Commit | Line | Data |
---|---|---|
b411b363 PR |
1 | /* |
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> | |
5a0e3ad6 | 29 | #include <linux/slab.h> |
b411b363 | 30 | #include <asm/kmap_types.h> |
f0ff1357 | 31 | |
b411b363 PR |
32 | #include "drbd_int.h" |
33 | ||
95a0f10c | 34 | |
b411b363 PR |
35 | /* 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". | |
4b0715f0 LE |
41 | */ |
42 | ||
b411b363 | 43 | |
4b0715f0 LE |
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 | * | |
24c4830c | 77 | * We plan to reduce the amount of in-core bitmap pages by paging them in |
4b0715f0 LE |
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. | |
b411b363 PR |
81 | */ |
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 | */ | |
93 | struct drbd_bitmap { | |
94 | struct page **bm_pages; | |
95 | spinlock_t bm_lock; | |
4b0715f0 LE |
96 | |
97 | /* see LIMITATIONS: above */ | |
98 | ||
b411b363 PR |
99 | 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; | |
8a03ae2a | 104 | struct mutex bm_change; /* serializes resize operations */ |
b411b363 | 105 | |
19f843aa | 106 | wait_queue_head_t bm_io_wait; /* used to serialize IO of single pages */ |
b411b363 | 107 | |
20ceb2b2 | 108 | enum bm_flag bm_flags; |
b411b363 PR |
109 | |
110 | /* debugging aid, in case we are still racy somewhere */ | |
111 | char *bm_why; | |
112 | struct task_struct *bm_task; | |
113 | }; | |
114 | ||
b4ee79da | 115 | static int __bm_change_bits_to(struct drbd_conf *mdev, const unsigned long s, |
fd76438c PR |
116 | unsigned long e, int val, const enum km_type km); |
117 | ||
b411b363 PR |
118 | #define bm_print_lock_info(m) __bm_print_lock_info(m, __func__) |
119 | static void __bm_print_lock_info(struct drbd_conf *mdev, const char *func) | |
120 | { | |
121 | struct drbd_bitmap *b = mdev->bitmap; | |
122 | if (!__ratelimit(&drbd_ratelimit_state)) | |
123 | return; | |
124 | dev_err(DEV, "FIXME %s in %s, bitmap locked for '%s' by %s\n", | |
125 | current == mdev->receiver.task ? "receiver" : | |
126 | current == mdev->asender.task ? "asender" : | |
127 | current == mdev->worker.task ? "worker" : current->comm, | |
128 | func, b->bm_why ?: "?", | |
129 | b->bm_task == mdev->receiver.task ? "receiver" : | |
130 | b->bm_task == mdev->asender.task ? "asender" : | |
131 | b->bm_task == mdev->worker.task ? "worker" : "?"); | |
132 | } | |
133 | ||
20ceb2b2 | 134 | void drbd_bm_lock(struct drbd_conf *mdev, char *why, enum bm_flag flags) |
b411b363 PR |
135 | { |
136 | struct drbd_bitmap *b = mdev->bitmap; | |
137 | int trylock_failed; | |
138 | ||
139 | if (!b) { | |
140 | dev_err(DEV, "FIXME no bitmap in drbd_bm_lock!?\n"); | |
141 | return; | |
142 | } | |
143 | ||
8a03ae2a | 144 | trylock_failed = !mutex_trylock(&b->bm_change); |
b411b363 PR |
145 | |
146 | if (trylock_failed) { | |
147 | dev_warn(DEV, "%s going to '%s' but bitmap already locked for '%s' by %s\n", | |
148 | current == mdev->receiver.task ? "receiver" : | |
149 | current == mdev->asender.task ? "asender" : | |
150 | current == mdev->worker.task ? "worker" : current->comm, | |
151 | why, b->bm_why ?: "?", | |
152 | b->bm_task == mdev->receiver.task ? "receiver" : | |
153 | b->bm_task == mdev->asender.task ? "asender" : | |
154 | b->bm_task == mdev->worker.task ? "worker" : "?"); | |
8a03ae2a | 155 | mutex_lock(&b->bm_change); |
b411b363 | 156 | } |
20ceb2b2 | 157 | if (BM_LOCKED_MASK & b->bm_flags) |
b411b363 | 158 | dev_err(DEV, "FIXME bitmap already locked in bm_lock\n"); |
20ceb2b2 | 159 | b->bm_flags |= flags & BM_LOCKED_MASK; |
b411b363 PR |
160 | |
161 | b->bm_why = why; | |
162 | b->bm_task = current; | |
163 | } | |
164 | ||
165 | void drbd_bm_unlock(struct drbd_conf *mdev) | |
166 | { | |
167 | struct drbd_bitmap *b = mdev->bitmap; | |
168 | if (!b) { | |
169 | dev_err(DEV, "FIXME no bitmap in drbd_bm_unlock!?\n"); | |
170 | return; | |
171 | } | |
172 | ||
20ceb2b2 | 173 | if (!(BM_LOCKED_MASK & mdev->bitmap->bm_flags)) |
b411b363 PR |
174 | dev_err(DEV, "FIXME bitmap not locked in bm_unlock\n"); |
175 | ||
20ceb2b2 | 176 | b->bm_flags &= ~BM_LOCKED_MASK; |
b411b363 PR |
177 | b->bm_why = NULL; |
178 | b->bm_task = NULL; | |
8a03ae2a | 179 | mutex_unlock(&b->bm_change); |
b411b363 PR |
180 | } |
181 | ||
19f843aa LE |
182 | /* we store some "meta" info about our pages in page->private */ |
183 | /* at a granularity of 4k storage per bitmap bit: | |
184 | * one peta byte storage: 1<<50 byte, 1<<38 * 4k storage blocks | |
185 | * 1<<38 bits, | |
186 | * 1<<23 4k bitmap pages. | |
187 | * Use 24 bits as page index, covers 2 peta byte storage | |
188 | * at a granularity of 4k per bit. | |
189 | * Used to report the failed page idx on io error from the endio handlers. | |
190 | */ | |
191 | #define BM_PAGE_IDX_MASK ((1UL<<24)-1) | |
192 | /* this page is currently read in, or written back */ | |
193 | #define BM_PAGE_IO_LOCK 31 | |
194 | /* if there has been an IO error for this page */ | |
195 | #define BM_PAGE_IO_ERROR 30 | |
196 | /* this is to be able to intelligently skip disk IO, | |
197 | * set if bits have been set since last IO. */ | |
198 | #define BM_PAGE_NEED_WRITEOUT 29 | |
199 | /* to mark for lazy writeout once syncer cleared all clearable bits, | |
200 | * we if bits have been cleared since last IO. */ | |
201 | #define BM_PAGE_LAZY_WRITEOUT 28 | |
202 | ||
24c4830c | 203 | /* store_page_idx uses non-atomic assignment. It is only used directly after |
19f843aa LE |
204 | * allocating the page. All other bm_set_page_* and bm_clear_page_* need to |
205 | * use atomic bit manipulation, as set_out_of_sync (and therefore bitmap | |
206 | * changes) may happen from various contexts, and wait_on_bit/wake_up_bit | |
207 | * requires it all to be atomic as well. */ | |
208 | static void bm_store_page_idx(struct page *page, unsigned long idx) | |
95a0f10c | 209 | { |
19f843aa LE |
210 | BUG_ON(0 != (idx & ~BM_PAGE_IDX_MASK)); |
211 | page_private(page) |= idx; | |
95a0f10c LE |
212 | } |
213 | ||
19f843aa | 214 | static unsigned long bm_page_to_idx(struct page *page) |
b411b363 | 215 | { |
19f843aa LE |
216 | return page_private(page) & BM_PAGE_IDX_MASK; |
217 | } | |
218 | ||
219 | /* As is very unlikely that the same page is under IO from more than one | |
220 | * context, we can get away with a bit per page and one wait queue per bitmap. | |
221 | */ | |
222 | static void bm_page_lock_io(struct drbd_conf *mdev, int page_nr) | |
223 | { | |
224 | struct drbd_bitmap *b = mdev->bitmap; | |
225 | void *addr = &page_private(b->bm_pages[page_nr]); | |
226 | wait_event(b->bm_io_wait, !test_and_set_bit(BM_PAGE_IO_LOCK, addr)); | |
227 | } | |
228 | ||
229 | static void bm_page_unlock_io(struct drbd_conf *mdev, int page_nr) | |
230 | { | |
231 | struct drbd_bitmap *b = mdev->bitmap; | |
232 | void *addr = &page_private(b->bm_pages[page_nr]); | |
233 | clear_bit(BM_PAGE_IO_LOCK, addr); | |
234 | smp_mb__after_clear_bit(); | |
235 | wake_up(&mdev->bitmap->bm_io_wait); | |
236 | } | |
237 | ||
238 | /* set _before_ submit_io, so it may be reset due to being changed | |
239 | * while this page is in flight... will get submitted later again */ | |
240 | static void bm_set_page_unchanged(struct page *page) | |
241 | { | |
242 | /* use cmpxchg? */ | |
243 | clear_bit(BM_PAGE_NEED_WRITEOUT, &page_private(page)); | |
244 | clear_bit(BM_PAGE_LAZY_WRITEOUT, &page_private(page)); | |
245 | } | |
246 | ||
