blob: 8dc29502dc08c172ce8d693e78a1578c72bc4562 [file] [log] [blame]
Philipp Reisnerb411b362009-09-25 16:07:19 -07001/*
2 drbd_bitmap.c
3
4 This file is part of DRBD by Philipp Reisner and Lars Ellenberg.
5
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>.
9
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)
13 any later version.
14
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.
19
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.
23 */
24
25#include <linux/bitops.h>
26#include <linux/vmalloc.h>
27#include <linux/string.h>
28#include <linux/drbd.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090029#include <linux/slab.h>
Philipp Reisnerb411b362009-09-25 16:07:19 -070030#include <asm/kmap_types.h>
Stephen Rothwellf0ff1352011-03-17 15:02:51 +010031
Philipp Reisnerb411b362009-09-25 16:07:19 -070032#include "drbd_int.h"
33
Lars Ellenberg95a0f102010-12-15 08:59:09 +010034
Philipp Reisnerb411b362009-09-25 16:07:19 -070035/* OPAQUE outside this file!
36 * interface defined in drbd_int.h
37
38 * convention:
39 * function name drbd_bm_... => used elsewhere, "public".
40 * function name bm_... => internal to implementation, "private".
Lars Ellenberg4b0715f2010-12-14 15:13:04 +010041 */
Philipp Reisnerb411b362009-09-25 16:07:19 -070042
Lars Ellenberg4b0715f2010-12-14 15:13:04 +010043
44/*
45 * LIMITATIONS:
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.
61 * 1 << (35 - 12)
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.
65 *
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.
69 *
70
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
75 * seems excessive.
76 *
Bart Van Assche24c48302011-05-21 18:32:29 +020077 * We plan to reduce the amount of in-core bitmap pages by paging them in
Lars Ellenberg4b0715f2010-12-14 15:13:04 +010078 * 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.
Philipp Reisnerb411b362009-09-25 16:07:19 -070081 */
82
83/*
84 * NOTE
85 * Access to the *bm_pages is protected by bm_lock.
86 * It is safe to read the other members within the lock.
87 *
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.
92 */
93struct drbd_bitmap {
94 struct page **bm_pages;
95 spinlock_t bm_lock;
Lars Ellenberg4b0715f2010-12-14 15:13:04 +010096
97 /* see LIMITATIONS: above */
98
Philipp Reisnerb411b362009-09-25 16:07:19 -070099 unsigned long bm_set; /* nr of set bits; THINK maybe atomic_t? */
100 unsigned long bm_bits;
101 size_t bm_words;
102 size_t bm_number_of_pages;
103 sector_t bm_dev_capacity;
Thomas Gleixner8a03ae22010-01-29 20:39:07 +0000104 struct mutex bm_change; /* serializes resize operations */
Philipp Reisnerb411b362009-09-25 16:07:19 -0700105
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100106 wait_queue_head_t bm_io_wait; /* used to serialize IO of single pages */
Philipp Reisnerb411b362009-09-25 16:07:19 -0700107
Lars Ellenberg20ceb2b2011-01-21 10:56:44 +0100108 enum bm_flag bm_flags;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700109
110 /* debugging aid, in case we are still racy somewhere */
111 char *bm_why;
112 struct task_struct *bm_task;
113};
114
Philipp Reisnerb411b362009-09-25 16:07:19 -0700115#define bm_print_lock_info(m) __bm_print_lock_info(m, __func__)
116static void __bm_print_lock_info(struct drbd_conf *mdev, const char *func)
117{
118 struct drbd_bitmap *b = mdev->bitmap;
119 if (!__ratelimit(&drbd_ratelimit_state))
120 return;
121 dev_err(DEV, "FIXME %s in %s, bitmap locked for '%s' by %s\n",
Philipp Reisner392c8802011-02-09 10:33:31 +0100122 drbd_task_to_thread_name(mdev->tconn, current),
Philipp Reisnerbed879a2011-02-04 14:00:37 +0100123 func, b->bm_why ?: "?",
Philipp Reisner392c8802011-02-09 10:33:31 +0100124 drbd_task_to_thread_name(mdev->tconn, b->bm_task));
Philipp Reisnerb411b362009-09-25 16:07:19 -0700125}
126
Lars Ellenberg20ceb2b2011-01-21 10:56:44 +0100127void drbd_bm_lock(struct drbd_conf *mdev, char *why, enum bm_flag flags)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700128{
129 struct drbd_bitmap *b = mdev->bitmap;
130 int trylock_failed;
131
132 if (!b) {
133 dev_err(DEV, "FIXME no bitmap in drbd_bm_lock!?\n");
134 return;
135 }
136
Thomas Gleixner8a03ae22010-01-29 20:39:07 +0000137 trylock_failed = !mutex_trylock(&b->bm_change);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700138
139 if (trylock_failed) {
140 dev_warn(DEV, "%s going to '%s' but bitmap already locked for '%s' by %s\n",
Philipp Reisner392c8802011-02-09 10:33:31 +0100141 drbd_task_to_thread_name(mdev->tconn, current),
Philipp Reisnerbed879a2011-02-04 14:00:37 +0100142 why, b->bm_why ?: "?",
Philipp Reisner392c8802011-02-09 10:33:31 +0100143 drbd_task_to_thread_name(mdev->tconn, b->bm_task));
Thomas Gleixner8a03ae22010-01-29 20:39:07 +0000144 mutex_lock(&b->bm_change);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700145 }
Lars Ellenberg20ceb2b2011-01-21 10:56:44 +0100146 if (BM_LOCKED_MASK & b->bm_flags)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700147 dev_err(DEV, "FIXME bitmap already locked in bm_lock\n");
Lars Ellenberg20ceb2b2011-01-21 10:56:44 +0100148 b->bm_flags |= flags & BM_LOCKED_MASK;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700149
150 b->bm_why = why;
151 b->bm_task = current;
152}
153
154void drbd_bm_unlock(struct drbd_conf *mdev)
155{
156 struct drbd_bitmap *b = mdev->bitmap;
157 if (!b) {
158 dev_err(DEV, "FIXME no bitmap in drbd_bm_unlock!?\n");
159 return;
160 }
161
Lars Ellenberg20ceb2b2011-01-21 10:56:44 +0100162 if (!(BM_LOCKED_MASK & mdev->bitmap->bm_flags))
Philipp Reisnerb411b362009-09-25 16:07:19 -0700163 dev_err(DEV, "FIXME bitmap not locked in bm_unlock\n");
164
Lars Ellenberg20ceb2b2011-01-21 10:56:44 +0100165 b->bm_flags &= ~BM_LOCKED_MASK;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700166 b->bm_why = NULL;
167 b->bm_task = NULL;
Thomas Gleixner8a03ae22010-01-29 20:39:07 +0000168 mutex_unlock(&b->bm_change);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700169}
170
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100171/* 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
174 * 1<<38 bits,
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.
179 */
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
Lars Ellenberg45dfffe2011-02-21 13:21:00 +0100191/* pages marked with this "HINT" will be considered for writeout
192 * on activity log transactions */
193#define BM_PAGE_HINT_WRITEOUT 27
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100194
Bart Van Assche24c48302011-05-21 18:32:29 +0200195/* store_page_idx uses non-atomic assignment. It is only used directly after
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100196 * allocating the page. All other bm_set_page_* and bm_clear_page_* need to
197 * use atomic bit manipulation, as set_out_of_sync (and therefore bitmap
198 * changes) may happen from various contexts, and wait_on_bit/wake_up_bit
199 * requires it all to be atomic as well. */
200static void bm_store_page_idx(struct page *page, unsigned long idx)
Lars Ellenberg95a0f102010-12-15 08:59:09 +0100201{
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100202 BUG_ON(0 != (idx & ~BM_PAGE_IDX_MASK));
Arne Redlich0c7db272012-03-16 08:19:33 +0100203 set_page_private(page, idx);
Lars Ellenberg95a0f102010-12-15 08:59:09 +0100204}
205
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100206static unsigned long bm_page_to_idx(struct page *page)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700207{
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100208 return page_private(page) & BM_PAGE_IDX_MASK;
209}
Philipp Reisnerb411b362009-09-25 16:07:19 -0700210
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100211/* As is very unlikely that the same page is under IO from more than one
212 * context, we can get away with a bit per page and one wait queue per bitmap.
213 */
214static void bm_page_lock_io(struct drbd_conf *mdev, int page_nr)
215{
216 struct drbd_bitmap *b = mdev->bitmap;
217 void *addr = &page_private(b->bm_pages[page_nr]);
218 wait_event(b->bm_io_wait, !test_and_set_bit(BM_PAGE_IO_LOCK, addr));
219}
220
221static void bm_page_unlock_io(struct drbd_conf *mdev, int page_nr)
222{
223 struct drbd_bitmap *b = mdev->bitmap;
224 void *addr = &page_private(b->bm_pages[page_nr]);
Lars Ellenberg4738fa12011-02-21 13:20:55 +0100225 clear_bit_unlock(BM_PAGE_IO_LOCK, addr);
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100226 wake_up(&mdev->bitmap->bm_io_wait);
227}
228
229/* set _before_ submit_io, so it may be reset due to being changed
230 * while this page is in flight... will get submitted later again */
231static void bm_set_page_unchanged(struct page *page)
232{
233 /* use cmpxchg? */
234 clear_bit(BM_PAGE_NEED_WRITEOUT, &page_private(page));
235 clear_bit(BM_PAGE_LAZY_WRITEOUT, &page_private(page));
236}
237
238static void bm_set_page_need_writeout(struct page *page)
239{
240 set_bit(BM_PAGE_NEED_WRITEOUT, &page_private(page));
241}
242
Lars Ellenberg45dfffe2011-02-21 13:21:00 +0100243/**
244 * drbd_bm_mark_for_writeout() - mark a page with a "hint" to be considered for writeout
245 * @mdev: DRBD device.
246 * @page_nr: the bitmap page to mark with the "hint" flag
247 *
248 * From within an activity log transaction, we mark a few pages with these
249 * hints, then call drbd_bm_write_hinted(), which will only write out changed
250 * pages which are flagged with this mark.
