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>
35 /* OPAQUE outside this file!
36 * interface defined in drbd_int.h
39 * function name drbd_bm_... => used elsewhere, "public".
40 * function name bm_... => internal to implementation, "private".
46 * We want to support >= peta byte of backend storage, while for now still using
47 * a granularity of one bit per 4KiB of storage.
48 * 1 << 50 bytes backend storage (1 PiB)
49 * 1 << (50 - 12) bits needed
50 * 38 --> we need u64 to index and count bits
51 * 1 << (38 - 3) bitmap bytes needed
52 * 35 --> we still need u64 to index and count bytes
53 * (that's 32 GiB of bitmap for 1 PiB storage)
54 * 1 << (35 - 2) 32bit longs needed
55 * 33 --> we'd even need u64 to index and count 32bit long words.
56 * 1 << (35 - 3) 64bit longs needed
57 * 32 --> we could get away with a 32bit unsigned int to index and count
58 * 64bit long words, but I rather stay with unsigned long for now.
59 * We probably should neither count nor point to bytes or long words
60 * directly, but either by bitnumber, or by page index and offset.
62 * 22 --> we need that much 4KiB pages of bitmap.
63 * 1 << (22 + 3) --> on a 64bit arch,
64 * we need 32 MiB to store the array of page pointers.
66 * Because I'm lazy, and because the resulting patch was too large, too ugly
67 * and still incomplete, on 32bit we still "only" support 16 TiB (minus some),
68 * (1 << 32) bits * 4k storage.
71 * bitmap storage and IO:
72 * Bitmap is stored little endian on disk, and is kept little endian in
73 * core memory. Currently we still hold the full bitmap in core as long
74 * as we are "attached" to a local disk, which at 32 GiB for 1PiB storage
77 * We plan to reduce the amount of in-core bitmap pages by paging them in
78 * and out against their on-disk location as necessary, but need to make
79 * sure we don't cause too much meta data IO, and must not deadlock in
80 * tight memory situations. This needs some more work.
85 * Access to the *bm_pages is protected by bm_lock.
86 * It is safe to read the other members within the lock.
88 * drbd_bm_set_bits is called from bio_endio callbacks,
89 * We may be called with irq already disabled,
90 * so we need spin_lock_irqsave().
91 * And we need the kmap_atomic.
94 struct page **bm_pages;
97 /* see LIMITATIONS: above */
99 unsigned long bm_set; /* nr of set bits; THINK maybe atomic_t? */
100 unsigned long bm_bits;
102 size_t bm_number_of_pages;
103 sector_t bm_dev_capacity;
104 struct mutex bm_change; /* serializes resize operations */
106 wait_queue_head_t bm_io_wait; /* used to serialize IO of single pages */
108 enum bm_flag bm_flags;
110 /* debugging aid, in case we are still racy somewhere */
112 struct task_struct *bm_task;
115 #define bm_print_lock_info(m) __bm_print_lock_info(m, __func__)
116 static void __bm_print_lock_info(struct drbd_conf *mdev, const char *func)
118 struct drbd_bitmap *b = mdev->bitmap;
119 if (!__ratelimit(&drbd_ratelimit_state))
121 dev_err(DEV, "FIXME %s in %s, bitmap locked for '%s' by %s\n",
122 drbd_task_to_thread_name(mdev->tconn, current),
123 func, b->bm_why ?: "?",
124 drbd_task_to_thread_name(mdev->tconn, b->bm_task));
127 void drbd_bm_lock(struct drbd_conf *mdev, char *why, enum bm_flag flags)
129 struct drbd_bitmap *b = mdev->bitmap;
133 dev_err(DEV, "FIXME no bitmap in drbd_bm_lock!?\n");
137 trylock_failed = !mutex_trylock(&b->bm_change);
139 if (trylock_failed) {
140 dev_warn(DEV, "%s going to '%s' but bitmap already locked for '%s' by %s\n",
141 drbd_task_to_thread_name(mdev->tconn, current),
142 why, b->bm_why ?: "?",
143 drbd_task_to_thread_name(mdev->tconn, b->bm_task));
144 mutex_lock(&b->bm_change);
146 if (BM_LOCKED_MASK & b->bm_flags)
147 dev_err(DEV, "FIXME bitmap already locked in bm_lock\n");
148 b->bm_flags |= flags & BM_LOCKED_MASK;
151 b->bm_task = current;
154 void drbd_bm_unlock(struct drbd_conf *mdev)
156 struct drbd_bitmap *b = mdev->bitmap;
158 dev_err(DEV, "FIXME no bitmap in drbd_bm_unlock!?\n");
162 if (!(BM_LOCKED_MASK & mdev->bitmap->bm_flags))
163 dev_err(DEV, "FIXME bitmap not locked in bm_unlock\n");
165 b->bm_flags &= ~BM_LOCKED_MASK;
168 mutex_unlock(&b->bm_change);
171 /* we store some "meta" info about our pages in page->private */
172 /* at a granularity of 4k storage per bitmap bit:
173 * one peta byte storage: 1<<50 byte, 1<<38 * 4k storage blocks
175 * 1<<23 4k bitmap pages.
176 * Use 24 bits as page index, covers 2 peta byte storage
177 * at a granularity of 4k per bit.
178 * Used to report the failed page idx on io error from the endio handlers.
180 #define BM_PAGE_IDX_MASK ((1UL<<24)-1)
181 /* this page is currently read in, or written back */
182 #define BM_PAGE_IO_LOCK 31
183 /* if there has been an IO error for this page */
184 #define BM_PAGE_IO_ERROR 30
185 /* this is to be able to intelligently skip disk IO,
186 * set if bits have been set since last IO. */
187 #define BM_PAGE_NEED_WRITEOUT 29
188 /* to mark for lazy writeout once syncer cleared all clearable bits,
189 * we if bits have been cleared since last IO. */
190 #define BM_PAGE_LAZY_WRITEOUT 28
192 /* store_page_idx uses non-atomic assignment. It is only used directly after
193 * allocating the page. All other bm_set_page_* and bm_clear_page_* need to
194 * use atomic bit manipulation, as set_out_of_sync (and therefore bitmap
195 * changes) may happen from various contexts, and wait_on_bit/wake_up_bit
196 * requires it all to be atomic as well. */
197 static void bm_store_page_idx(struct page *page, unsigned long idx)
199 BUG_ON(0 != (idx & ~BM_PAGE_IDX_MASK));
200 page_private(page) |= idx;
203 static unsigned long bm_page_to_idx(struct page *page)
205 return page_private(page) & BM_PAGE_IDX_MASK;
208 /* As is very unlikely that the same page is under IO from more than one
209 * context, we can get away with a bit per page and one wait queue per bitmap.
