4 This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
6 Copyright (C) 2004-2008, LINBIT Information Technologies GmbH.
7 Copyright (C) 2004-2008, Philipp Reisner <philipp.reisner@linbit.com>.
8 Copyright (C) 2004-2008, Lars Ellenberg <lars.ellenberg@linbit.com>.
10 drbd is free software; you can redistribute it and/or modify
11 it under the terms of the GNU General Public License as published by
12 the Free Software Foundation; either version 2, or (at your option)
15 drbd is distributed in the hope that it will be useful,
16 but WITHOUT ANY WARRANTY; without even the implied warranty of
17 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 GNU General Public License for more details.
20 You should have received a copy of the GNU General Public License
21 along with drbd; see the file COPYING. If not, write to
22 the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
25 #include <linux/bitops.h>
26 #include <linux/vmalloc.h>
27 #include <linux/string.h>
28 #include <linux/drbd.h>
29 #include <linux/slab.h>
30 #include <asm/kmap_types.h>
33 /* OPAQUE outside this file!
34 * interface defined in drbd_int.h
37 * function name drbd_bm_... => used elsewhere, "public".
38 * function name bm_... => internal to implementation, "private".
40 * Note that since find_first_bit returns int, at the current granularity of
41 * the bitmap (4KB per byte), this implementation "only" supports up to
42 * 1<<(32+12) == 16 TB...
47 * Access to the *bm_pages is protected by bm_lock.
48 * It is safe to read the other members within the lock.
50 * drbd_bm_set_bits is called from bio_endio callbacks,
51 * We may be called with irq already disabled,
52 * so we need spin_lock_irqsave().
53 * And we need the kmap_atomic.
56 struct page **bm_pages;
58 /* WARNING unsigned long bm_*:
59 * 32bit number of bit offset is just enough for 512 MB bitmap.
60 * it will blow up if we make the bitmap bigger...
61 * not that it makes much sense to have a bitmap that large,
62 * rather change the granularity to 16k or 64k or something.
63 * (that implies other problems, however...)
65 unsigned long bm_set; /* nr of set bits; THINK maybe atomic_t? */
66 unsigned long bm_bits;
68 size_t bm_number_of_pages;
69 sector_t bm_dev_capacity;
70 struct semaphore bm_change; /* serializes resize operations */
73 wait_queue_head_t bm_io_wait;
75 unsigned long bm_flags;
77 /* debugging aid, in case we are still racy somewhere */
79 struct task_struct *bm_task;
82 /* definition of bits in bm_flags */
84 #define BM_MD_IO_ERROR 1
85 #define BM_P_VMALLOCED 2
87 static int bm_is_locked(struct drbd_bitmap *b)
89 return test_bit(BM_LOCKED, &b->bm_flags);
92 #define bm_print_lock_info(m) __bm_print_lock_info(m, __func__)
93 static void __bm_print_lock_info(struct drbd_conf *mdev, const char *func)
95 struct drbd_bitmap *b = mdev->bitmap;
96 if (!__ratelimit(&drbd_ratelimit_state))
98 dev_err(DEV, "FIXME %s in %s, bitmap locked for '%s' by %s\n",
99 current == mdev->receiver.task ? "receiver" :
100 current == mdev->asender.task ? "asender" :
101 current == mdev->worker.task ? "worker" : current->comm,
102 func, b->bm_why ?: "?",
103 b->bm_task == mdev->receiver.task ? "receiver" :
104 b->bm_task == mdev->asender.task ? "asender" :
105 b->bm_task == mdev->worker.task ? "worker" : "?");
108 void drbd_bm_lock(struct drbd_conf *mdev, char *why)
110 struct drbd_bitmap *b = mdev->bitmap;
114 dev_err(DEV, "FIXME no bitmap in drbd_bm_lock!?\n");
118 trylock_failed = down_trylock(&b->bm_change);
120 if (trylock_failed) {
121 dev_warn(DEV, "%s going to '%s' but bitmap already locked for '%s' by %s\n",
122 current == mdev->receiver.task ? "receiver" :
123 current == mdev->asender.task ? "asender" :
124 current == mdev->worker.task ? "worker" : current->comm,
125 why, b->bm_why ?: "?",
126 b->bm_task == mdev->receiver.task ? "receiver" :
127 b->bm_task == mdev->asender.task ? "asender" :
128 b->bm_task == mdev->worker.task ? "worker" : "?");
131 if (__test_and_set_bit(BM_LOCKED, &b->bm_flags))
132 dev_err(DEV, "FIXME bitmap already locked in bm_lock\n");
135 b->bm_task = current;
138 void drbd_bm_unlock(struct drbd_conf *mdev)
140 struct drbd_bitmap *b = mdev->bitmap;
142 dev_err(DEV, "FIXME no bitmap in drbd_bm_unlock!?\n");
146 if (!__test_and_clear_bit(BM_LOCKED, &mdev->bitmap->bm_flags))
147 dev_err(DEV, "FIXME bitmap not locked in bm_unlock\n");
154 /* word offset to long pointer */
155 static unsigned long *__bm_map_paddr(struct drbd_bitmap *b, unsigned long offset, const enum km_type km)
158 unsigned long page_nr;
160 /* page_nr = (word*sizeof(long)) >> PAGE_SHIFT; */
161 page_nr = offset >> (PAGE_SHIFT - LN2_BPL + 3);
162 BUG_ON(page_nr >= b->bm_number_of_pages);
163 page = b->bm_pages[page_nr];
165 return (unsigned long *) kmap_atomic(page, km);
168 static unsigned long * bm_map_paddr(struct drbd_bitmap *b, unsigned long offset)
170 return __bm_map_paddr(b, offset, KM_IRQ1);
173 static void __bm_unmap(unsigned long *p_addr, const enum km_type km)
175 kunmap_atomic(p_addr, km);
178 static void bm_unmap(unsigned long *p_addr)
180 return __bm_unmap(p_addr, KM_IRQ1);
183 /* long word offset of _bitmap_ sector */
184 #define S2W(s) ((s)<<(BM_EXT_SHIFT-BM_BLOCK_SHIFT-LN2_BPL))
185 /* word offset from start of bitmap to word number _in_page_
186 * modulo longs per page
187 #define MLPP(X) ((X) % (PAGE_SIZE/sizeof(long))
188 hm, well, Philipp thinks gcc might not optimze the % into & (... - 1)
191 #define MLPP(X) ((X) & ((PAGE_SIZE/sizeof(long))-1))
193 /* Long words per page */
194 #define LWPP (PAGE_SIZE/sizeof(long))
197 * actually most functions herein should take a struct drbd_bitmap*, not a
198 * struct drbd_conf*, but for the debug macros I like to have the mdev around
199 * to be able to report device specific.
