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 | ||
f88c5d90 LE |
25 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
26 | ||
b411b363 PR |
27 | #include <linux/bitops.h> |
28 | #include <linux/vmalloc.h> | |
29 | #include <linux/string.h> | |
30 | #include <linux/drbd.h> | |
5a0e3ad6 | 31 | #include <linux/slab.h> |
b411b363 | 32 | #include <asm/kmap_types.h> |
f0ff1357 | 33 | |
b411b363 PR |
34 | #include "drbd_int.h" |
35 | ||
95a0f10c | 36 | |
b411b363 PR |
37 | /* OPAQUE outside this file! |
38 | * interface defined in drbd_int.h | |
39 | ||
40 | * convention: | |
41 | * function name drbd_bm_... => used elsewhere, "public". | |
42 | * function name bm_... => internal to implementation, "private". | |
4b0715f0 LE |
43 | */ |
44 | ||
b411b363 | 45 | |
4b0715f0 LE |
46 | /* |
47 | * LIMITATIONS: | |
48 | * We want to support >= peta byte of backend storage, while for now still using | |
49 | * a granularity of one bit per 4KiB of storage. | |
50 | * 1 << 50 bytes backend storage (1 PiB) | |
51 | * 1 << (50 - 12) bits needed | |
52 | * 38 --> we need u64 to index and count bits | |
53 | * 1 << (38 - 3) bitmap bytes needed | |
54 | * 35 --> we still need u64 to index and count bytes | |
55 | * (that's 32 GiB of bitmap for 1 PiB storage) | |
56 | * 1 << (35 - 2) 32bit longs needed | |
57 | * 33 --> we'd even need u64 to index and count 32bit long words. | |
58 | * 1 << (35 - 3) 64bit longs needed | |
59 | * 32 --> we could get away with a 32bit unsigned int to index and count | |
60 | * 64bit long words, but I rather stay with unsigned long for now. | |
61 | * We probably should neither count nor point to bytes or long words | |
62 | * directly, but either by bitnumber, or by page index and offset. | |
63 | * 1 << (35 - 12) | |
64 | * 22 --> we need that much 4KiB pages of bitmap. | |
65 | * 1 << (22 + 3) --> on a 64bit arch, | |
66 | * we need 32 MiB to store the array of page pointers. | |
67 | * | |
68 | * Because I'm lazy, and because the resulting patch was too large, too ugly | |
69 | * and still incomplete, on 32bit we still "only" support 16 TiB (minus some), | |
70 | * (1 << 32) bits * 4k storage. | |
71 | * | |
72 | ||
73 | * bitmap storage and IO: | |
74 | * Bitmap is stored little endian on disk, and is kept little endian in | |
75 | * core memory. Currently we still hold the full bitmap in core as long | |
76 | * as we are "attached" to a local disk, which at 32 GiB for 1PiB storage | |
77 | * seems excessive. | |
78 | * | |
24c4830c | 79 | * We plan to reduce the amount of in-core bitmap pages by paging them in |
4b0715f0 LE |
80 | * and out against their on-disk location as necessary, but need to make |
81 | * sure we don't cause too much meta data IO, and must not deadlock in | |
82 | * tight memory situations. This needs some more work. | |
b411b363 PR |
83 | */ |
84 | ||
85 | /* | |
86 | * NOTE | |
87 | * Access to the *bm_pages is protected by bm_lock. | |
88 | * It is safe to read the other members within the lock. | |
89 | * | |
90 | * drbd_bm_set_bits is called from bio_endio callbacks, | |
91 | * We may be called with irq already disabled, | |
92 | * so we need spin_lock_irqsave(). | |
93 | * And we need the kmap_atomic. | |
94 | */ | |
95 | struct drbd_bitmap { | |
96 | struct page **bm_pages; | |
97 | spinlock_t bm_lock; | |
4b0715f0 LE |
98 | |
99 | /* see LIMITATIONS: above */ | |
100 | ||
b411b363 PR |
101 | unsigned long bm_set; /* nr of set bits; THINK maybe atomic_t? */ |
102 | unsigned long bm_bits; | |
103 | size_t bm_words; | |
104 | size_t bm_number_of_pages; | |
105 | sector_t bm_dev_capacity; | |
8a03ae2a | 106 | struct mutex bm_change; /* serializes resize operations */ |
b411b363 | 107 | |
19f843aa | 108 | wait_queue_head_t bm_io_wait; /* used to serialize IO of single pages */ |
b411b363 | 109 | |
20ceb2b2 | 110 | enum bm_flag bm_flags; |
b411b363 PR |
111 | |
112 | /* debugging aid, in case we are still racy somewhere */ | |
113 | char *bm_why; | |
114 | struct task_struct *bm_task; | |
115 | }; | |
116 | ||
b411b363 | 117 | #define bm_print_lock_info(m) __bm_print_lock_info(m, __func__) |
b30ab791 | 118 | static void __bm_print_lock_info(struct drbd_device *device, const char *func) |
b411b363 | 119 | { |
b30ab791 | 120 | struct drbd_bitmap *b = device->bitmap; |
b411b363 PR |
121 | if (!__ratelimit(&drbd_ratelimit_state)) |
122 | return; | |
c60b0251 AG |
123 | drbd_err(device, "FIXME %s[%d] in %s, bitmap locked for '%s' by %s[%d]\n", |
124 | current->comm, task_pid_nr(current), | |
125 | func, b->bm_why ?: "?", | |
126 | b->bm_task->comm, task_pid_nr(b->bm_task)); | |
b411b363 PR |
127 | } |
128 | ||
b30ab791 | 129 | void drbd_bm_lock(struct drbd_device *device, char *why, enum bm_flag flags) |
b411b363 | 130 | { |
b30ab791 | 131 | struct drbd_bitmap *b = device->bitmap; |
b411b363 PR |
132 | int trylock_failed; |
133 | ||
134 | if (!b) { | |
d0180171 | 135 | drbd_err(device, "FIXME no bitmap in drbd_bm_lock!?\n"); |
b411b363 PR |
136 | return; |
137 | } | |
138 | ||
8a03ae2a | 139 | trylock_failed = !mutex_trylock(&b->bm_change); |
b411b363 PR |
140 | |
141 | if (trylock_failed) { | |
c60b0251 AG |
142 | drbd_warn(device, "%s[%d] going to '%s' but bitmap already locked for '%s' by %s[%d]\n", |
143 | current->comm, task_pid_nr(current), | |
