storage.file_pages; i++)
1891 set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
1892 bitmap_write_all(bitmap);
1893 bitmap_unplug(bitmap);
1894 }
1895 *low = lo;
1896 *high = hi;
1897 err:
1898 bitmap_free(bitmap);
1899 return rv;
1900 }
1901 EXPORT_SYMBOL_GPL(bitmap_copy_from_slot);
1902
1903
1904 void bitmap_status(struct seq_file *seq, struct bitmap *bitmap)
1905 {
1906 unsigned long chunk_kb;
1907 struct bitmap_counts *counts;
1908
1909 if (!bitmap)
1910 return;
1911
1912 counts = &bitmap->counts;
1913
1914 chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10;
1915 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
1916 "%lu%s chunk",
1917 counts->pages - counts->missing_pages,
1918 counts->pages,
1919 (counts->pages - counts->missing_pages)
1920 << (PAGE_SHIFT - 10),
1921 chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize,
1922 chunk_kb ? "KB" : "B");
1923 if (bitmap->storage.file) {
1924 seq_printf(seq, ", file: ");
1925 seq_path(seq, &bitmap->storage.file->f_path, " \t\n");
1926 }
1927
1928 seq_printf(seq, "\n");
1929 }
1930
1931 int bitmap_resize(struct bitmap *bitmap, sector_t blocks,
1932 int chunksize, int init)
1933 {
1934 /* If chunk_size is 0, choose an appropriate chunk size.
1935 * Then possibly allocate new storage space.
1936 * Then quiesce, copy bits, replace bitmap, and re-start
1937 *
1938 * This function is called both to set up the initial bitmap
1939 * and to resize the bitmap while the array is active.
1940 * If this happens as a result of the array being resized,
1941 * chunksize will be zero, and we need to choose a suitable
1942 * chunksize, otherwise we use what we are given.
1943 */
1944 struct bitmap_storage store;
1945 struct bitmap_counts old_counts;
1946 unsigned long chunks;
1947 sector_t block;
1948 sector_t old_blocks, new_blocks;
1949 int chunkshift;
1950 int ret = 0;
1951 long pages;
1952 struct bitmap_page *new_bp;
1953
1954 if (chunksize == 0) {
1955 /* If there is enough space, leave the chunk size unchanged,
1956 * else increase by factor of two until there is enough space.
1957 */
1958 long bytes;
1959 long space = bitmap->mddev->bitmap_info.space;
1960
1961 if (space == 0) {
1962 /* We don't know how much space there is, so limit
1963 * to current size - in sectors.
1964 */
1965 bytes = DIV_ROUND_UP(bitmap->counts.chunks, 8);
1966 if (!bitmap->mddev->bitmap_info.external)
1967 bytes += sizeof(bitmap_super_t);
1968 space = DIV_ROUND_UP(bytes, 512);
1969 bitmap->mddev->bitmap_info.space = space;
1970 }
1971 chunkshift = bitmap->counts.chunkshift;
1972 chunkshift--;
1973 do {
1974 /* 'chunkshift' is shift from block size to chunk size */
1975 chunkshift++;
1976 chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
1977 bytes = DIV_ROUND_UP(chunks, 8);
1978 if (!bitmap->mddev->bitmap_info.external)
1979 bytes += sizeof(bitmap_super_t);
1980 } while (bytes > (space << 9));
1981 } else
1982 chunkshift = ffz(~chunksize) - BITMAP_BLOCK_SHIFT;
1983
1984 chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
1985 memset(&store, 0, sizeof(store));
1986 if (bitmap->mddev->bitmap_info.offset || bitmap->mddev->bitmap_info.file)
1987 ret = bitmap_storage_alloc(&store, chunks,
1988 !bitmap->mddev->bitmap_info.external,
1989 bitmap->cluster_slot);
1990 if (ret)
1991 goto err;
1992
1993 pages = DIV_ROUND_UP(chunks, PAGE_COUNTER_RATIO);
1994
1995 new_bp = kzalloc(pages * sizeof(*new_bp), GFP_KERNEL);
1996 ret = -ENOMEM;
1997 if (!new_bp) {
1998 bitmap_file_unmap(&store);
1999 goto err;
2000 }
2001
2002 if (!init)
2003 bitmap->mddev->pers->quiesce(bitmap->mddev, 1);
2004
2005 store.file = bitmap->storage.file;
2006 bitmap->storage.file = NULL;
2007
2008 if (store.sb_page && bitmap->storage.sb_page)
2009 memcpy(page_address(store.sb_page),
2010 page_address(bitmap->storage.sb_page),
2011 sizeof(bitmap_super_t));
2012 bitmap_file_unmap(&bitmap->storage);