211 static void bm_page_lock_io(struct drbd_conf *mdev, int page_nr)
213 struct drbd_bitmap *b = mdev->bitmap;
214 void *addr = &page_private(b->bm_pages[page_nr]);
215 wait_event(b->bm_io_wait, !test_and_set_bit(BM_PAGE_IO_LOCK, addr));
218 static void bm_page_unlock_io(struct drbd_conf *mdev, int page_nr)
220 struct drbd_bitmap *b = mdev->bitmap;
221 void *addr = &page_private(b->bm_pages[page_nr]);
222 clear_bit(BM_PAGE_IO_LOCK, addr);
223 smp_mb__after_clear_bit();
224 wake_up(&mdev->bitmap->bm_io_wait);
227 /* set _before_ submit_io, so it may be reset due to being changed
228 * while this page is in flight... will get submitted later again */
229 static void bm_set_page_unchanged(struct page *page)
232 clear_bit(BM_PAGE_NEED_WRITEOUT, &page_private(page));
233 clear_bit(BM_PAGE_LAZY_WRITEOUT, &page_private(page));
236 static void bm_set_page_need_writeout(struct page *page)
238 set_bit(BM_PAGE_NEED_WRITEOUT, &page_private(page));
241 static int bm_test_page_unchanged(struct page *page)
243 volatile const unsigned long *addr = &page_private(page);
244 return (*addr & ((1UL<<BM_PAGE_NEED_WRITEOUT)|(1UL<<BM_PAGE_LAZY_WRITEOUT))) == 0;
247 static void bm_set_page_io_err(struct page *page)
249 set_bit(BM_PAGE_IO_ERROR, &page_private(page));
252 static void bm_clear_page_io_err(struct page *page)
254 clear_bit(BM_PAGE_IO_ERROR, &page_private(page));
257 static void bm_set_page_lazy_writeout(struct page *page)
259 set_bit(BM_PAGE_LAZY_WRITEOUT, &page_private(page));
262 static int bm_test_page_lazy_writeout(struct page *page)
264 return test_bit(BM_PAGE_LAZY_WRITEOUT, &page_private(page));
267 /* on a 32bit box, this would allow for exactly (2<<38) bits. */
268 static unsigned int bm_word_to_page_idx(struct drbd_bitmap *b, unsigned long long_nr)
270 /* page_nr = (word*sizeof(long)) >> PAGE_SHIFT; */
271 unsigned int page_nr = long_nr >> (PAGE_SHIFT - LN2_BPL + 3);
272 BUG_ON(page_nr >= b->bm_number_of_pages);
276 static unsigned int bm_bit_to_page_idx(struct drbd_bitmap *b, u64 bitnr)
278 /* page_nr = (bitnr/8) >> PAGE_SHIFT; */
279 unsigned int page_nr = bitnr >> (PAGE_SHIFT + 3);
280 BUG_ON(page_nr >= b->bm_number_of_pages);
284 static unsigned long *__bm_map_pidx(struct drbd_bitmap *b, unsigned int idx, const enum km_type km)
286 struct page *page = b->bm_pages[idx];
287 return (unsigned long *) kmap_atomic(page, km);
290 static unsigned long *bm_map_pidx(struct drbd_bitmap *b, unsigned int idx)
292 return __bm_map_pidx(b, idx, KM_IRQ1);
295 static void __bm_unmap(unsigned long *p_addr, const enum km_type km)
297 kunmap_atomic(p_addr, km);
300 static void bm_unmap(unsigned long *p_addr)
302 return __bm_unmap(p_addr, KM_IRQ1);
305 /* long word offset of _bitmap_ sector */
306 #define S2W(s) ((s)<<(BM_EXT_SHIFT-BM_BLOCK_SHIFT-LN2_BPL))
307 /* word offset from start of bitmap to word number _in_page_
308 * modulo longs per page
309 #define MLPP(X) ((X) % (PAGE_SIZE/sizeof(long))
310 hm, well, Philipp thinks gcc might not optimize the % into & (... - 1)
313 #define MLPP(X) ((X) & ((PAGE_SIZE/sizeof(long))-1))
315 /* Long words per page */
316 #define LWPP (PAGE_SIZE/sizeof(long))
319 * actually most functions herein should take a struct drbd_bitmap*, not a
320 * struct drbd_conf*, but for the debug macros I like to have the mdev around
321 * to be able to report device specific.
325 static void bm_free_pages(struct page **pages, unsigned long number)
331 for (i = 0; i < number; i++) {
333 printk(KERN_ALERT "drbd: bm_free_pages tried to free "
334 "a NULL pointer; i=%lu n=%lu\n",
338 __free_page(pages[i]);
343 static void bm_vk_free(void *ptr, int v)
352 * "have" and "want" are NUMBER OF PAGES.
354 static struct page **bm_realloc_pages(struct drbd_bitmap *b, unsigned long want)
356 struct page **old_pages = b->bm_pages;
357 struct page **new_pages, *page;
358 unsigned int i, bytes, vmalloced = 0;
359 unsigned long have = b->bm_number_of_pages;
361 BUG_ON(have == 0 && old_pages != NULL);
362 BUG_ON(have != 0 && old_pages == NULL);
367 /* Trying kmalloc first, falling back to vmalloc.
368 * GFP_KERNEL is ok, as this is done when a lower level disk is
369 * "attached" to the drbd. Context is receiver thread or cqueue
370 * thread. As we have no disk yet, we are not in the IO path,
371 * not even the IO path of the peer. */
372 bytes = sizeof(struct page *)*want;
373 new_pages = kmalloc(bytes, GFP_KERNEL);
375 new_pages = vmalloc(bytes);
381 memset(new_pages, 0, bytes);
383 for (i = 0; i < have; i++)
384 new_pages[i] = old_pages[i];
385 for (; i < want; i++) {
386 page = alloc_page(GFP_HIGHUSER);
388 bm_free_pages(new_pages + have, i - have);
389 bm_vk_free(new_pages, vmalloced);
392 /* we want to know which page it is
393 * from the endio handlers */
394 bm_store_page_idx(page, i);
398 for (i = 0; i < want; i++)
399 new_pages[i] = old_pages[i];
400 /* NOT HERE, we are outside the spinlock!
401 bm_free_pages(old_pages + want, have - want);
406 b->bm_flags |= BM_P_VMALLOCED;
408 b->bm_flags &= ~BM_P_VMALLOCED;
414 * called on driver init only. TODO call when a device is created.
415 * allocates the drbd_bitmap, and stores it in mdev->bitmap.