247 | static void bm_set_page_need_writeout(struct page *page) | |
248 | { | |
249 | set_bit(BM_PAGE_NEED_WRITEOUT, &page_private(page)); | |
250 | } | |
251 | ||
252 | static int bm_test_page_unchanged(struct page *page) | |
253 | { | |
254 | volatile const unsigned long *addr = &page_private(page); | |
255 | return (*addr & ((1UL<<BM_PAGE_NEED_WRITEOUT)|(1UL<<BM_PAGE_LAZY_WRITEOUT))) == 0; | |
256 | } | |
b411b363 | 257 | |
19f843aa LE |
258 | static void bm_set_page_io_err(struct page *page) |
259 | { | |
260 | set_bit(BM_PAGE_IO_ERROR, &page_private(page)); | |
261 | } | |
262 | ||
263 | static void bm_clear_page_io_err(struct page *page) | |
264 | { | |
265 | clear_bit(BM_PAGE_IO_ERROR, &page_private(page)); | |
266 | } | |
267 | ||
268 | static void bm_set_page_lazy_writeout(struct page *page) | |
269 | { | |
270 | set_bit(BM_PAGE_LAZY_WRITEOUT, &page_private(page)); | |
271 | } | |
272 | ||
273 | static int bm_test_page_lazy_writeout(struct page *page) | |
274 | { | |
275 | return test_bit(BM_PAGE_LAZY_WRITEOUT, &page_private(page)); | |
276 | } | |
277 | ||
278 | /* on a 32bit box, this would allow for exactly (2<<38) bits. */ | |
279 | static unsigned int bm_word_to_page_idx(struct drbd_bitmap *b, unsigned long long_nr) | |
280 | { | |
b411b363 | 281 | /* page_nr = (word*sizeof(long)) >> PAGE_SHIFT; */ |
19f843aa | 282 | unsigned int page_nr = long_nr >> (PAGE_SHIFT - LN2_BPL + 3); |
b411b363 | 283 | BUG_ON(page_nr >= b->bm_number_of_pages); |
19f843aa LE |
284 | return page_nr; |
285 | } | |
b411b363 | 286 | |
19f843aa LE |
287 | static unsigned int bm_bit_to_page_idx(struct drbd_bitmap *b, u64 bitnr) |
288 | { | |
289 | /* page_nr = (bitnr/8) >> PAGE_SHIFT; */ | |
290 | unsigned int page_nr = bitnr >> (PAGE_SHIFT + 3); | |
291 | BUG_ON(page_nr >= b->bm_number_of_pages); | |
292 | return page_nr; | |
b411b363 PR |
293 | } |
294 | ||
95a0f10c LE |
295 | static unsigned long *__bm_map_pidx(struct drbd_bitmap *b, unsigned int idx, const enum km_type km) |
296 | { | |
297 | struct page *page = b->bm_pages[idx]; | |
298 | return (unsigned long *) kmap_atomic(page, km); | |
299 | } | |
300 | ||
301 | static unsigned long *bm_map_pidx(struct drbd_bitmap *b, unsigned int idx) | |
302 | { | |
303 | return __bm_map_pidx(b, idx, KM_IRQ1); | |
304 | } | |
305 | ||
b411b363 PR |
306 | static void __bm_unmap(unsigned long *p_addr, const enum km_type km) |
307 | { | |
308 | kunmap_atomic(p_addr, km); | |
309 | }; | |
310 | ||
311 | static void bm_unmap(unsigned long *p_addr) | |
312 | { | |
313 | return __bm_unmap(p_addr, KM_IRQ1); | |
314 | } | |
315 | ||
316 | /* long word offset of _bitmap_ sector */ | |
317 | #define S2W(s) ((s)<<(BM_EXT_SHIFT-BM_BLOCK_SHIFT-LN2_BPL)) | |
318 | /* word offset from start of bitmap to word number _in_page_ | |
319 | * modulo longs per page | |
320 | #define MLPP(X) ((X) % (PAGE_SIZE/sizeof(long)) | |
24c4830c | 321 | hm, well, Philipp thinks gcc might not optimize the % into & (... - 1) |
b411b363 PR |
322 | so do it explicitly: |
323 | */ | |
324 | #define MLPP(X) ((X) & ((PAGE_SIZE/sizeof(long))-1)) | |
325 | ||
326 | /* Long words per page */ | |
327 | #define LWPP (PAGE_SIZE/sizeof(long)) | |
328 | ||
329 | /* | |
330 | * actually most functions herein should take a struct drbd_bitmap*, not a | |
331 | * struct drbd_conf*, but for the debug macros I like to have the mdev around | |
332 | * to be able to report device specific. | |
333 | */ | |
334 | ||
19f843aa | 335 | |
b411b363 PR |
336 | static void bm_free_pages(struct page **pages, unsigned long number) |
337 | { | |
338 | unsigned long i; | |
339 | if (!pages) | |
340 | return; | |
341 | ||
342 | for (i = 0; i < number; i++) { | |
343 | if (!pages[i]) { | |
344 | printk(KERN_ALERT "drbd: bm_free_pages tried to free " | |
345 | "a NULL pointer; i=%lu n=%lu\n", | |
346 | i, number); | |
347 | continue; | |
348 | } | |
349 | __free_page(pages[i]); | |
350 | pages[i] = NULL; | |
351 | } | |
352 | } | |
353 | ||
354 | static void bm_vk_free(void *ptr, int v) | |
355 | { | |
356 | if (v) | |
357 | vfree(ptr); | |
358 | else | |
359 | kfree(ptr); | |
360 | } | |
361 | ||
362 | /* | |
363 | * "have" and "want" are NUMBER OF PAGES. | |
364 | */ | |
365 | static struct page **bm_realloc_pages(struct drbd_bitmap *b, unsigned long want) | |
366 | { | |
367 | struct page **old_pages = b->bm_pages; | |
368 | struct page **new_pages, *page; | |
369 | unsigned int i, bytes, vmalloced = 0; | |
370 | unsigned long have = b->bm_number_of_pages; | |
371 | ||
372 | BUG_ON(have == 0 && old_pages != NULL); | |
373 | BUG_ON(have != 0 && old_pages == NULL); | |
374 | ||
375 | if (have == want) | |
376 | return old_pages; | |
377 | ||
378 | /* Trying kmalloc first, falling back to vmalloc. | |
379 | * GFP_KERNEL is ok, as this is done when a lower level disk is | |
380 | * "attached" to the drbd. Context is receiver thread or cqueue | |
381 | * thread. As we have no disk yet, we are not in the IO path, | |
382 | * not even the IO path of the peer. */ | |
383 | bytes = sizeof(struct page *)*want; | |
384 | new_pages = kmalloc(bytes, GFP_KERNEL); | |
385 | if (!new_pages) { | |
386 | new_pages = vmalloc(bytes); | |
387 | if (!new_pages) | |
388 | return NULL; | |
389 | vmalloced = 1; | |
390 | } | |
391 | ||
392 | memset(new_pages, 0, bytes); | |
393 | if (want >= have) { | |
394 | for (i = 0; i < have; i++) | |
395 | new_pages[i] = old_pages[i]; | |
396 | for (; i < want; i++) { | |
397 | page = alloc_page(GFP_HIGHUSER); | |
398 | if (!page) { | |
399 | bm_free_pages(new_pages + have, i - have); | |
400 | bm_vk_free(new_pages, vmalloced); | |
401 | return NULL; | |
402 | } | |
19f843aa LE |
403 | /* we want to know which page it is |
404 | * from the endio handlers */ | |
405 | bm_store_page_idx(page, i); | |
b411b363 PR |
406 | new_pages[i] = page; |
407 | } | |
408 | } else { | |
409 | for (i = 0; i < want; i++) | |
410 | new_pages[i] = old_pages[i]; | |
411 | /* NOT HERE, we are outside the spinlock! | |
412 | bm_free_pages(old_pages + want, have - want); | |
413 | */ | |
414 | } | |
415 | ||
416 | if (vmalloced) | |
20ceb2b2 | 417 | b->bm_flags |= BM_P_VMALLOCED; |
b411b363 | 418 | else |
20ceb2b2 | 419 | b->bm_flags &= ~BM_P_VMALLOCED; |
b411b363 PR |
420 | |
421 | return new_pages; | |
422 | } | |
423 | ||
424 | /* | |
425 | * called on driver init only. TODO call when a device is created. | |
426 | * allocates the drbd_bitmap, and stores it in mdev->bitmap. | |
427 | */ | |
428 | int drbd_bm_init(struct drbd_conf *mdev) | |
429 | { | |
430 | struct drbd_bitmap *b = mdev->bitmap; | |
431 | WARN_ON(b != NULL); | |
432 | b = kzalloc(sizeof(struct drbd_bitmap), GFP_KERNEL); | |
433 | if (!b) | |
434 | return -ENOMEM; | |
435 | spin_lock_init(&b->bm_lock); | |
8a03ae2a | 436 | mutex_init(&b->bm_change); |
b411b363 PR |
437 | init_waitqueue_head(&b->bm_io_wait); |
438 | ||
439 | mdev->bitmap = b; | |
440 | ||
441 | return 0; | |
442 | } | |
443 | ||
444 | sector_t drbd_bm_capacity(struct drbd_conf *mdev) | |
445 | { | |
446 | ERR_IF(!mdev->bitmap) return 0; | |
447 | return mdev->bitmap->bm_dev_capacity; | |
448 | } | |
449 | ||
450 | /* called on driver unload. TODO: call when a device is destroyed. | |
451 | */ | |
452 | void drbd_bm_cleanup(struct drbd_conf *mdev) | |
453 | { | |
454 | ERR_IF (!mdev->bitmap) return; | |
455 | bm_free_pages(mdev->bitmap->bm_pages, mdev->bitmap->bm_number_of_pages); | |