144 | why, b->bm_why ?: "?", | |
145 | b->bm_task->comm, task_pid_nr(b->bm_task)); | |
8a03ae2a | 146 | mutex_lock(&b->bm_change); |
b411b363 | 147 | } |
20ceb2b2 | 148 | if (BM_LOCKED_MASK & b->bm_flags) |
d0180171 | 149 | drbd_err(device, "FIXME bitmap already locked in bm_lock\n"); |
20ceb2b2 | 150 | b->bm_flags |= flags & BM_LOCKED_MASK; |
b411b363 PR |
151 | |
152 | b->bm_why = why; | |
153 | b->bm_task = current; | |
154 | } | |
155 | ||
b30ab791 | 156 | void drbd_bm_unlock(struct drbd_device *device) |
b411b363 | 157 | { |
b30ab791 | 158 | struct drbd_bitmap *b = device->bitmap; |
b411b363 | 159 | if (!b) { |
d0180171 | 160 | drbd_err(device, "FIXME no bitmap in drbd_bm_unlock!?\n"); |
b411b363 PR |
161 | return; |
162 | } | |
163 | ||
b30ab791 | 164 | if (!(BM_LOCKED_MASK & device->bitmap->bm_flags)) |
d0180171 | 165 | drbd_err(device, "FIXME bitmap not locked in bm_unlock\n"); |
b411b363 | 166 | |
20ceb2b2 | 167 | b->bm_flags &= ~BM_LOCKED_MASK; |
b411b363 PR |
168 | b->bm_why = NULL; |
169 | b->bm_task = NULL; | |
8a03ae2a | 170 | mutex_unlock(&b->bm_change); |
b411b363 PR |
171 | } |
172 | ||
19f843aa LE |
173 | /* we store some "meta" info about our pages in page->private */ |
174 | /* at a granularity of 4k storage per bitmap bit: | |
175 | * one peta byte storage: 1<<50 byte, 1<<38 * 4k storage blocks | |
176 | * 1<<38 bits, | |
177 | * 1<<23 4k bitmap pages. | |
178 | * Use 24 bits as page index, covers 2 peta byte storage | |
179 | * at a granularity of 4k per bit. | |
180 | * Used to report the failed page idx on io error from the endio handlers. | |
181 | */ | |
182 | #define BM_PAGE_IDX_MASK ((1UL<<24)-1) | |
183 | /* this page is currently read in, or written back */ | |
184 | #define BM_PAGE_IO_LOCK 31 | |
185 | /* if there has been an IO error for this page */ | |
186 | #define BM_PAGE_IO_ERROR 30 | |
187 | /* this is to be able to intelligently skip disk IO, | |
188 | * set if bits have been set since last IO. */ | |
189 | #define BM_PAGE_NEED_WRITEOUT 29 | |
190 | /* to mark for lazy writeout once syncer cleared all clearable bits, | |
191 | * we if bits have been cleared since last IO. */ | |
192 | #define BM_PAGE_LAZY_WRITEOUT 28 | |
45dfffeb LE |
193 | /* pages marked with this "HINT" will be considered for writeout |
194 | * on activity log transactions */ | |
195 | #define BM_PAGE_HINT_WRITEOUT 27 | |
19f843aa | 196 | |
24c4830c | 197 | /* store_page_idx uses non-atomic assignment. It is only used directly after |
19f843aa LE |
198 | * allocating the page. All other bm_set_page_* and bm_clear_page_* need to |
199 | * use atomic bit manipulation, as set_out_of_sync (and therefore bitmap | |
200 | * changes) may happen from various contexts, and wait_on_bit/wake_up_bit | |
201 | * requires it all to be atomic as well. */ | |
202 | static void bm_store_page_idx(struct page *page, unsigned long idx) | |
95a0f10c | 203 | { |
19f843aa | 204 | BUG_ON(0 != (idx & ~BM_PAGE_IDX_MASK)); |
0c7db279 | 205 | set_page_private(page, idx); |
95a0f10c LE |
206 | } |
207 | ||
19f843aa | 208 | static unsigned long bm_page_to_idx(struct page *page) |
b411b363 | 209 | { |
19f843aa LE |
210 | return page_private(page) & BM_PAGE_IDX_MASK; |
211 | } | |
212 | ||
213 | /* As is very unlikely that the same page is under IO from more than one | |
214 | * context, we can get away with a bit per page and one wait queue per bitmap. | |
215 | */ | |
b30ab791 | 216 | static void bm_page_lock_io(struct drbd_device *device, int page_nr) |
19f843aa | 217 | { |
b30ab791 | 218 | struct drbd_bitmap *b = device->bitmap; |
19f843aa LE |
219 | void *addr = &page_private(b->bm_pages[page_nr]); |
220 | wait_event(b->bm_io_wait, !test_and_set_bit(BM_PAGE_IO_LOCK, addr)); | |
221 | } | |
222 | ||
b30ab791 | 223 | static void bm_page_unlock_io(struct drbd_device *device, int page_nr) |
19f843aa | 224 | { |
b30ab791 | 225 | struct drbd_bitmap *b = device->bitmap; |
19f843aa | 226 | void *addr = &page_private(b->bm_pages[page_nr]); |
4738fa16 | 227 | clear_bit_unlock(BM_PAGE_IO_LOCK, addr); |
b30ab791 | 228 | wake_up(&device->bitmap->bm_io_wait); |
19f843aa LE |
229 | } |
230 | ||
231 | /* set _before_ submit_io, so it may be reset due to being changed | |
232 | * while this page is in flight... will get submitted later again */ | |
233 | static void bm_set_page_unchanged(struct page *page) | |
234 | { | |
235 | /* use cmpxchg? */ | |
236 | clear_bit(BM_PAGE_NEED_WRITEOUT, &page_private(page)); | |
237 | clear_bit(BM_PAGE_LAZY_WRITEOUT, &page_private(page)); | |
238 | } | |
239 | ||
240 | static void bm_set_page_need_writeout(struct page *page) | |
241 | { | |
242 | set_bit(BM_PAGE_NEED_WRITEOUT, &page_private(page)); | |
243 | } | |
244 | ||
45dfffeb LE |
245 | /** |
246 | * drbd_bm_mark_for_writeout() - mark a page with a "hint" to be considered for writeout | |
b30ab791 | 247 | * @device: DRBD device. |
45dfffeb LE |
248 | * @page_nr: the bitmap page to mark with the "hint" flag |
249 | * | |
250 | * From within an activity log transaction, we mark a few pages with these | |
251 | * hints, then call drbd_bm_write_hinted(), which will only write out changed | |
252 | * pages which are flagged with this mark. | |
253 | */ | |
b30ab791 | 254 | void drbd_bm_mark_for_writeout(struct drbd_device *device, int page_nr) |
45dfffeb LE |