417 int drbd_bm_init(struct drbd_conf *mdev)
419 struct drbd_bitmap *b = mdev->bitmap;
421 b = kzalloc(sizeof(struct drbd_bitmap), GFP_KERNEL);
424 spin_lock_init(&b->bm_lock);
425 mutex_init(&b->bm_change);
426 init_waitqueue_head(&b->bm_io_wait);
433 sector_t drbd_bm_capacity(struct drbd_conf *mdev)
435 if (!expect(mdev->bitmap))
437 return mdev->bitmap->bm_dev_capacity;
440 /* called on driver unload. TODO: call when a device is destroyed.
442 void drbd_bm_cleanup(struct drbd_conf *mdev)
444 if (!expect(mdev->bitmap))
446 bm_free_pages(mdev->bitmap->bm_pages, mdev->bitmap->bm_number_of_pages);
447 bm_vk_free(mdev->bitmap->bm_pages, (BM_P_VMALLOCED & mdev->bitmap->bm_flags));
453 * since (b->bm_bits % BITS_PER_LONG) != 0,
454 * this masks out the remaining bits.
455 * Returns the number of bits cleared.
457 #define BITS_PER_PAGE (1UL << (PAGE_SHIFT + 3))
458 #define BITS_PER_PAGE_MASK (BITS_PER_PAGE - 1)
459 #define BITS_PER_LONG_MASK (BITS_PER_LONG - 1)
460 static int bm_clear_surplus(struct drbd_bitmap *b)
463 unsigned long *p_addr, *bm;
467 /* number of bits modulo bits per page */
468 tmp = (b->bm_bits & BITS_PER_PAGE_MASK);
469 /* mask the used bits of the word containing the last bit */
470 mask = (1UL << (tmp & BITS_PER_LONG_MASK)) -1;
471 /* bitmap is always stored little endian,
472 * on disk and in core memory alike */
473 mask = cpu_to_lel(mask);
475 p_addr = bm_map_pidx(b, b->bm_number_of_pages - 1);
476 bm = p_addr + (tmp/BITS_PER_LONG);
478 /* If mask != 0, we are not exactly aligned, so bm now points
479 * to the long containing the last bit.
480 * If mask == 0, bm already points to the word immediately
481 * after the last (long word aligned) bit. */
482 cleared = hweight_long(*bm & ~mask);
487 if (BITS_PER_LONG == 32 && ((bm - p_addr) & 1) == 1) {
488 /* on a 32bit arch, we may need to zero out
489 * a padding long to align with a 64bit remote */
490 cleared += hweight_long(*bm);
497 static void bm_set_surplus(struct drbd_bitmap *b)
500 unsigned long *p_addr, *bm;
503 /* number of bits modulo bits per page */
504 tmp = (b->bm_bits & BITS_PER_PAGE_MASK);
505 /* mask the used bits of the word containing the last bit */
506 mask = (1UL << (tmp & BITS_PER_LONG_MASK)) -1;
507 /* bitmap is always stored little endian,
508 * on disk and in core memory alike */
509 mask = cpu_to_lel(mask);
511 p_addr = bm_map_pidx(b, b->bm_number_of_pages - 1);
512 bm = p_addr + (tmp/BITS_PER_LONG);
514 /* If mask != 0, we are not exactly aligned, so bm now points
515 * to the long containing the last bit.
516 * If mask == 0, bm already points to the word immediately
517 * after the last (long word aligned) bit. */
522 if (BITS_PER_LONG == 32 && ((bm - p_addr) & 1) == 1) {
523 /* on a 32bit arch, we may need to zero out
524 * a padding long to align with a 64bit remote */
530 /* you better not modify the bitmap while this is running,
531 * or its results will be stale */
532 static unsigned long bm_count_bits(struct drbd_bitmap *b)
534 unsigned long *p_addr;
535 unsigned long bits = 0;
536 unsigned long mask = (1UL << (b->bm_bits & BITS_PER_LONG_MASK)) -1;
537 int idx, i, last_word;
539 /* all but last page */
540 for (idx = 0; idx < b->bm_number_of_pages - 1; idx++) {
541 p_addr = __bm_map_pidx(b, idx, KM_USER0);
542 for (i = 0; i < LWPP; i++)
543 bits += hweight_long(p_addr[i]);
544 __bm_unmap(p_addr, KM_USER0);
547 /* last (or only) page */
548 last_word = ((b->bm_bits - 1) & BITS_PER_PAGE_MASK) >> LN2_BPL;
549 p_addr = __bm_map_pidx(b, idx, KM_USER0);
550 for (i = 0; i < last_word; i++)
551 bits += hweight_long(p_addr[i]);
552 p_addr[last_word] &= cpu_to_lel(mask);
553 bits += hweight_long(p_addr[last_word]);
554 /* 32bit arch, may have an unused padding long */
555 if (BITS_PER_LONG == 32 && (last_word & 1) == 0)
556 p_addr[last_word+1] = 0;
557 __bm_unmap(p_addr, KM_USER0);
561 /* offset and len in long words.*/
562 static void bm_memset(struct drbd_bitmap *b, size_t offset, int c, size_t len)
564 unsigned long *p_addr, *bm;
570 if (end > b->bm_words) {
571 printk(KERN_ALERT "drbd: bm_memset end > bm_words\n");
575 while (offset < end) {
576 do_now = min_t(size_t, ALIGN(offset + 1, LWPP), end) - offset;
577 idx = bm_word_to_page_idx(b, offset);
578 p_addr = bm_map_pidx(b, idx);
579 bm = p_addr + MLPP(offset);
580 if (bm+do_now > p_addr + LWPP) {
581 printk(KERN_ALERT "drbd: BUG BUG BUG! p_addr:%p bm:%p do_now:%d\n",
582 p_addr, bm, (int)do_now);
584 memset(bm, c, do_now * sizeof(long));
586 bm_set_page_need_writeout(b->bm_pages[idx]);
592 * make sure the bitmap has enough room for the attached storage,
593 * if necessary, resize.
594 * called whenever we may have changed the device size.
595 * returns -ENOMEM if we could not allocate enough memory, 0 on success.
596 * In case this is actually a resize, we copy the old bitmap into the new one.
597 * Otherwise, the bitmap is initialized to all bits set.