251 */
252void drbd_bm_mark_for_writeout(struct drbd_conf *mdev, int page_nr)
253{
254 struct page *page;
255 if (page_nr >= mdev->bitmap->bm_number_of_pages) {
256 dev_warn(DEV, "BAD: page_nr: %u, number_of_pages: %u\n",
257 page_nr, (int)mdev->bitmap->bm_number_of_pages);
258 return;
259 }
260 page = mdev->bitmap->bm_pages[page_nr];
261 set_bit(BM_PAGE_HINT_WRITEOUT, &page_private(page));
262}
263
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100264static int bm_test_page_unchanged(struct page *page)
265{
266 volatile const unsigned long *addr = &page_private(page);
267 return (*addr & ((1UL<<BM_PAGE_NEED_WRITEOUT)|(1UL<<BM_PAGE_LAZY_WRITEOUT))) == 0;
268}
269
270static void bm_set_page_io_err(struct page *page)
271{
272 set_bit(BM_PAGE_IO_ERROR, &page_private(page));
273}
274
275static void bm_clear_page_io_err(struct page *page)
276{
277 clear_bit(BM_PAGE_IO_ERROR, &page_private(page));
278}
279
280static void bm_set_page_lazy_writeout(struct page *page)
281{
282 set_bit(BM_PAGE_LAZY_WRITEOUT, &page_private(page));
283}
284
285static int bm_test_page_lazy_writeout(struct page *page)
286{
287 return test_bit(BM_PAGE_LAZY_WRITEOUT, &page_private(page));
288}
289
290/* on a 32bit box, this would allow for exactly (2<<38) bits. */
291static unsigned int bm_word_to_page_idx(struct drbd_bitmap *b, unsigned long long_nr)
292{
Philipp Reisnerb411b362009-09-25 16:07:19 -0700293 /* page_nr = (word*sizeof(long)) >> PAGE_SHIFT; */
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100294 unsigned int page_nr = long_nr >> (PAGE_SHIFT - LN2_BPL + 3);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700295 BUG_ON(page_nr >= b->bm_number_of_pages);
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100296 return page_nr;
297}
Philipp Reisnerb411b362009-09-25 16:07:19 -0700298
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100299static unsigned int bm_bit_to_page_idx(struct drbd_bitmap *b, u64 bitnr)
300{
301 /* page_nr = (bitnr/8) >> PAGE_SHIFT; */
302 unsigned int page_nr = bitnr >> (PAGE_SHIFT + 3);
303 BUG_ON(page_nr >= b->bm_number_of_pages);
304 return page_nr;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700305}
306
Cong Wang589973a2011-11-25 23:51:58 +0800307static unsigned long *__bm_map_pidx(struct drbd_bitmap *b, unsigned int idx)
Lars Ellenberg95a0f102010-12-15 08:59:09 +0100308{
309 struct page *page = b->bm_pages[idx];
Cong Wangcfd80052011-11-25 23:14:18 +0800310 return (unsigned long *) kmap_atomic(page);
Lars Ellenberg95a0f102010-12-15 08:59:09 +0100311}
312
313static unsigned long *bm_map_pidx(struct drbd_bitmap *b, unsigned int idx)
314{
Cong Wangcfd80052011-11-25 23:14:18 +0800315 return __bm_map_pidx(b, idx);
Lars Ellenberg95a0f102010-12-15 08:59:09 +0100316}
317
Cong Wangcfd80052011-11-25 23:14:18 +0800318static void __bm_unmap(unsigned long *p_addr)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700319{
Cong Wangcfd80052011-11-25 23:14:18 +0800320 kunmap_atomic(p_addr);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700321};
322
323static void bm_unmap(unsigned long *p_addr)
324{
Cong Wangcfd80052011-11-25 23:14:18 +0800325 return __bm_unmap(p_addr);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700326}
327
328/* long word offset of _bitmap_ sector */
329#define S2W(s) ((s)<<(BM_EXT_SHIFT-BM_BLOCK_SHIFT-LN2_BPL))
330/* word offset from start of bitmap to word number _in_page_
331 * modulo longs per page
332#define MLPP(X) ((X) % (PAGE_SIZE/sizeof(long))
Bart Van Assche24c48302011-05-21 18:32:29 +0200333 hm, well, Philipp thinks gcc might not optimize the % into & (... - 1)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700334 so do it explicitly:
335 */
336#define MLPP(X) ((X) & ((PAGE_SIZE/sizeof(long))-1))
337
338/* Long words per page */
339#define LWPP (PAGE_SIZE/sizeof(long))
340
341/*
342 * actually most functions herein should take a struct drbd_bitmap*, not a
343 * struct drbd_conf*, but for the debug macros I like to have the mdev around
344 * to be able to report device specific.
345 */
346
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100347
Philipp Reisnerb411b362009-09-25 16:07:19 -0700348static void bm_free_pages(struct page **pages, unsigned long number)
349{
350 unsigned long i;
351 if (!pages)
352 return;
353
354 for (i = 0; i < number; i++) {
355 if (!pages[i]) {
356 printk(KERN_ALERT "drbd: bm_free_pages tried to free "
357 "a NULL pointer; i=%lu n=%lu\n",
358 i, number);
359 continue;
360 }
361 __free_page(pages[i]);
362 pages[i] = NULL;
363 }
364}
365
366static void bm_vk_free(void *ptr, int v)
367{
368 if (v)
369 vfree(ptr);
370 else
371 kfree(ptr);
372}
373
374/*
375 * "have" and "want" are NUMBER OF PAGES.
376 */
377static struct page **bm_realloc_pages(struct drbd_bitmap *b, unsigned long want)
378{
379 struct page **old_pages = b->bm_pages;
380 struct page **new_pages, *page;
381 unsigned int i, bytes, vmalloced = 0;
382 unsigned long have = b->bm_number_of_pages;
383
384 BUG_ON(have == 0 && old_pages != NULL);
385 BUG_ON(have != 0 && old_pages == NULL);
386
387 if (have == want)
388 return old_pages;
389
390 /* Trying kmalloc first, falling back to vmalloc.
Lars Ellenberg0b143d42012-09-20 14:05:39 +0200391 * GFP_NOIO, as this is called while drbd IO is "suspended",
392 * and during resize or attach on diskless Primary,
393 * we must not block on IO to ourselves.
Lars Ellenbergbc891c92012-09-26 14:18:51 +0200394 * Context is receiver thread or dmsetup. */
Philipp Reisnerb411b362009-09-25 16:07:19 -0700395 bytes = sizeof(struct page *)*want;
Lars Ellenberg0b143d42012-09-20 14:05:39 +0200396 new_pages = kzalloc(bytes, GFP_NOIO);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700397 if (!new_pages) {
Lars Ellenberg0b143d42012-09-20 14:05:39 +0200398 new_pages = __vmalloc(bytes,
399 GFP_NOIO | __GFP_HIGHMEM | __GFP_ZERO,
400 PAGE_KERNEL);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700401 if (!new_pages)
402 return NULL;
403 vmalloced = 1;
404 }
405
Philipp Reisnerb411b362009-09-25 16:07:19 -0700406 if (want >= have) {
407 for (i = 0; i < have; i++)
408 new_pages[i] = old_pages[i];
409 for (; i < want; i++) {
Lars Ellenberg0b143d42012-09-20 14:05:39 +0200410 page = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700411 if (!page) {
412 bm_free_pages(new_pages + have, i - have);
413 bm_vk_free(new_pages, vmalloced);
414 return NULL;
415 }
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100416 /* we want to know which page it is
417 * from the endio handlers */
418 bm_store_page_idx(page, i);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700419 new_pages[i] = page;
420 }
421 } else {
422 for (i = 0; i < want; i++)
423 new_pages[i] = old_pages[i];
424 /* NOT HERE, we are outside the spinlock!
425 bm_free_pages(old_pages + want, have - want);
426 */
427 }
428
429 if (vmalloced)
Lars Ellenberg20ceb2b2011-01-21 10:56:44 +0100430 b->bm_flags |= BM_P_VMALLOCED;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700431 else
Lars Ellenberg20ceb2b2011-01-21 10:56:44 +0100432 b->bm_flags &= ~BM_P_VMALLOCED;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700433
434 return new_pages;
435}
436
437/*
438 * called on driver init only. TODO call when a device is created.
439 * allocates the drbd_bitmap, and stores it in mdev->bitmap.
440 */
441int drbd_bm_init(struct drbd_conf *mdev)
442{
443 struct drbd_bitmap *b = mdev->bitmap;
444 WARN_ON(b != NULL);
445 b = kzalloc(sizeof(struct drbd_bitmap), GFP_KERNEL);
446 if (!b)
447 return -ENOMEM;
448 spin_lock_init(&b->bm_lock);
Thomas Gleixner8a03ae22010-01-29 20:39:07 +0000449 mutex_init(&b->bm_change);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700450 init_waitqueue_head(&b->bm_io_wait);
451
452 mdev->bitmap = b;
453
454 return 0;
455}
456
457sector_t drbd_bm_capacity(struct drbd_conf *mdev)
458{
Andreas Gruenbacher841ce242010-12-15 19:31:20 +0100459 if (!expect(mdev->bitmap))
460 return 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700461 return mdev->bitmap->bm_dev_capacity;
462}
463
464/* called on driver unload. TODO: call when a device is destroyed.
465 */
466void drbd_bm_cleanup(struct drbd_conf *mdev)
467{
Andreas Gruenbacher841ce242010-12-15 19:31:20 +0100468 if (!expect(mdev->bitmap))
469 return;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700470 bm_free_pages(mdev->bitmap->bm_pages, mdev->bitmap->bm_number_of_pages);
Lars Ellenberg20ceb2b2011-01-21 10:56:44 +0100471 bm_vk_free(mdev->bitmap->bm_pages, (BM_P_VMALLOCED & mdev->bitmap->bm_flags));
Philipp Reisnerb411b362009-09-25 16:07:19 -0700472 kfree(mdev->bitmap);
473 mdev->bitmap = NULL;
474}
475
476/*
477 * since (b->bm_bits % BITS_PER_LONG) != 0,
478 * this masks out the remaining bits.
479 * Returns the number of bits cleared.
480 */
Lars Ellenberg95a0f102010-12-15 08:59:09 +0100481#define BITS_PER_PAGE (1UL << (PAGE_SHIFT + 3))
482#define BITS_PER_PAGE_MASK (BITS_PER_PAGE - 1)
483#define BITS_PER_LONG_MASK (BITS_PER_LONG - 1)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700484static int bm_clear_surplus(struct drbd_bitmap *b)
485{
Lars Ellenberg95a0f102010-12-15 08:59:09 +0100486 unsigned long mask;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700487 unsigned long *p_addr, *bm;
Lars Ellenberg95a0f102010-12-15 08:59:09 +0100488 int tmp;
489 int cleared = 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700490
Lars Ellenberg95a0f102010-12-15 08:59:09 +0100491 /* number of bits modulo bits per page */
492 tmp = (b->bm_bits & BITS_PER_PAGE_MASK);
493 /* mask the used bits of the word containing the last bit */
494 mask = (1UL << (tmp & BITS_PER_LONG_MASK)) -1;
495 /* bitmap is always stored little endian,
496 * on disk and in core memory alike */
497 mask = cpu_to_lel(mask);
498
Lars Ellenberg6850c442010-12-16 00:32:38 +0100499 p_addr = bm_map_pidx(b, b->bm_number_of_pages - 1);
Lars Ellenberg95a0f102010-12-15 08:59:09 +0100500 bm = p_addr + (tmp/BITS_PER_LONG);
501 if (mask) {
502 /* If mask != 0, we are not exactly aligned, so bm now points
503 * to the long containing the last bit.