20ceb2b2 | 456 | bm_vk_free(mdev->bitmap->bm_pages, (BM_P_VMALLOCED & mdev->bitmap->bm_flags)); |
b411b363 PR |
457 | kfree(mdev->bitmap); |
458 | mdev->bitmap = NULL; | |
459 | } | |
460 | ||
461 | /* | |
462 | * since (b->bm_bits % BITS_PER_LONG) != 0, | |
463 | * this masks out the remaining bits. | |
464 | * Returns the number of bits cleared. | |
465 | */ | |
95a0f10c LE |
466 | #define BITS_PER_PAGE (1UL << (PAGE_SHIFT + 3)) |
467 | #define BITS_PER_PAGE_MASK (BITS_PER_PAGE - 1) | |
468 | #define BITS_PER_LONG_MASK (BITS_PER_LONG - 1) | |
b411b363 PR |
469 | static int bm_clear_surplus(struct drbd_bitmap *b) |
470 | { | |
95a0f10c | 471 | unsigned long mask; |
b411b363 | 472 | unsigned long *p_addr, *bm; |
95a0f10c LE |
473 | int tmp; |
474 | int cleared = 0; | |
b411b363 | 475 | |
95a0f10c LE |
476 | /* number of bits modulo bits per page */ |
477 | tmp = (b->bm_bits & BITS_PER_PAGE_MASK); | |
478 | /* mask the used bits of the word containing the last bit */ | |
479 | mask = (1UL << (tmp & BITS_PER_LONG_MASK)) -1; | |
480 | /* bitmap is always stored little endian, | |
481 | * on disk and in core memory alike */ | |
482 | mask = cpu_to_lel(mask); | |
483 | ||
6850c442 | 484 | p_addr = bm_map_pidx(b, b->bm_number_of_pages - 1); |
95a0f10c LE |
485 | bm = p_addr + (tmp/BITS_PER_LONG); |
486 | if (mask) { | |
487 | /* If mask != 0, we are not exactly aligned, so bm now points | |
488 | * to the long containing the last bit. | |
489 | * If mask == 0, bm already points to the word immediately | |
490 | * after the last (long word aligned) bit. */ | |
b411b363 PR |
491 | cleared = hweight_long(*bm & ~mask); |
492 | *bm &= mask; | |
95a0f10c | 493 | bm++; |
b411b363 PR |
494 | } |
495 | ||
95a0f10c LE |
496 | if (BITS_PER_LONG == 32 && ((bm - p_addr) & 1) == 1) { |
497 | /* on a 32bit arch, we may need to zero out | |
498 | * a padding long to align with a 64bit remote */ | |
b411b363 PR |
499 | cleared += hweight_long(*bm); |
500 | *bm = 0; | |
501 | } | |
502 | bm_unmap(p_addr); | |
503 | return cleared; | |
504 | } | |
505 | ||
506 | static void bm_set_surplus(struct drbd_bitmap *b) | |
507 | { | |
95a0f10c | 508 | unsigned long mask; |
b411b363 | 509 | unsigned long *p_addr, *bm; |
95a0f10c LE |
510 | int tmp; |
511 | ||
512 | /* number of bits modulo bits per page */ | |
513 | tmp = (b->bm_bits & BITS_PER_PAGE_MASK); | |
514 | /* mask the used bits of the word containing the last bit */ | |
515 | mask = (1UL << (tmp & BITS_PER_LONG_MASK)) -1; | |
516 | /* bitmap is always stored little endian, | |
517 | * on disk and in core memory alike */ | |
518 | mask = cpu_to_lel(mask); | |
519 | ||
6850c442 | 520 | p_addr = bm_map_pidx(b, b->bm_number_of_pages - 1); |
95a0f10c LE |
521 | bm = p_addr + (tmp/BITS_PER_LONG); |
522 | if (mask) { | |
523 | /* If mask != 0, we are not exactly aligned, so bm now points | |
524 | * to the long containing the last bit. | |
525 | * If mask == 0, bm already points to the word immediately | |
526 | * after the last (long word aligned) bit. */ | |
b411b363 | 527 | *bm |= ~mask; |
95a0f10c | 528 | bm++; |
b411b363 PR |
529 | } |
530 | ||
95a0f10c LE |
531 | if (BITS_PER_LONG == 32 && ((bm - p_addr) & 1) == 1) { |
532 | /* on a 32bit arch, we may need to zero out | |
533 | * a padding long to align with a 64bit remote */ | |
534 | *bm = ~0UL; | |
b411b363 PR |
535 | } |
536 | bm_unmap(p_addr); | |
537 | } | |
538 | ||
4b0715f0 LE |
539 | /* you better not modify the bitmap while this is running, |
540 | * or its results will be stale */ | |
95a0f10c | 541 | static unsigned long bm_count_bits(struct drbd_bitmap *b) |
b411b363 | 542 | { |
4b0715f0 | 543 | unsigned long *p_addr; |
b411b363 | 544 | unsigned long bits = 0; |
4b0715f0 | 545 | unsigned long mask = (1UL << (b->bm_bits & BITS_PER_LONG_MASK)) -1; |
6850c442 | 546 | int idx, i, last_word; |
4b0715f0 LE |
547 | |
548 | /* all but last page */ | |
6850c442 | 549 | for (idx = 0; idx < b->bm_number_of_pages - 1; idx++) { |
4b0715f0 LE |
550 | p_addr = __bm_map_pidx(b, idx, KM_USER0); |
551 | for (i = 0; i < LWPP; i++) | |
552 | bits += hweight_long(p_addr[i]); | |
7777a8ba | 553 | __bm_unmap(p_addr, KM_USER0); |
b411b363 PR |
554 | cond_resched(); |
555 | } | |
4b0715f0 LE |
556 | /* last (or only) page */ |
557 | last_word = ((b->bm_bits - 1) & BITS_PER_PAGE_MASK) >> LN2_BPL; | |
558 | p_addr = __bm_map_pidx(b, idx, KM_USER0); | |
559 | for (i = 0; i < last_word; i++) | |
560 | bits += hweight_long(p_addr[i]); | |
561 | p_addr[last_word] &= cpu_to_lel(mask); | |
562 | bits += hweight_long(p_addr[last_word]); | |
563 | /* 32bit arch, may have an unused padding long */ | |
564 | if (BITS_PER_LONG == 32 && (last_word & 1) == 0) | |
565 | p_addr[last_word+1] = 0; | |
566 | __bm_unmap(p_addr, KM_USER0); | |
b411b363 PR |
567 | return bits; |
568 | } | |
569 | ||
b411b363 PR |
570 | /* offset and len in long words.*/ |
571 | static void bm_memset(struct drbd_bitmap *b, size_t offset, int c, size_t len) | |
572 | { | |
573 | unsigned long *p_addr, *bm; | |
19f843aa | 574 | unsigned int idx; |
b411b363 PR |
575 | size_t do_now, end; |
576 | ||
b411b363 PR |
577 | end = offset + len; |
578 | ||
579 | if (end > b->bm_words) { | |
580 | printk(KERN_ALERT "drbd: bm_memset end > bm_words\n"); | |
581 | return; | |
582 | } | |
583 | ||
584 | while (offset < end) { | |
585 | do_now = min_t(size_t, ALIGN(offset + 1, LWPP), end) - offset; | |
19f843aa LE |
586 | idx = bm_word_to_page_idx(b, offset); |
587 | p_addr = bm_map_pidx(b, idx); | |
b411b363 PR |
588 | bm = p_addr + MLPP(offset); |
589 | if (bm+do_now > p_addr + LWPP) { | |
590 | printk(KERN_ALERT "drbd: BUG BUG BUG! p_addr:%p bm:%p do_now:%d\n", | |
591 | p_addr, bm, (int)do_now); | |
84e7c0f7 LE |
592 | } else |
593 | memset(bm, c, do_now * sizeof(long)); | |
b411b363 | 594 | bm_unmap(p_addr); |
19f843aa | 595 | bm_set_page_need_writeout(b->bm_pages[idx]); |
b411b363 PR |
596 | offset += do_now; |
597 | } | |
598 | } | |
599 | ||
600 | /* | |
601 | * make sure the bitmap has enough room for the attached storage, | |
602 | * if necessary, resize. | |
603 | * called whenever we may have changed the device size. | |
604 | * returns -ENOMEM if we could not allocate enough memory, 0 on success. | |
605 | * In case this is actually a resize, we copy the old bitmap into the new one. | |
606 | * Otherwise, the bitmap is initialized to all bits set. | |
607 | */ | |
02d9a94b | 608 | int drbd_bm_resize(struct drbd_conf *mdev, sector_t capacity, int set_new_bits) |
b411b363 PR |
609 | { |
610 | struct drbd_bitmap *b = mdev->bitmap; | |
6850c442 | 611 | unsigned long bits, words, owords, obits; |
b411b363 PR |
612 | unsigned long want, have, onpages; /* number of pages */ |
613 | struct page **npages, **opages = NULL; | |
614 | int err = 0, growing; | |
615 | int opages_vmalloced; | |
616 | ||
617 | ERR_IF(!b) return -ENOMEM; | |
618 | ||
20ceb2b2 | 619 | drbd_bm_lock(mdev, "resize", BM_LOCKED_MASK); |
b411b363 PR |
620 | |
621 | dev_info(DEV, "drbd_bm_resize called with capacity == %llu\n", | |
622 | (unsigned long long)capacity); | |
623 | ||
624 | if (capacity == b->bm_dev_capacity) | |
625 | goto out; | |
626 | ||
20ceb2b2 | 627 | opages_vmalloced = (BM_P_VMALLOCED & b->bm_flags); |
b411b363 PR |
628 | |
629 | if (capacity == 0) { | |
630 | spin_lock_irq(&b->bm_lock); | |
631 | opages = b->bm_pages; | |
632 | onpages = b->bm_number_of_pages; | |