2013 bitmap->storage = store;
2014
2015 old_counts = bitmap->counts;
2016 bitmap->counts.bp = new_bp;
2017 bitmap->counts.pages = pages;
2018 bitmap->counts.missing_pages = pages;
2019 bitmap->counts.chunkshift = chunkshift;
2020 bitmap->counts.chunks = chunks;
2021 bitmap->mddev->bitmap_info.chunksize = 1 << (chunkshift +
2022 BITMAP_BLOCK_SHIFT);
2023
2024 blocks = min(old_counts.chunks << old_counts.chunkshift,
2025 chunks << chunkshift);
2026
2027 spin_lock_irq(&bitmap->counts.lock);
2028 for (block = 0; block < blocks; ) {
2029 bitmap_counter_t *bmc_old, *bmc_new;
2030 int set;
2031
2032 bmc_old = bitmap_get_counter(&old_counts, block,
2033 &old_blocks, 0);
2034 set = bmc_old && NEEDED(*bmc_old);
2035
2036 if (set) {
2037 bmc_new = bitmap_get_counter(&bitmap->counts, block,
2038 &new_blocks, 1);
2039 if (*bmc_new == 0) {
2040 /* need to set on-disk bits too. */
2041 sector_t end = block + new_blocks;
2042 sector_t start = block >> chunkshift;
2043 start <<= chunkshift;
2044 while (start < end) {
2045 bitmap_file_set_bit(bitmap, block);
2046 start += 1 << chunkshift;
2047 }
2048 *bmc_new = 2;
2049 bitmap_count_page(&bitmap->counts,
2050 block, 1);
2051 bitmap_set_pending(&bitmap->counts,
2052 block);
2053 }
2054 *bmc_new |= NEEDED_MASK;
2055 if (new_blocks < old_blocks)
2056 old_blocks = new_blocks;
2057 }
2058 block += old_blocks;
2059 }
2060
2061 if (!init) {
2062 int i;
2063 while (block < (chunks << chunkshift)) {
2064 bitmap_counter_t *bmc;
2065 bmc = bitmap_get_counter(&bitmap->counts, block,
2066 &new_blocks, 1);
2067 if (bmc) {
2068 /* new space. It needs to be resynced, so
2069 * we set NEEDED_MASK.
2070 */
2071 if (*bmc == 0) {
2072 *bmc = NEEDED_MASK | 2;
2073 bitmap_count_page(&bitmap->counts,
2074 block, 1);
2075 bitmap_set_pending(&bitmap->counts,
2076 block);
2077 }
2078 }
2079 block += new_blocks;
2080 }
2081 for (i = 0; i < bitmap->storage.file_pages; i++)
2082 set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
2083 }
2084 spin_unlock_irq(&bitmap->counts.lock);
2085
2086 if (!init) {
2087 bitmap_unplug(bitmap);
2088 bitmap->mddev->pers->quiesce(bitmap->mddev, 0);
2089 }
2090 ret = 0;
2091 err:
2092 return ret;
2093 }
2094 EXPORT_SYMBOL_GPL(bitmap_resize);
2095
2096 static ssize_t
2097 location_show(struct mddev *mddev, char *page)
2098 {
2099 ssize_t len;
2100 if (mddev->bitmap_info.file)
2101 len = sprintf(page, "file");
2102 else if (mddev->bitmap_info.offset)
2103 len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset);
2104 else
2105 len = sprintf(page, "none");
2106 len += sprintf(page+len, "\n");
2107 return len;
2108 }
2109
2110 static ssize_t
2111 location_store(struct mddev *mddev, const char *buf, size_t len)
2112 {
2113
2114 if (mddev->pers) {
2115 if (!mddev->pers->quiesce)
2116 return -EBUSY;
2117 if (mddev->recovery || mddev->sync_thread)
2118 return -EBUSY;
2119 }
2120
2121 if (mddev->bitmap || mddev->bitmap_info.file ||
2122 mddev->bitmap_info.offset) {
2123 /* bitmap already configured. Only option is to clear it */
2124 if (strncmp(buf, "none", 4) != 0)
2125 return -EBUSY;
2126 if (mddev->pers) {
2127 mddev->pers->quiesce(mddev, 1);
2128 bitmap_destroy(mddev);
2129 mddev->pers->quiesce(mddev, 0);
2130 }
2131 mddev->bitmap_info.offset = 0;
2132 if (mddev->bitmap_info.file) {
2133 struct file *f = mddev->bitmap_info.file;
2134 mddev->bitmap_info.file = NULL;
2135 fput(f);
2136 }
2137 } else {
2138 /* No bitmap, OK to set a location */
2139 long long offset;
2140 if (strncmp(buf, "none", 4) == 0)
2141 /* nothing to be done */;
2142 else if (strncmp(buf, "file:", 5) == 0) {
2143 /* Not supported yet */
2144 return -EINVAL;
2145 } else {
2146 int rv;
2147 if (buf[0] == '+')
2148 rv = kstrtoll(buf+1, 10, &offset);
2149 else
2150 rv = kstrtoll(buf, 10, &offset);
2151 if (rv)
2152 return rv;
2153 if (offset == 0)
2154 return -EINVAL;