504 * If mask == 0, bm already points to the word immediately
505 * after the last (long word aligned) bit. */
Philipp Reisnerb411b362009-09-25 16:07:19 -0700506 cleared = hweight_long(*bm & ~mask);
507 *bm &= mask;
Lars Ellenberg95a0f102010-12-15 08:59:09 +0100508 bm++;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700509 }
510
Lars Ellenberg95a0f102010-12-15 08:59:09 +0100511 if (BITS_PER_LONG == 32 && ((bm - p_addr) & 1) == 1) {
512 /* on a 32bit arch, we may need to zero out
513 * a padding long to align with a 64bit remote */
Philipp Reisnerb411b362009-09-25 16:07:19 -0700514 cleared += hweight_long(*bm);
515 *bm = 0;
516 }
517 bm_unmap(p_addr);
518 return cleared;
519}
520
521static void bm_set_surplus(struct drbd_bitmap *b)
522{
Lars Ellenberg95a0f102010-12-15 08:59:09 +0100523 unsigned long mask;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700524 unsigned long *p_addr, *bm;
Lars Ellenberg95a0f102010-12-15 08:59:09 +0100525 int tmp;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700526
Lars Ellenberg95a0f102010-12-15 08:59:09 +0100527 /* number of bits modulo bits per page */
528 tmp = (b->bm_bits & BITS_PER_PAGE_MASK);
529 /* mask the used bits of the word containing the last bit */
530 mask = (1UL << (tmp & BITS_PER_LONG_MASK)) -1;
531 /* bitmap is always stored little endian,
532 * on disk and in core memory alike */
533 mask = cpu_to_lel(mask);
534
Lars Ellenberg6850c442010-12-16 00:32:38 +0100535 p_addr = bm_map_pidx(b, b->bm_number_of_pages - 1);
Lars Ellenberg95a0f102010-12-15 08:59:09 +0100536 bm = p_addr + (tmp/BITS_PER_LONG);
537 if (mask) {
538 /* If mask != 0, we are not exactly aligned, so bm now points
539 * to the long containing the last bit.
540 * If mask == 0, bm already points to the word immediately
541 * after the last (long word aligned) bit. */
Philipp Reisnerb411b362009-09-25 16:07:19 -0700542 *bm |= ~mask;
Lars Ellenberg95a0f102010-12-15 08:59:09 +0100543 bm++;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700544 }
545
Lars Ellenberg95a0f102010-12-15 08:59:09 +0100546 if (BITS_PER_LONG == 32 && ((bm - p_addr) & 1) == 1) {
547 /* on a 32bit arch, we may need to zero out
548 * a padding long to align with a 64bit remote */
549 *bm = ~0UL;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700550 }
551 bm_unmap(p_addr);
552}
553
Lars Ellenberg4b0715f2010-12-14 15:13:04 +0100554/* you better not modify the bitmap while this is running,
555 * or its results will be stale */
Lars Ellenberg95a0f102010-12-15 08:59:09 +0100556static unsigned long bm_count_bits(struct drbd_bitmap *b)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700557{
Lars Ellenberg4b0715f2010-12-14 15:13:04 +0100558 unsigned long *p_addr;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700559 unsigned long bits = 0;
Lars Ellenberg4b0715f2010-12-14 15:13:04 +0100560 unsigned long mask = (1UL << (b->bm_bits & BITS_PER_LONG_MASK)) -1;
Lars Ellenberg6850c442010-12-16 00:32:38 +0100561 int idx, i, last_word;
Lars Ellenberg7777a8b2010-12-15 23:21:39 +0100562
Lars Ellenberg4b0715f2010-12-14 15:13:04 +0100563 /* all but last page */
Lars Ellenberg6850c442010-12-16 00:32:38 +0100564 for (idx = 0; idx < b->bm_number_of_pages - 1; idx++) {
Cong Wangcfd80052011-11-25 23:14:18 +0800565 p_addr = __bm_map_pidx(b, idx);
Lars Ellenberg4b0715f2010-12-14 15:13:04 +0100566 for (i = 0; i < LWPP; i++)
567 bits += hweight_long(p_addr[i]);
Cong Wangcfd80052011-11-25 23:14:18 +0800568 __bm_unmap(p_addr);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700569 cond_resched();
570 }
Lars Ellenberg4b0715f2010-12-14 15:13:04 +0100571 /* last (or only) page */
572 last_word = ((b->bm_bits - 1) & BITS_PER_PAGE_MASK) >> LN2_BPL;
Cong Wang589973a2011-11-25 23:51:58 +0800573 p_addr = __bm_map_pidx(b, idx);
Lars Ellenberg4b0715f2010-12-14 15:13:04 +0100574 for (i = 0; i < last_word; i++)
575 bits += hweight_long(p_addr[i]);
576 p_addr[last_word] &= cpu_to_lel(mask);
577 bits += hweight_long(p_addr[last_word]);
578 /* 32bit arch, may have an unused padding long */
579 if (BITS_PER_LONG == 32 && (last_word & 1) == 0)
580 p_addr[last_word+1] = 0;
Cong Wang589973a2011-11-25 23:51:58 +0800581 __bm_unmap(p_addr);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700582 return bits;
583}
584
Philipp Reisnerb411b362009-09-25 16:07:19 -0700585/* offset and len in long words.*/
586static void bm_memset(struct drbd_bitmap *b, size_t offset, int c, size_t len)
587{
588 unsigned long *p_addr, *bm;
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100589 unsigned int idx;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700590 size_t do_now, end;
591
Philipp Reisnerb411b362009-09-25 16:07:19 -0700592 end = offset + len;
593
594 if (end > b->bm_words) {
595 printk(KERN_ALERT "drbd: bm_memset end > bm_words\n");
596 return;
597 }
598
599 while (offset < end) {
600 do_now = min_t(size_t, ALIGN(offset + 1, LWPP), end) - offset;
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100601 idx = bm_word_to_page_idx(b, offset);
602 p_addr = bm_map_pidx(b, idx);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700603 bm = p_addr + MLPP(offset);
604 if (bm+do_now > p_addr + LWPP) {
605 printk(KERN_ALERT "drbd: BUG BUG BUG! p_addr:%p bm:%p do_now:%d\n",
606 p_addr, bm, (int)do_now);
Lars Ellenberg84e7c0f2010-12-16 00:37:57 +0100607 } else
608 memset(bm, c, do_now * sizeof(long));
Philipp Reisnerb411b362009-09-25 16:07:19 -0700609 bm_unmap(p_addr);
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100610 bm_set_page_need_writeout(b->bm_pages[idx]);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700611 offset += do_now;
612 }
613}
614
615/*
616 * make sure the bitmap has enough room for the attached storage,
617 * if necessary, resize.
618 * called whenever we may have changed the device size.
619 * returns -ENOMEM if we could not allocate enough memory, 0 on success.
620 * In case this is actually a resize, we copy the old bitmap into the new one.
621 * Otherwise, the bitmap is initialized to all bits set.
622 */
Philipp Reisner02d9a942010-03-24 16:23:03 +0100623int drbd_bm_resize(struct drbd_conf *mdev, sector_t capacity, int set_new_bits)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700624{
625 struct drbd_bitmap *b = mdev->bitmap;
Lars Ellenberg6850c442010-12-16 00:32:38 +0100626 unsigned long bits, words, owords, obits;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700627 unsigned long want, have, onpages; /* number of pages */
628 struct page **npages, **opages = NULL;
629 int err = 0, growing;
630 int opages_vmalloced;
631
Andreas Gruenbacher841ce242010-12-15 19:31:20 +0100632 if (!expect(b))
633 return -ENOMEM;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700634
Lars Ellenberg20ceb2b2011-01-21 10:56:44 +0100635 drbd_bm_lock(mdev, "resize", BM_LOCKED_MASK);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700636
637 dev_info(DEV, "drbd_bm_resize called with capacity == %llu\n",
638 (unsigned long long)capacity);
639
640 if (capacity == b->bm_dev_capacity)
641 goto out;
642
Lars Ellenberg20ceb2b2011-01-21 10:56:44 +0100643 opages_vmalloced = (BM_P_VMALLOCED & b->bm_flags);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700644
645 if (capacity == 0) {
646 spin_lock_irq(&b->bm_lock);
647 opages = b->bm_pages;
648 onpages = b->bm_number_of_pages;
649 owords = b->bm_words;
650 b->bm_pages = NULL;
651 b->bm_number_of_pages =
652 b->bm_set =
653 b->bm_bits =
654 b->bm_words =
655 b->bm_dev_capacity = 0;
656 spin_unlock_irq(&b->bm_lock);
657 bm_free_pages(opages, onpages);
658 bm_vk_free(opages, opages_vmalloced);
659 goto out;
660 }
661 bits = BM_SECT_TO_BIT(ALIGN(capacity, BM_SECT_PER_BIT));
662
663 /* if we would use
664 words = ALIGN(bits,BITS_PER_LONG) >> LN2_BPL;
665 a 32bit host could present the wrong number of words
666 to a 64bit host.
667 */
668 words = ALIGN(bits, 64) >> LN2_BPL;
669
670 if (get_ldev(mdev)) {
Lars Ellenberg4b0715f2010-12-14 15:13:04 +0100671 u64 bits_on_disk = ((u64)mdev->ldev->md.md_size_sect-MD_BM_OFFSET) << 12;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700672 put_ldev(mdev);
Lars Ellenberg4b0715f2010-12-14 15:13:04 +0100673 if (bits > bits_on_disk) {
674 dev_info(DEV, "bits = %lu\n", bits);
675 dev_info(DEV, "bits_on_disk = %llu\n", bits_on_disk);
676 err = -ENOSPC;
677 goto out;
678 }
Philipp Reisnerb411b362009-09-25 16:07:19 -0700679 }
680
Lars Ellenberg6850c442010-12-16 00:32:38 +0100681 want = ALIGN(words*sizeof(long), PAGE_SIZE) >> PAGE_SHIFT;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700682 have = b->bm_number_of_pages;
683 if (want == have) {
684 D_ASSERT(b->bm_pages != NULL);
685 npages = b->bm_pages;
686 } else {
Andreas Gruenbacher0cf9d272010-12-07 10:43:29 +0100687 if (drbd_insert_fault(mdev, DRBD_FAULT_BM_ALLOC))
Philipp Reisnerb411b362009-09-25 16:07:19 -0700688 npages = NULL;
689 else
690 npages = bm_realloc_pages(b, want);
691 }
692
693 if (!npages) {
694 err = -ENOMEM;
695 goto out;
696 }
697
698 spin_lock_irq(&b->bm_lock);
699 opages = b->bm_pages;
700 owords = b->bm_words;
701 obits = b->bm_bits;
702
703 growing = bits > obits;
Philipp Reisner52236712010-04-28 14:46:57 +0200704 if (opages && growing && set_new_bits)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700705 bm_set_surplus(b);
706
707 b->bm_pages = npages;
708 b->bm_number_of_pages = want;
709 b->bm_bits = bits;
710 b->bm_words = words;
711 b->bm_dev_capacity = capacity;
712
713 if (growing) {
Philipp Reisner02d9a942010-03-24 16:23:03 +0100714 if (set_new_bits) {
715 bm_memset(b, owords, 0xff, words-owords);
716 b->bm_set += bits - obits;
717 } else
718 bm_memset(b, owords, 0x00, words-owords);
719
Philipp Reisnerb411b362009-09-25 16:07:19 -0700720 }
721
722 if (want < have) {
723 /* implicit: (opages != NULL) && (opages != npages) */
724 bm_free_pages(opages + want, have - want);
725 }
726
Philipp Reisnerb411b362009-09-25 16:07:19 -0700727 (void)bm_clear_surplus(b);
728
729 spin_unlock_irq(&b->bm_lock);
730 if (opages != npages)
731 bm_vk_free(opages, opages_vmalloced);
732 if (!growing)
733 b->bm_set = bm_count_bits(b);
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100734 dev_info(DEV, "resync bitmap: bits=%lu words=%lu pages=%lu\n", bits, words, want);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700735
736 out:
737 drbd_bm_unlock(mdev);
738 return err;
739}
740
741/* inherently racy:
742 * if not protected by other means, return value may be out of date when
743 * leaving this function...