599 int drbd_bm_resize(struct drbd_conf *mdev, sector_t capacity, int set_new_bits)
601 struct drbd_bitmap *b = mdev->bitmap;
602 unsigned long bits, words, owords, obits;
603 unsigned long want, have, onpages; /* number of pages */
604 struct page **npages, **opages = NULL;
605 int err = 0, growing;
606 int opages_vmalloced;
611 drbd_bm_lock(mdev, "resize", BM_LOCKED_MASK);
613 dev_info(DEV, "drbd_bm_resize called with capacity == %llu\n",
614 (unsigned long long)capacity);
616 if (capacity == b->bm_dev_capacity)
619 opages_vmalloced = (BM_P_VMALLOCED & b->bm_flags);
622 spin_lock_irq(&b->bm_lock);
623 opages = b->bm_pages;
624 onpages = b->bm_number_of_pages;
625 owords = b->bm_words;
627 b->bm_number_of_pages =
631 b->bm_dev_capacity = 0;
632 spin_unlock_irq(&b->bm_lock);
633 bm_free_pages(opages, onpages);
634 bm_vk_free(opages, opages_vmalloced);
637 bits = BM_SECT_TO_BIT(ALIGN(capacity, BM_SECT_PER_BIT));
640 words = ALIGN(bits,BITS_PER_LONG) >> LN2_BPL;
641 a 32bit host could present the wrong number of words
644 words = ALIGN(bits, 64) >> LN2_BPL;
646 if (get_ldev(mdev)) {
647 u64 bits_on_disk = ((u64)mdev->ldev->md.md_size_sect-MD_BM_OFFSET) << 12;
649 if (bits > bits_on_disk) {
650 dev_info(DEV, "bits = %lu\n", bits);
651 dev_info(DEV, "bits_on_disk = %llu\n", bits_on_disk);
657 want = ALIGN(words*sizeof(long), PAGE_SIZE) >> PAGE_SHIFT;
658 have = b->bm_number_of_pages;
660 D_ASSERT(b->bm_pages != NULL);
661 npages = b->bm_pages;
663 if (drbd_insert_fault(mdev, DRBD_FAULT_BM_ALLOC))
666 npages = bm_realloc_pages(b, want);
674 spin_lock_irq(&b->bm_lock);
675 opages = b->bm_pages;
676 owords = b->bm_words;
679 growing = bits > obits;
680 if (opages && growing && set_new_bits)
683 b->bm_pages = npages;
684 b->bm_number_of_pages = want;
687 b->bm_dev_capacity = capacity;
691 bm_memset(b, owords, 0xff, words-owords);
692 b->bm_set += bits - obits;
694 bm_memset(b, owords, 0x00, words-owords);
699 /* implicit: (opages != NULL) && (opages != npages) */
700 bm_free_pages(opages + want, have - want);
703 (void)bm_clear_surplus(b);
705 spin_unlock_irq(&b->bm_lock);
706 if (opages != npages)
707 bm_vk_free(opages, opages_vmalloced);
709 b->bm_set = bm_count_bits(b);
710 dev_info(DEV, "resync bitmap: bits=%lu words=%lu pages=%lu\n", bits, words, want);
713 drbd_bm_unlock(mdev);
718 * if not protected by other means, return value may be out of date when
719 * leaving this function...
720 * we still need to lock it, since it is important that this returns
721 * bm_set == 0 precisely.
723 * maybe bm_set should be atomic_t ?
725 unsigned long _drbd_bm_total_weight(struct drbd_conf *mdev)
727 struct drbd_bitmap *b = mdev->bitmap;
733 if (!expect(b->bm_pages))
736 spin_lock_irqsave(&b->bm_lock, flags);
738 spin_unlock_irqrestore(&b->bm_lock, flags);
743 unsigned long drbd_bm_total_weight(struct drbd_conf *mdev)
746 /* if I don't have a disk, I don't know about out-of-sync status */
747 if (!get_ldev_if_state(mdev, D_NEGOTIATING))
749 s = _drbd_bm_total_weight(mdev);
754 size_t drbd_bm_words(struct drbd_conf *mdev)
756 struct drbd_bitmap *b = mdev->bitmap;
759 if (!expect(b->bm_pages))
765 unsigned long drbd_bm_bits(struct drbd_conf *mdev)
767 struct drbd_bitmap *b = mdev->bitmap;
774 /* merge number words from buffer into the bitmap starting at offset.
775 * buffer[i] is expected to be little endian unsigned long.
776 * bitmap must be locked by drbd_bm_lock.
777 * currently only used from receive_bitmap.
779 void drbd_bm_merge_lel(struct drbd_conf *mdev, size_t offset, size_t number,
780 unsigned long *buffer)
782 struct drbd_bitmap *b = mdev->bitmap;
783 unsigned long *p_addr, *bm;
784 unsigned long word, bits;
788 end = offset + number;
792 if (!expect(b->bm_pages))
796 WARN_ON(offset >= b->bm_words);
797 WARN_ON(end > b->bm_words);
799 spin_lock_irq(&b->bm_lock);
800 while (offset < end) {
801 do_now = min_t(size_t, ALIGN(offset+1, LWPP), end) - offset;
802 idx = bm_word_to_page_idx(b, offset);
803 p_addr = bm_map_pidx(b, idx);
804 bm = p_addr + MLPP(offset);
807 bits = hweight_long(*bm);
808 word = *bm | *buffer++;
810 b->bm_set += hweight_long(word) - bits;
813 bm_set_page_need_writeout(b->bm_pages[idx]);
815 /* with 32bit <-> 64bit cross-platform connect
816 * this is only correct for current usage,
817 * where we _know_ that we are 64 bit aligned,
818 * and know that this function is used in this way, too...
820 if (end == b->bm_words)
821 b->bm_set -= bm_clear_surplus(b);
822 spin_unlock_irq(&b->bm_lock);
825 /* copy number words from the bitmap starting at offset into the buffer.
826 * buffer[i] will be little endian unsigned long.