202 static void bm_free_pages(struct page **pages, unsigned long number)
208 for (i = 0; i < number; i++) {
210 printk(KERN_ALERT "drbd: bm_free_pages tried to free "
211 "a NULL pointer; i=%lu n=%lu\n",
215 __free_page(pages[i]);
220 static void bm_vk_free(void *ptr, int v)
229 * "have" and "want" are NUMBER OF PAGES.
231 static struct page **bm_realloc_pages(struct drbd_bitmap *b, unsigned long want)
233 struct page **old_pages = b->bm_pages;
234 struct page **new_pages, *page;
235 unsigned int i, bytes, vmalloced = 0;
236 unsigned long have = b->bm_number_of_pages;
238 BUG_ON(have == 0 && old_pages != NULL);
239 BUG_ON(have != 0 && old_pages == NULL);
244 /* Trying kmalloc first, falling back to vmalloc.
245 * GFP_KERNEL is ok, as this is done when a lower level disk is
246 * "attached" to the drbd. Context is receiver thread or cqueue
247 * thread. As we have no disk yet, we are not in the IO path,
248 * not even the IO path of the peer. */
249 bytes = sizeof(struct page *)*want;
250 new_pages = kmalloc(bytes, GFP_KERNEL);
252 new_pages = vmalloc(bytes);
258 memset(new_pages, 0, bytes);
260 for (i = 0; i < have; i++)
261 new_pages[i] = old_pages[i];
262 for (; i < want; i++) {
263 page = alloc_page(GFP_HIGHUSER);
265 bm_free_pages(new_pages + have, i - have);
266 bm_vk_free(new_pages, vmalloced);
272 for (i = 0; i < want; i++)
273 new_pages[i] = old_pages[i];
274 /* NOT HERE, we are outside the spinlock!
275 bm_free_pages(old_pages + want, have - want);
280 set_bit(BM_P_VMALLOCED, &b->bm_flags);
282 clear_bit(BM_P_VMALLOCED, &b->bm_flags);
288 * called on driver init only. TODO call when a device is created.
289 * allocates the drbd_bitmap, and stores it in mdev->bitmap.
291 int drbd_bm_init(struct drbd_conf *mdev)
293 struct drbd_bitmap *b = mdev->bitmap;
295 b = kzalloc(sizeof(struct drbd_bitmap), GFP_KERNEL);
298 spin_lock_init(&b->bm_lock);
299 init_MUTEX(&b->bm_change);
300 init_waitqueue_head(&b->bm_io_wait);
307 sector_t drbd_bm_capacity(struct drbd_conf *mdev)
309 ERR_IF(!mdev->bitmap) return 0;
310 return mdev->bitmap->bm_dev_capacity;
313 /* called on driver unload. TODO: call when a device is destroyed.
315 void drbd_bm_cleanup(struct drbd_conf *mdev)
317 ERR_IF (!mdev->bitmap) return;
318 bm_free_pages(mdev->bitmap->bm_pages, mdev->bitmap->bm_number_of_pages);
319 bm_vk_free(mdev->bitmap->bm_pages, test_bit(BM_P_VMALLOCED, &mdev->bitmap->bm_flags));
325 * since (b->bm_bits % BITS_PER_LONG) != 0,
326 * this masks out the remaining bits.
327 * Returns the number of bits cleared.