633 | owords = b->bm_words; | |
634 | b->bm_pages = NULL; | |
635 | b->bm_number_of_pages = | |
636 | b->bm_set = | |
637 | b->bm_bits = | |
638 | b->bm_words = | |
639 | b->bm_dev_capacity = 0; | |
640 | spin_unlock_irq(&b->bm_lock); | |
641 | bm_free_pages(opages, onpages); | |
642 | bm_vk_free(opages, opages_vmalloced); | |
643 | goto out; | |
644 | } | |
645 | bits = BM_SECT_TO_BIT(ALIGN(capacity, BM_SECT_PER_BIT)); | |
646 | ||
647 | /* if we would use | |
648 | words = ALIGN(bits,BITS_PER_LONG) >> LN2_BPL; | |
649 | a 32bit host could present the wrong number of words | |
650 | to a 64bit host. | |
651 | */ | |
652 | words = ALIGN(bits, 64) >> LN2_BPL; | |
653 | ||
654 | if (get_ldev(mdev)) { | |
4b0715f0 | 655 | u64 bits_on_disk = ((u64)mdev->ldev->md.md_size_sect-MD_BM_OFFSET) << 12; |
b411b363 | 656 | put_ldev(mdev); |
4b0715f0 LE |
657 | if (bits > bits_on_disk) { |
658 | dev_info(DEV, "bits = %lu\n", bits); | |
659 | dev_info(DEV, "bits_on_disk = %llu\n", bits_on_disk); | |
660 | err = -ENOSPC; | |
661 | goto out; | |
662 | } | |
b411b363 PR |
663 | } |
664 | ||
6850c442 | 665 | want = ALIGN(words*sizeof(long), PAGE_SIZE) >> PAGE_SHIFT; |
b411b363 PR |
666 | have = b->bm_number_of_pages; |
667 | if (want == have) { | |
668 | D_ASSERT(b->bm_pages != NULL); | |
669 | npages = b->bm_pages; | |
670 | } else { | |
0cf9d27e | 671 | if (drbd_insert_fault(mdev, DRBD_FAULT_BM_ALLOC)) |
b411b363 PR |
672 | npages = NULL; |
673 | else | |
674 | npages = bm_realloc_pages(b, want); | |
675 | } | |
676 | ||
677 | if (!npages) { | |
678 | err = -ENOMEM; | |
679 | goto out; | |
680 | } | |
681 | ||
682 | spin_lock_irq(&b->bm_lock); | |
683 | opages = b->bm_pages; | |
684 | owords = b->bm_words; | |
685 | obits = b->bm_bits; | |
686 | ||
687 | growing = bits > obits; | |
5223671b | 688 | if (opages && growing && set_new_bits) |
b411b363 PR |
689 | bm_set_surplus(b); |
690 | ||
691 | b->bm_pages = npages; | |
692 | b->bm_number_of_pages = want; | |
693 | b->bm_bits = bits; | |
694 | b->bm_words = words; | |
695 | b->bm_dev_capacity = capacity; | |
696 | ||
697 | if (growing) { | |
02d9a94b PR |
698 | if (set_new_bits) { |
699 | bm_memset(b, owords, 0xff, words-owords); | |
700 | b->bm_set += bits - obits; | |
701 | } else | |
702 | bm_memset(b, owords, 0x00, words-owords); | |
703 | ||
b411b363 PR |
704 | } |
705 | ||
706 | if (want < have) { | |
707 | /* implicit: (opages != NULL) && (opages != npages) */ | |
708 | bm_free_pages(opages + want, have - want); | |
709 | } | |
710 | ||
b411b363 PR |
711 | (void)bm_clear_surplus(b); |
712 | ||
713 | spin_unlock_irq(&b->bm_lock); | |
714 | if (opages != npages) | |
715 | bm_vk_free(opages, opages_vmalloced); | |
716 | if (!growing) | |
717 | b->bm_set = bm_count_bits(b); | |
19f843aa | 718 | dev_info(DEV, "resync bitmap: bits=%lu words=%lu pages=%lu\n", bits, words, want); |
b411b363 PR |
719 | |
720 | out: | |
721 | drbd_bm_unlock(mdev); | |
722 | return err; | |
723 | } | |
724 | ||
725 | /* inherently racy: | |
726 | * if not protected by other means, return value may be out of date when | |
727 | * leaving this function... | |
728 | * we still need to lock it, since it is important that this returns | |
729 | * bm_set == 0 precisely. | |
730 | * | |
731 | * maybe bm_set should be atomic_t ? | |
732 | */ | |
0778286a | 733 | unsigned long _drbd_bm_total_weight(struct drbd_conf *mdev) |
b411b363 PR |
734 | { |
735 | struct drbd_bitmap *b = mdev->bitmap; | |
736 | unsigned long s; | |
737 | unsigned long flags; | |
738 | ||
739 | ERR_IF(!b) return 0; | |
740 | ERR_IF(!b->bm_pages) return 0; | |
741 | ||
742 | spin_lock_irqsave(&b->bm_lock, flags); | |
743 | s = b->bm_set; | |
744 | spin_unlock_irqrestore(&b->bm_lock, flags); | |
745 | ||
746 | return s; | |
747 | } | |
748 | ||
749 | unsigned long drbd_bm_total_weight(struct drbd_conf *mdev) | |
750 | { | |
751 | unsigned long s; | |
752 | /* if I don't have a disk, I don't know about out-of-sync status */ | |
753 | if (!get_ldev_if_state(mdev, D_NEGOTIATING)) | |
754 | return 0; | |
755 | s = _drbd_bm_total_weight(mdev); | |
756 | put_ldev(mdev); | |
757 | return s; | |
758 | } | |
759 | ||
760 | size_t drbd_bm_words(struct drbd_conf *mdev) | |
761 | { | |
762 | struct drbd_bitmap *b = mdev->bitmap; | |
763 | ERR_IF(!b) return 0; | |
764 | ERR_IF(!b->bm_pages) return 0; | |
765 | ||
766 | return b->bm_words; | |
767 | } | |
768 | ||
769 | unsigned long drbd_bm_bits(struct drbd_conf *mdev) | |
770 | { | |
771 | struct drbd_bitmap *b = mdev->bitmap; | |
772 | ERR_IF(!b) return 0; | |
773 | ||
774 | return b->bm_bits; | |
775 | } | |
776 | ||
777 | /* merge number words from buffer into the bitmap starting at offset. | |
778 | * buffer[i] is expected to be little endian unsigned long. | |
779 | * bitmap must be locked by drbd_bm_lock. | |
780 | * currently only used from receive_bitmap. | |
781 | */ | |
782 | void drbd_bm_merge_lel(struct drbd_conf *mdev, size_t offset, size_t number, | |
783 | unsigned long *buffer) | |
784 | { | |
785 | struct drbd_bitmap *b = mdev->bitmap; | |
786 | unsigned long *p_addr, *bm; | |
787 | unsigned long word, bits; | |
19f843aa | 788 | unsigned int idx; |
b411b363 PR |
789 | size_t end, do_now; |
790 | ||
791 | end = offset + number; | |
792 | ||
793 | ERR_IF(!b) return; | |
794 | ERR_IF(!b->bm_pages) return; | |
795 | if (number == 0) | |
796 | return; | |
797 | WARN_ON(offset >= b->bm_words); | |
798 | WARN_ON(end > b->bm_words); | |
799 | ||
800 | spin_lock_irq(&b->bm_lock); | |
801 | while (offset < end) { | |
802 | do_now = min_t(size_t, ALIGN(offset+1, LWPP), end) - offset; | |
19f843aa LE |
803 | idx = bm_word_to_page_idx(b, offset); |
804 | p_addr = bm_map_pidx(b, idx); | |
b411b363 PR |
805 | bm = p_addr + MLPP(offset); |
806 | offset += do_now; | |
807 | while (do_now--) { | |
808 | bits = hweight_long(*bm); | |
95a0f10c | 809 | word = *bm | *buffer++; |
b411b363 PR |
810 | *bm++ = word; |
811 | b->bm_set += hweight_long(word) - bits; | |
812 | } | |
813 | bm_unmap(p_addr); | |
19f843aa | 814 | bm_set_page_need_writeout(b->bm_pages[idx]); |
b411b363 PR |
815 | } |
816 | /* with 32bit <-> 64bit cross-platform connect | |
817 | * this is only correct for current usage, | |
818 | * where we _know_ that we are 64 bit aligned, | |
819 | * and know that this function is used in this way, too... | |
820 | */ | |
821 | if (end == b->bm_words) | |
822 | b->bm_set -= bm_clear_surplus(b); | |
b411b363 PR |
823 | spin_unlock_irq(&b->bm_lock); |
824 | } | |
825 | ||
826 | /* copy number words from the bitmap starting at offset into the buffer. | |
827 | * buffer[i] will be little endian unsigned long. | |
828 | */ | |
829 | void drbd_bm_get_lel(struct drbd_conf *mdev, size_t offset, size_t number, | |
830 | unsigned long *buffer) | |
831 | { | |
832 | struct drbd_bitmap *b = mdev->bitmap; | |
833 | unsigned long *p_addr, *bm; | |
834 | size_t end, do_now; | |
835 | ||
836 | end = offset + number; | |
837 | ||
838 | ERR_IF(!b) return; | |
839 | ERR_IF(!b->bm_pages) return; | |
840 | ||
841 | spin_lock_irq(&b->bm_lock); | |
842 | if ((offset >= b->bm_words) || | |
843 | (end > b->bm_words) || | |
844 | (number <= 0)) | |
845 | dev_err(DEV, "offset=%lu number=%lu bm_words=%lu\n", | |
846 | (unsigned long) offset, | |
847 | (unsigned long) number, | |
848 | (unsigned long) b->bm_words); | |
849 | else { | |
850 | while (offset < end) { | |