828 void drbd_bm_get_lel(struct drbd_conf *mdev, size_t offset, size_t number,
829 unsigned long *buffer)
831 struct drbd_bitmap *b = mdev->bitmap;
832 unsigned long *p_addr, *bm;
835 end = offset + number;
839 if (!expect(b->bm_pages))
842 spin_lock_irq(&b->bm_lock);
843 if ((offset >= b->bm_words) ||
844 (end > b->bm_words) ||
846 dev_err(DEV, "offset=%lu number=%lu bm_words=%lu\n",
847 (unsigned long) offset,
848 (unsigned long) number,
849 (unsigned long) b->bm_words);
851 while (offset < end) {
852 do_now = min_t(size_t, ALIGN(offset+1, LWPP), end) - offset;
853 p_addr = bm_map_pidx(b, bm_word_to_page_idx(b, offset));
854 bm = p_addr + MLPP(offset);
861 spin_unlock_irq(&b->bm_lock);
864 /* set all bits in the bitmap */
865 void drbd_bm_set_all(struct drbd_conf *mdev)
867 struct drbd_bitmap *b = mdev->bitmap;
870 if (!expect(b->bm_pages))
873 spin_lock_irq(&b->bm_lock);
874 bm_memset(b, 0, 0xff, b->bm_words);
875 (void)bm_clear_surplus(b);
876 b->bm_set = b->bm_bits;
877 spin_unlock_irq(&b->bm_lock);
880 /* clear all bits in the bitmap */
881 void drbd_bm_clear_all(struct drbd_conf *mdev)
883 struct drbd_bitmap *b = mdev->bitmap;
886 if (!expect(b->bm_pages))
889 spin_lock_irq(&b->bm_lock);
890 bm_memset(b, 0, 0, b->bm_words);
892 spin_unlock_irq(&b->bm_lock);
896 struct drbd_conf *mdev;
898 struct completion done;
900 #define BM_AIO_COPY_PAGES 1
904 /* bv_page may be a copy, or may be the original */
905 static void bm_async_io_complete(struct bio *bio, int error)
907 struct bm_aio_ctx *ctx = bio->bi_private;
908 struct drbd_conf *mdev = ctx->mdev;
909 struct drbd_bitmap *b = mdev->bitmap;
910 unsigned int idx = bm_page_to_idx(bio->bi_io_vec[0].bv_page);
911 int uptodate = bio_flagged(bio, BIO_UPTODATE);
914 /* strange behavior of some lower level drivers...
915 * fail the request by clearing the uptodate flag,
916 * but do not return any error?!
917 * do we want to WARN() on this? */
918 if (!error && !uptodate)
921 if ((ctx->flags & BM_AIO_COPY_PAGES) == 0 &&
922 !bm_test_page_unchanged(b->bm_pages[idx]))
923 dev_warn(DEV, "bitmap page idx %u changed during IO!\n", idx);
926 /* ctx error will hold the completed-last non-zero error code,
927 * in case error codes differ. */
929 bm_set_page_io_err(b->bm_pages[idx]);
930 /* Not identical to on disk version of it.
931 * Is BM_PAGE_IO_ERROR enough? */
932 if (__ratelimit(&drbd_ratelimit_state))
933 dev_err(DEV, "IO ERROR %d on bitmap page idx %u\n",
936 bm_clear_page_io_err(b->bm_pages[idx]);
937 dynamic_dev_dbg(DEV, "bitmap page idx %u completed\n", idx);
940 bm_page_unlock_io(mdev, idx);
942 /* FIXME give back to page pool */
943 if (ctx->flags & BM_AIO_COPY_PAGES)
944 put_page(bio->bi_io_vec[0].bv_page);
948 if (atomic_dec_and_test(&ctx->in_flight))
949 complete(&ctx->done);
952 static void bm_page_io_async(struct bm_aio_ctx *ctx, int page_nr, int rw) __must_hold(local)
954 /* we are process context. we always get a bio */
955 struct bio *bio = bio_alloc(GFP_KERNEL, 1);
956 struct drbd_conf *mdev = ctx->mdev;
957 struct drbd_bitmap *b = mdev->bitmap;
961 sector_t on_disk_sector =
962 mdev->ldev->md.md_offset + mdev->ldev->md.bm_offset;
963 on_disk_sector += ((sector_t)page_nr) << (PAGE_SHIFT-9);
965 /* this might happen with very small
966 * flexible external meta data device,
967 * or with PAGE_SIZE > 4k */
968 len = min_t(unsigned int, PAGE_SIZE,
969 (drbd_md_last_sector(mdev->ldev) - on_disk_sector + 1)<<9);
971 /* serialize IO on this page */
972 bm_page_lock_io(mdev, page_nr);
973 /* before memcpy and submit,
974 * so it can be redirtied any time */
975 bm_set_page_unchanged(b->bm_pages[page_nr]);
977 if (ctx->flags & BM_AIO_COPY_PAGES) {
978 /* FIXME alloc_page is good enough for now, but actually needs
979 * to use pre-allocated page pool */
981 page = alloc_page(__GFP_HIGHMEM|__GFP_WAIT);
982 dest = kmap_atomic(page, KM_USER0);
983 src = kmap_atomic(b->bm_pages[page_nr], KM_USER1);
984 memcpy(dest, src, PAGE_SIZE);
985 kunmap_atomic(src, KM_USER1);
986 kunmap_atomic(dest, KM_USER0);
987 bm_store_page_idx(page, page_nr);
989 page = b->bm_pages[page_nr];
991 bio->bi_bdev = mdev->ldev->md_bdev;
992 bio->bi_sector = on_disk_sector;
993 bio_add_page(bio, page, len, 0);
994 bio->bi_private = ctx;
995 bio->bi_end_io = bm_async_io_complete;
997 if (drbd_insert_fault(mdev, (rw & WRITE) ? DRBD_FAULT_MD_WR : DRBD_FAULT_MD_RD)) {
999 bio_endio(bio, -EIO);
1001 submit_bio(rw, bio);
1002 /* this should not count as user activity and cause the
1003 * resync to throttle -- see drbd_rs_should_slow_down(). */
1004 atomic_add(len >> 9, &mdev->rs_sect_ev);
1009 * bm_rw: read/write the whole bitmap from/to its on disk location.
1011 static int bm_rw(struct drbd_conf *mdev, int rw, unsigned lazy_writeout_upper_idx) __must_hold(local)
1013 struct bm_aio_ctx ctx = {
1015 .in_flight = ATOMIC_INIT(1),
1016 .done = COMPLETION_INITIALIZER_ONSTACK(ctx.done),
1017 .flags = lazy_writeout_upper_idx ? BM_AIO_COPY_PAGES : 0,
1019 struct drbd_bitmap *b = mdev->bitmap;
1020 int num_pages, i, count = 0;
1026 * We are protected against bitmap disappearing/resizing by holding an
1027 * ldev reference (caller must have called get_ldev()).
1028 * For read/write, we are protected against changes to the bitmap by
1029 * the bitmap lock (see drbd_bitmap_io).
1030 * For lazy writeout, we don't care for ongoing changes to the bitmap,
1031 * as we submit copies of pages anyways.