744 * we still need to lock it, since it is important that this returns
745 * bm_set == 0 precisely.
746 *
747 * maybe bm_set should be atomic_t ?
748 */
Philipp Reisner07782862010-08-31 12:00:50 +0200749unsigned long _drbd_bm_total_weight(struct drbd_conf *mdev)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700750{
751 struct drbd_bitmap *b = mdev->bitmap;
752 unsigned long s;
753 unsigned long flags;
754
Andreas Gruenbacher841ce242010-12-15 19:31:20 +0100755 if (!expect(b))
756 return 0;
757 if (!expect(b->bm_pages))
758 return 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700759
760 spin_lock_irqsave(&b->bm_lock, flags);
761 s = b->bm_set;
762 spin_unlock_irqrestore(&b->bm_lock, flags);
763
764 return s;
765}
766
767unsigned long drbd_bm_total_weight(struct drbd_conf *mdev)
768{
769 unsigned long s;
770 /* if I don't have a disk, I don't know about out-of-sync status */
771 if (!get_ldev_if_state(mdev, D_NEGOTIATING))
772 return 0;
773 s = _drbd_bm_total_weight(mdev);
774 put_ldev(mdev);
775 return s;
776}
777
778size_t drbd_bm_words(struct drbd_conf *mdev)
779{
780 struct drbd_bitmap *b = mdev->bitmap;
Andreas Gruenbacher841ce242010-12-15 19:31:20 +0100781 if (!expect(b))
782 return 0;
783 if (!expect(b->bm_pages))
784 return 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700785
786 return b->bm_words;
787}
788
789unsigned long drbd_bm_bits(struct drbd_conf *mdev)
790{
791 struct drbd_bitmap *b = mdev->bitmap;
Andreas Gruenbacher841ce242010-12-15 19:31:20 +0100792 if (!expect(b))
793 return 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700794
795 return b->bm_bits;
796}
797
798/* merge number words from buffer into the bitmap starting at offset.
799 * buffer[i] is expected to be little endian unsigned long.
800 * bitmap must be locked by drbd_bm_lock.
801 * currently only used from receive_bitmap.
802 */
803void drbd_bm_merge_lel(struct drbd_conf *mdev, size_t offset, size_t number,
804 unsigned long *buffer)
805{
806 struct drbd_bitmap *b = mdev->bitmap;
807 unsigned long *p_addr, *bm;
808 unsigned long word, bits;
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100809 unsigned int idx;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700810 size_t end, do_now;
811
812 end = offset + number;
813
Andreas Gruenbacher841ce242010-12-15 19:31:20 +0100814 if (!expect(b))
815 return;
816 if (!expect(b->bm_pages))
817 return;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700818 if (number == 0)
819 return;
820 WARN_ON(offset >= b->bm_words);
821 WARN_ON(end > b->bm_words);
822
823 spin_lock_irq(&b->bm_lock);
824 while (offset < end) {
825 do_now = min_t(size_t, ALIGN(offset+1, LWPP), end) - offset;
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100826 idx = bm_word_to_page_idx(b, offset);
827 p_addr = bm_map_pidx(b, idx);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700828 bm = p_addr + MLPP(offset);
829 offset += do_now;
830 while (do_now--) {
831 bits = hweight_long(*bm);
Lars Ellenberg95a0f102010-12-15 08:59:09 +0100832 word = *bm | *buffer++;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700833 *bm++ = word;
834 b->bm_set += hweight_long(word) - bits;
835 }
836 bm_unmap(p_addr);
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100837 bm_set_page_need_writeout(b->bm_pages[idx]);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700838 }
839 /* with 32bit <-> 64bit cross-platform connect
840 * this is only correct for current usage,
841 * where we _know_ that we are 64 bit aligned,
842 * and know that this function is used in this way, too...
843 */
844 if (end == b->bm_words)
845 b->bm_set -= bm_clear_surplus(b);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700846 spin_unlock_irq(&b->bm_lock);
847}
848
849/* copy number words from the bitmap starting at offset into the buffer.
850 * buffer[i] will be little endian unsigned long.
851 */
852void drbd_bm_get_lel(struct drbd_conf *mdev, size_t offset, size_t number,
853 unsigned long *buffer)
854{
855 struct drbd_bitmap *b = mdev->bitmap;
856 unsigned long *p_addr, *bm;
857 size_t end, do_now;
858
859 end = offset + number;
860
Andreas Gruenbacher841ce242010-12-15 19:31:20 +0100861 if (!expect(b))
862 return;
863 if (!expect(b->bm_pages))
864 return;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700865
866 spin_lock_irq(&b->bm_lock);
867 if ((offset >= b->bm_words) ||
868 (end > b->bm_words) ||
869 (number <= 0))
870 dev_err(DEV, "offset=%lu number=%lu bm_words=%lu\n",
871 (unsigned long) offset,
872 (unsigned long) number,
873 (unsigned long) b->bm_words);
874 else {
875 while (offset < end) {
876 do_now = min_t(size_t, ALIGN(offset+1, LWPP), end) - offset;
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100877 p_addr = bm_map_pidx(b, bm_word_to_page_idx(b, offset));
Philipp Reisnerb411b362009-09-25 16:07:19 -0700878 bm = p_addr + MLPP(offset);
879 offset += do_now;
880 while (do_now--)
Lars Ellenberg95a0f102010-12-15 08:59:09 +0100881 *buffer++ = *bm++;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700882 bm_unmap(p_addr);
883 }
884 }
885 spin_unlock_irq(&b->bm_lock);
886}
887
888/* set all bits in the bitmap */
889void drbd_bm_set_all(struct drbd_conf *mdev)
890{
891 struct drbd_bitmap *b = mdev->bitmap;
Andreas Gruenbacher841ce242010-12-15 19:31:20 +0100892 if (!expect(b))
893 return;
894 if (!expect(b->bm_pages))
895 return;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700896
897 spin_lock_irq(&b->bm_lock);
898 bm_memset(b, 0, 0xff, b->bm_words);
899 (void)bm_clear_surplus(b);
900 b->bm_set = b->bm_bits;
901 spin_unlock_irq(&b->bm_lock);
902}
903
904/* clear all bits in the bitmap */
905void drbd_bm_clear_all(struct drbd_conf *mdev)
906{
907 struct drbd_bitmap *b = mdev->bitmap;
Andreas Gruenbacher841ce242010-12-15 19:31:20 +0100908 if (!expect(b))
909 return;
910 if (!expect(b->bm_pages))
911 return;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700912
913 spin_lock_irq(&b->bm_lock);
914 bm_memset(b, 0, 0, b->bm_words);
915 b->bm_set = 0;
916 spin_unlock_irq(&b->bm_lock);
917}
918
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100919struct bm_aio_ctx {
920 struct drbd_conf *mdev;
921 atomic_t in_flight;
Philipp Reisner9e58c4d2011-06-27 15:29:16 +0200922 unsigned int done;
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100923 unsigned flags;
924#define BM_AIO_COPY_PAGES 1
Lars Ellenberg45dfffe2011-02-21 13:21:00 +0100925#define BM_AIO_WRITE_HINTED 2
Philipp Reisnerfef45d22012-08-14 11:46:59 +0200926#define BM_WRITE_ALL_PAGES 4
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100927 int error;
Philipp Reisnerd1f37792011-06-27 14:56:52 +0200928 struct kref kref;
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100929};
930
Philipp Reisnerd1f37792011-06-27 14:56:52 +0200931static void bm_aio_ctx_destroy(struct kref *kref)
932{
933 struct bm_aio_ctx *ctx = container_of(kref, struct bm_aio_ctx, kref);
934
Philipp Reisner9e58c4d2011-06-27 15:29:16 +0200935 put_ldev(ctx->mdev);
Philipp Reisnerd1f37792011-06-27 14:56:52 +0200936 kfree(ctx);
937}
938
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100939/* bv_page may be a copy, or may be the original */
Philipp Reisnerb411b362009-09-25 16:07:19 -0700940static void bm_async_io_complete(struct bio *bio, int error)
941{
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100942 struct bm_aio_ctx *ctx = bio->bi_private;
943 struct drbd_conf *mdev = ctx->mdev;
944 struct drbd_bitmap *b = mdev->bitmap;
945 unsigned int idx = bm_page_to_idx(bio->bi_io_vec[0].bv_page);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700946 int uptodate = bio_flagged(bio, BIO_UPTODATE);
947
948
949 /* strange behavior of some lower level drivers...
950 * fail the request by clearing the uptodate flag,
951 * but do not return any error?!
952 * do we want to WARN() on this? */
953 if (!error && !uptodate)
954 error = -EIO;
955
Lars Ellenberg7648cdf2010-12-17 23:58:41 +0100956 if ((ctx->flags & BM_AIO_COPY_PAGES) == 0 &&
957 !bm_test_page_unchanged(b->bm_pages[idx]))
958 dev_warn(DEV, "bitmap page idx %u changed during IO!\n", idx);
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100959
Philipp Reisnerb411b362009-09-25 16:07:19 -0700960 if (error) {
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100961 /* ctx error will hold the completed-last non-zero error code,
962 * in case error codes differ. */
963 ctx->error = error;
964 bm_set_page_io_err(b->bm_pages[idx]);
965 /* Not identical to on disk version of it.