329 static int bm_clear_surplus(struct drbd_bitmap *b)
331 const unsigned long mask = (1UL << (b->bm_bits & (BITS_PER_LONG-1))) - 1;
332 size_t w = b->bm_bits >> LN2_BPL;
334 unsigned long *p_addr, *bm;
336 p_addr = bm_map_paddr(b, w);
337 bm = p_addr + MLPP(w);
338 if (w < b->bm_words) {
339 cleared = hweight_long(*bm & ~mask);
344 if (w < b->bm_words) {
345 cleared += hweight_long(*bm);
352 static void bm_set_surplus(struct drbd_bitmap *b)
354 const unsigned long mask = (1UL << (b->bm_bits & (BITS_PER_LONG-1))) - 1;
355 size_t w = b->bm_bits >> LN2_BPL;
356 unsigned long *p_addr, *bm;
358 p_addr = bm_map_paddr(b, w);
359 bm = p_addr + MLPP(w);
360 if (w < b->bm_words) {
365 if (w < b->bm_words) {
371 static unsigned long __bm_count_bits(struct drbd_bitmap *b, const int swap_endian)
373 unsigned long *p_addr, *bm, offset = 0;
374 unsigned long bits = 0;
375 unsigned long i, do_now;
377 while (offset < b->bm_words) {
378 i = do_now = min_t(size_t, b->bm_words-offset, LWPP);
379 p_addr = __bm_map_paddr(b, offset, KM_USER0);
380 bm = p_addr + MLPP(offset);
382 #ifndef __LITTLE_ENDIAN
384 *bm = lel_to_cpu(*bm);
386 bits += hweight_long(*bm++);
388 __bm_unmap(p_addr, KM_USER0);
396 static unsigned long bm_count_bits(struct drbd_bitmap *b)
398 return __bm_count_bits(b, 0);
401 static unsigned long bm_count_bits_swap_endian(struct drbd_bitmap *b)
403 return __bm_count_bits(b, 1);
406 /* offset and len in long words.*/
407 static void bm_memset(struct drbd_bitmap *b, size_t offset, int c, size_t len)
409 unsigned long *p_addr, *bm;
412 #define BM_SECTORS_PER_BIT (BM_BLOCK_SIZE/512)
416 if (end > b->bm_words) {
417 printk(KERN_ALERT "drbd: bm_memset end > bm_words\n");
421 while (offset < end) {
422 do_now = min_t(size_t, ALIGN(offset + 1, LWPP), end) - offset;
423 p_addr = bm_map_paddr(b, offset);
424 bm = p_addr + MLPP(offset);
425 if (bm+do_now > p_addr + LWPP) {
426 printk(KERN_ALERT "drbd: BUG BUG BUG! p_addr:%p bm:%p do_now:%d\n",
427 p_addr, bm, (int)do_now);
428 break; /* breaks to after catch_oob_access_end() only! */
430 memset(bm, c, do_now * sizeof(long));
437 * make sure the bitmap has enough room for the attached storage,
438 * if necessary, resize.
439 * called whenever we may have changed the device size.
440 * returns -ENOMEM if we could not allocate enough memory, 0 on success.
441 * In case this is actually a resize, we copy the old bitmap into the new one.
442 * Otherwise, the bitmap is initialized to all bits set.
444 int drbd_bm_resize(struct drbd_conf *mdev, sector_t capacity)
446 struct drbd_bitmap *b = mdev->bitmap;
447 unsigned long bits, words, owords, obits, *p_addr, *bm;
448 unsigned long want, have, onpages; /* number of pages */
449 struct page **npages, **opages = NULL;
450 int err = 0, growing;
451 int opages_vmalloced;
453 ERR_IF(!b) return -ENOMEM;
455 drbd_bm_lock(mdev, "resize");
457 dev_info(DEV, "drbd_bm_resize called with capacity == %llu\n",
458 (unsigned long long)capacity);
460 if (capacity == b->bm_dev_capacity)
463 opages_vmalloced = test_bit(BM_P_VMALLOCED, &b->bm_flags);
466 spin_lock_irq(&b->bm_lock);
467 opages = b->bm_pages;
468 onpages = b->bm_number_of_pages;
469 owords = b->bm_words;
471 b->bm_number_of_pages =
475 b->bm_dev_capacity = 0;
476 spin_unlock_irq(&b->bm_lock);
477 bm_free_pages(opages, onpages);
478 bm_vk_free(opages, opages_vmalloced);
481 bits = BM_SECT_TO_BIT(ALIGN(capacity, BM_SECT_PER_BIT));
484 words = ALIGN(bits,BITS_PER_LONG) >> LN2_BPL;
485 a 32bit host could present the wrong number of words
488 words = ALIGN(bits, 64) >> LN2_BPL;
490 if (get_ldev(mdev)) {
491 D_ASSERT((u64)bits <= (((u64)mdev->ldev->md.md_size_sect-MD_BM_OFFSET) << 12));
495 /* one extra long to catch off by one errors */
496 want = ALIGN((words+1)*sizeof(long), PAGE_SIZE) >> PAGE_SHIFT;
497 have = b->bm_number_of_pages;
499 D_ASSERT(b->bm_pages != NULL);
500 npages = b->bm_pages;
502 if (FAULT_ACTIVE(mdev, DRBD_FAULT_BM_ALLOC))
505 npages = bm_realloc_pages(b, want);
513 spin_lock_irq(&b->bm_lock);
514 opages = b->bm_pages;
515 owords = b->bm_words;
518 growing = bits > obits;
522 b->bm_pages = npages;
523 b->bm_number_of_pages = want;
526 b->bm_dev_capacity = capacity;
529 bm_memset(b, owords, 0xff, words-owords);
530 b->bm_set += bits - obits;
534 /* implicit: (opages != NULL) && (opages != npages) */
535 bm_free_pages(opages + want, have - want);
538 p_addr = bm_map_paddr(b, words);
539 bm = p_addr + MLPP(words);
543 (void)bm_clear_surplus(b);
545 spin_unlock_irq(&b->bm_lock);
546 if (opages != npages)
547 bm_vk_free(opages, opages_vmalloced);
549 b->bm_set = bm_count_bits(b);
550 dev_info(DEV, "resync bitmap: bits=%lu words=%lu\n", bits, words);
553 drbd_bm_unlock(mdev);
558 * if not protected by other means, return value may be out of date when
559 * leaving this function...