255 | { |
256 | struct page *page; | |
b30ab791 | 257 | if (page_nr >= device->bitmap->bm_number_of_pages) { |
d0180171 | 258 | drbd_warn(device, "BAD: page_nr: %u, number_of_pages: %u\n", |
b30ab791 | 259 | page_nr, (int)device->bitmap->bm_number_of_pages); |
45dfffeb LE |
260 | return; |
261 | } | |
b30ab791 | 262 | page = device->bitmap->bm_pages[page_nr]; |
45dfffeb LE |
263 | set_bit(BM_PAGE_HINT_WRITEOUT, &page_private(page)); |
264 | } | |
265 | ||
19f843aa LE |
266 | static int bm_test_page_unchanged(struct page *page) |
267 | { | |
268 | volatile const unsigned long *addr = &page_private(page); | |
269 | return (*addr & ((1UL<<BM_PAGE_NEED_WRITEOUT)|(1UL<<BM_PAGE_LAZY_WRITEOUT))) == 0; | |
270 | } | |
b411b363 | 271 | |
19f843aa LE |
272 | static void bm_set_page_io_err(struct page *page) |
273 | { | |
274 | set_bit(BM_PAGE_IO_ERROR, &page_private(page)); | |
275 | } | |
276 | ||
277 | static void bm_clear_page_io_err(struct page *page) | |
278 | { | |
279 | clear_bit(BM_PAGE_IO_ERROR, &page_private(page)); | |
280 | } | |
281 | ||
282 | static void bm_set_page_lazy_writeout(struct page *page) | |
283 | { | |
284 | set_bit(BM_PAGE_LAZY_WRITEOUT, &page_private(page)); | |
285 | } | |
286 | ||
287 | static int bm_test_page_lazy_writeout(struct page *page) | |
288 | { | |
289 | return test_bit(BM_PAGE_LAZY_WRITEOUT, &page_private(page)); | |
290 | } | |
291 | ||
292 | /* on a 32bit box, this would allow for exactly (2<<38) bits. */ | |
293 | static unsigned int bm_word_to_page_idx(struct drbd_bitmap *b, unsigned long long_nr) | |
294 | { | |
b411b363 | 295 | /* page_nr = (word*sizeof(long)) >> PAGE_SHIFT; */ |
19f843aa | 296 | unsigned int page_nr = long_nr >> (PAGE_SHIFT - LN2_BPL + 3); |
b411b363 | 297 | BUG_ON(page_nr >= b->bm_number_of_pages); |
19f843aa LE |
298 | return page_nr; |
299 | } | |
b411b363 | 300 | |
19f843aa LE |
301 | static unsigned int bm_bit_to_page_idx(struct drbd_bitmap *b, u64 bitnr) |
302 | { | |
303 | /* page_nr = (bitnr/8) >> PAGE_SHIFT; */ | |
304 | unsigned int page_nr = bitnr >> (PAGE_SHIFT + 3); | |
305 | BUG_ON(page_nr >= b->bm_number_of_pages); | |
306 | return page_nr; | |
b411b363 PR |
307 | } |
308 | ||
589973a7 | 309 | static unsigned long *__bm_map_pidx(struct drbd_bitmap *b, unsigned int idx) |
95a0f10c LE |
310 | { |
311 | struct page *page = b->bm_pages[idx]; | |
cfd8005c | 312 | return (unsigned long *) kmap_atomic(page); |
95a0f10c LE |
313 | } |
314 | ||
315 | static unsigned long *bm_map_pidx(struct drbd_bitmap *b, unsigned int idx) | |
316 | { | |
cfd8005c | 317 | return __bm_map_pidx(b, idx); |
95a0f10c LE |
318 | } |
319 | ||
cfd8005c | 320 | static void __bm_unmap(unsigned long *p_addr) |
b411b363 | 321 | { |
cfd8005c | 322 | kunmap_atomic(p_addr); |
b411b363 PR |
323 | }; |
324 | ||
325 | static void bm_unmap(unsigned long *p_addr) | |
326 | { | |
cfd8005c | 327 | return __bm_unmap(p_addr); |
b411b363 PR |
328 | } |
329 | ||
330 | /* long word offset of _bitmap_ sector */ | |
331 | #define S2W(s) ((s)<<(BM_EXT_SHIFT-BM_BLOCK_SHIFT-LN2_BPL)) | |
332 | /* word offset from start of bitmap to word number _in_page_ | |
333 | * modulo longs per page | |
334 | #define MLPP(X) ((X) % (PAGE_SIZE/sizeof(long)) | |
24c4830c | 335 | hm, well, Philipp thinks gcc might not optimize the % into & (... - 1) |
b411b363 PR |
336 | so do it explicitly: |
337 | */ | |
338 | #define MLPP(X) ((X) & ((PAGE_SIZE/sizeof(long))-1)) | |
339 | ||
340 | /* Long words per page */ | |
341 | #define LWPP (PAGE_SIZE/sizeof(long)) | |
342 | ||
343 | /* | |
344 | * actually most functions herein should take a struct drbd_bitmap*, not a | |
b30ab791 | 345 | * struct drbd_device*, but for the debug macros I like to have the device around |
b411b363 PR |
346 | * to be able to report device specific. |
347 | */ | |
348 | ||
19f843aa | 349 | |
b411b363 PR |
350 | static void bm_free_pages(struct page **pages, unsigned long number) |
351 | { | |
352 | unsigned long i; | |
353 | if (!pages) | |
354 | return; | |
355 | ||
356 | for (i = 0; i < number; i++) { | |
357 | if (!pages[i]) { | |
f88c5d90 LE |
358 | pr_alert("bm_free_pages tried to free a NULL pointer; i=%lu n=%lu\n", |
359 | i, number); | |
b411b363 PR |
360 | continue; |
361 | } | |
362 | __free_page(pages[i]); | |
363 | pages[i] = NULL; | |
364 | } | |
365 | } | |
366 | ||
367 | static void bm_vk_free(void *ptr, int v) | |
368 | { | |
369 | if (v) | |
370 | vfree(ptr); | |
371 | else | |
372 | kfree(ptr); | |
373 | } | |
374 | ||
375 | /* | |
376 | * "have" and "want" are NUMBER OF PAGES. | |
377 | */ | |
378 | static struct page **bm_realloc_pages(struct drbd_bitmap *b, unsigned long want) | |
379 | { | |
380 | struct page **old_pages = b->bm_pages; | |
381 | struct page **new_pages, *page; | |
382 | unsigned int i, bytes, vmalloced = 0; | |
383 | unsigned long have = b->bm_number_of_pages; | |
384 | ||
385 | BUG_ON(have == 0 && old_pages != NULL); | |
386 | BUG_ON(have != 0 && old_pages == NULL); | |
387 | ||
388 | if (have == want) | |
389 | return old_pages; | |
390 | ||
391 | /* Trying kmalloc first, falling back to vmalloc. | |
0b143d43 LE |
392 | * GFP_NOIO, as this is called while drbd IO is "suspended", |
393 | * and during resize or attach on diskless Primary, | |
394 | * we must not block on IO to ourselves. | |