966 * Is BM_PAGE_IO_ERROR enough? */
967 if (__ratelimit(&drbd_ratelimit_state))
968 dev_err(DEV, "IO ERROR %d on bitmap page idx %u\n",
969 error, idx);
970 } else {
971 bm_clear_page_io_err(b->bm_pages[idx]);
972 dynamic_dev_dbg(DEV, "bitmap page idx %u completed\n", idx);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700973 }
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100974
975 bm_page_unlock_io(mdev, idx);
976
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100977 if (ctx->flags & BM_AIO_COPY_PAGES)
Lars Ellenberg4d95a102011-02-23 15:38:47 +0100978 mempool_free(bio->bi_io_vec[0].bv_page, drbd_md_io_page_pool);
Philipp Reisnerb411b362009-09-25 16:07:19 -0700979
980 bio_put(bio);
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100981
Philipp Reisnerd1f37792011-06-27 14:56:52 +0200982 if (atomic_dec_and_test(&ctx->in_flight)) {
Philipp Reisner9e58c4d2011-06-27 15:29:16 +0200983 ctx->done = 1;
984 wake_up(&mdev->misc_wait);
Philipp Reisnerd1f37792011-06-27 14:56:52 +0200985 kref_put(&ctx->kref, &bm_aio_ctx_destroy);
986 }
Philipp Reisnerb411b362009-09-25 16:07:19 -0700987}
988
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100989static void bm_page_io_async(struct bm_aio_ctx *ctx, int page_nr, int rw) __must_hold(local)
Philipp Reisnerb411b362009-09-25 16:07:19 -0700990{
Lars Ellenberg9476f392011-02-23 17:02:01 +0100991 struct bio *bio = bio_alloc_drbd(GFP_NOIO);
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100992 struct drbd_conf *mdev = ctx->mdev;
993 struct drbd_bitmap *b = mdev->bitmap;
994 struct page *page;
Philipp Reisnerb411b362009-09-25 16:07:19 -0700995 unsigned int len;
Lars Ellenberg19f843a2010-12-15 08:59:11 +0100996
Philipp Reisnerb411b362009-09-25 16:07:19 -0700997 sector_t on_disk_sector =
998 mdev->ldev->md.md_offset + mdev->ldev->md.bm_offset;
999 on_disk_sector += ((sector_t)page_nr) << (PAGE_SHIFT-9);
1000
1001 /* this might happen with very small
Lars Ellenberg19f843a2010-12-15 08:59:11 +01001002 * flexible external meta data device,
1003 * or with PAGE_SIZE > 4k */
Philipp Reisnerb411b362009-09-25 16:07:19 -07001004 len = min_t(unsigned int, PAGE_SIZE,
1005 (drbd_md_last_sector(mdev->ldev) - on_disk_sector + 1)<<9);
1006
Lars Ellenberg19f843a2010-12-15 08:59:11 +01001007 /* serialize IO on this page */
1008 bm_page_lock_io(mdev, page_nr);
1009 /* before memcpy and submit,
1010 * so it can be redirtied any time */
1011 bm_set_page_unchanged(b->bm_pages[page_nr]);
1012
1013 if (ctx->flags & BM_AIO_COPY_PAGES) {
Lars Ellenberg4d95a102011-02-23 15:38:47 +01001014 page = mempool_alloc(drbd_md_io_page_pool, __GFP_HIGHMEM|__GFP_WAIT);
Akinobu Mitaf1d6a322012-11-08 16:12:31 -08001015 copy_highpage(page, b->bm_pages[page_nr]);
Lars Ellenberg19f843a2010-12-15 08:59:11 +01001016 bm_store_page_idx(page, page_nr);
1017 } else
1018 page = b->bm_pages[page_nr];
Philipp Reisnerb411b362009-09-25 16:07:19 -07001019 bio->bi_bdev = mdev->ldev->md_bdev;
1020 bio->bi_sector = on_disk_sector;
Lars Ellenberg4d95a102011-02-23 15:38:47 +01001021 /* bio_add_page of a single page to an empty bio will always succeed,
1022 * according to api. Do we want to assert that? */
Lars Ellenberg19f843a2010-12-15 08:59:11 +01001023 bio_add_page(bio, page, len, 0);
1024 bio->bi_private = ctx;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001025 bio->bi_end_io = bm_async_io_complete;
1026
Andreas Gruenbacher0cf9d272010-12-07 10:43:29 +01001027 if (drbd_insert_fault(mdev, (rw & WRITE) ? DRBD_FAULT_MD_WR : DRBD_FAULT_MD_RD)) {
Philipp Reisnerb411b362009-09-25 16:07:19 -07001028 bio->bi_rw |= rw;
1029 bio_endio(bio, -EIO);
1030 } else {
1031 submit_bio(rw, bio);
Lars Ellenberg5a8b4242011-06-14 14:18:23 +02001032 /* this should not count as user activity and cause the
1033 * resync to throttle -- see drbd_rs_should_slow_down(). */
1034 atomic_add(len >> 9, &mdev->rs_sect_ev);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001035 }
1036}
1037
Philipp Reisnerb411b362009-09-25 16:07:19 -07001038/*
1039 * bm_rw: read/write the whole bitmap from/to its on disk location.
1040 */
Lars Ellenberg0e8488a2012-04-25 23:06:45 +02001041static int bm_rw(struct drbd_conf *mdev, int rw, unsigned flags, unsigned lazy_writeout_upper_idx) __must_hold(local)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001042{
Philipp Reisnerd1f37792011-06-27 14:56:52 +02001043 struct bm_aio_ctx *ctx;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001044 struct drbd_bitmap *b = mdev->bitmap;
Lars Ellenberg6850c442010-12-16 00:32:38 +01001045 int num_pages, i, count = 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001046 unsigned long now;
1047 char ppb[10];
1048 int err = 0;
1049
Lars Ellenberg19f843a2010-12-15 08:59:11 +01001050 /*
1051 * We are protected against bitmap disappearing/resizing by holding an
1052 * ldev reference (caller must have called get_ldev()).
1053 * For read/write, we are protected against changes to the bitmap by
1054 * the bitmap lock (see drbd_bitmap_io).
1055 * For lazy writeout, we don't care for ongoing changes to the bitmap,
1056 * as we submit copies of pages anyways.
1057 */
Philipp Reisnerd1f37792011-06-27 14:56:52 +02001058
Lars Ellenberg22f46ce2011-07-11 17:32:26 +02001059 ctx = kmalloc(sizeof(struct bm_aio_ctx), GFP_NOIO);
Philipp Reisnerd1f37792011-06-27 14:56:52 +02001060 if (!ctx)
1061 return -ENOMEM;
1062
1063 *ctx = (struct bm_aio_ctx) {
1064 .mdev = mdev,
1065 .in_flight = ATOMIC_INIT(1),
Philipp Reisner9e58c4d2011-06-27 15:29:16 +02001066 .done = 0,
Lars Ellenberg0e8488a2012-04-25 23:06:45 +02001067 .flags = flags,
Philipp Reisnerd1f37792011-06-27 14:56:52 +02001068 .error = 0,
1069 .kref = { ATOMIC_INIT(2) },
1070 };
1071
Philipp Reisner9e58c4d2011-06-27 15:29:16 +02001072 if (!get_ldev_if_state(mdev, D_ATTACHING)) { /* put is in bm_aio_ctx_destroy() */
1073 dev_err(DEV, "ASSERT FAILED: get_ldev_if_state() == 1 in bm_rw()\n");
1074 kfree(ctx);
1075 return -ENODEV;
1076 }
1077
Philipp Reisnerd1f37792011-06-27 14:56:52 +02001078 if (!ctx->flags)
Lars Ellenberg20ceb2b2011-01-21 10:56:44 +01001079 WARN_ON(!(BM_LOCKED_MASK & b->bm_flags));
Philipp Reisnerb411b362009-09-25 16:07:19 -07001080
Lars Ellenberg6850c442010-12-16 00:32:38 +01001081 num_pages = b->bm_number_of_pages;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001082
Philipp Reisnerb411b362009-09-25 16:07:19 -07001083 now = jiffies;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001084
1085 /* let the layers below us try to merge these bios... */
Lars Ellenberg6850c442010-12-16 00:32:38 +01001086 for (i = 0; i < num_pages; i++) {
Lars Ellenberg19f843a2010-12-15 08:59:11 +01001087 /* ignore completely unchanged pages */
1088 if (lazy_writeout_upper_idx && i == lazy_writeout_upper_idx)
1089 break;
1090 if (rw & WRITE) {
Lars Ellenberg45dfffe2011-02-21 13:21:00 +01001091 if ((flags & BM_AIO_WRITE_HINTED) &&
1092 !test_and_clear_bit(BM_PAGE_HINT_WRITEOUT,
1093 &page_private(b->bm_pages[i])))
1094 continue;
Philipp Reisnerfef45d22012-08-14 11:46:59 +02001095
Philipp Reisnerd1aa4d02012-08-08 21:19:09 +02001096 if (!(flags & BM_WRITE_ALL_PAGES) &&
1097 bm_test_page_unchanged(b->bm_pages[i])) {
Lars Ellenberg19f843a2010-12-15 08:59:11 +01001098 dynamic_dev_dbg(DEV, "skipped bm write for idx %u\n", i);
1099 continue;
1100 }
1101 /* during lazy writeout,
1102 * ignore those pages not marked for lazy writeout. */
1103 if (lazy_writeout_upper_idx &&
1104 !bm_test_page_lazy_writeout(b->bm_pages[i])) {
1105 dynamic_dev_dbg(DEV, "skipped bm lazy write for idx %u\n", i);
1106 continue;
1107 }
1108 }
Philipp Reisnerd1f37792011-06-27 14:56:52 +02001109 atomic_inc(&ctx->in_flight);
1110 bm_page_io_async(ctx, i, rw);
Lars Ellenberg19f843a2010-12-15 08:59:11 +01001111 ++count;
1112 cond_resched();
1113 }
Philipp Reisnerb411b362009-09-25 16:07:19 -07001114
Lars Ellenberg725a97e2010-12-19 11:29:55 +01001115 /*
Philipp Reisnerd1f37792011-06-27 14:56:52 +02001116 * We initialize ctx->in_flight to one to make sure bm_async_io_complete
Philipp Reisner9e58c4d2011-06-27 15:29:16 +02001117 * will not set ctx->done early, and decrement / test it here. If there
Lars Ellenberg725a97e2010-12-19 11:29:55 +01001118 * are still some bios in flight, we need to wait for them here.
Philipp Reisner9e58c4d2011-06-27 15:29:16 +02001119 * If all IO is done already (or nothing had been submitted), there is
1120 * no need to wait. Still, we need to put the kref associated with the
1121 * "in_flight reached zero, all done" event.