560 * we still need to lock it, since it is important that this returns
561 * bm_set == 0 precisely.
563 * maybe bm_set should be atomic_t ?
565 static unsigned long _drbd_bm_total_weight(struct drbd_conf *mdev)
567 struct drbd_bitmap *b = mdev->bitmap;
572 ERR_IF(!b->bm_pages) return 0;
574 spin_lock_irqsave(&b->bm_lock, flags);
576 spin_unlock_irqrestore(&b->bm_lock, flags);
581 unsigned long drbd_bm_total_weight(struct drbd_conf *mdev)
584 /* if I don't have a disk, I don't know about out-of-sync status */
585 if (!get_ldev_if_state(mdev, D_NEGOTIATING))
587 s = _drbd_bm_total_weight(mdev);
592 size_t drbd_bm_words(struct drbd_conf *mdev)
594 struct drbd_bitmap *b = mdev->bitmap;
596 ERR_IF(!b->bm_pages) return 0;
601 unsigned long drbd_bm_bits(struct drbd_conf *mdev)
603 struct drbd_bitmap *b = mdev->bitmap;
609 /* merge number words from buffer into the bitmap starting at offset.
610 * buffer[i] is expected to be little endian unsigned long.
611 * bitmap must be locked by drbd_bm_lock.
612 * currently only used from receive_bitmap.
614 void drbd_bm_merge_lel(struct drbd_conf *mdev, size_t offset, size_t number,
615 unsigned long *buffer)
617 struct drbd_bitmap *b = mdev->bitmap;
618 unsigned long *p_addr, *bm;
619 unsigned long word, bits;
622 end = offset + number;
625 ERR_IF(!b->bm_pages) return;
628 WARN_ON(offset >= b->bm_words);
629 WARN_ON(end > b->bm_words);
631 spin_lock_irq(&b->bm_lock);
632 while (offset < end) {
633 do_now = min_t(size_t, ALIGN(offset+1, LWPP), end) - offset;
634 p_addr = bm_map_paddr(b, offset);
635 bm = p_addr + MLPP(offset);
638 bits = hweight_long(*bm);
639 word = *bm | lel_to_cpu(*buffer++);
641 b->bm_set += hweight_long(word) - bits;
645 /* with 32bit <-> 64bit cross-platform connect
646 * this is only correct for current usage,
647 * where we _know_ that we are 64 bit aligned,
648 * and know that this function is used in this way, too...
650 if (end == b->bm_words)
651 b->bm_set -= bm_clear_surplus(b);
653 spin_unlock_irq(&b->bm_lock);
656 /* copy number words from the bitmap starting at offset into the buffer.
657 * buffer[i] will be little endian unsigned long.
659 void drbd_bm_get_lel(struct drbd_conf *mdev, size_t offset, size_t number,
660 unsigned long *buffer)
662 struct drbd_bitmap *b = mdev->bitmap;
663 unsigned long *p_addr, *bm;
666 end = offset + number;
669 ERR_IF(!b->bm_pages) return;
671 spin_lock_irq(&b->bm_lock);
672 if ((offset >= b->bm_words) ||
673 (end > b->bm_words) ||
675 dev_err(DEV, "offset=%lu number=%lu bm_words=%lu\n",
676 (unsigned long) offset,
677 (unsigned long) number,
678 (unsigned long) b->bm_words);
680 while (offset < end) {
681 do_now = min_t(size_t, ALIGN(offset+1, LWPP), end) - offset;
682 p_addr = bm_map_paddr(b, offset);
683 bm = p_addr + MLPP(offset);
686 *buffer++ = cpu_to_lel(*bm++);
690 spin_unlock_irq(&b->bm_lock);
693 /* set all bits in the bitmap */
694 void drbd_bm_set_all(struct drbd_conf *mdev)
696 struct drbd_bitmap *b = mdev->bitmap;
698 ERR_IF(!b->bm_pages) return;
700 spin_lock_irq(&b->bm_lock);
701 bm_memset(b, 0, 0xff, b->bm_words);
702 (void)bm_clear_surplus(b);
703 b->bm_set = b->bm_bits;
704 spin_unlock_irq(&b->bm_lock);
707 /* clear all bits in the bitmap */
708 void drbd_bm_clear_all(struct drbd_conf *mdev)
710 struct drbd_bitmap *b = mdev->bitmap;
712 ERR_IF(!b->bm_pages) return;
714 spin_lock_irq(&b->bm_lock);
715 bm_memset(b, 0, 0, b->bm_words);
717 spin_unlock_irq(&b->bm_lock);
720 static void bm_async_io_complete(struct bio *bio, int error)
722 struct drbd_bitmap *b = bio->bi_private;
723 int uptodate = bio_flagged(bio, BIO_UPTODATE);
726 /* strange behavior of some lower level drivers...
727 * fail the request by clearing the uptodate flag,
728 * but do not return any error?!