1034 WARN_ON(!(BM_LOCKED_MASK & b->bm_flags));
1036 num_pages = b->bm_number_of_pages;
1040 /* let the layers below us try to merge these bios... */
1041 for (i = 0; i < num_pages; i++) {
1042 /* ignore completely unchanged pages */
1043 if (lazy_writeout_upper_idx && i == lazy_writeout_upper_idx)
1046 if (bm_test_page_unchanged(b->bm_pages[i])) {
1047 dynamic_dev_dbg(DEV, "skipped bm write for idx %u\n", i);
1050 /* during lazy writeout,
1051 * ignore those pages not marked for lazy writeout. */
1052 if (lazy_writeout_upper_idx &&
1053 !bm_test_page_lazy_writeout(b->bm_pages[i])) {
1054 dynamic_dev_dbg(DEV, "skipped bm lazy write for idx %u\n", i);
1058 atomic_inc(&ctx.in_flight);
1059 bm_page_io_async(&ctx, i, rw);
1065 * We initialize ctx.in_flight to one to make sure bm_async_io_complete
1066 * will not complete() early, and decrement / test it here. If there
1067 * are still some bios in flight, we need to wait for them here.
1069 if (!atomic_dec_and_test(&ctx.in_flight))
1070 wait_for_completion(&ctx.done);
1071 dev_info(DEV, "bitmap %s of %u pages took %lu jiffies\n",
1072 rw == WRITE ? "WRITE" : "READ",
1073 count, jiffies - now);
1076 dev_alert(DEV, "we had at least one MD IO ERROR during bitmap IO\n");
1077 drbd_chk_io_error(mdev, 1, true);
1078 err = -EIO; /* ctx.error ? */
1083 drbd_md_flush(mdev);
1084 } else /* rw == READ */ {
1085 b->bm_set = bm_count_bits(b);
1086 dev_info(DEV, "recounting of set bits took additional %lu jiffies\n",
1091 dev_info(DEV, "%s (%lu bits) marked out-of-sync by on disk bit-map.\n",
1092 ppsize(ppb, now << (BM_BLOCK_SHIFT-10)), now);
1098 * drbd_bm_read() - Read the whole bitmap from its on disk location.
1099 * @mdev: DRBD device.
1101 int drbd_bm_read(struct drbd_conf *mdev) __must_hold(local)
1103 return bm_rw(mdev, READ, 0);
1107 * drbd_bm_write() - Write the whole bitmap to its on disk location.
1108 * @mdev: DRBD device.
1110 * Will only write pages that have changed since last IO.
1112 int drbd_bm_write(struct drbd_conf *mdev) __must_hold(local)
1114 return bm_rw(mdev, WRITE, 0);
1118 * drbd_bm_lazy_write_out() - Write bitmap pages 0 to @upper_idx-1, if they have changed.
1119 * @mdev: DRBD device.
1120 * @upper_idx: 0: write all changed pages; +ve: page index to stop scanning for changed pages
1122 int drbd_bm_write_lazy(struct drbd_conf *mdev, unsigned upper_idx) __must_hold(local)
1124 return bm_rw(mdev, WRITE, upper_idx);
1129 * drbd_bm_write_page: Writes a PAGE_SIZE aligned piece of bitmap
1130 * @mdev: DRBD device.
1131 * @idx: bitmap page index
1133 * We don't want to special case on logical_block_size of the backend device,
1134 * so we submit PAGE_SIZE aligned pieces.
1135 * Note that on "most" systems, PAGE_SIZE is 4k.
1137 * In case this becomes an issue on systems with larger PAGE_SIZE,
1138 * we may want to change this again to write 4k aligned 4k pieces.
1140 int drbd_bm_write_page(struct drbd_conf *mdev, unsigned int idx) __must_hold(local)
1142 struct bm_aio_ctx ctx = {
1144 .in_flight = ATOMIC_INIT(1),
1145 .done = COMPLETION_INITIALIZER_ONSTACK(ctx.done),
1146 .flags = BM_AIO_COPY_PAGES,
1149 if (bm_test_page_unchanged(mdev->bitmap->bm_pages[idx])) {
1150 dynamic_dev_dbg(DEV, "skipped bm page write for idx %u\n", idx);
1154 bm_page_io_async(&ctx, idx, WRITE_SYNC);
1155 wait_for_completion(&ctx.done);
1158 drbd_chk_io_error(mdev, 1, true);
1159 /* that should force detach, so the in memory bitmap will be
1160 * gone in a moment as well. */
1162 mdev->bm_writ_cnt++;
1167 * find_first_bit returns int, we return unsigned long.
1168 * For this to work on 32bit arch with bitnumbers > (1<<32),
1169 * we'd need to return u64, and get a whole lot of other places
1170 * fixed where we still use unsigned long.
1172 * this returns a bit number, NOT a sector!
1174 static unsigned long __bm_find_next(struct drbd_conf *mdev, unsigned long bm_fo,
1175 const int find_zero_bit, const enum km_type km)
1177 struct drbd_bitmap *b = mdev->bitmap;
1178 unsigned long *p_addr;
1179 unsigned long bit_offset;
1183 if (bm_fo > b->bm_bits) {
1184 dev_err(DEV, "bm_fo=%lu bm_bits=%lu\n", bm_fo, b->bm_bits);
1185 bm_fo = DRBD_END_OF_BITMAP;
1187 while (bm_fo < b->bm_bits) {
1188 /* bit offset of the first bit in the page */
1189 bit_offset = bm_fo & ~BITS_PER_PAGE_MASK;
1190 p_addr = __bm_map_pidx(b, bm_bit_to_page_idx(b, bm_fo), km);
1193 i = find_next_zero_bit_le(p_addr,
1194 PAGE_SIZE*8, bm_fo & BITS_PER_PAGE_MASK);
1196 i = find_next_bit_le(p_addr,
1197 PAGE_SIZE*8, bm_fo & BITS_PER_PAGE_MASK);
1199 __bm_unmap(p_addr, km);
1200 if (i < PAGE_SIZE*8) {
1201 bm_fo = bit_offset + i;
1202 if (bm_fo >= b->bm_bits)
1206 bm_fo = bit_offset + PAGE_SIZE*8;
1208 bm_fo = DRBD_END_OF_BITMAP;
1214 static unsigned long bm_find_next(struct drbd_conf *mdev,
1215 unsigned long bm_fo, const int find_zero_bit)
1217 struct drbd_bitmap *b = mdev->bitmap;
1218 unsigned long i = DRBD_END_OF_BITMAP;
1222 if (!expect(b->bm_pages))
1225 spin_lock_irq(&b->bm_lock);
1226 if (BM_DONT_TEST & b->bm_flags)
1227 bm_print_lock_info(mdev);
1229 i = __bm_find_next(mdev, bm_fo, find_zero_bit, KM_IRQ1);
1231 spin_unlock_irq(&b->bm_lock);
1235 unsigned long drbd_bm_find_next(struct drbd_conf *mdev, unsigned long bm_fo)
1237 return bm_find_next(mdev, bm_fo, 0);
1241 /* not yet needed for anything. */
1242 unsigned long drbd_bm_find_next_zero(struct drbd_conf *mdev, unsigned long bm_fo)
1244 return bm_find_next(mdev, bm_fo, 1);
1248 /* does not spin_lock_irqsave.