851 | do_now = min_t(size_t, ALIGN(offset+1, LWPP), end) - offset; | |
19f843aa | 852 | p_addr = bm_map_pidx(b, bm_word_to_page_idx(b, offset)); |
b411b363 PR |
853 | bm = p_addr + MLPP(offset); |
854 | offset += do_now; | |
855 | while (do_now--) | |
95a0f10c | 856 | *buffer++ = *bm++; |
b411b363 PR |
857 | bm_unmap(p_addr); |
858 | } | |
859 | } | |
860 | spin_unlock_irq(&b->bm_lock); | |
861 | } | |
862 | ||
863 | /* set all bits in the bitmap */ | |
864 | void drbd_bm_set_all(struct drbd_conf *mdev) | |
865 | { | |
866 | struct drbd_bitmap *b = mdev->bitmap; | |
867 | ERR_IF(!b) return; | |
868 | ERR_IF(!b->bm_pages) return; | |
869 | ||
870 | spin_lock_irq(&b->bm_lock); | |
871 | bm_memset(b, 0, 0xff, b->bm_words); | |
872 | (void)bm_clear_surplus(b); | |
873 | b->bm_set = b->bm_bits; | |
874 | spin_unlock_irq(&b->bm_lock); | |
875 | } | |
876 | ||
877 | /* clear all bits in the bitmap */ | |
878 | void drbd_bm_clear_all(struct drbd_conf *mdev) | |
879 | { | |
880 | struct drbd_bitmap *b = mdev->bitmap; | |
881 | ERR_IF(!b) return; | |
882 | ERR_IF(!b->bm_pages) return; | |
883 | ||
884 | spin_lock_irq(&b->bm_lock); | |
885 | bm_memset(b, 0, 0, b->bm_words); | |
886 | b->bm_set = 0; | |
887 | spin_unlock_irq(&b->bm_lock); | |
888 | } | |
889 | ||
19f843aa LE |
890 | struct bm_aio_ctx { |
891 | struct drbd_conf *mdev; | |
892 | atomic_t in_flight; | |
725a97e4 | 893 | struct completion done; |
19f843aa LE |
894 | unsigned flags; |
895 | #define BM_AIO_COPY_PAGES 1 | |
896 | int error; | |
897 | }; | |
898 | ||
899 | /* bv_page may be a copy, or may be the original */ | |
b411b363 PR |
900 | static void bm_async_io_complete(struct bio *bio, int error) |
901 | { | |
19f843aa LE |
902 | struct bm_aio_ctx *ctx = bio->bi_private; |
903 | struct drbd_conf *mdev = ctx->mdev; | |
904 | struct drbd_bitmap *b = mdev->bitmap; | |
905 | unsigned int idx = bm_page_to_idx(bio->bi_io_vec[0].bv_page); | |
b411b363 PR |
906 | int uptodate = bio_flagged(bio, BIO_UPTODATE); |
907 | ||
908 | ||
909 | /* strange behavior of some lower level drivers... | |
910 | * fail the request by clearing the uptodate flag, | |
911 | * but do not return any error?! | |
912 | * do we want to WARN() on this? */ | |
913 | if (!error && !uptodate) | |
914 | error = -EIO; | |
915 | ||
7648cdfe LE |
916 | if ((ctx->flags & BM_AIO_COPY_PAGES) == 0 && |
917 | !bm_test_page_unchanged(b->bm_pages[idx])) | |
918 | dev_warn(DEV, "bitmap page idx %u changed during IO!\n", idx); | |
19f843aa | 919 | |
b411b363 | 920 | if (error) { |
19f843aa LE |
921 | /* ctx error will hold the completed-last non-zero error code, |
922 | * in case error codes differ. */ | |
923 | ctx->error = error; | |
924 | bm_set_page_io_err(b->bm_pages[idx]); | |
925 | /* Not identical to on disk version of it. | |
926 | * Is BM_PAGE_IO_ERROR enough? */ | |
927 | if (__ratelimit(&drbd_ratelimit_state)) | |
928 | dev_err(DEV, "IO ERROR %d on bitmap page idx %u\n", | |
929 | error, idx); | |
930 | } else { | |
931 | bm_clear_page_io_err(b->bm_pages[idx]); | |
932 | dynamic_dev_dbg(DEV, "bitmap page idx %u completed\n", idx); | |
b411b363 | 933 | } |
19f843aa LE |
934 | |
935 | bm_page_unlock_io(mdev, idx); | |
936 | ||
937 | /* FIXME give back to page pool */ | |
938 | if (ctx->flags & BM_AIO_COPY_PAGES) | |
939 | put_page(bio->bi_io_vec[0].bv_page); | |
b411b363 PR |
940 | |
941 | bio_put(bio); | |
19f843aa LE |
942 | |
943 | if (atomic_dec_and_test(&ctx->in_flight)) | |
725a97e4 | 944 | complete(&ctx->done); |
b411b363 PR |
945 | } |
946 | ||
19f843aa | 947 | static void bm_page_io_async(struct bm_aio_ctx *ctx, int page_nr, int rw) __must_hold(local) |
b411b363 PR |
948 | { |
949 | /* we are process context. we always get a bio */ | |
950 | struct bio *bio = bio_alloc(GFP_KERNEL, 1); | |
19f843aa LE |
951 | struct drbd_conf *mdev = ctx->mdev; |
952 | struct drbd_bitmap *b = mdev->bitmap; | |
953 | struct page *page; | |
b411b363 | 954 | unsigned int len; |
19f843aa | 955 | |
b411b363 PR |
956 | sector_t on_disk_sector = |
957 | mdev->ldev->md.md_offset + mdev->ldev->md.bm_offset; | |
958 | on_disk_sector += ((sector_t)page_nr) << (PAGE_SHIFT-9); | |
959 | ||
960 | /* this might happen with very small | |
19f843aa LE |
961 | * flexible external meta data device, |
962 | * or with PAGE_SIZE > 4k */ | |
b411b363 PR |
963 | len = min_t(unsigned int, PAGE_SIZE, |
964 | (drbd_md_last_sector(mdev->ldev) - on_disk_sector + 1)<<9); | |
965 | ||
19f843aa LE |
966 | /* serialize IO on this page */ |
967 | bm_page_lock_io(mdev, page_nr); | |
968 | /* before memcpy and submit, | |
969 | * so it can be redirtied any time */ | |
970 | bm_set_page_unchanged(b->bm_pages[page_nr]); | |
971 | ||
972 | if (ctx->flags & BM_AIO_COPY_PAGES) { | |
973 | /* FIXME alloc_page is good enough for now, but actually needs | |
974 | * to use pre-allocated page pool */ | |
975 | void *src, *dest; | |
976 | page = alloc_page(__GFP_HIGHMEM|__GFP_WAIT); | |
977 | dest = kmap_atomic(page, KM_USER0); | |
978 | src = kmap_atomic(b->bm_pages[page_nr], KM_USER1); | |
979 | memcpy(dest, src, PAGE_SIZE); | |
980 | kunmap_atomic(src, KM_USER1); | |
981 | kunmap_atomic(dest, KM_USER0); | |
982 | bm_store_page_idx(page, page_nr); | |
983 | } else | |
984 | page = b->bm_pages[page_nr]; | |
985 | ||
b411b363 PR |
986 | bio->bi_bdev = mdev->ldev->md_bdev; |
987 | bio->bi_sector = on_disk_sector; | |
19f843aa LE |
988 | bio_add_page(bio, page, len, 0); |
989 | bio->bi_private = ctx; | |
b411b363 PR |
990 | bio->bi_end_io = bm_async_io_complete; |
991 | ||
0cf9d27e | 992 | if (drbd_insert_fault(mdev, (rw & WRITE) ? DRBD_FAULT_MD_WR : DRBD_FAULT_MD_RD)) { |
b411b363 PR |
993 | bio->bi_rw |= rw; |
994 | bio_endio(bio, -EIO); | |
995 | } else { | |
996 | submit_bio(rw, bio); | |
997 | } | |
998 | } | |
999 | ||
b411b363 PR |
1000 | /* |
1001 | * bm_rw: read/write the whole bitmap from/to its on disk location. | |
1002 | */ | |
19f843aa | 1003 | static int bm_rw(struct drbd_conf *mdev, int rw, unsigned lazy_writeout_upper_idx) __must_hold(local) |
b411b363 | 1004 | { |
725a97e4 LE |
1005 | struct bm_aio_ctx ctx = { |
1006 | .mdev = mdev, | |
1007 | .in_flight = ATOMIC_INIT(1), | |
1008 | .done = COMPLETION_INITIALIZER_ONSTACK(ctx.done), | |
1009 | .flags = lazy_writeout_upper_idx ? BM_AIO_COPY_PAGES : 0, | |
1010 | }; | |
b411b363 | 1011 | struct drbd_bitmap *b = mdev->bitmap; |
6850c442 | 1012 | int num_pages, i, count = 0; |
b411b363 PR |
1013 | unsigned long now; |
1014 | char ppb[10]; | |
1015 | int err = 0; | |
1016 | ||
19f843aa LE |
1017 | /* |
1018 | * We are protected against bitmap disappearing/resizing by holding an | |
1019 | * ldev reference (caller must have called get_ldev()). | |
1020 | * For read/write, we are protected against changes to the bitmap by | |
1021 | * the bitmap lock (see drbd_bitmap_io). | |
1022 | * For lazy writeout, we don't care for ongoing changes to the bitmap, | |
1023 | * as we submit copies of pages anyways. | |
1024 | */ | |
1025 | if (!ctx.flags) | |
20ceb2b2 | 1026 | WARN_ON(!(BM_LOCKED_MASK & b->bm_flags)); |
b411b363 | 1027 | |
6850c442 | 1028 | num_pages = b->bm_number_of_pages; |
b411b363 | 1029 | |
b411b363 | 1030 | now = jiffies; |
b411b363 PR |
1031 | |
1032 | /* let the layers below us try to merge these bios... */ | |
6850c442 | 1033 | for (i = 0; i < num_pages; i++) { |
19f843aa LE |
1034 | /* ignore completely unchanged pages */ |