729 * do we want to WARN() on this? */
730 if (!error && !uptodate)
735 * for now, set all bits, and flag MD_IO_ERROR */
736 __set_bit(BM_MD_IO_ERROR, &b->bm_flags);
738 if (atomic_dec_and_test(&b->bm_async_io))
739 wake_up(&b->bm_io_wait);
744 static void bm_page_io_async(struct drbd_conf *mdev, struct drbd_bitmap *b, int page_nr, int rw) __must_hold(local)
746 /* we are process context. we always get a bio */
747 struct bio *bio = bio_alloc(GFP_KERNEL, 1);
749 sector_t on_disk_sector =
750 mdev->ldev->md.md_offset + mdev->ldev->md.bm_offset;
751 on_disk_sector += ((sector_t)page_nr) << (PAGE_SHIFT-9);
753 /* this might happen with very small
754 * flexible external meta data device */
755 len = min_t(unsigned int, PAGE_SIZE,
756 (drbd_md_last_sector(mdev->ldev) - on_disk_sector + 1)<<9);
758 bio->bi_bdev = mdev->ldev->md_bdev;
759 bio->bi_sector = on_disk_sector;
760 bio_add_page(bio, b->bm_pages[page_nr], len, 0);
762 bio->bi_end_io = bm_async_io_complete;
764 if (FAULT_ACTIVE(mdev, (rw & WRITE) ? DRBD_FAULT_MD_WR : DRBD_FAULT_MD_RD)) {
766 bio_endio(bio, -EIO);
772 # if defined(__LITTLE_ENDIAN)
773 /* nothing to do, on disk == in memory */
774 # define bm_cpu_to_lel(x) ((void)0)
776 void bm_cpu_to_lel(struct drbd_bitmap *b)
778 /* need to cpu_to_lel all the pages ...
779 * this may be optimized by using
780 * cpu_to_lel(-1) == -1 and cpu_to_lel(0) == 0;
781 * the following is still not optimal, but better than nothing */
783 unsigned long *p_addr, *bm;
784 if (b->bm_set == 0) {
785 /* no page at all; avoid swap if all is 0 */
786 i = b->bm_number_of_pages;
787 } else if (b->bm_set == b->bm_bits) {
788 /* only the last page */
789 i = b->bm_number_of_pages - 1;
794 for (; i < b->bm_number_of_pages; i++) {
795 p_addr = kmap_atomic(b->bm_pages[i], KM_USER0);
796 for (bm = p_addr; bm < p_addr + PAGE_SIZE/sizeof(long); bm++)
797 *bm = cpu_to_lel(*bm);
798 kunmap_atomic(p_addr, KM_USER0);
802 /* lel_to_cpu == cpu_to_lel */
803 # define bm_lel_to_cpu(x) bm_cpu_to_lel(x)
806 * bm_rw: read/write the whole bitmap from/to its on disk location.
808 static int bm_rw(struct drbd_conf *mdev, int rw) __must_hold(local)
810 struct drbd_bitmap *b = mdev->bitmap;
811 /* sector_t sector; */
812 int bm_words, num_pages, i;
817 WARN_ON(!bm_is_locked(b));
819 /* no spinlock here, the drbd_bm_lock should be enough! */
821 bm_words = drbd_bm_words(mdev);
822 num_pages = (bm_words*sizeof(long) + PAGE_SIZE-1) >> PAGE_SHIFT;
824 /* on disk bitmap is little endian */
829 atomic_set(&b->bm_async_io, num_pages);
830 __clear_bit(BM_MD_IO_ERROR, &b->bm_flags);
832 /* let the layers below us try to merge these bios... */
833 for (i = 0; i < num_pages; i++)
834 bm_page_io_async(mdev, b, i, rw);
836 drbd_blk_run_queue(bdev_get_queue(mdev->ldev->md_bdev));
837 wait_event(b->bm_io_wait, atomic_read(&b->bm_async_io) == 0);
839 if (test_bit(BM_MD_IO_ERROR, &b->bm_flags)) {
840 dev_alert(DEV, "we had at least one MD IO ERROR during bitmap IO\n");
841 drbd_chk_io_error(mdev, 1, TRUE);
847 /* swap back endianness */
849 /* flush bitmap to stable storage */
851 } else /* rw == READ */ {
852 /* just read, if necessary adjust endianness */
853 b->bm_set = bm_count_bits_swap_endian(b);
854 dev_info(DEV, "recounting of set bits took additional %lu jiffies\n",
859 dev_info(DEV, "%s (%lu bits) marked out-of-sync by on disk bit-map.\n",
860 ppsize(ppb, now << (BM_BLOCK_SHIFT-10)), now);
866 * drbd_bm_read() - Read the whole bitmap from its on disk location.
867 * @mdev: DRBD device.
869 int drbd_bm_read(struct drbd_conf *mdev) __must_hold(local)
871 return bm_rw(mdev, READ);
875 * drbd_bm_write() - Write the whole bitmap to its on disk location.
876 * @mdev: DRBD device.
878 int drbd_bm_write(struct drbd_conf *mdev) __must_hold(local)
880 return bm_rw(mdev, WRITE);
884 * drbd_bm_write_sect: Writes a 512 (MD_SECTOR_SIZE) byte piece of the bitmap
885 * @mdev: DRBD device.
886 * @enr: Extent number in the resync lru (happens to be sector offset)
888 * The BM_EXT_SIZE is on purpose exactly the amount of the bitmap covered
889 * by a single sector write. Therefore enr == sector offset from the
890 * start of the bitmap.