Lars Ellenberg725a97e2010-12-19 11:29:55 +01001122 */
Philipp Reisnerd1f37792011-06-27 14:56:52 +02001123 if (!atomic_dec_and_test(&ctx->in_flight))
Lars Ellenberg44edfb02012-09-27 13:03:45 +02001124 wait_until_done_or_force_detached(mdev, mdev->ldev, &ctx->done);
Philipp Reisner9e58c4d2011-06-27 15:29:16 +02001125 else
1126 kref_put(&ctx->kref, &bm_aio_ctx_destroy);
Philipp Reisnerd1f37792011-06-27 14:56:52 +02001127
Lars Ellenbergc9d963a2011-02-21 15:10:23 +01001128 /* summary for global bitmap IO */
1129 if (flags == 0)
1130 dev_info(DEV, "bitmap %s of %u pages took %lu jiffies\n",
Philipp Reisnercdfda632011-07-05 15:38:59 +02001131 rw == WRITE ? "WRITE" : "READ",
1132 count, jiffies - now);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001133
Philipp Reisnerd1f37792011-06-27 14:56:52 +02001134 if (ctx->error) {
Philipp Reisnerb411b362009-09-25 16:07:19 -07001135 dev_alert(DEV, "we had at least one MD IO ERROR during bitmap IO\n");
Lars Ellenberg383606e2012-06-14 14:21:32 +02001136 drbd_chk_io_error(mdev, 1, DRBD_META_IO_ERROR);
Philipp Reisnerd1f37792011-06-27 14:56:52 +02001137 err = -EIO; /* ctx->error ? */
Philipp Reisnerb411b362009-09-25 16:07:19 -07001138 }
1139
Philipp Reisner9e58c4d2011-06-27 15:29:16 +02001140 if (atomic_read(&ctx->in_flight))
Lars Ellenberg44edfb02012-09-27 13:03:45 +02001141 err = -EIO; /* Disk timeout/force-detach during IO... */
Philipp Reisner9e58c4d2011-06-27 15:29:16 +02001142
Philipp Reisnerb411b362009-09-25 16:07:19 -07001143 now = jiffies;
1144 if (rw == WRITE) {
Philipp Reisnerb411b362009-09-25 16:07:19 -07001145 drbd_md_flush(mdev);
1146 } else /* rw == READ */ {
Lars Ellenberg95a0f102010-12-15 08:59:09 +01001147 b->bm_set = bm_count_bits(b);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001148 dev_info(DEV, "recounting of set bits took additional %lu jiffies\n",
1149 jiffies - now);
1150 }
1151 now = b->bm_set;
1152
Lars Ellenbergc9d963a2011-02-21 15:10:23 +01001153 if (flags == 0)
1154 dev_info(DEV, "%s (%lu bits) marked out-of-sync by on disk bit-map.\n",
1155 ppsize(ppb, now << (BM_BLOCK_SHIFT-10)), now);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001156
Philipp Reisnerd1f37792011-06-27 14:56:52 +02001157 kref_put(&ctx->kref, &bm_aio_ctx_destroy);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001158 return err;
1159}
1160
1161/**
1162 * drbd_bm_read() - Read the whole bitmap from its on disk location.
1163 * @mdev: DRBD device.
1164 */
1165int drbd_bm_read(struct drbd_conf *mdev) __must_hold(local)
1166{
Lars Ellenberg0e8488a2012-04-25 23:06:45 +02001167 return bm_rw(mdev, READ, 0, 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001168}
1169
1170/**
1171 * drbd_bm_write() - Write the whole bitmap to its on disk location.
1172 * @mdev: DRBD device.
Lars Ellenberg19f843a2010-12-15 08:59:11 +01001173 *
1174 * Will only write pages that have changed since last IO.
Philipp Reisnerb411b362009-09-25 16:07:19 -07001175 */
1176int drbd_bm_write(struct drbd_conf *mdev) __must_hold(local)
1177{
Lars Ellenberg0e8488a2012-04-25 23:06:45 +02001178 return bm_rw(mdev, WRITE, 0, 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001179}
1180
1181/**
Philipp Reisnerd1aa4d02012-08-08 21:19:09 +02001182 * drbd_bm_write_all() - Write the whole bitmap to its on disk location.
1183 * @mdev: DRBD device.
1184 *
1185 * Will write all pages.
1186 */
1187int drbd_bm_write_all(struct drbd_conf *mdev) __must_hold(local)
1188{
1189 return bm_rw(mdev, WRITE, BM_WRITE_ALL_PAGES, 0);
1190}
1191
1192/**
Lars Ellenberg19f843a2010-12-15 08:59:11 +01001193 * drbd_bm_lazy_write_out() - Write bitmap pages 0 to @upper_idx-1, if they have changed.
Philipp Reisnerb411b362009-09-25 16:07:19 -07001194 * @mdev: DRBD device.
Lars Ellenberg19f843a2010-12-15 08:59:11 +01001195 * @upper_idx: 0: write all changed pages; +ve: page index to stop scanning for changed pages
Philipp Reisnerb411b362009-09-25 16:07:19 -07001196 */
Lars Ellenberg19f843a2010-12-15 08:59:11 +01001197int drbd_bm_write_lazy(struct drbd_conf *mdev, unsigned upper_idx) __must_hold(local)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001198{
Lars Ellenberg0e8488a2012-04-25 23:06:45 +02001199 return bm_rw(mdev, WRITE, BM_AIO_COPY_PAGES, upper_idx);
1200}
1201
1202/**
1203 * drbd_bm_write_copy_pages() - Write the whole bitmap to its on disk location.
1204 * @mdev: DRBD device.
1205 *
1206 * Will only write pages that have changed since last IO.
1207 * In contrast to drbd_bm_write(), this will copy the bitmap pages
1208 * to temporary writeout pages. It is intended to trigger a full write-out
1209 * while still allowing the bitmap to change, for example if a resync or online
1210 * verify is aborted due to a failed peer disk, while local IO continues, or
1211 * pending resync acks are still being processed.
1212 */
1213int drbd_bm_write_copy_pages(struct drbd_conf *mdev) __must_hold(local)
1214{
1215 return bm_rw(mdev, WRITE, BM_AIO_COPY_PAGES, 0);
Lars Ellenberg19f843a2010-12-15 08:59:11 +01001216}
Philipp Reisnerb411b362009-09-25 16:07:19 -07001217
Lars Ellenberga220d292012-05-07 12:07:18 +02001218/**
Lars Ellenberg45dfffe2011-02-21 13:21:00 +01001219 * drbd_bm_write_hinted() - Write bitmap pages with "hint" marks, if they have changed.
1220 * @mdev: DRBD device.
1221 */
1222int drbd_bm_write_hinted(struct drbd_conf *mdev) __must_hold(local)
1223{
1224 return bm_rw(mdev, WRITE, BM_AIO_WRITE_HINTED | BM_AIO_COPY_PAGES, 0);
1225}
Lars Ellenberg19f843a2010-12-15 08:59:11 +01001226
1227/**
Lars Ellenberg45dfffe2011-02-21 13:21:00 +01001228 * drbd_bm_write_page() - Writes a PAGE_SIZE aligned piece of bitmap
Lars Ellenberg19f843a2010-12-15 08:59:11 +01001229 * @mdev: DRBD device.
1230 * @idx: bitmap page index
1231 *
Lars Ellenberg4b0715f2010-12-14 15:13:04 +01001232 * We don't want to special case on logical_block_size of the backend device,
1233 * so we submit PAGE_SIZE aligned pieces.
Lars Ellenberg19f843a2010-12-15 08:59:11 +01001234 * Note that on "most" systems, PAGE_SIZE is 4k.
Lars Ellenberg4b0715f2010-12-14 15:13:04 +01001235 *
1236 * In case this becomes an issue on systems with larger PAGE_SIZE,
1237 * we may want to change this again to write 4k aligned 4k pieces.
Lars Ellenberg19f843a2010-12-15 08:59:11 +01001238 */
1239int drbd_bm_write_page(struct drbd_conf *mdev, unsigned int idx) __must_hold(local)
1240{
Philipp Reisnerd1f37792011-06-27 14:56:52 +02001241 struct bm_aio_ctx *ctx;
1242 int err;
Lars Ellenberg19f843a2010-12-15 08:59:11 +01001243
1244 if (bm_test_page_unchanged(mdev->bitmap->bm_pages[idx])) {
Lars Ellenberg7648cdf2010-12-17 23:58:41 +01001245 dynamic_dev_dbg(DEV, "skipped bm page write for idx %u\n", idx);
Lars Ellenberg19f843a2010-12-15 08:59:11 +01001246 return 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001247 }
Lars Ellenberg19f843a2010-12-15 08:59:11 +01001248
Lars Ellenberg22f46ce2011-07-11 17:32:26 +02001249 ctx = kmalloc(sizeof(struct bm_aio_ctx), GFP_NOIO);
Philipp Reisnerd1f37792011-06-27 14:56:52 +02001250 if (!ctx)
1251 return -ENOMEM;
Lars Ellenberg19f843a2010-12-15 08:59:11 +01001252
Philipp Reisnerd1f37792011-06-27 14:56:52 +02001253 *ctx = (struct bm_aio_ctx) {
1254 .mdev = mdev,
1255 .in_flight = ATOMIC_INIT(1),
Philipp Reisner9e58c4d2011-06-27 15:29:16 +02001256 .done = 0,
Philipp Reisnerd1f37792011-06-27 14:56:52 +02001257 .flags = BM_AIO_COPY_PAGES,
1258 .error = 0,
1259 .kref = { ATOMIC_INIT(2) },
1260 };
1261
Philipp Reisner9e58c4d2011-06-27 15:29:16 +02001262 if (!get_ldev_if_state(mdev, D_ATTACHING)) { /* put is in bm_aio_ctx_destroy() */
1263 dev_err(DEV, "ASSERT FAILED: get_ldev_if_state() == 1 in drbd_bm_write_page()\n");
1264 kfree(ctx);
1265 return -ENODEV;
1266 }
1267
Philipp Reisnerd1f37792011-06-27 14:56:52 +02001268 bm_page_io_async(ctx, idx, WRITE_SYNC);
Lars Ellenberg44edfb02012-09-27 13:03:45 +02001269 wait_until_done_or_force_detached(mdev, mdev->ldev, &ctx->done);
Philipp Reisnerd1f37792011-06-27 14:56:52 +02001270
1271 if (ctx->error)
Lars Ellenberg383606e2012-06-14 14:21:32 +02001272 drbd_chk_io_error(mdev, 1, DRBD_META_IO_ERROR);
Lars Ellenberg44edfb02012-09-27 13:03:45 +02001273 /* that causes us to detach, so the in memory bitmap will be
Lars Ellenberg19f843a2010-12-15 08:59:11 +01001274 * gone in a moment as well. */
1275
Philipp Reisnerb411b362009-09-25 16:07:19 -07001276 mdev->bm_writ_cnt++;
Philipp Reisner9e58c4d2011-06-27 15:29:16 +02001277 err = atomic_read(&ctx->in_flight) ? -EIO : ctx->error;
Philipp Reisnerd1f37792011-06-27 14:56:52 +02001278 kref_put(&ctx->kref, &bm_aio_ctx_destroy);
Philipp Reisnerd1f37792011-06-27 14:56:52 +02001279 return err;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001280}
1281
1282/* NOTE
1283 * find_first_bit returns int, we return unsigned long.
Lars Ellenberg4b0715f2010-12-14 15:13:04 +01001284 * For this to work on 32bit arch with bitnumbers > (1<<32),
1285 * we'd need to return u64, and get a whole lot of other places
1286 * fixed where we still use unsigned long.
Philipp Reisnerb411b362009-09-25 16:07:19 -07001287 *
1288 * this returns a bit number, NOT a sector!