1035 | if (lazy_writeout_upper_idx && i == lazy_writeout_upper_idx) | |
1036 | break; | |
1037 | if (rw & WRITE) { | |
1038 | if (bm_test_page_unchanged(b->bm_pages[i])) { | |
1039 | dynamic_dev_dbg(DEV, "skipped bm write for idx %u\n", i); | |
1040 | continue; | |
1041 | } | |
1042 | /* during lazy writeout, | |
1043 | * ignore those pages not marked for lazy writeout. */ | |
1044 | if (lazy_writeout_upper_idx && | |
1045 | !bm_test_page_lazy_writeout(b->bm_pages[i])) { | |
1046 | dynamic_dev_dbg(DEV, "skipped bm lazy write for idx %u\n", i); | |
1047 | continue; | |
1048 | } | |
1049 | } | |
1050 | atomic_inc(&ctx.in_flight); | |
1051 | bm_page_io_async(&ctx, i, rw); | |
1052 | ++count; | |
1053 | cond_resched(); | |
1054 | } | |
b411b363 | 1055 | |
725a97e4 LE |
1056 | /* |
1057 | * We initialize ctx.in_flight to one to make sure bm_async_io_complete | |
1058 | * will not complete() early, and decrement / test it here. If there | |
1059 | * are still some bios in flight, we need to wait for them here. | |
1060 | */ | |
1061 | if (!atomic_dec_and_test(&ctx.in_flight)) | |
1062 | wait_for_completion(&ctx.done); | |
19f843aa LE |
1063 | dev_info(DEV, "bitmap %s of %u pages took %lu jiffies\n", |
1064 | rw == WRITE ? "WRITE" : "READ", | |
1065 | count, jiffies - now); | |
b411b363 | 1066 | |
19f843aa | 1067 | if (ctx.error) { |
b411b363 | 1068 | dev_alert(DEV, "we had at least one MD IO ERROR during bitmap IO\n"); |
81e84650 | 1069 | drbd_chk_io_error(mdev, 1, true); |
19f843aa | 1070 | err = -EIO; /* ctx.error ? */ |
b411b363 PR |
1071 | } |
1072 | ||
1073 | now = jiffies; | |
1074 | if (rw == WRITE) { | |
b411b363 PR |
1075 | drbd_md_flush(mdev); |
1076 | } else /* rw == READ */ { | |
95a0f10c | 1077 | b->bm_set = bm_count_bits(b); |
b411b363 PR |
1078 | dev_info(DEV, "recounting of set bits took additional %lu jiffies\n", |
1079 | jiffies - now); | |
1080 | } | |
1081 | now = b->bm_set; | |
1082 | ||
1083 | dev_info(DEV, "%s (%lu bits) marked out-of-sync by on disk bit-map.\n", | |
1084 | ppsize(ppb, now << (BM_BLOCK_SHIFT-10)), now); | |
1085 | ||
1086 | return err; | |
1087 | } | |
1088 | ||
1089 | /** | |
1090 | * drbd_bm_read() - Read the whole bitmap from its on disk location. | |
1091 | * @mdev: DRBD device. | |
1092 | */ | |
1093 | int drbd_bm_read(struct drbd_conf *mdev) __must_hold(local) | |
1094 | { | |
19f843aa | 1095 | return bm_rw(mdev, READ, 0); |
b411b363 PR |
1096 | } |
1097 | ||
1098 | /** | |
1099 | * drbd_bm_write() - Write the whole bitmap to its on disk location. | |
1100 | * @mdev: DRBD device. | |
19f843aa LE |
1101 | * |
1102 | * Will only write pages that have changed since last IO. | |
b411b363 PR |
1103 | */ |
1104 | int drbd_bm_write(struct drbd_conf *mdev) __must_hold(local) | |
1105 | { | |
19f843aa | 1106 | return bm_rw(mdev, WRITE, 0); |
b411b363 PR |
1107 | } |
1108 | ||
1109 | /** | |
19f843aa | 1110 | * drbd_bm_lazy_write_out() - Write bitmap pages 0 to @upper_idx-1, if they have changed. |
b411b363 | 1111 | * @mdev: DRBD device. |
19f843aa LE |
1112 | * @upper_idx: 0: write all changed pages; +ve: page index to stop scanning for changed pages |
1113 | */ | |
1114 | int drbd_bm_write_lazy(struct drbd_conf *mdev, unsigned upper_idx) __must_hold(local) | |
1115 | { | |
1116 | return bm_rw(mdev, WRITE, upper_idx); | |
1117 | } | |
1118 | ||
1119 | ||
1120 | /** | |
1121 | * drbd_bm_write_page: Writes a PAGE_SIZE aligned piece of bitmap | |
1122 | * @mdev: DRBD device. | |
1123 | * @idx: bitmap page index | |
b411b363 | 1124 | * |
4b0715f0 LE |
1125 | * We don't want to special case on logical_block_size of the backend device, |
1126 | * so we submit PAGE_SIZE aligned pieces. | |
19f843aa | 1127 | * Note that on "most" systems, PAGE_SIZE is 4k. |
4b0715f0 LE |
1128 | * |
1129 | * In case this becomes an issue on systems with larger PAGE_SIZE, | |
1130 | * we may want to change this again to write 4k aligned 4k pieces. | |
b411b363 | 1131 | */ |
19f843aa | 1132 | int drbd_bm_write_page(struct drbd_conf *mdev, unsigned int idx) __must_hold(local) |
b411b363 | 1133 | { |
725a97e4 LE |
1134 | struct bm_aio_ctx ctx = { |
1135 | .mdev = mdev, | |
1136 | .in_flight = ATOMIC_INIT(1), | |
1137 | .done = COMPLETION_INITIALIZER_ONSTACK(ctx.done), | |
1138 | .flags = BM_AIO_COPY_PAGES, | |
1139 | }; | |
b411b363 | 1140 | |
19f843aa | 1141 | if (bm_test_page_unchanged(mdev->bitmap->bm_pages[idx])) { |
7648cdfe | 1142 | dynamic_dev_dbg(DEV, "skipped bm page write for idx %u\n", idx); |
19f843aa | 1143 | return 0; |
b411b363 | 1144 | } |
19f843aa | 1145 | |
19f843aa | 1146 | bm_page_io_async(&ctx, idx, WRITE_SYNC); |
725a97e4 | 1147 | wait_for_completion(&ctx.done); |
19f843aa LE |
1148 | |
1149 | if (ctx.error) | |
1150 | drbd_chk_io_error(mdev, 1, true); | |
1151 | /* that should force detach, so the in memory bitmap will be | |
1152 | * gone in a moment as well. */ | |
1153 | ||
b411b363 | 1154 | mdev->bm_writ_cnt++; |
19f843aa | 1155 | return ctx.error; |
b411b363 PR |
1156 | } |
1157 | ||
1158 | /* NOTE | |
1159 | * find_first_bit returns int, we return unsigned long. | |
4b0715f0 LE |
1160 | * For this to work on 32bit arch with bitnumbers > (1<<32), |
1161 | * we'd need to return u64, and get a whole lot of other places | |
1162 | * fixed where we still use unsigned long. | |
b411b363 PR |
1163 | * |
1164 | * this returns a bit number, NOT a sector! | |
1165 | */ | |
b411b363 PR |
1166 | static unsigned long __bm_find_next(struct drbd_conf *mdev, unsigned long bm_fo, |
1167 | const int find_zero_bit, const enum km_type km) | |
1168 | { | |
1169 | struct drbd_bitmap *b = mdev->bitmap; | |
b411b363 | 1170 | unsigned long *p_addr; |
4b0715f0 LE |
1171 | unsigned long bit_offset; |
1172 | unsigned i; | |
1173 | ||
b411b363 PR |
1174 | |
1175 | if (bm_fo > b->bm_bits) { | |
1176 | dev_err(DEV, "bm_fo=%lu bm_bits=%lu\n", bm_fo, b->bm_bits); | |
4b0715f0 | 1177 | bm_fo = DRBD_END_OF_BITMAP; |
b411b363 PR |
1178 | } else { |
1179 | while (bm_fo < b->bm_bits) { | |
19f843aa | 1180 | /* bit offset of the first bit in the page */ |
4b0715f0 | 1181 | bit_offset = bm_fo & ~BITS_PER_PAGE_MASK; |
19f843aa | 1182 | p_addr = __bm_map_pidx(b, bm_bit_to_page_idx(b, bm_fo), km); |
b411b363 PR |
1183 | |
1184 | if (find_zero_bit) | |
7e599e6e | 1185 | i = find_next_zero_bit_le(p_addr, |
4b0715f0 | 1186 | PAGE_SIZE*8, bm_fo & BITS_PER_PAGE_MASK); |
b411b363 | 1187 | else |
7e599e6e | 1188 | i = find_next_bit_le(p_addr, |
4b0715f0 | 1189 | PAGE_SIZE*8, bm_fo & BITS_PER_PAGE_MASK); |
b411b363 PR |
1190 | |
1191 | __bm_unmap(p_addr, km); | |
1192 | if (i < PAGE_SIZE*8) { | |
4b0715f0 LE |
1193 | bm_fo = bit_offset + i; |
1194 | if (bm_fo >= b->bm_bits) | |
b411b363 PR |
1195 | break; |
1196 | goto found; | |
1197 | } | |
1198 | bm_fo = bit_offset + PAGE_SIZE*8; | |
1199 | } | |
4b0715f0 | 1200 | bm_fo = DRBD_END_OF_BITMAP; |
b411b363 PR |
1201 | } |
1202 | found: | |
4b0715f0 | 1203 | return bm_fo; |
b411b363 PR |
1204 | } |
1205 | ||
1206 | static unsigned long bm_find_next(struct drbd_conf *mdev, | |
1207 | unsigned long bm_fo, const int find_zero_bit) | |
1208 | { | |
1209 | struct drbd_bitmap *b = mdev->bitmap; | |
4b0715f0 | 1210 | unsigned long i = DRBD_END_OF_BITMAP; |
b411b363 PR |
1211 | |
1212 | ERR_IF(!b) return i; | |
1213 | ERR_IF(!b->bm_pages) return i; | |
1214 | ||
1215 | spin_lock_irq(&b->bm_lock); | |
20ceb2b2 | 1216 | if (BM_DONT_TEST & b->bm_flags) |