bc891c9a | 395 | * Context is receiver thread or dmsetup. */ |
b411b363 | 396 | bytes = sizeof(struct page *)*want; |
8be04b93 | 397 | new_pages = kzalloc(bytes, GFP_NOIO | __GFP_NOWARN); |
b411b363 | 398 | if (!new_pages) { |
0b143d43 LE |
399 | new_pages = __vmalloc(bytes, |
400 | GFP_NOIO | __GFP_HIGHMEM | __GFP_ZERO, | |
401 | PAGE_KERNEL); | |
b411b363 PR |
402 | if (!new_pages) |
403 | return NULL; | |
404 | vmalloced = 1; | |
405 | } | |
406 | ||
b411b363 PR |
407 | if (want >= have) { |
408 | for (i = 0; i < have; i++) | |
409 | new_pages[i] = old_pages[i]; | |
410 | for (; i < want; i++) { | |
0b143d43 | 411 | page = alloc_page(GFP_NOIO | __GFP_HIGHMEM); |
b411b363 PR |
412 | if (!page) { |
413 | bm_free_pages(new_pages + have, i - have); | |
414 | bm_vk_free(new_pages, vmalloced); | |
415 | return NULL; | |
416 | } | |
19f843aa LE |
417 | /* we want to know which page it is |
418 | * from the endio handlers */ | |
419 | bm_store_page_idx(page, i); | |
b411b363 PR |
420 | new_pages[i] = page; |
421 | } | |
422 | } else { | |
423 | for (i = 0; i < want; i++) | |
424 | new_pages[i] = old_pages[i]; | |
425 | /* NOT HERE, we are outside the spinlock! | |
426 | bm_free_pages(old_pages + want, have - want); | |
427 | */ | |
428 | } | |
429 | ||
430 | if (vmalloced) | |
20ceb2b2 | 431 | b->bm_flags |= BM_P_VMALLOCED; |
b411b363 | 432 | else |
20ceb2b2 | 433 | b->bm_flags &= ~BM_P_VMALLOCED; |
b411b363 PR |
434 | |
435 | return new_pages; | |
436 | } | |
437 | ||
438 | /* | |
439 | * called on driver init only. TODO call when a device is created. | |
b30ab791 | 440 | * allocates the drbd_bitmap, and stores it in device->bitmap. |
b411b363 | 441 | */ |
b30ab791 | 442 | int drbd_bm_init(struct drbd_device *device) |
b411b363 | 443 | { |
b30ab791 | 444 | struct drbd_bitmap *b = device->bitmap; |
b411b363 PR |
445 | WARN_ON(b != NULL); |
446 | b = kzalloc(sizeof(struct drbd_bitmap), GFP_KERNEL); | |
447 | if (!b) | |
448 | return -ENOMEM; | |
449 | spin_lock_init(&b->bm_lock); | |
8a03ae2a | 450 | mutex_init(&b->bm_change); |
b411b363 PR |
451 | init_waitqueue_head(&b->bm_io_wait); |
452 | ||
b30ab791 | 453 | device->bitmap = b; |
b411b363 PR |
454 | |
455 | return 0; | |
456 | } | |
457 | ||
b30ab791 | 458 | sector_t drbd_bm_capacity(struct drbd_device *device) |
b411b363 | 459 | { |
b30ab791 | 460 | if (!expect(device->bitmap)) |
841ce241 | 461 | return 0; |
b30ab791 | 462 | return device->bitmap->bm_dev_capacity; |
b411b363 PR |
463 | } |
464 | ||
465 | /* called on driver unload. TODO: call when a device is destroyed. | |
466 | */ | |
b30ab791 | 467 | void drbd_bm_cleanup(struct drbd_device *device) |
b411b363 | 468 | { |
b30ab791 | 469 | if (!expect(device->bitmap)) |
841ce241 | 470 | return; |
b30ab791 AG |
471 | bm_free_pages(device->bitmap->bm_pages, device->bitmap->bm_number_of_pages); |
472 | bm_vk_free(device->bitmap->bm_pages, (BM_P_VMALLOCED & device->bitmap->bm_flags)); | |
473 | kfree(device->bitmap); | |
474 | device->bitmap = NULL; | |
b411b363 PR |
475 | } |
476 | ||
477 | /* | |
478 | * since (b->bm_bits % BITS_PER_LONG) != 0, | |
479 | * this masks out the remaining bits. | |
480 | * Returns the number of bits cleared. | |
481 | */ | |
95a0f10c LE |
482 | #define BITS_PER_PAGE (1UL << (PAGE_SHIFT + 3)) |
483 | #define BITS_PER_PAGE_MASK (BITS_PER_PAGE - 1) | |
484 | #define BITS_PER_LONG_MASK (BITS_PER_LONG - 1) | |
b411b363 PR |
485 | static int bm_clear_surplus(struct drbd_bitmap *b) |
486 | { | |
95a0f10c | 487 | unsigned long mask; |
b411b363 | 488 | unsigned long *p_addr, *bm; |
95a0f10c LE |
489 | int tmp; |
490 | int cleared = 0; | |
b411b363 | 491 | |
95a0f10c LE |
492 | /* number of bits modulo bits per page */ |
493 | tmp = (b->bm_bits & BITS_PER_PAGE_MASK); | |
494 | /* mask the used bits of the word containing the last bit */ | |
495 | mask = (1UL << (tmp & BITS_PER_LONG_MASK)) -1; | |
496 | /* bitmap is always stored little endian, | |
497 | * on disk and in core memory alike */ | |
498 | mask = cpu_to_lel(mask); | |
499 | ||
6850c442 | 500 | p_addr = bm_map_pidx(b, b->bm_number_of_pages - 1); |
95a0f10c LE |
501 | bm = p_addr + (tmp/BITS_PER_LONG); |
502 | if (mask) { | |
503 | /* If mask != 0, we are not exactly aligned, so bm now points | |
504 | * to the long containing the last bit. | |
505 | * If mask == 0, bm already points to the word immediately | |
506 | * after the last (long word aligned) bit. */ | |
b411b363 PR |
507 | cleared = hweight_long(*bm & ~mask); |
508 | *bm &= mask; | |
95a0f10c | 509 | bm++; |
b411b363 PR |
510 | } |
511 | ||
95a0f10c LE |
512 | if (BITS_PER_LONG == 32 && ((bm - p_addr) & 1) == 1) { |
513 | /* on a 32bit arch, we may need to zero out | |
514 | * a padding long to align with a 64bit remote */ | |
b411b363 PR |
515 | cleared += hweight_long(*bm); |
516 | *bm = 0; | |
517 | } | |
518 | bm_unmap(p_addr); | |
519 | return cleared; | |
520 | } | |
521 | ||
522 | static void bm_set_surplus(struct drbd_bitmap *b) | |
523 | { | |
95a0f10c | 524 | unsigned long mask; |
b411b363 | 525 | unsigned long *p_addr, *bm; |
95a0f10c LE |
526 | int tmp; |
527 | ||
528 | /* number of bits modulo bits per page */ | |
529 | tmp = (b->bm_bits & BITS_PER_PAGE_MASK); | |
530 | /* mask the used bits of the word containing the last bit */ | |