1289 */
Philipp Reisnerb411b362009-09-25 16:07:19 -07001290static unsigned long __bm_find_next(struct drbd_conf *mdev, unsigned long bm_fo,
Cong Wangcfd80052011-11-25 23:14:18 +08001291 const int find_zero_bit)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001292{
1293 struct drbd_bitmap *b = mdev->bitmap;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001294 unsigned long *p_addr;
Lars Ellenberg4b0715f2010-12-14 15:13:04 +01001295 unsigned long bit_offset;
1296 unsigned i;
1297
Philipp Reisnerb411b362009-09-25 16:07:19 -07001298
1299 if (bm_fo > b->bm_bits) {
1300 dev_err(DEV, "bm_fo=%lu bm_bits=%lu\n", bm_fo, b->bm_bits);
Lars Ellenberg4b0715f2010-12-14 15:13:04 +01001301 bm_fo = DRBD_END_OF_BITMAP;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001302 } else {
1303 while (bm_fo < b->bm_bits) {
Lars Ellenberg19f843a2010-12-15 08:59:11 +01001304 /* bit offset of the first bit in the page */
Lars Ellenberg4b0715f2010-12-14 15:13:04 +01001305 bit_offset = bm_fo & ~BITS_PER_PAGE_MASK;
Cong Wangcfd80052011-11-25 23:14:18 +08001306 p_addr = __bm_map_pidx(b, bm_bit_to_page_idx(b, bm_fo));
Philipp Reisnerb411b362009-09-25 16:07:19 -07001307
1308 if (find_zero_bit)
Linus Torvalds7e599e62011-03-28 07:42:58 -07001309 i = find_next_zero_bit_le(p_addr,
Lars Ellenberg4b0715f2010-12-14 15:13:04 +01001310 PAGE_SIZE*8, bm_fo & BITS_PER_PAGE_MASK);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001311 else
Linus Torvalds7e599e62011-03-28 07:42:58 -07001312 i = find_next_bit_le(p_addr,
Lars Ellenberg4b0715f2010-12-14 15:13:04 +01001313 PAGE_SIZE*8, bm_fo & BITS_PER_PAGE_MASK);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001314
Cong Wangcfd80052011-11-25 23:14:18 +08001315 __bm_unmap(p_addr);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001316 if (i < PAGE_SIZE*8) {
Lars Ellenberg4b0715f2010-12-14 15:13:04 +01001317 bm_fo = bit_offset + i;
1318 if (bm_fo >= b->bm_bits)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001319 break;
1320 goto found;
1321 }
1322 bm_fo = bit_offset + PAGE_SIZE*8;
1323 }
Lars Ellenberg4b0715f2010-12-14 15:13:04 +01001324 bm_fo = DRBD_END_OF_BITMAP;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001325 }
1326 found:
Lars Ellenberg4b0715f2010-12-14 15:13:04 +01001327 return bm_fo;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001328}
1329
1330static unsigned long bm_find_next(struct drbd_conf *mdev,
1331 unsigned long bm_fo, const int find_zero_bit)
1332{
1333 struct drbd_bitmap *b = mdev->bitmap;
Lars Ellenberg4b0715f2010-12-14 15:13:04 +01001334 unsigned long i = DRBD_END_OF_BITMAP;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001335
Andreas Gruenbacher841ce242010-12-15 19:31:20 +01001336 if (!expect(b))
1337 return i;
1338 if (!expect(b->bm_pages))
1339 return i;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001340
1341 spin_lock_irq(&b->bm_lock);
Lars Ellenberg20ceb2b2011-01-21 10:56:44 +01001342 if (BM_DONT_TEST & b->bm_flags)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001343 bm_print_lock_info(mdev);
1344
Cong Wangcfd80052011-11-25 23:14:18 +08001345 i = __bm_find_next(mdev, bm_fo, find_zero_bit);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001346
1347 spin_unlock_irq(&b->bm_lock);
1348 return i;
1349}
1350
1351unsigned long drbd_bm_find_next(struct drbd_conf *mdev, unsigned long bm_fo)
1352{
1353 return bm_find_next(mdev, bm_fo, 0);
1354}
1355
1356#if 0
1357/* not yet needed for anything. */
1358unsigned long drbd_bm_find_next_zero(struct drbd_conf *mdev, unsigned long bm_fo)
1359{
1360 return bm_find_next(mdev, bm_fo, 1);
1361}
1362#endif
1363
1364/* does not spin_lock_irqsave.
1365 * you must take drbd_bm_lock() first */
1366unsigned long _drbd_bm_find_next(struct drbd_conf *mdev, unsigned long bm_fo)
1367{
Lars Ellenberg20ceb2b2011-01-21 10:56:44 +01001368 /* WARN_ON(!(BM_DONT_SET & mdev->b->bm_flags)); */
Cong Wangcfd80052011-11-25 23:14:18 +08001369 return __bm_find_next(mdev, bm_fo, 0);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001370}
1371
1372unsigned long _drbd_bm_find_next_zero(struct drbd_conf *mdev, unsigned long bm_fo)
1373{
Lars Ellenberg20ceb2b2011-01-21 10:56:44 +01001374 /* WARN_ON(!(BM_DONT_SET & mdev->b->bm_flags)); */
Cong Wangcfd80052011-11-25 23:14:18 +08001375 return __bm_find_next(mdev, bm_fo, 1);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001376}
1377
1378/* returns number of bits actually changed.
1379 * for val != 0, we change 0 -> 1, return code positive
1380 * for val == 0, we change 1 -> 0, return code negative
1381 * wants bitnr, not sector.
1382 * expected to be called for only a few bits (e - s about BITS_PER_LONG).
1383 * Must hold bitmap lock already. */
Philipp Reisnerb4ee79d2010-04-01 09:57:40 +02001384static int __bm_change_bits_to(struct drbd_conf *mdev, const unsigned long s,
Lars Ellenberg829c6082011-06-03 21:18:13 +02001385 unsigned long e, int val)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001386{
1387 struct drbd_bitmap *b = mdev->bitmap;
1388 unsigned long *p_addr = NULL;
1389 unsigned long bitnr;
Lars Ellenberg19f843a2010-12-15 08:59:11 +01001390 unsigned int last_page_nr = -1U;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001391 int c = 0;
Lars Ellenberg19f843a2010-12-15 08:59:11 +01001392 int changed_total = 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001393
1394 if (e >= b->bm_bits) {
1395 dev_err(DEV, "ASSERT FAILED: bit_s=%lu bit_e=%lu bm_bits=%lu\n",
1396 s, e, b->bm_bits);
1397 e = b->bm_bits ? b->bm_bits -1 : 0;
1398 }
1399 for (bitnr = s; bitnr <= e; bitnr++) {
Lars Ellenberg19f843a2010-12-15 08:59:11 +01001400 unsigned int page_nr = bm_bit_to_page_idx(b, bitnr);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001401 if (page_nr != last_page_nr) {
1402 if (p_addr)
Cong Wangcfd80052011-11-25 23:14:18 +08001403 __bm_unmap(p_addr);
Lars Ellenberg19f843a2010-12-15 08:59:11 +01001404 if (c < 0)
1405 bm_set_page_lazy_writeout(b->bm_pages[last_page_nr]);
1406 else if (c > 0)
1407 bm_set_page_need_writeout(b->bm_pages[last_page_nr]);
1408 changed_total += c;
1409 c = 0;
Cong Wangcfd80052011-11-25 23:14:18 +08001410 p_addr = __bm_map_pidx(b, page_nr);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001411 last_page_nr = page_nr;
1412 }
1413 if (val)
Linus Torvalds7e599e62011-03-28 07:42:58 -07001414 c += (0 == __test_and_set_bit_le(bitnr & BITS_PER_PAGE_MASK, p_addr));
Philipp Reisnerb411b362009-09-25 16:07:19 -07001415 else
Linus Torvalds7e599e62011-03-28 07:42:58 -07001416 c -= (0 != __test_and_clear_bit_le(bitnr & BITS_PER_PAGE_MASK, p_addr));
Philipp Reisnerb411b362009-09-25 16:07:19 -07001417 }
1418 if (p_addr)
Cong Wangcfd80052011-11-25 23:14:18 +08001419 __bm_unmap(p_addr);
Lars Ellenberg19f843a2010-12-15 08:59:11 +01001420 if (c < 0)
1421 bm_set_page_lazy_writeout(b->bm_pages[last_page_nr]);
1422 else if (c > 0)
1423 bm_set_page_need_writeout(b->bm_pages[last_page_nr]);
1424 changed_total += c;
1425 b->bm_set += changed_total;
1426 return changed_total;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001427}
1428
1429/* returns number of bits actually changed.
1430 * for val != 0, we change 0 -> 1, return code positive
1431 * for val == 0, we change 1 -> 0, return code negative
1432 * wants bitnr, not sector */
Philipp Reisnerb4ee79d2010-04-01 09:57:40 +02001433static int bm_change_bits_to(struct drbd_conf *mdev, const unsigned long s,
Philipp Reisnerb411b362009-09-25 16:07:19 -07001434 const unsigned long e, int val)
1435{
1436 unsigned long flags;
1437 struct drbd_bitmap *b = mdev->bitmap;
1438 int c = 0;
1439
Andreas Gruenbacher841ce242010-12-15 19:31:20 +01001440 if (!expect(b))
1441 return 1;
1442 if (!expect(b->bm_pages))
1443 return 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001444
1445 spin_lock_irqsave(&b->bm_lock, flags);
Lars Ellenberg20ceb2b2011-01-21 10:56:44 +01001446 if ((val ? BM_DONT_SET : BM_DONT_CLEAR) & b->bm_flags)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001447 bm_print_lock_info(mdev);
1448
Lars Ellenberg829c6082011-06-03 21:18:13 +02001449 c = __bm_change_bits_to(mdev, s, e, val);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001450
1451 spin_unlock_irqrestore(&b->bm_lock, flags);
1452 return c;
1453}
1454
1455/* returns number of bits changed 0 -> 1 */
1456int drbd_bm_set_bits(struct drbd_conf *mdev, const unsigned long s, const unsigned long e)
1457{
1458 return bm_change_bits_to(mdev, s, e, 1);
1459}
1460
1461/* returns number of bits changed 1 -> 0 */
1462int drbd_bm_clear_bits(struct drbd_conf *mdev, const unsigned long s, const unsigned long e)
1463{
1464 return -bm_change_bits_to(mdev, s, e, 0);
1465}
1466
1467/* sets all bits in full words,
1468 * from first_word up to, but not including, last_word */
1469static inline void bm_set_full_words_within_one_page(struct drbd_bitmap *b,
1470 int page_nr, int first_word, int last_word)
1471{
1472 int i;
1473 int bits;
Lars Ellenberg22d81142011-09-26 09:58:04 +02001474 int changed = 0;
Cong Wangcfd80052011-11-25 23:14:18 +08001475 unsigned long *paddr = kmap_atomic(b->bm_pages[page_nr]);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001476 for (i = first_word; i < last_word; i++) {
1477 bits = hweight_long(paddr[i]);
1478 paddr[i] = ~0UL;
Lars Ellenberg22d81142011-09-26 09:58:04 +02001479 changed += BITS_PER_LONG - bits;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001480 }
Cong Wangcfd80052011-11-25 23:14:18 +08001481 kunmap_atomic(paddr);
Lars Ellenberg22d81142011-09-26 09:58:04 +02001482 if (changed) {
1483 /* We only need lazy writeout, the information is still in the
1484 * remote bitmap as well, and is reconstructed during the next
1485 * bitmap exchange, if lost locally due to a crash. */
1486 bm_set_page_lazy_writeout(b->bm_pages[page_nr]);
1487 b->bm_set += changed;
1488 }
Philipp Reisnerb411b362009-09-25 16:07:19 -07001489}
1490
Lars Ellenberg829c6082011-06-03 21:18:13 +02001491/* Same thing as drbd_bm_set_bits,
1492 * but more efficient for a large bit range.