b411b363 PR |
1217 | bm_print_lock_info(mdev); |
1218 | ||
1219 | i = __bm_find_next(mdev, bm_fo, find_zero_bit, KM_IRQ1); | |
1220 | ||
1221 | spin_unlock_irq(&b->bm_lock); | |
1222 | return i; | |
1223 | } | |
1224 | ||
1225 | unsigned long drbd_bm_find_next(struct drbd_conf *mdev, unsigned long bm_fo) | |
1226 | { | |
1227 | return bm_find_next(mdev, bm_fo, 0); | |
1228 | } | |
1229 | ||
1230 | #if 0 | |
1231 | /* not yet needed for anything. */ | |
1232 | unsigned long drbd_bm_find_next_zero(struct drbd_conf *mdev, unsigned long bm_fo) | |
1233 | { | |
1234 | return bm_find_next(mdev, bm_fo, 1); | |
1235 | } | |
1236 | #endif | |
1237 | ||
1238 | /* does not spin_lock_irqsave. | |
1239 | * you must take drbd_bm_lock() first */ | |
1240 | unsigned long _drbd_bm_find_next(struct drbd_conf *mdev, unsigned long bm_fo) | |
1241 | { | |
20ceb2b2 | 1242 | /* WARN_ON(!(BM_DONT_SET & mdev->b->bm_flags)); */ |
b411b363 PR |
1243 | return __bm_find_next(mdev, bm_fo, 0, KM_USER1); |
1244 | } | |
1245 | ||
1246 | unsigned long _drbd_bm_find_next_zero(struct drbd_conf *mdev, unsigned long bm_fo) | |
1247 | { | |
20ceb2b2 | 1248 | /* WARN_ON(!(BM_DONT_SET & mdev->b->bm_flags)); */ |
b411b363 PR |
1249 | return __bm_find_next(mdev, bm_fo, 1, KM_USER1); |
1250 | } | |
1251 | ||
1252 | /* returns number of bits actually changed. | |
1253 | * for val != 0, we change 0 -> 1, return code positive | |
1254 | * for val == 0, we change 1 -> 0, return code negative | |
1255 | * wants bitnr, not sector. | |
1256 | * expected to be called for only a few bits (e - s about BITS_PER_LONG). | |
1257 | * Must hold bitmap lock already. */ | |
b4ee79da | 1258 | static int __bm_change_bits_to(struct drbd_conf *mdev, const unsigned long s, |
b411b363 PR |
1259 | unsigned long e, int val, const enum km_type km) |
1260 | { | |
1261 | struct drbd_bitmap *b = mdev->bitmap; | |
1262 | unsigned long *p_addr = NULL; | |
1263 | unsigned long bitnr; | |
19f843aa | 1264 | unsigned int last_page_nr = -1U; |
b411b363 | 1265 | int c = 0; |
19f843aa | 1266 | int changed_total = 0; |
b411b363 PR |
1267 | |
1268 | if (e >= b->bm_bits) { | |
1269 | dev_err(DEV, "ASSERT FAILED: bit_s=%lu bit_e=%lu bm_bits=%lu\n", | |
1270 | s, e, b->bm_bits); | |
1271 | e = b->bm_bits ? b->bm_bits -1 : 0; | |
1272 | } | |
1273 | for (bitnr = s; bitnr <= e; bitnr++) { | |
19f843aa | 1274 | unsigned int page_nr = bm_bit_to_page_idx(b, bitnr); |
b411b363 PR |
1275 | if (page_nr != last_page_nr) { |
1276 | if (p_addr) | |
1277 | __bm_unmap(p_addr, km); | |
19f843aa LE |
1278 | if (c < 0) |
1279 | bm_set_page_lazy_writeout(b->bm_pages[last_page_nr]); | |
1280 | else if (c > 0) | |
1281 | bm_set_page_need_writeout(b->bm_pages[last_page_nr]); | |
1282 | changed_total += c; | |
1283 | c = 0; | |
1284 | p_addr = __bm_map_pidx(b, page_nr, km); | |
b411b363 PR |
1285 | last_page_nr = page_nr; |
1286 | } | |
1287 | if (val) | |
7e599e6e | 1288 | c += (0 == __test_and_set_bit_le(bitnr & BITS_PER_PAGE_MASK, p_addr)); |
b411b363 | 1289 | else |
7e599e6e | 1290 | c -= (0 != __test_and_clear_bit_le(bitnr & BITS_PER_PAGE_MASK, p_addr)); |
b411b363 PR |
1291 | } |
1292 | if (p_addr) | |
1293 | __bm_unmap(p_addr, km); | |
19f843aa LE |
1294 | if (c < 0) |
1295 | bm_set_page_lazy_writeout(b->bm_pages[last_page_nr]); | |
1296 | else if (c > 0) | |
1297 | bm_set_page_need_writeout(b->bm_pages[last_page_nr]); | |
1298 | changed_total += c; | |
1299 | b->bm_set += changed_total; | |
1300 | return changed_total; | |
b411b363 PR |
1301 | } |
1302 | ||
1303 | /* returns number of bits actually changed. | |
1304 | * for val != 0, we change 0 -> 1, return code positive | |
1305 | * for val == 0, we change 1 -> 0, return code negative | |
1306 | * wants bitnr, not sector */ | |
b4ee79da | 1307 | static int bm_change_bits_to(struct drbd_conf *mdev, const unsigned long s, |
b411b363 PR |
1308 | const unsigned long e, int val) |
1309 | { | |
1310 | unsigned long flags; | |
1311 | struct drbd_bitmap *b = mdev->bitmap; | |
1312 | int c = 0; | |
1313 | ||
1314 | ERR_IF(!b) return 1; | |
1315 | ERR_IF(!b->bm_pages) return 0; | |
1316 | ||
1317 | spin_lock_irqsave(&b->bm_lock, flags); | |
20ceb2b2 | 1318 | if ((val ? BM_DONT_SET : BM_DONT_CLEAR) & b->bm_flags) |
b411b363 PR |
1319 | bm_print_lock_info(mdev); |
1320 | ||
1321 | c = __bm_change_bits_to(mdev, s, e, val, KM_IRQ1); | |
1322 | ||
1323 | spin_unlock_irqrestore(&b->bm_lock, flags); | |
1324 | return c; | |
1325 | } | |
1326 | ||
1327 | /* returns number of bits changed 0 -> 1 */ | |
1328 | int drbd_bm_set_bits(struct drbd_conf *mdev, const unsigned long s, const unsigned long e) | |
1329 | { | |
1330 | return bm_change_bits_to(mdev, s, e, 1); | |
1331 | } | |
1332 | ||
1333 | /* returns number of bits changed 1 -> 0 */ | |
1334 | int drbd_bm_clear_bits(struct drbd_conf *mdev, const unsigned long s, const unsigned long e) | |
1335 | { | |
1336 | return -bm_change_bits_to(mdev, s, e, 0); | |
1337 | } | |
1338 | ||
1339 | /* sets all bits in full words, | |
1340 | * from first_word up to, but not including, last_word */ | |
1341 | static inline void bm_set_full_words_within_one_page(struct drbd_bitmap *b, | |
1342 | int page_nr, int first_word, int last_word) | |
1343 | { | |
1344 | int i; | |
1345 | int bits; | |
1346 | unsigned long *paddr = kmap_atomic(b->bm_pages[page_nr], KM_USER0); | |
1347 | for (i = first_word; i < last_word; i++) { | |
1348 | bits = hweight_long(paddr[i]); | |
1349 | paddr[i] = ~0UL; | |
1350 | b->bm_set += BITS_PER_LONG - bits; | |
1351 | } | |
1352 | kunmap_atomic(paddr, KM_USER0); | |
1353 | } | |
1354 | ||
1355 | /* Same thing as drbd_bm_set_bits, but without taking the spin_lock_irqsave. | |
1356 | * You must first drbd_bm_lock(). | |
1357 | * Can be called to set the whole bitmap in one go. | |
1358 | * Sets bits from s to e _inclusive_. */ | |
1359 | void _drbd_bm_set_bits(struct drbd_conf *mdev, const unsigned long s, const unsigned long e) | |
1360 | { | |
1361 | /* First set_bit from the first bit (s) | |
1362 | * up to the next long boundary (sl), | |
1363 | * then assign full words up to the last long boundary (el), | |
1364 | * then set_bit up to and including the last bit (e). | |
1365 | * | |
1366 | * Do not use memset, because we must account for changes, | |
1367 | * so we need to loop over the words with hweight() anyways. | |
1368 | */ | |
1369 | unsigned long sl = ALIGN(s,BITS_PER_LONG); | |
1370 | unsigned long el = (e+1) & ~((unsigned long)BITS_PER_LONG-1); | |
1371 | int first_page; | |
1372 | int last_page; | |
1373 | int page_nr; | |
1374 | int first_word; | |
1375 | int last_word; | |
1376 | ||
1377 | if (e - s <= 3*BITS_PER_LONG) { | |
1378 | /* don't bother; el and sl may even be wrong. */ | |
1379 | __bm_change_bits_to(mdev, s, e, 1, KM_USER0); | |
1380 | return; | |
1381 | } | |
1382 | ||
1383 | /* difference is large enough that we can trust sl and el */ | |
1384 | ||
1385 | /* bits filling the current long */ | |
1386 | if (sl) | |
1387 | __bm_change_bits_to(mdev, s, sl-1, 1, KM_USER0); | |
1388 | ||
1389 | first_page = sl >> (3 + PAGE_SHIFT); | |
1390 | last_page = el >> (3 + PAGE_SHIFT); | |
1391 | ||
1392 | /* MLPP: modulo longs per page */ | |
1393 | /* LWPP: long words per page */ | |
1394 | first_word = MLPP(sl >> LN2_BPL); | |
1395 | last_word = LWPP; | |
1396 | ||
1397 | /* first and full pages, unless first page == last page */ | |