1249 * you must take drbd_bm_lock() first */
1250 unsigned long _drbd_bm_find_next(struct drbd_conf *mdev, unsigned long bm_fo)
1252 /* WARN_ON(!(BM_DONT_SET & mdev->b->bm_flags)); */
1253 return __bm_find_next(mdev, bm_fo, 0, KM_USER1);
1256 unsigned long _drbd_bm_find_next_zero(struct drbd_conf *mdev, unsigned long bm_fo)
1258 /* WARN_ON(!(BM_DONT_SET & mdev->b->bm_flags)); */
1259 return __bm_find_next(mdev, bm_fo, 1, KM_USER1);
1262 /* returns number of bits actually changed.
1263 * for val != 0, we change 0 -> 1, return code positive
1264 * for val == 0, we change 1 -> 0, return code negative
1265 * wants bitnr, not sector.
1266 * expected to be called for only a few bits (e - s about BITS_PER_LONG).
1267 * Must hold bitmap lock already. */
1268 static int __bm_change_bits_to(struct drbd_conf *mdev, const unsigned long s,
1269 unsigned long e, int val)
1271 struct drbd_bitmap *b = mdev->bitmap;
1272 unsigned long *p_addr = NULL;
1273 unsigned long bitnr;
1274 unsigned int last_page_nr = -1U;
1276 int changed_total = 0;
1278 if (e >= b->bm_bits) {
1279 dev_err(DEV, "ASSERT FAILED: bit_s=%lu bit_e=%lu bm_bits=%lu\n",
1281 e = b->bm_bits ? b->bm_bits -1 : 0;
1283 for (bitnr = s; bitnr <= e; bitnr++) {
1284 unsigned int page_nr = bm_bit_to_page_idx(b, bitnr);
1285 if (page_nr != last_page_nr) {
1287 __bm_unmap(p_addr, KM_IRQ1);
1289 bm_set_page_lazy_writeout(b->bm_pages[last_page_nr]);
1291 bm_set_page_need_writeout(b->bm_pages[last_page_nr]);
1294 p_addr = __bm_map_pidx(b, page_nr, KM_IRQ1);
1295 last_page_nr = page_nr;
1298 c += (0 == __test_and_set_bit_le(bitnr & BITS_PER_PAGE_MASK, p_addr));
1300 c -= (0 != __test_and_clear_bit_le(bitnr & BITS_PER_PAGE_MASK, p_addr));
1303 __bm_unmap(p_addr, KM_IRQ1);
1305 bm_set_page_lazy_writeout(b->bm_pages[last_page_nr]);
1307 bm_set_page_need_writeout(b->bm_pages[last_page_nr]);
1309 b->bm_set += changed_total;
1310 return changed_total;
1313 /* returns number of bits actually changed.
1314 * for val != 0, we change 0 -> 1, return code positive
1315 * for val == 0, we change 1 -> 0, return code negative
1316 * wants bitnr, not sector */
1317 static int bm_change_bits_to(struct drbd_conf *mdev, const unsigned long s,
1318 const unsigned long e, int val)
1320 unsigned long flags;
1321 struct drbd_bitmap *b = mdev->bitmap;
1326 if (!expect(b->bm_pages))
1329 spin_lock_irqsave(&b->bm_lock, flags);
1330 if ((val ? BM_DONT_SET : BM_DONT_CLEAR) & b->bm_flags)
1331 bm_print_lock_info(mdev);
1333 c = __bm_change_bits_to(mdev, s, e, val);
1335 spin_unlock_irqrestore(&b->bm_lock, flags);
1339 /* returns number of bits changed 0 -> 1 */
1340 int drbd_bm_set_bits(struct drbd_conf *mdev, const unsigned long s, const unsigned long e)
1342 return bm_change_bits_to(mdev, s, e, 1);
1345 /* returns number of bits changed 1 -> 0 */
1346 int drbd_bm_clear_bits(struct drbd_conf *mdev, const unsigned long s, const unsigned long e)
1348 return -bm_change_bits_to(mdev, s, e, 0);
1351 /* sets all bits in full words,
1352 * from first_word up to, but not including, last_word */
1353 static inline void bm_set_full_words_within_one_page(struct drbd_bitmap *b,
1354 int page_nr, int first_word, int last_word)
1358 unsigned long *paddr = kmap_atomic(b->bm_pages[page_nr], KM_IRQ1);
1359 for (i = first_word; i < last_word; i++) {
1360 bits = hweight_long(paddr[i]);
1362 b->bm_set += BITS_PER_LONG - bits;
1364 kunmap_atomic(paddr, KM_IRQ1);
1367 /* Same thing as drbd_bm_set_bits,
1368 * but more efficient for a large bit range.
1369 * You must first drbd_bm_lock().
1370 * Can be called to set the whole bitmap in one go.
1371 * Sets bits from s to e _inclusive_. */
1372 void _drbd_bm_set_bits(struct drbd_conf *mdev, const unsigned long s, const unsigned long e)
1374 /* First set_bit from the first bit (s)
1375 * up to the next long boundary (sl),
1376 * then assign full words up to the last long boundary (el),
1377 * then set_bit up to and including the last bit (e).
1379 * Do not use memset, because we must account for changes,
1380 * so we need to loop over the words with hweight() anyways.
1382 struct drbd_bitmap *b = mdev->bitmap;
1383 unsigned long sl = ALIGN(s,BITS_PER_LONG);
1384 unsigned long el = (e+1) & ~((unsigned long)BITS_PER_LONG-1);
1391 if (e - s <= 3*BITS_PER_LONG) {
1392 /* don't bother; el and sl may even be wrong. */
1393 spin_lock_irq(&b->bm_lock);
1394 __bm_change_bits_to(mdev, s, e, 1);
1395 spin_unlock_irq(&b->bm_lock);
1399 /* difference is large enough that we can trust sl and el */
1401 spin_lock_irq(&b->bm_lock);
1403 /* bits filling the current long */
1405 __bm_change_bits_to(mdev, s, sl-1, 1);
1407 first_page = sl >> (3 + PAGE_SHIFT);
1408 last_page = el >> (3 + PAGE_SHIFT);
1410 /* MLPP: modulo longs per page */
1411 /* LWPP: long words per page */
1412 first_word = MLPP(sl >> LN2_BPL);
1415 /* first and full pages, unless first page == last page */
1416 for (page_nr = first_page; page_nr < last_page; page_nr++) {
1417 bm_set_full_words_within_one_page(mdev->bitmap, page_nr, first_word, last_word);
1418 spin_unlock_irq(&b->bm_lock);
1421 spin_lock_irq(&b->bm_lock);
1424 /* last page (respectively only page, for first page == last page) */
1425 last_word = MLPP(el >> LN2_BPL);
1426 bm_set_full_words_within_one_page(mdev->bitmap, last_page, first_word, last_word);
1428 /* possibly trailing bits.