531 | mask = (1UL << (tmp & BITS_PER_LONG_MASK)) -1; | |
532 | /* bitmap is always stored little endian, | |
533 | * on disk and in core memory alike */ | |
534 | mask = cpu_to_lel(mask); | |
535 | ||
6850c442 | 536 | p_addr = bm_map_pidx(b, b->bm_number_of_pages - 1); |
95a0f10c LE |
537 | bm = p_addr + (tmp/BITS_PER_LONG); |
538 | if (mask) { | |
539 | /* If mask != 0, we are not exactly aligned, so bm now points | |
540 | * to the long containing the last bit. | |
541 | * If mask == 0, bm already points to the word immediately | |
542 | * after the last (long word aligned) bit. */ | |
b411b363 | 543 | *bm |= ~mask; |
95a0f10c | 544 | bm++; |
b411b363 PR |
545 | } |
546 | ||
95a0f10c LE |
547 | if (BITS_PER_LONG == 32 && ((bm - p_addr) & 1) == 1) { |
548 | /* on a 32bit arch, we may need to zero out | |
549 | * a padding long to align with a 64bit remote */ | |
550 | *bm = ~0UL; | |
b411b363 PR |
551 | } |
552 | bm_unmap(p_addr); | |
553 | } | |
554 | ||
4b0715f0 LE |
555 | /* you better not modify the bitmap while this is running, |
556 | * or its results will be stale */ | |
95a0f10c | 557 | static unsigned long bm_count_bits(struct drbd_bitmap *b) |
b411b363 | 558 | { |
4b0715f0 | 559 | unsigned long *p_addr; |
b411b363 | 560 | unsigned long bits = 0; |
4b0715f0 | 561 | unsigned long mask = (1UL << (b->bm_bits & BITS_PER_LONG_MASK)) -1; |
6850c442 | 562 | int idx, i, last_word; |
4b0715f0 LE |
563 | |
564 | /* all but last page */ | |
6850c442 | 565 | for (idx = 0; idx < b->bm_number_of_pages - 1; idx++) { |
cfd8005c | 566 | p_addr = __bm_map_pidx(b, idx); |
4b0715f0 LE |
567 | for (i = 0; i < LWPP; i++) |
568 | bits += hweight_long(p_addr[i]); | |
cfd8005c | 569 | __bm_unmap(p_addr); |
b411b363 PR |
570 | cond_resched(); |
571 | } | |
4b0715f0 LE |
572 | /* last (or only) page */ |
573 | last_word = ((b->bm_bits - 1) & BITS_PER_PAGE_MASK) >> LN2_BPL; | |
589973a7 | 574 | p_addr = __bm_map_pidx(b, idx); |
4b0715f0 LE |
575 | for (i = 0; i < last_word; i++) |
576 | bits += hweight_long(p_addr[i]); | |
577 | p_addr[last_word] &= cpu_to_lel(mask); | |
578 | bits += hweight_long(p_addr[last_word]); | |
579 | /* 32bit arch, may have an unused padding long */ | |
580 | if (BITS_PER_LONG == 32 && (last_word & 1) == 0) | |
581 | p_addr[last_word+1] = 0; | |
589973a7 | 582 | __bm_unmap(p_addr); |
b411b363 PR |
583 | return bits; |
584 | } | |
585 | ||
b411b363 PR |
586 | /* offset and len in long words.*/ |
587 | static void bm_memset(struct drbd_bitmap *b, size_t offset, int c, size_t len) | |
588 | { | |
589 | unsigned long *p_addr, *bm; | |
19f843aa | 590 | unsigned int idx; |
b411b363 PR |
591 | size_t do_now, end; |
592 | ||
b411b363 PR |
593 | end = offset + len; |
594 | ||
595 | if (end > b->bm_words) { | |
f88c5d90 | 596 | pr_alert("bm_memset end > bm_words\n"); |
b411b363 PR |
597 | return; |
598 | } | |
599 | ||
600 | while (offset < end) { | |
601 | do_now = min_t(size_t, ALIGN(offset + 1, LWPP), end) - offset; | |
19f843aa LE |
602 | idx = bm_word_to_page_idx(b, offset); |
603 | p_addr = bm_map_pidx(b, idx); | |
b411b363 PR |
604 | bm = p_addr + MLPP(offset); |
605 | if (bm+do_now > p_addr + LWPP) { | |
f88c5d90 | 606 | pr_alert("BUG BUG BUG! p_addr:%p bm:%p do_now:%d\n", |
b411b363 | 607 | p_addr, bm, (int)do_now); |
84e7c0f7 LE |
608 | } else |
609 | memset(bm, c, do_now * sizeof(long)); | |
b411b363 | 610 | bm_unmap(p_addr); |
19f843aa | 611 | bm_set_page_need_writeout(b->bm_pages[idx]); |
b411b363 PR |
612 | offset += do_now; |
613 | } | |
614 | } | |
615 | ||
ae8bf312 LE |
616 | /* For the layout, see comment above drbd_md_set_sector_offsets(). */ |
617 | static u64 drbd_md_on_disk_bits(struct drbd_backing_dev *ldev) | |
618 | { | |
619 | u64 bitmap_sectors; | |
620 | if (ldev->md.al_offset == 8) | |
621 | bitmap_sectors = ldev->md.md_size_sect - ldev->md.bm_offset; | |
622 | else | |
623 | bitmap_sectors = ldev->md.al_offset - ldev->md.bm_offset; | |
624 | return bitmap_sectors << (9 + 3); | |
625 | } | |
626 | ||
b411b363 PR |
627 | /* |
628 | * make sure the bitmap has enough room for the attached storage, | |
629 | * if necessary, resize. | |
630 | * called whenever we may have changed the device size. | |
631 | * returns -ENOMEM if we could not allocate enough memory, 0 on success. | |
632 | * In case this is actually a resize, we copy the old bitmap into the new one. | |
633 | * Otherwise, the bitmap is initialized to all bits set. | |
634 | */ | |
b30ab791 | 635 | int drbd_bm_resize(struct drbd_device *device, sector_t capacity, int set_new_bits) |
b411b363 | 636 | { |
b30ab791 | 637 | struct drbd_bitmap *b = device->bitmap; |
6850c442 | 638 | unsigned long bits, words, owords, obits; |
b411b363 PR |
639 | unsigned long want, have, onpages; /* number of pages */ |
640 | struct page **npages, **opages = NULL; | |
641 | int err = 0, growing; | |
642 | int opages_vmalloced; | |
643 | ||
841ce241 AG |
644 | if (!expect(b)) |
645 | return -ENOMEM; | |
b411b363 | 646 | |
b30ab791 | 647 | drbd_bm_lock(device, "resize", BM_LOCKED_MASK); |
b411b363 | 648 | |
d0180171 | 649 | drbd_info(device, "drbd_bm_resize called with capacity == %llu\n", |
b411b363 PR |
650 | (unsigned long long)capacity); |
651 | ||
652 | if (capacity == b->bm_dev_capacity) | |
653 | goto out; | |
654 | ||
20ceb2b2 | 655 | opages_vmalloced = (BM_P_VMALLOCED & b->bm_flags); |