892 int drbd_bm_write_sect(struct drbd_conf *mdev, unsigned long enr) __must_hold(local)
894 sector_t on_disk_sector = enr + mdev->ldev->md.md_offset
895 + mdev->ldev->md.bm_offset;
896 int bm_words, num_words, offset;
899 mutex_lock(&mdev->md_io_mutex);
900 bm_words = drbd_bm_words(mdev);
901 offset = S2W(enr); /* word offset into bitmap */
902 num_words = min(S2W(1), bm_words - offset);
903 if (num_words < S2W(1))
904 memset(page_address(mdev->md_io_page), 0, MD_SECTOR_SIZE);
905 drbd_bm_get_lel(mdev, offset, num_words,
906 page_address(mdev->md_io_page));
907 if (!drbd_md_sync_page_io(mdev, mdev->ldev, on_disk_sector, WRITE)) {
910 dev_err(DEV, "IO ERROR writing bitmap sector %lu "
911 "(meta-disk sector %llus)\n",
912 enr, (unsigned long long)on_disk_sector);
913 drbd_chk_io_error(mdev, 1, TRUE);
914 for (i = 0; i < AL_EXT_PER_BM_SECT; i++)
915 drbd_bm_ALe_set_all(mdev, enr*AL_EXT_PER_BM_SECT+i);
918 mutex_unlock(&mdev->md_io_mutex);
923 * find_first_bit returns int, we return unsigned long.
924 * should not make much difference anyways, but ...
926 * this returns a bit number, NOT a sector!
928 #define BPP_MASK ((1UL << (PAGE_SHIFT+3)) - 1)
929 static unsigned long __bm_find_next(struct drbd_conf *mdev, unsigned long bm_fo,
930 const int find_zero_bit, const enum km_type km)
932 struct drbd_bitmap *b = mdev->bitmap;
933 unsigned long i = -1UL;
934 unsigned long *p_addr;
935 unsigned long bit_offset; /* bit offset of the mapped page. */
937 if (bm_fo > b->bm_bits) {
938 dev_err(DEV, "bm_fo=%lu bm_bits=%lu\n", bm_fo, b->bm_bits);
940 while (bm_fo < b->bm_bits) {
941 unsigned long offset;
942 bit_offset = bm_fo & ~BPP_MASK; /* bit offset of the page */
943 offset = bit_offset >> LN2_BPL; /* word offset of the page */
944 p_addr = __bm_map_paddr(b, offset, km);
947 i = find_next_zero_bit(p_addr, PAGE_SIZE*8, bm_fo & BPP_MASK);
949 i = find_next_bit(p_addr, PAGE_SIZE*8, bm_fo & BPP_MASK);
951 __bm_unmap(p_addr, km);
952 if (i < PAGE_SIZE*8) {
958 bm_fo = bit_offset + PAGE_SIZE*8;
966 static unsigned long bm_find_next(struct drbd_conf *mdev,
967 unsigned long bm_fo, const int find_zero_bit)
969 struct drbd_bitmap *b = mdev->bitmap;
970 unsigned long i = -1UL;
973 ERR_IF(!b->bm_pages) return i;
975 spin_lock_irq(&b->bm_lock);
977 bm_print_lock_info(mdev);
979 i = __bm_find_next(mdev, bm_fo, find_zero_bit, KM_IRQ1);
981 spin_unlock_irq(&b->bm_lock);
985 unsigned long drbd_bm_find_next(struct drbd_conf *mdev, unsigned long bm_fo)
987 return bm_find_next(mdev, bm_fo, 0);
991 /* not yet needed for anything. */
992 unsigned long drbd_bm_find_next_zero(struct drbd_conf *mdev, unsigned long bm_fo)
994 return bm_find_next(mdev, bm_fo, 1);
998 /* does not spin_lock_irqsave.
999 * you must take drbd_bm_lock() first */
1000 unsigned long _drbd_bm_find_next(struct drbd_conf *mdev, unsigned long bm_fo)
1002 /* WARN_ON(!bm_is_locked(mdev)); */
1003 return __bm_find_next(mdev, bm_fo, 0, KM_USER1);
1006 unsigned long _drbd_bm_find_next_zero(struct drbd_conf *mdev, unsigned long bm_fo)
1008 /* WARN_ON(!bm_is_locked(mdev)); */
1009 return __bm_find_next(mdev, bm_fo, 1, KM_USER1);
1012 /* returns number of bits actually changed.
1013 * for val != 0, we change 0 -> 1, return code positive
1014 * for val == 0, we change 1 -> 0, return code negative
1015 * wants bitnr, not sector.
1016 * expected to be called for only a few bits (e - s about BITS_PER_LONG).