1429 * example: (e & 63) == 63, el will be e+1.
1430 * if that even was the very last bit,
1431 * it would trigger an assert in __bm_change_bits_to()
1434 __bm_change_bits_to(mdev, el, e, 1);
1435 spin_unlock_irq(&b->bm_lock);
1438 /* returns bit state
1439 * wants bitnr, NOT sector.
1440 * inherently racy... area needs to be locked by means of {al,rs}_lru
1443 * -1 ... first out of bounds access, stop testing for bits!
1445 int drbd_bm_test_bit(struct drbd_conf *mdev, const unsigned long bitnr)
1447 unsigned long flags;
1448 struct drbd_bitmap *b = mdev->bitmap;
1449 unsigned long *p_addr;
1454 if (!expect(b->bm_pages))
1457 spin_lock_irqsave(&b->bm_lock, flags);
1458 if (BM_DONT_TEST & b->bm_flags)
1459 bm_print_lock_info(mdev);
1460 if (bitnr < b->bm_bits) {
1461 p_addr = bm_map_pidx(b, bm_bit_to_page_idx(b, bitnr));
1462 i = test_bit_le(bitnr & BITS_PER_PAGE_MASK, p_addr) ? 1 : 0;
1464 } else if (bitnr == b->bm_bits) {
1466 } else { /* (bitnr > b->bm_bits) */
1467 dev_err(DEV, "bitnr=%lu > bm_bits=%lu\n", bitnr, b->bm_bits);
1471 spin_unlock_irqrestore(&b->bm_lock, flags);
1475 /* returns number of bits set in the range [s, e] */
1476 int drbd_bm_count_bits(struct drbd_conf *mdev, const unsigned long s, const unsigned long e)
1478 unsigned long flags;
1479 struct drbd_bitmap *b = mdev->bitmap;
1480 unsigned long *p_addr = NULL;
1481 unsigned long bitnr;
1482 unsigned int page_nr = -1U;
1485 /* If this is called without a bitmap, that is a bug. But just to be
1486 * robust in case we screwed up elsewhere, in that case pretend there
1487 * was one dirty bit in the requested area, so we won't try to do a
1488 * local read there (no bitmap probably implies no disk) */
1491 if (!expect(b->bm_pages))
1494 spin_lock_irqsave(&b->bm_lock, flags);
1495 if (BM_DONT_TEST & b->bm_flags)
1496 bm_print_lock_info(mdev);
1497 for (bitnr = s; bitnr <= e; bitnr++) {
1498 unsigned int idx = bm_bit_to_page_idx(b, bitnr);
1499 if (page_nr != idx) {
1503 p_addr = bm_map_pidx(b, idx);
1505 if (expect(bitnr < b->bm_bits))
1506 c += (0 != test_bit_le(bitnr - (page_nr << (PAGE_SHIFT+3)), p_addr));
1508 dev_err(DEV, "bitnr=%lu bm_bits=%lu\n", bitnr, b->bm_bits);
1512 spin_unlock_irqrestore(&b->bm_lock, flags);
1517 /* inherently racy...
1518 * return value may be already out-of-date when this function returns.
1519 * but the general usage is that this is only use during a cstate when bits are
1520 * only cleared, not set, and typically only care for the case when the return
1521 * value is zero, or we already "locked" this "bitmap extent" by other means.
1523 * enr is bm-extent number, since we chose to name one sector (512 bytes)
1524 * worth of the bitmap a "bitmap extent".
1527 * I think since we use it like a reference count, we should use the real
1528 * reference count of some bitmap extent element from some lru instead...
1531 int drbd_bm_e_weight(struct drbd_conf *mdev, unsigned long enr)
1533 struct drbd_bitmap *b = mdev->bitmap;
1535 unsigned long flags;
1536 unsigned long *p_addr, *bm;
1540 if (!expect(b->bm_pages))
1543 spin_lock_irqsave(&b->bm_lock, flags);
1544 if (BM_DONT_TEST & b->bm_flags)
1545 bm_print_lock_info(mdev);
1548 e = min((size_t)S2W(enr+1), b->bm_words);
1550 if (s < b->bm_words) {
1552 p_addr = bm_map_pidx(b, bm_word_to_page_idx(b, s));
1553 bm = p_addr + MLPP(s);
1555 count += hweight_long(*bm++);
1558 dev_err(DEV, "start offset (%d) too large in drbd_bm_e_weight\n", s);
1560 spin_unlock_irqrestore(&b->bm_lock, flags);
1564 /* Set all bits covered by the AL-extent al_enr.
1565 * Returns number of bits changed. */
1566 unsigned long drbd_bm_ALe_set_all(struct drbd_conf *mdev, unsigned long al_enr)
1568 struct drbd_bitmap *b = mdev->bitmap;
1569 unsigned long *p_addr, *bm;
1570 unsigned long weight;
1572 int count, i, do_now;
1575 if (!expect(b->bm_pages))
1578 spin_lock_irq(&b->bm_lock);
1579 if (BM_DONT_SET & b->bm_flags)
1580 bm_print_lock_info(mdev);
1583 s = al_enr * BM_WORDS_PER_AL_EXT;
1584 e = min_t(size_t, s + BM_WORDS_PER_AL_EXT, b->bm_words);
1585 /* assert that s and e are on the same page */
1586 D_ASSERT((e-1) >> (PAGE_SHIFT - LN2_BPL + 3)
1587 == s >> (PAGE_SHIFT - LN2_BPL + 3));
1589 if (s < b->bm_words) {
1591 p_addr = bm_map_pidx(b, bm_word_to_page_idx(b, s));
1592 bm = p_addr + MLPP(s);
1594 count += hweight_long(*bm);
1599 b->bm_set += do_now*BITS_PER_LONG - count;
1600 if (e == b->bm_words)
1601 b->bm_set -= bm_clear_surplus(b);
1603 dev_err(DEV, "start offset (%lu) too large in drbd_bm_ALe_set_all\n", s);
1605 weight = b->bm_set - weight;
1606 spin_unlock_irq(&b->bm_lock);