2155 if (mddev->bitmap_info.external == 0 &&
2156 mddev->major_version == 0 &&
2157 offset != mddev->bitmap_info.default_offset)
2158 return -EINVAL;
2159 mddev->bitmap_info.offset = offset;
2160 if (mddev->pers) {
2161 struct bitmap *bitmap;
2162 mddev->pers->quiesce(mddev, 1);
2163 bitmap = bitmap_create(mddev, -1);
2164 if (IS_ERR(bitmap))
2165 rv = PTR_ERR(bitmap);
2166 else {
2167 mddev->bitmap = bitmap;
2168 rv = bitmap_load(mddev);
2169 if (rv) {
2170 bitmap_destroy(mddev);
2171 mddev->bitmap_info.offset = 0;
2172 }
2173 }
2174 mddev->pers->quiesce(mddev, 0);
2175 if (rv)
2176 return rv;
2177 }
2178 }
2179 }
2180 if (!mddev->external) {
2181 /* Ensure new bitmap info is stored in
2182 * metadata promptly.
2183 */
2184 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2185 md_wakeup_thread(mddev->thread);
2186 }
2187 return len;
2188 }
2189
2190 static struct md_sysfs_entry bitmap_location =
2191 __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store);
2192
2193 /* 'bitmap/space' is the space available at 'location' for the
2194 * bitmap. This allows the kernel to know when it is safe to
2195 * resize the bitmap to match a resized array.
2196 */
2197 static ssize_t
2198 space_show(struct mddev *mddev, char *page)
2199 {
2200 return sprintf(page, "%lu\n", mddev->bitmap_info.space);
2201 }
2202
2203 static ssize_t
2204 space_store(struct mddev *mddev, const char *buf, size_t len)
2205 {
2206 unsigned long sectors;
2207 int rv;
2208
2209 rv = kstrtoul(buf, 10, &sectors);
2210 if (rv)
2211 return rv;
2212
2213 if (sectors == 0)
2214 return -EINVAL;
2215
2216 if (mddev->bitmap &&
2217 sectors < (mddev->bitmap->storage.bytes + 511) >> 9)
2218 return -EFBIG; /* Bitmap is too big for this small space */
2219
2220 /* could make sure it isn't too big, but that isn't really
2221 * needed - user-space should be careful.
2222 */
2223 mddev->bitmap_info.space = sectors;
2224 return len;
2225 }
2226
2227 static struct md_sysfs_entry bitmap_space =
2228 __ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store);
2229
2230 static ssize_t
2231 timeout_show(struct mddev *mddev, char *page)
2232 {
2233 ssize_t len;
2234 unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ;
2235 unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ;
2236
2237 len = sprintf(page, "%lu", secs);
2238 if (jifs)
2239 len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs));
2240 len += sprintf(page+len, "\n");
2241 return len;
2242 }
2243
2244 static ssize_t
2245 timeout_store(struct mddev *mddev, const char *buf, size_t len)
2246 {
2247 /* timeout can be set at any time */
2248 unsigned long timeout;
2249 int rv = strict_strtoul_scaled(buf, &timeout, 4);
2250 if (rv)
2251 return rv;
2252
2253 /* just to make sure we don't overflow... */
2254 if (timeout >= LONG_MAX / HZ)
2255 return -EINVAL;
2256
2257 timeout = timeout * HZ / 10000;
2258
2259 if (timeout >= MAX_SCHEDULE_TIMEOUT)
2260 timeout = MAX_SCHEDULE_TIMEOUT-1;
2261 if (timeout < 1)
2262 timeout = 1;
2263 mddev->bitmap_info.daemon_sleep = timeout;
2264 if (mddev->thread) {
2265 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
2266 * the bitmap is all clean and we don't need to
2267 * adjust the timeout right now
2268 */
2269 if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) {
2270 mddev->thread->timeout = timeout;
2271 md_wakeup_thread(mddev->thread);
2272 }
2273 }
2274 return len;
2275 }
2276
2277 static struct md_sysfs_entry bitmap_timeout =
2278 __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store);
2279
2280 static ssize_t
2281 backlog_show(struct mddev *mddev, char *page)
2282 {
2283 return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind);
2284 }
2285
2286 static ssize_t
2287 backlog_store(struct mddev *mddev, const char *buf, size_t len)
2288 {
2289 unsigned long backlog;
2290 int rv = kstrtoul(buf, 10, &backlog);
2291 if (rv)