1017 * Must hold bitmap lock already. */
1018 int __bm_change_bits_to(struct drbd_conf *mdev, const unsigned long s,
1019 unsigned long e, int val, const enum km_type km)
1021 struct drbd_bitmap *b = mdev->bitmap;
1022 unsigned long *p_addr = NULL;
1023 unsigned long bitnr;
1024 unsigned long last_page_nr = -1UL;
1027 if (e >= b->bm_bits) {
1028 dev_err(DEV, "ASSERT FAILED: bit_s=%lu bit_e=%lu bm_bits=%lu\n",
1030 e = b->bm_bits ? b->bm_bits -1 : 0;
1032 for (bitnr = s; bitnr <= e; bitnr++) {
1033 unsigned long offset = bitnr>>LN2_BPL;
1034 unsigned long page_nr = offset >> (PAGE_SHIFT - LN2_BPL + 3);
1035 if (page_nr != last_page_nr) {
1037 __bm_unmap(p_addr, km);
1038 p_addr = __bm_map_paddr(b, offset, km);
1039 last_page_nr = page_nr;
1042 c += (0 == __test_and_set_bit(bitnr & BPP_MASK, p_addr));
1044 c -= (0 != __test_and_clear_bit(bitnr & BPP_MASK, p_addr));
1047 __bm_unmap(p_addr, km);
1052 /* returns number of bits actually changed.
1053 * for val != 0, we change 0 -> 1, return code positive
1054 * for val == 0, we change 1 -> 0, return code negative
1055 * wants bitnr, not sector */
1056 int bm_change_bits_to(struct drbd_conf *mdev, const unsigned long s,
1057 const unsigned long e, int val)
1059 unsigned long flags;
1060 struct drbd_bitmap *b = mdev->bitmap;
1063 ERR_IF(!b) return 1;
1064 ERR_IF(!b->bm_pages) return 0;
1066 spin_lock_irqsave(&b->bm_lock, flags);
1067 if (bm_is_locked(b))
1068 bm_print_lock_info(mdev);
1070 c = __bm_change_bits_to(mdev, s, e, val, KM_IRQ1);
1072 spin_unlock_irqrestore(&b->bm_lock, flags);
1076 /* returns number of bits changed 0 -> 1 */
1077 int drbd_bm_set_bits(struct drbd_conf *mdev, const unsigned long s, const unsigned long e)
1079 return bm_change_bits_to(mdev, s, e, 1);
1082 /* returns number of bits changed 1 -> 0 */
1083 int drbd_bm_clear_bits(struct drbd_conf *mdev, const unsigned long s, const unsigned long e)
1085 return -bm_change_bits_to(mdev, s, e, 0);
1088 /* sets all bits in full words,
1089 * from first_word up to, but not including, last_word */
1090 static inline void bm_set_full_words_within_one_page(struct drbd_bitmap *b,
1091 int page_nr, int first_word, int last_word)
1095 unsigned long *paddr = kmap_atomic(b->bm_pages[page_nr], KM_USER0);
1096 for (i = first_word; i < last_word; i++) {
1097 bits = hweight_long(paddr[i]);
1099 b->bm_set += BITS_PER_LONG - bits;
1101 kunmap_atomic(paddr, KM_USER0);
1104 /* Same thing as drbd_bm_set_bits, but without taking the spin_lock_irqsave.
1105 * You must first drbd_bm_lock().
1106 * Can be called to set the whole bitmap in one go.
1107 * Sets bits from s to e _inclusive_. */
1108 void _drbd_bm_set_bits(struct drbd_conf *mdev, const unsigned long s, const unsigned long e)
1110 /* First set_bit from the first bit (s)
1111 * up to the next long boundary (sl),
1112 * then assign full words up to the last long boundary (el),
1113 * then set_bit up to and including the last bit (e).
1115 * Do not use memset, because we must account for changes,
1116 * so we need to loop over the words with hweight() anyways.
1118 unsigned long sl = ALIGN(s,BITS_PER_LONG);
1119 unsigned long el = (e+1) & ~((unsigned long)BITS_PER_LONG-1);
1126 if (e - s <= 3*BITS_PER_LONG) {
1127 /* don't bother; el and sl may even be wrong. */
1128 __bm_change_bits_to(mdev, s, e, 1, KM_USER0);
1132 /* difference is large enough that we can trust sl and el */
1134 /* bits filling the current long */
1136 __bm_change_bits_to(mdev, s, sl-1, 1, KM_USER0);
1138 first_page = sl >> (3 + PAGE_SHIFT);
1139 last_page = el >> (3 + PAGE_SHIFT);
1141 /* MLPP: modulo longs per page */
1142 /* LWPP: long words per page */
1143 first_word = MLPP(sl >> LN2_BPL);
1146 /* first and full pages, unless first page == last page */
1147 for (page_nr = first_page; page_nr < last_page; page_nr++) {
1148 bm_set_full_words_within_one_page(mdev->bitmap, page_nr, first_word, last_word);
1153 /* last page (respectively only page, for first page == last page) */
1154 last_word = MLPP(el >> LN2_BPL);
1155 bm_set_full_words_within_one_page(mdev->bitmap, last_page, first_word, last_word);
1157 /* possibly trailing bits.
1158 * example: (e & 63) == 63, el will be e+1.
1159 * if that even was the very last bit,
1160 * it would trigger an assert in __bm_change_bits_to()
1163 __bm_change_bits_to(mdev, el, e, 1, KM_USER0);
1166 /* returns bit state
1167 * wants bitnr, NOT sector.
1168 * inherently racy... area needs to be locked by means of {al,rs}_lru
1171 * -1 ... first out of bounds access, stop testing for bits!