b411b363 PR |
656 | |
657 | if (capacity == 0) { | |
658 | spin_lock_irq(&b->bm_lock); | |
659 | opages = b->bm_pages; | |
660 | onpages = b->bm_number_of_pages; | |
661 | owords = b->bm_words; | |
662 | b->bm_pages = NULL; | |
663 | b->bm_number_of_pages = | |
664 | b->bm_set = | |
665 | b->bm_bits = | |
666 | b->bm_words = | |
667 | b->bm_dev_capacity = 0; | |
668 | spin_unlock_irq(&b->bm_lock); | |
669 | bm_free_pages(opages, onpages); | |
670 | bm_vk_free(opages, opages_vmalloced); | |
671 | goto out; | |
672 | } | |
673 | bits = BM_SECT_TO_BIT(ALIGN(capacity, BM_SECT_PER_BIT)); | |
674 | ||
675 | /* if we would use | |
676 | words = ALIGN(bits,BITS_PER_LONG) >> LN2_BPL; | |
677 | a 32bit host could present the wrong number of words | |
678 | to a 64bit host. | |
679 | */ | |
680 | words = ALIGN(bits, 64) >> LN2_BPL; | |
681 | ||
b30ab791 AG |
682 | if (get_ldev(device)) { |
683 | u64 bits_on_disk = drbd_md_on_disk_bits(device->ldev); | |
684 | put_ldev(device); | |
4b0715f0 | 685 | if (bits > bits_on_disk) { |
d0180171 AG |
686 | drbd_info(device, "bits = %lu\n", bits); |
687 | drbd_info(device, "bits_on_disk = %llu\n", bits_on_disk); | |
4b0715f0 LE |
688 | err = -ENOSPC; |
689 | goto out; | |
690 | } | |
b411b363 PR |
691 | } |
692 | ||
6850c442 | 693 | want = ALIGN(words*sizeof(long), PAGE_SIZE) >> PAGE_SHIFT; |
b411b363 PR |
694 | have = b->bm_number_of_pages; |
695 | if (want == have) { | |
0b0ba1ef | 696 | D_ASSERT(device, b->bm_pages != NULL); |
b411b363 PR |
697 | npages = b->bm_pages; |
698 | } else { | |
b30ab791 | 699 | if (drbd_insert_fault(device, DRBD_FAULT_BM_ALLOC)) |
b411b363 PR |
700 | npages = NULL; |
701 | else | |
702 | npages = bm_realloc_pages(b, want); | |
703 | } | |
704 | ||
705 | if (!npages) { | |
706 | err = -ENOMEM; | |
707 | goto out; | |
708 | } | |
709 | ||
710 | spin_lock_irq(&b->bm_lock); | |
711 | opages = b->bm_pages; | |
712 | owords = b->bm_words; | |
713 | obits = b->bm_bits; | |
714 | ||
715 | growing = bits > obits; | |
5223671b | 716 | if (opages && growing && set_new_bits) |
b411b363 PR |
717 | bm_set_surplus(b); |
718 | ||
719 | b->bm_pages = npages; | |
720 | b->bm_number_of_pages = want; | |
721 | b->bm_bits = bits; | |
722 | b->bm_words = words; | |
723 | b->bm_dev_capacity = capacity; | |
724 | ||
725 | if (growing) { | |
02d9a94b PR |
726 | if (set_new_bits) { |
727 | bm_memset(b, owords, 0xff, words-owords); | |
728 | b->bm_set += bits - obits; | |
729 | } else | |
730 | bm_memset(b, owords, 0x00, words-owords); | |
731 | ||
b411b363 PR |
732 | } |
733 | ||
734 | if (want < have) { | |
735 | /* implicit: (opages != NULL) && (opages != npages) */ | |
736 | bm_free_pages(opages + want, have - want); | |
737 | } | |
738 | ||
b411b363 PR |
739 | (void)bm_clear_surplus(b); |
740 | ||
741 | spin_unlock_irq(&b->bm_lock); | |
742 | if (opages != npages) | |
743 | bm_vk_free(opages, opages_vmalloced); | |
744 | if (!growing) | |
745 | b->bm_set = bm_count_bits(b); | |
d0180171 | 746 | drbd_info(device, "resync bitmap: bits=%lu words=%lu pages=%lu\n", bits, words, want); |
b411b363 PR |
747 | |
748 | out: | |
b30ab791 | 749 | drbd_bm_unlock(device); |
b411b363 PR |
750 | return err; |
751 | } | |
752 | ||
753 | /* inherently racy: | |
754 | * if not protected by other means, return value may be out of date when | |
755 | * leaving this function... | |
756 | * we still need to lock it, since it is important that this returns | |
757 | * bm_set == 0 precisely. | |
758 | * | |
759 | * maybe bm_set should be atomic_t ? | |
760 | */ | |
b30ab791 | 761 | unsigned long _drbd_bm_total_weight(struct drbd_device *device) |
b411b363 | 762 | { |
b30ab791 | 763 | struct drbd_bitmap *b = device->bitmap; |
b411b363 PR |
764 | unsigned long s; |
765 | unsigned long flags; | |
766 | ||
841ce241 AG |
767 | if (!expect(b)) |
768 | return 0; | |
769 | if (!expect(b->bm_pages)) | |
770 | return 0; | |
b411b363 PR |
771 | |
772 | spin_lock_irqsave(&b->bm_lock, flags); | |
773 | s = b->bm_set; | |
774 | spin_unlock_irqrestore(&b->bm_lock, flags); | |
775 | ||
776 | return s; | |
777 | } | |
778 | ||
b30ab791 | 779 | unsigned long drbd_bm_total_weight(struct drbd_device *device) |
b411b363 PR |
780 | { |
781 | unsigned long s; | |
782 | /* if I don't have a disk, I don't know about out-of-sync status */ | |
b30ab791 | 783 | if (!get_ldev_if_state(device, D_NEGOTIATING)) |
b411b363 | 784 | return 0; |
b30ab791 AG |
785 | s = _drbd_bm_total_weight(device); |
786 | put_ldev(device); | |
b411b363 PR |
787 | return s; |
788 | } | |
789 | ||
b30ab791 | 790 | size_t drbd_bm_words(struct drbd_device *device) |
b411b363 | 791 | { |
b30ab791 | 792 | struct drbd_bitmap *b = device->bitmap; |
841ce241 AG |
793 | if (!expect(b)) |
794 | return 0; | |
795 | if (!expect(b->bm_pages)) | |
796 | return 0; | |
b411b363 PR |
797 | |
798 | return b->bm_words; | |
799 | } | |
800 | ||
b30ab791 | 801 | unsigned long drbd_bm_bits(struct drbd_device *device) |
b411b363 | 802 | { |
b30ab791 | 803 | struct drbd_bitmap *b = device->bitmap; |
841ce241 AG |
804 | if (!expect(b)) |
805 | return 0; | |
b411b363 PR |
806 | |
807 | return b->bm_bits; | |
808 | } | |
809 | ||
810 | /* merge number words from buffer into the bitmap starting at offset. | |