Philipp Reisnerb411b362009-09-25 16:07:19 -07001493 * You must first drbd_bm_lock().
1494 * Can be called to set the whole bitmap in one go.
1495 * Sets bits from s to e _inclusive_. */
1496void _drbd_bm_set_bits(struct drbd_conf *mdev, const unsigned long s, const unsigned long e)
1497{
1498 /* First set_bit from the first bit (s)
1499 * up to the next long boundary (sl),
1500 * then assign full words up to the last long boundary (el),
1501 * then set_bit up to and including the last bit (e).
1502 *
1503 * Do not use memset, because we must account for changes,
1504 * so we need to loop over the words with hweight() anyways.
1505 */
Lars Ellenberg829c6082011-06-03 21:18:13 +02001506 struct drbd_bitmap *b = mdev->bitmap;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001507 unsigned long sl = ALIGN(s,BITS_PER_LONG);
1508 unsigned long el = (e+1) & ~((unsigned long)BITS_PER_LONG-1);
1509 int first_page;
1510 int last_page;
1511 int page_nr;
1512 int first_word;
1513 int last_word;
1514
1515 if (e - s <= 3*BITS_PER_LONG) {
1516 /* don't bother; el and sl may even be wrong. */
Lars Ellenberg829c6082011-06-03 21:18:13 +02001517 spin_lock_irq(&b->bm_lock);
1518 __bm_change_bits_to(mdev, s, e, 1);
1519 spin_unlock_irq(&b->bm_lock);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001520 return;
1521 }
1522
1523 /* difference is large enough that we can trust sl and el */
1524
Lars Ellenberg829c6082011-06-03 21:18:13 +02001525 spin_lock_irq(&b->bm_lock);
1526
Philipp Reisnerb411b362009-09-25 16:07:19 -07001527 /* bits filling the current long */
1528 if (sl)
Lars Ellenberg829c6082011-06-03 21:18:13 +02001529 __bm_change_bits_to(mdev, s, sl-1, 1);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001530
1531 first_page = sl >> (3 + PAGE_SHIFT);
1532 last_page = el >> (3 + PAGE_SHIFT);
1533
1534 /* MLPP: modulo longs per page */
1535 /* LWPP: long words per page */
1536 first_word = MLPP(sl >> LN2_BPL);
1537 last_word = LWPP;
1538
1539 /* first and full pages, unless first page == last page */
1540 for (page_nr = first_page; page_nr < last_page; page_nr++) {
1541 bm_set_full_words_within_one_page(mdev->bitmap, page_nr, first_word, last_word);
Lars Ellenberg8ccee202011-06-06 11:31:42 +02001542 spin_unlock_irq(&b->bm_lock);
1543 cond_resched();
Philipp Reisnerb411b362009-09-25 16:07:19 -07001544 first_word = 0;
Lars Ellenberg8ccee202011-06-06 11:31:42 +02001545 spin_lock_irq(&b->bm_lock);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001546 }
Philipp Reisnerb411b362009-09-25 16:07:19 -07001547 /* last page (respectively only page, for first page == last page) */
1548 last_word = MLPP(el >> LN2_BPL);
Lars Ellenberg4eccc572012-06-08 13:18:51 +02001549
1550 /* consider bitmap->bm_bits = 32768, bitmap->bm_number_of_pages = 1. (or multiples).
1551 * ==> e = 32767, el = 32768, last_page = 2,
1552 * and now last_word = 0.
1553 * We do not want to touch last_page in this case,
1554 * as we did not allocate it, it is not present in bitmap->bm_pages.
1555 */
1556 if (last_word)
1557 bm_set_full_words_within_one_page(mdev->bitmap, last_page, first_word, last_word);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001558
1559 /* possibly trailing bits.
1560 * example: (e & 63) == 63, el will be e+1.
1561 * if that even was the very last bit,
1562 * it would trigger an assert in __bm_change_bits_to()
1563 */
1564 if (el <= e)
Lars Ellenberg829c6082011-06-03 21:18:13 +02001565 __bm_change_bits_to(mdev, el, e, 1);
1566 spin_unlock_irq(&b->bm_lock);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001567}
1568
1569/* returns bit state
1570 * wants bitnr, NOT sector.
1571 * inherently racy... area needs to be locked by means of {al,rs}_lru
1572 * 1 ... bit set
1573 * 0 ... bit not set
1574 * -1 ... first out of bounds access, stop testing for bits!
1575 */
1576int drbd_bm_test_bit(struct drbd_conf *mdev, const unsigned long bitnr)
1577{
1578 unsigned long flags;
1579 struct drbd_bitmap *b = mdev->bitmap;
1580 unsigned long *p_addr;
1581 int i;
1582
Andreas Gruenbacher841ce242010-12-15 19:31:20 +01001583 if (!expect(b))
1584 return 0;
1585 if (!expect(b->bm_pages))
1586 return 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001587
1588 spin_lock_irqsave(&b->bm_lock, flags);
Lars Ellenberg20ceb2b2011-01-21 10:56:44 +01001589 if (BM_DONT_TEST & b->bm_flags)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001590 bm_print_lock_info(mdev);
1591 if (bitnr < b->bm_bits) {
Lars Ellenberg19f843a2010-12-15 08:59:11 +01001592 p_addr = bm_map_pidx(b, bm_bit_to_page_idx(b, bitnr));
Linus Torvalds7e599e62011-03-28 07:42:58 -07001593 i = test_bit_le(bitnr & BITS_PER_PAGE_MASK, p_addr) ? 1 : 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001594 bm_unmap(p_addr);
1595 } else if (bitnr == b->bm_bits) {
1596 i = -1;
1597 } else { /* (bitnr > b->bm_bits) */
1598 dev_err(DEV, "bitnr=%lu > bm_bits=%lu\n", bitnr, b->bm_bits);
1599 i = 0;
1600 }
1601
1602 spin_unlock_irqrestore(&b->bm_lock, flags);
1603 return i;
1604}
1605
1606/* returns number of bits set in the range [s, e] */
1607int drbd_bm_count_bits(struct drbd_conf *mdev, const unsigned long s, const unsigned long e)
1608{
1609 unsigned long flags;
1610 struct drbd_bitmap *b = mdev->bitmap;
Lars Ellenberg19f843a2010-12-15 08:59:11 +01001611 unsigned long *p_addr = NULL;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001612 unsigned long bitnr;
Lars Ellenberg19f843a2010-12-15 08:59:11 +01001613 unsigned int page_nr = -1U;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001614 int c = 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001615
1616 /* If this is called without a bitmap, that is a bug. But just to be
1617 * robust in case we screwed up elsewhere, in that case pretend there
1618 * was one dirty bit in the requested area, so we won't try to do a
1619 * local read there (no bitmap probably implies no disk) */
Andreas Gruenbacher841ce242010-12-15 19:31:20 +01001620 if (!expect(b))
1621 return 1;
1622 if (!expect(b->bm_pages))
1623 return 1;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001624
1625 spin_lock_irqsave(&b->bm_lock, flags);
Lars Ellenberg20ceb2b2011-01-21 10:56:44 +01001626 if (BM_DONT_TEST & b->bm_flags)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001627 bm_print_lock_info(mdev);
1628 for (bitnr = s; bitnr <= e; bitnr++) {
Lars Ellenberg19f843a2010-12-15 08:59:11 +01001629 unsigned int idx = bm_bit_to_page_idx(b, bitnr);
1630 if (page_nr != idx) {
1631 page_nr = idx;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001632 if (p_addr)
1633 bm_unmap(p_addr);
Lars Ellenberg19f843a2010-12-15 08:59:11 +01001634 p_addr = bm_map_pidx(b, idx);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001635 }
Andreas Gruenbacher841ce242010-12-15 19:31:20 +01001636 if (expect(bitnr < b->bm_bits))
Linus Torvalds7e599e62011-03-28 07:42:58 -07001637 c += (0 != test_bit_le(bitnr - (page_nr << (PAGE_SHIFT+3)), p_addr));
Andreas Gruenbacher841ce242010-12-15 19:31:20 +01001638 else
1639 dev_err(DEV, "bitnr=%lu bm_bits=%lu\n", bitnr, b->bm_bits);
Philipp Reisnerb411b362009-09-25 16:07:19 -07001640 }
1641 if (p_addr)
1642 bm_unmap(p_addr);
1643 spin_unlock_irqrestore(&b->bm_lock, flags);
1644 return c;
1645}
1646
1647
1648/* inherently racy...
1649 * return value may be already out-of-date when this function returns.
1650 * but the general usage is that this is only use during a cstate when bits are
1651 * only cleared, not set, and typically only care for the case when the return
1652 * value is zero, or we already "locked" this "bitmap extent" by other means.
1653 *
1654 * enr is bm-extent number, since we chose to name one sector (512 bytes)
1655 * worth of the bitmap a "bitmap extent".
1656 *
1657 * TODO
1658 * I think since we use it like a reference count, we should use the real
1659 * reference count of some bitmap extent element from some lru instead...
1660 *
1661 */
1662int drbd_bm_e_weight(struct drbd_conf *mdev, unsigned long enr)
1663{
1664 struct drbd_bitmap *b = mdev->bitmap;
1665 int count, s, e;
1666 unsigned long flags;
1667 unsigned long *p_addr, *bm;
1668
Andreas Gruenbacher841ce242010-12-15 19:31:20 +01001669 if (!expect(b))
1670 return 0;
1671 if (!expect(b->bm_pages))
1672 return 0;
Philipp Reisnerb411b362009-09-25 16:07:19 -07001673
1674 spin_lock_irqsave(&b->bm_lock, flags);
Lars Ellenberg20ceb2b2011-01-21 10:56:44 +01001675 if (BM_DONT_TEST & b->bm_flags)
Philipp Reisnerb411b362009-09-25 16:07:19 -07001676 bm_print_lock_info(mdev);
1677
1678 s = S2W(enr);
1679 e = min((size_t)S2W(enr+1), b->bm_words);
1680 count = 0;
1681 if (s < b->bm_words) {
1682 int n = e-s;
Lars Ellenberg19f843a2010-12-15 08:59:11 +01001683 p_addr = bm_map_pidx(b, bm_word_to_page_idx(b, s));
Philipp Reisnerb411b362009-09-25 16:07:19 -07001684 bm = p_addr + MLPP(s);
1685 while (n--)
1686 count += hweight_long(*bm++);
1687 bm_unmap(p_addr);
1688 } else {
1689 dev_err(DEV, "start offset (%d) too large in drbd_bm_e_weight\n", s);
1690 }
1691 spin_unlock_irqrestore(&b->bm_lock, flags);
1692 return count;
1693}