1173 int drbd_bm_test_bit(struct drbd_conf *mdev, const unsigned long bitnr)
1175 unsigned long flags;
1176 struct drbd_bitmap *b = mdev->bitmap;
1177 unsigned long *p_addr;
1180 ERR_IF(!b) return 0;
1181 ERR_IF(!b->bm_pages) return 0;
1183 spin_lock_irqsave(&b->bm_lock, flags);
1184 if (bm_is_locked(b))
1185 bm_print_lock_info(mdev);
1186 if (bitnr < b->bm_bits) {
1187 unsigned long offset = bitnr>>LN2_BPL;
1188 p_addr = bm_map_paddr(b, offset);
1189 i = test_bit(bitnr & BPP_MASK, p_addr) ? 1 : 0;
1191 } else if (bitnr == b->bm_bits) {
1193 } else { /* (bitnr > b->bm_bits) */
1194 dev_err(DEV, "bitnr=%lu > bm_bits=%lu\n", bitnr, b->bm_bits);
1198 spin_unlock_irqrestore(&b->bm_lock, flags);
1202 /* returns number of bits set in the range [s, e] */
1203 int drbd_bm_count_bits(struct drbd_conf *mdev, const unsigned long s, const unsigned long e)
1205 unsigned long flags;
1206 struct drbd_bitmap *b = mdev->bitmap;
1207 unsigned long *p_addr = NULL, page_nr = -1;
1208 unsigned long bitnr;
1212 /* If this is called without a bitmap, that is a bug. But just to be
1213 * robust in case we screwed up elsewhere, in that case pretend there
1214 * was one dirty bit in the requested area, so we won't try to do a
1215 * local read there (no bitmap probably implies no disk) */
1216 ERR_IF(!b) return 1;
1217 ERR_IF(!b->bm_pages) return 1;
1219 spin_lock_irqsave(&b->bm_lock, flags);
1220 if (bm_is_locked(b))
1221 bm_print_lock_info(mdev);
1222 for (bitnr = s; bitnr <= e; bitnr++) {
1223 w = bitnr >> LN2_BPL;
1224 if (page_nr != w >> (PAGE_SHIFT - LN2_BPL + 3)) {
1225 page_nr = w >> (PAGE_SHIFT - LN2_BPL + 3);
1228 p_addr = bm_map_paddr(b, w);
1230 ERR_IF (bitnr >= b->bm_bits) {
1231 dev_err(DEV, "bitnr=%lu bm_bits=%lu\n", bitnr, b->bm_bits);
1233 c += (0 != test_bit(bitnr - (page_nr << (PAGE_SHIFT+3)), p_addr));
1238 spin_unlock_irqrestore(&b->bm_lock, flags);
1243 /* inherently racy...
1244 * return value may be already out-of-date when this function returns.
1245 * but the general usage is that this is only use during a cstate when bits are
1246 * only cleared, not set, and typically only care for the case when the return
1247 * value is zero, or we already "locked" this "bitmap extent" by other means.
1249 * enr is bm-extent number, since we chose to name one sector (512 bytes)
1250 * worth of the bitmap a "bitmap extent".
1253 * I think since we use it like a reference count, we should use the real
1254 * reference count of some bitmap extent element from some lru instead...
1257 int drbd_bm_e_weight(struct drbd_conf *mdev, unsigned long enr)
1259 struct drbd_bitmap *b = mdev->bitmap;
1261 unsigned long flags;
1262 unsigned long *p_addr, *bm;
1264 ERR_IF(!b) return 0;
1265 ERR_IF(!b->bm_pages) return 0;
1267 spin_lock_irqsave(&b->bm_lock, flags);
1268 if (bm_is_locked(b))
1269 bm_print_lock_info(mdev);
1272 e = min((size_t)S2W(enr+1), b->bm_words);
1274 if (s < b->bm_words) {
1276 p_addr = bm_map_paddr(b, s);
1277 bm = p_addr + MLPP(s);
1279 count += hweight_long(*bm++);
1282 dev_err(DEV, "start offset (%d) too large in drbd_bm_e_weight\n", s);
1284 spin_unlock_irqrestore(&b->bm_lock, flags);
1288 /* set all bits covered by the AL-extent al_enr */
1289 unsigned long drbd_bm_ALe_set_all(struct drbd_conf *mdev, unsigned long al_enr)
1291 struct drbd_bitmap *b = mdev->bitmap;
1292 unsigned long *p_addr, *bm;
1293 unsigned long weight;
1294 int count, s, e, i, do_now;
1295 ERR_IF(!b) return 0;
1296 ERR_IF(!b->bm_pages) return 0;
1298 spin_lock_irq(&b->bm_lock);
1299 if (bm_is_locked(b))
1300 bm_print_lock_info(mdev);
1303 s = al_enr * BM_WORDS_PER_AL_EXT;
1304 e = min_t(size_t, s + BM_WORDS_PER_AL_EXT, b->bm_words);
1305 /* assert that s and e are on the same page */
1306 D_ASSERT((e-1) >> (PAGE_SHIFT - LN2_BPL + 3)
1307 == s >> (PAGE_SHIFT - LN2_BPL + 3));
1309 if (s < b->bm_words) {
1311 p_addr = bm_map_paddr(b, s);
1312 bm = p_addr + MLPP(s);
1314 count += hweight_long(*bm);
1319 b->bm_set += do_now*BITS_PER_LONG - count;
1320 if (e == b->bm_words)
1321 b->bm_set -= bm_clear_surplus(b);
1323 dev_err(DEV, "start offset (%d) too large in drbd_bm_ALe_set_all\n", s);
1325 weight = b->bm_set - weight;
1326 spin_unlock_irq(&b->bm_lock);