1398 | for (page_nr = first_page; page_nr < last_page; page_nr++) { | |
1399 | bm_set_full_words_within_one_page(mdev->bitmap, page_nr, first_word, last_word); | |
1400 | cond_resched(); | |
1401 | first_word = 0; | |
1402 | } | |
1403 | ||
1404 | /* last page (respectively only page, for first page == last page) */ | |
1405 | last_word = MLPP(el >> LN2_BPL); | |
1406 | bm_set_full_words_within_one_page(mdev->bitmap, last_page, first_word, last_word); | |
1407 | ||
1408 | /* possibly trailing bits. | |
1409 | * example: (e & 63) == 63, el will be e+1. | |
1410 | * if that even was the very last bit, | |
1411 | * it would trigger an assert in __bm_change_bits_to() | |
1412 | */ | |
1413 | if (el <= e) | |
1414 | __bm_change_bits_to(mdev, el, e, 1, KM_USER0); | |
1415 | } | |
1416 | ||
1417 | /* returns bit state | |
1418 | * wants bitnr, NOT sector. | |
1419 | * inherently racy... area needs to be locked by means of {al,rs}_lru | |
1420 | * 1 ... bit set | |
1421 | * 0 ... bit not set | |
1422 | * -1 ... first out of bounds access, stop testing for bits! | |
1423 | */ | |
1424 | int drbd_bm_test_bit(struct drbd_conf *mdev, const unsigned long bitnr) | |
1425 | { | |
1426 | unsigned long flags; | |
1427 | struct drbd_bitmap *b = mdev->bitmap; | |
1428 | unsigned long *p_addr; | |
1429 | int i; | |
1430 | ||
1431 | ERR_IF(!b) return 0; | |
1432 | ERR_IF(!b->bm_pages) return 0; | |
1433 | ||
1434 | spin_lock_irqsave(&b->bm_lock, flags); | |
20ceb2b2 | 1435 | if (BM_DONT_TEST & b->bm_flags) |
b411b363 PR |
1436 | bm_print_lock_info(mdev); |
1437 | if (bitnr < b->bm_bits) { | |
19f843aa | 1438 | p_addr = bm_map_pidx(b, bm_bit_to_page_idx(b, bitnr)); |
7e599e6e | 1439 | i = test_bit_le(bitnr & BITS_PER_PAGE_MASK, p_addr) ? 1 : 0; |
b411b363 PR |
1440 | bm_unmap(p_addr); |
1441 | } else if (bitnr == b->bm_bits) { | |
1442 | i = -1; | |
1443 | } else { /* (bitnr > b->bm_bits) */ | |
1444 | dev_err(DEV, "bitnr=%lu > bm_bits=%lu\n", bitnr, b->bm_bits); | |
1445 | i = 0; | |
1446 | } | |
1447 | ||
1448 | spin_unlock_irqrestore(&b->bm_lock, flags); | |
1449 | return i; | |
1450 | } | |
1451 | ||
1452 | /* returns number of bits set in the range [s, e] */ | |
1453 | int drbd_bm_count_bits(struct drbd_conf *mdev, const unsigned long s, const unsigned long e) | |
1454 | { | |
1455 | unsigned long flags; | |
1456 | struct drbd_bitmap *b = mdev->bitmap; | |
19f843aa | 1457 | unsigned long *p_addr = NULL; |
b411b363 | 1458 | unsigned long bitnr; |
19f843aa | 1459 | unsigned int page_nr = -1U; |
b411b363 | 1460 | int c = 0; |
b411b363 PR |
1461 | |
1462 | /* If this is called without a bitmap, that is a bug. But just to be | |
1463 | * robust in case we screwed up elsewhere, in that case pretend there | |
1464 | * was one dirty bit in the requested area, so we won't try to do a | |
1465 | * local read there (no bitmap probably implies no disk) */ | |
1466 | ERR_IF(!b) return 1; | |
1467 | ERR_IF(!b->bm_pages) return 1; | |
1468 | ||
1469 | spin_lock_irqsave(&b->bm_lock, flags); | |
20ceb2b2 | 1470 | if (BM_DONT_TEST & b->bm_flags) |
b411b363 PR |
1471 | bm_print_lock_info(mdev); |
1472 | for (bitnr = s; bitnr <= e; bitnr++) { | |
19f843aa LE |
1473 | unsigned int idx = bm_bit_to_page_idx(b, bitnr); |
1474 | if (page_nr != idx) { | |
1475 | page_nr = idx; | |
b411b363 PR |
1476 | if (p_addr) |
1477 | bm_unmap(p_addr); | |
19f843aa | 1478 | p_addr = bm_map_pidx(b, idx); |
b411b363 PR |
1479 | } |
1480 | ERR_IF (bitnr >= b->bm_bits) { | |
1481 | dev_err(DEV, "bitnr=%lu bm_bits=%lu\n", bitnr, b->bm_bits); | |
1482 | } else { | |
7e599e6e | 1483 | c += (0 != test_bit_le(bitnr - (page_nr << (PAGE_SHIFT+3)), p_addr)); |
b411b363 PR |
1484 | } |
1485 | } | |
1486 | if (p_addr) | |
1487 | bm_unmap(p_addr); | |
1488 | spin_unlock_irqrestore(&b->bm_lock, flags); | |
1489 | return c; | |
1490 | } | |
1491 | ||
1492 | ||
1493 | /* inherently racy... | |
1494 | * return value may be already out-of-date when this function returns. | |
1495 | * but the general usage is that this is only use during a cstate when bits are | |
1496 | * only cleared, not set, and typically only care for the case when the return | |
1497 | * value is zero, or we already "locked" this "bitmap extent" by other means. | |
1498 | * | |
1499 | * enr is bm-extent number, since we chose to name one sector (512 bytes) | |
1500 | * worth of the bitmap a "bitmap extent". | |
1501 | * | |
1502 | * TODO | |
1503 | * I think since we use it like a reference count, we should use the real | |
1504 | * reference count of some bitmap extent element from some lru instead... | |
1505 | * | |
1506 | */ | |
1507 | int drbd_bm_e_weight(struct drbd_conf *mdev, unsigned long enr) | |
1508 | { | |
1509 | struct drbd_bitmap *b = mdev->bitmap; | |
1510 | int count, s, e; | |
1511 | unsigned long flags; | |
1512 | unsigned long *p_addr, *bm; | |
1513 | ||
1514 | ERR_IF(!b) return 0; | |
1515 | ERR_IF(!b->bm_pages) return 0; | |
1516 | ||
1517 | spin_lock_irqsave(&b->bm_lock, flags); | |
20ceb2b2 | 1518 | if (BM_DONT_TEST & b->bm_flags) |
b411b363 PR |
1519 | bm_print_lock_info(mdev); |
1520 | ||
1521 | s = S2W(enr); | |
1522 | e = min((size_t)S2W(enr+1), b->bm_words); | |
1523 | count = 0; | |
1524 | if (s < b->bm_words) { | |
1525 | int n = e-s; | |
19f843aa | 1526 | p_addr = bm_map_pidx(b, bm_word_to_page_idx(b, s)); |
b411b363 PR |
1527 | bm = p_addr + MLPP(s); |
1528 | while (n--) | |
1529 | count += hweight_long(*bm++); | |
1530 | bm_unmap(p_addr); | |
1531 | } else { | |
1532 | dev_err(DEV, "start offset (%d) too large in drbd_bm_e_weight\n", s); | |
1533 | } | |
1534 | spin_unlock_irqrestore(&b->bm_lock, flags); | |
1535 | return count; | |
1536 | } | |
1537 | ||
4b0715f0 LE |
1538 | /* Set all bits covered by the AL-extent al_enr. |
1539 | * Returns number of bits changed. */ | |
b411b363 PR |
1540 | unsigned long drbd_bm_ALe_set_all(struct drbd_conf *mdev, unsigned long al_enr) |
1541 | { | |
1542 | struct drbd_bitmap *b = mdev->bitmap; | |
1543 | unsigned long *p_addr, *bm; | |
1544 | unsigned long weight; | |
4b0715f0 LE |
1545 | unsigned long s, e; |
1546 | int count, i, do_now; | |
b411b363 PR |
1547 | ERR_IF(!b) return 0; |
1548 | ERR_IF(!b->bm_pages) return 0; | |
1549 | ||
1550 | spin_lock_irq(&b->bm_lock); | |
20ceb2b2 | 1551 | if (BM_DONT_SET & b->bm_flags) |
b411b363 PR |
1552 | bm_print_lock_info(mdev); |
1553 | weight = b->bm_set; | |
1554 | ||
1555 | s = al_enr * BM_WORDS_PER_AL_EXT; | |
1556 | e = min_t(size_t, s + BM_WORDS_PER_AL_EXT, b->bm_words); | |
1557 | /* assert that s and e are on the same page */ | |
1558 | D_ASSERT((e-1) >> (PAGE_SHIFT - LN2_BPL + 3) | |
1559 | == s >> (PAGE_SHIFT - LN2_BPL + 3)); | |
1560 | count = 0; | |
1561 | if (s < b->bm_words) { | |
1562 | i = do_now = e-s; | |
19f843aa | 1563 | p_addr = bm_map_pidx(b, bm_word_to_page_idx(b, s)); |
b411b363 PR |
1564 | bm = p_addr + MLPP(s); |
1565 | while (i--) { | |
1566 | count += hweight_long(*bm); | |
1567 | *bm = -1UL; | |
1568 | bm++; | |
1569 | } | |
1570 | bm_unmap(p_addr); | |
1571 | b->bm_set += do_now*BITS_PER_LONG - count; | |
1572 | if (e == b->bm_words) | |
1573 | b->bm_set -= bm_clear_surplus(b); | |
1574 | } else { | |
4b0715f0 | 1575 | dev_err(DEV, "start offset (%lu) too large in drbd_bm_ALe_set_all\n", s); |
b411b363 PR |
1576 | } |
1577 | weight = b->bm_set - weight; | |
1578 | spin_unlock_irq(&b->bm_lock); | |
1579 | return weight; | |
1580 | } |