2292 return rv;
2293 if (backlog > COUNTER_MAX)
2294 return -EINVAL;
2295 mddev->bitmap_info.max_write_behind = backlog;
2296 return len;
2297 }
2298
2299 static struct md_sysfs_entry bitmap_backlog =
2300 __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store);
2301
2302 static ssize_t
2303 chunksize_show(struct mddev *mddev, char *page)
2304 {
2305 return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize);
2306 }
2307
2308 static ssize_t
2309 chunksize_store(struct mddev *mddev, const char *buf, size_t len)
2310 {
2311 /* Can only be changed when no bitmap is active */
2312 int rv;
2313 unsigned long csize;
2314 if (mddev->bitmap)
2315 return -EBUSY;
2316 rv = kstrtoul(buf, 10, &csize);
2317 if (rv)
2318 return rv;
2319 if (csize < 512 ||
2320 !is_power_of_2(csize))
2321 return -EINVAL;
2322 mddev->bitmap_info.chunksize = csize;
2323 return len;
2324 }
2325
2326 static struct md_sysfs_entry bitmap_chunksize =
2327 __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store);
2328
2329 static ssize_t metadata_show(struct mddev *mddev, char *page)
2330 {
2331 if (mddev_is_clustered(mddev))
2332 return sprintf(page, "clustered\n");
2333 return sprintf(page, "%s\n", (mddev->bitmap_info.external
2334 ? "external" : "internal"));
2335 }
2336
2337 static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len)
2338 {
2339 if (mddev->bitmap ||
2340 mddev->bitmap_info.file ||
2341 mddev->bitmap_info.offset)
2342 return -EBUSY;
2343 if (strncmp(buf, "external", 8) == 0)
2344 mddev->bitmap_info.external = 1;
2345 else if ((strncmp(buf, "internal", 8) == 0) ||
2346 (strncmp(buf, "clustered", 9) == 0))
2347 mddev->bitmap_info.external = 0;
2348 else
2349 return -EINVAL;
2350 return len;
2351 }
2352
2353 static struct md_sysfs_entry bitmap_metadata =
2354 __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2355
2356 static ssize_t can_clear_show(struct mddev *mddev, char *page)
2357 {
2358 int len;
2359 spin_lock(&mddev->lock);
2360 if (mddev->bitmap)
2361 len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ?
2362 "false" : "true"));
2363 else
2364 len = sprintf(page, "\n");
2365 spin_unlock(&mddev->lock);
2366 return len;
2367 }
2368
2369 static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len)
2370 {
2371 if (mddev->bitmap == NULL)
2372 return -ENOENT;
2373 if (strncmp(buf, "false", 5) == 0)
2374 mddev->bitmap->need_sync = 1;
2375 else if (strncmp(buf, "true", 4) == 0) {
2376 if (mddev->degraded)
2377 return -EBUSY;
2378 mddev->bitmap->need_sync = 0;
2379 } else
2380 return -EINVAL;
2381 return len;
2382 }
2383
2384 static struct md_sysfs_entry bitmap_can_clear =
2385 __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store);
2386
2387 static ssize_t
2388 behind_writes_used_show(struct mddev *mddev, char *page)
2389 {
2390 ssize_t ret;
2391 spin_lock(&mddev->lock);
2392 if (mddev->bitmap == NULL)
2393 ret = sprintf(page, "0\n");
2394 else
2395 ret = sprintf(page, "%lu\n",
2396 mddev->bitmap->behind_writes_used);
2397 spin_unlock(&mddev->lock);
2398 return ret;
2399 }
2400
2401 static ssize_t
2402 behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len)
2403 {
2404 if (mddev->bitmap)
2405 mddev->bitmap->behind_writes_used = 0;
2406 return len;
2407 }
2408
2409 static struct md_sysfs_entry max_backlog_used =
2410 __ATTR(max_backlog_used, S_IRUGO | S_IWUSR,
2411 behind_writes_used_show, behind_writes_used_reset);
2412
2413 static struct attribute *md_bitmap_attrs[] = {
2414 &bitmap_location.attr,
2415 &bitmap_space.attr,
2416 &bitmap_timeout.attr,
2417 &bitmap_backlog.attr,
2418 &bitmap_chunksize.attr,
2419 &bitmap_metadata.attr,
2420 &bitmap_can_clear.attr,
2421 &max_backlog_used.attr,
2422 NULL
2423 };
2424 struct attribute_group md_bitmap_group = {
2425 .name = "bitmap",
2426 .attrs = md_bitmap_attrs,
2427 };
2428
This page took 0.108485 seconds and 4 git commands to generate.