811 | * buffer[i] is expected to be little endian unsigned long. | |
812 | * bitmap must be locked by drbd_bm_lock. | |
813 | * currently only used from receive_bitmap. | |
814 | */ | |
b30ab791 | 815 | void drbd_bm_merge_lel(struct drbd_device *device, size_t offset, size_t number, |
b411b363 PR |
816 | unsigned long *buffer) |
817 | { | |
b30ab791 | 818 | struct drbd_bitmap *b = device->bitmap; |
b411b363 PR |
819 | unsigned long *p_addr, *bm; |
820 | unsigned long word, bits; | |
19f843aa | 821 | unsigned int idx; |
b411b363 PR |
822 | size_t end, do_now; |
823 | ||
824 | end = offset + number; | |
825 | ||
841ce241 AG |
826 | if (!expect(b)) |
827 | return; | |
828 | if (!expect(b->bm_pages)) | |
829 | return; | |
b411b363 PR |
830 | if (number == 0) |
831 | return; | |
832 | WARN_ON(offset >= b->bm_words); | |
833 | WARN_ON(end > b->bm_words); | |
834 | ||
835 | spin_lock_irq(&b->bm_lock); | |
836 | while (offset < end) { | |
837 | do_now = min_t(size_t, ALIGN(offset+1, LWPP), end) - offset; | |
19f843aa LE |
838 | idx = bm_word_to_page_idx(b, offset); |
839 | p_addr = bm_map_pidx(b, idx); | |
b411b363 PR |
840 | bm = p_addr + MLPP(offset); |
841 | offset += do_now; | |
842 | while (do_now--) { | |
843 | bits = hweight_long(*bm); | |
95a0f10c | 844 | word = *bm | *buffer++; |
b411b363 PR |
845 | *bm++ = word; |
846 | b->bm_set += hweight_long(word) - bits; | |
847 | } | |
848 | bm_unmap(p_addr); | |
19f843aa | 849 | bm_set_page_need_writeout(b->bm_pages[idx]); |
b411b363 PR |
850 | } |
851 | /* with 32bit <-> 64bit cross-platform connect | |
852 | * this is only correct for current usage, | |
853 | * where we _know_ that we are 64 bit aligned, | |
854 | * and know that this function is used in this way, too... | |
855 | */ | |
856 | if (end == b->bm_words) | |
857 | b->bm_set -= bm_clear_surplus(b); | |
b411b363 PR |
858 | spin_unlock_irq(&b->bm_lock); |
859 | } | |
860 | ||
861 | /* copy number words from the bitmap starting at offset into the buffer. | |
862 | * buffer[i] will be little endian unsigned long. | |
863 | */ | |
b30ab791 | 864 | void drbd_bm_get_lel(struct drbd_device *device, size_t offset, size_t number, |
b411b363 PR |
865 | unsigned long *buffer) |
866 | { | |
b30ab791 | 867 | struct drbd_bitmap *b = device->bitmap; |
b411b363 PR |
868 | unsigned long *p_addr, *bm; |
869 | size_t end, do_now; | |
870 | ||
871 | end = offset + number; | |
872 | ||
841ce241 AG |
873 | if (!expect(b)) |
874 | return; | |
875 | if (!expect(b->bm_pages)) | |
876 | return; | |
b411b363 PR |
877 | |
878 | spin_lock_irq(&b->bm_lock); | |
879 | if ((offset >= b->bm_words) || | |
880 | (end > b->bm_words) || | |
881 | (number <= 0)) | |
d0180171 | 882 | drbd_err(device, "offset=%lu number=%lu bm_words=%lu\n", |
b411b363 PR |
883 | (unsigned long) offset, |
884 | (unsigned long) number, | |
885 | (unsigned long) b->bm_words); | |
886 | else { | |
887 | while (offset < end) { | |
888 | do_now = min_t(size_t, ALIGN(offset+1, LWPP), end) - offset; | |
19f843aa | 889 | p_addr = bm_map_pidx(b, bm_word_to_page_idx(b, offset)); |
b411b363 PR |
890 | bm = p_addr + MLPP(offset); |
891 | offset += do_now; | |
892 | while (do_now--) | |
95a0f10c | 893 | *buffer++ = *bm++; |
b411b363 PR |
894 | bm_unmap(p_addr); |
895 | } | |
896 | } | |
897 | spin_unlock_irq(&b->bm_lock); | |
898 | } | |
899 | ||
900 | /* set all bits in the bitmap */ | |
b30ab791 | 901 | void drbd_bm_set_all(struct drbd_device *device) |
b411b363 | 902 | { |
b30ab791 | 903 | struct drbd_bitmap *b = device->bitmap; |
841ce241 AG |
904 | if (!expect(b)) |
905 | return; | |
906 | if (!expect(b->bm_pages)) | |
907 | return; | |
b411b363 PR |
908 | |
909 | spin_lock_irq(&b->bm_lock); | |
910 | bm_memset(b, 0, 0xff, b->bm_words); | |
911 | (void)bm_clear_surplus(b); | |
912 | b->bm_set = b->bm_bits; | |
913 | spin_unlock_irq(&b->bm_lock); | |
914 | } | |
915 | ||
916 | /* clear all bits in the bitmap */ | |
b30ab791 | 917 | void drbd_bm_clear_all(struct drbd_device *device) |
b411b363 | 918 | { |
b30ab791 | 919 | struct drbd_bitmap *b = device->bitmap; |
841ce241 AG |
920 | if (!expect(b)) |
921 | return; | |
922 | if (!expect(b->bm_pages)) | |
923 | return; | |
b411b363 PR |
924 | |
925 | spin_lock_irq(&b->bm_lock); | |
926 | bm_memset(b, 0, 0, b->bm_words); | |
927 | b->bm_set = 0; | |
928 | spin_unlock_irq(&b->bm_lock); | |
929 | } | |
930 | ||
4ce49266 | 931 | static void drbd_bm_aio_ctx_destroy(struct kref *kref) |
d1f3779b | 932 | { |
4ce49266 LE |
933 | struct drbd_bm_aio_ctx *ctx = container_of(kref, struct drbd_bm_aio_ctx, kref); |
934 | unsigned long flags; | |
d1f3779b | 935 | |
4ce49266 LE |
936 | spin_lock_irqsave(&ctx->device->resource->req_lock, flags); |
937 | list_del(&ctx->list); | |
938 | spin_unlock_irqrestore(&ctx->device->resource->req_lock, flags); | |
b30ab791 | 939 | put_ldev(ctx->device); |
d1f3779b PR |
940 | kfree(ctx); |
941 | } | |
942 | ||
19f843aa | 943 | /* bv_page may be a copy, or may be the original */ |
4246a0b6 | 944 | static void drbd_bm_endio(struct bio *bio) |
b411b363 | 945 | { |
4ce49266 | 946 | struct drbd_bm_aio_ctx *ctx = bio->bi_private; |
b30ab791 AG |
947 | struct drbd_device *device = ctx->device; |
948 | struct drbd_bitmap *b = device->bitmap; | |
19f843aa | 949 | unsigned int idx = bm_page_to_idx(bio->bi_io_vec[0].bv_page); |
b411b363 | 950 | |
7648cdfe LE |
951 | if ((ctx->flags & BM_AIO_COPY_PAGES) == 0 && |
952 | !bm_test_page_unchanged(b->bm_pages[idx])) | |
d0180171 | 953 | drbd_warn(device, "bitmap page idx %u changed during IO!\n", idx); |
19f843aa | 954 | |
4246a0b6 | 955 | if (bio->bi_error) { |
19f843aa LE |
956 | /* ctx error will hold the completed-last non-zero error code, |
957 | * in case error codes differ. */ | |
4246a0b6 | 958 | ctx->error = bio->bi_error; |
19f843aa LE |
959 | bm_set_page_io_err(b->bm_pages[idx]); |
960 | /* Not identical to on disk version of it. | |
961 | * Is BM_PAGE_IO_ERROR enough? */ | |
962 | if (__ratelimit(&drbd_ratelimit_state)) | |
d0180171 | 963 | drbd_err(device, "IO ERROR %d on bitmap page idx %u\n", |
4246a0b6 | 964 | bio->bi_error, idx); |
19f843aa LE |
965 | } else { |
966 | bm_clear_page_io_err(b->bm_pages[idx]); | |
d0180171 | 967 | dynamic_drbd_dbg(device, "bitmap page idx %u completed\n", idx); |
b411b363 | 968 | } |
19f843aa | 969 | |
b30ab791 | 970 | bm_page_unlock_io(device, idx); |
19f843aa | 971 | |
19f843aa | 972 | if (ctx->flags & BM_AIO_COPY_PAGES) |
4d95a10f | 973 | mempool_free(bio->bi_io_vec[0].bv_page, drbd_md_io_page_pool); |
b411b363 PR |
974 | |
975 | bio_put(bio); | |
19f843aa | 976 | |
d1f3779b | 977 | if (atomic_dec_and_test(&ctx->in_flight)) { |
9e58c4da | 978 | ctx->done = 1; |
b30ab791 | 979 | wake_up(&device->misc_wait); |
4ce49266 | 980 | kref_put(&ctx->kref, &drbd_bm_aio_ctx_destroy); |
d1f3779b | 981 | } |
b411b363 PR |
982 | } |
983 | ||
4ce49266 | 984 | static void bm_page_io_async(struct drbd_bm_aio_ctx *ctx, int page_nr) __must_hold(local) |
b411b363 | 985 | { |
9476f39d | 986 | struct bio *bio = bio_alloc_drbd(GFP_NOIO); |
b30ab791 AG |
987 | struct drbd_device *device = ctx->device; |
988 | struct drbd_bitmap *b = device->bitmap; | |
19f843aa | 989 | struct page *page; |
b411b363 | 990 | unsigned int len; |
4ce49266 | 991 | unsigned int rw = (ctx->flags & BM_AIO_READ) ? READ : WRITE; |
19f843aa | 992 | |
b411b363 | 993 | sector_t on_disk_sector = |
b30ab791 | 994 | device->ldev->md.md_offset + device->ldev->md.bm_offset; |
b411b363 PR |
995 | on_disk_sector += ((sector_t)page_nr) << (PAGE_SHIFT-9); |
996 | ||
997 | /* this might happen with very small | |
19f843aa LE |
998 | * flexible external meta data device, |
999 | * or with PAGE_SIZE > 4k */ | |
b411b363 | 1000 | len = min_t(unsigned int, PAGE_SIZE, |
b30ab791 | 1001 | (drbd_md_last_sector(device->ldev) - on_disk_sector + 1)<<9); |
b411b363 | 1002 | |
19f843aa | 1003 | /* serialize IO on this page */ |
b30ab791 | 1004 | bm_page_lock_io(device, page_nr); |
19f843aa LE |
1005 | /* before memcpy and submit, |
1006 | * so it can be redirtied any time */ | |
1007 | bm_set_page_unchanged(b->bm_pages[page_nr]); | |
1008 | ||
1009 | if (ctx->flags & BM_AIO_COPY_PAGES) { | |
4d95a10f | 1010 | page = mempool_alloc(drbd_md_io_page_pool, __GFP_HIGHMEM|__GFP_WAIT); |
f1d6a328 | 1011 | copy_highpage(page, b->bm_pages[page_nr]); |
19f843aa LE |
1012 | bm_store_page_idx(page, page_nr); |
1013 | } else | |
1014 | page = b->bm_pages[page_nr]; | |
b30ab791 | 1015 | bio->bi_bdev = device->ldev->md_bdev; |
4f024f37 | 1016 | bio->bi_iter.bi_sector = on_disk_sector; |
4d95a10f LE |
1017 | /* bio_add_page of a single page to an empty bio will always succeed, |
1018 | * according to api. Do we want to assert that? */ | |
19f843aa LE |
1019 | bio_add_page(bio, page, len, 0); |
1020 | bio->bi_private = ctx; | |
ed15b795 | 1021 | bio->bi_end_io = drbd_bm_endio; |
b411b363 | 1022 | |
b30ab791 | 1023 | if (drbd_insert_fault(device, (rw & WRITE) ? DRBD_FAULT_MD_WR : DRBD_FAULT_MD_RD)) { |
b411b363 | 1024 | bio->bi_rw |= rw; |
4246a0b6 | 1025 | bio_io_error(bio); |
b411b363 PR |
1026 | } else { |
1027 | submit_bio(rw, bio); | |
5a8b4242 LE |
1028 | /* this should not count as user activity and cause the |
1029 | * resync to throttle -- see drbd_rs_should_slow_down(). */ | |
b30ab791 | 1030 | atomic_add(len >> 9, &device->rs_sect_ev); |
b411b363 PR |
1031 | } |
1032 | } | |
1033 | ||
b411b363 PR |
1034 | /* |
1035 | * bm_rw: read/write the whole bitmap from/to its on disk location. | |
1036 | */ | |
4ce49266 | 1037 | static int bm_rw(struct drbd_device *device, const unsigned int flags, unsigned lazy_writeout_upper_idx) __must_hold(local) |
b411b363 | 1038 | { |
4ce49266 | 1039 | struct drbd_bm_aio_ctx *ctx; |
b30ab791 | 1040 | struct drbd_bitmap *b = device->bitmap; |
6850c442 | 1041 | int num_pages, i, count = 0; |
b411b363 PR |
1042 | unsigned long now; |
1043 | char ppb[10]; | |
1044 | int err = 0; | |
1045 | ||
19f843aa LE |
1046 | /* |
1047 | * We are protected against bitmap disappearing/resizing by holding an | |
1048 | * ldev reference (caller must have called get_ldev()). | |
1049 | * For read/write, we are protected against changes to the bitmap by | |
1050 | * the bitmap lock (see drbd_bitmap_io). | |
1051 | * For lazy writeout, we don't care for ongoing changes to the bitmap, | |
1052 | * as we submit copies of pages anyways. | |
1053 | */ | |
d1f3779b | 1054 | |
4ce49266 | 1055 | ctx = kmalloc(sizeof(struct drbd_bm_aio_ctx), GFP_NOIO); |
d1f3779b PR |
1056 | if (!ctx) |
1057 | return -ENOMEM; | |
1058 | ||
4ce49266 | 1059 | *ctx = (struct drbd_bm_aio_ctx) { |
b30ab791 | 1060 | .device = device, |
4ce49266 | 1061 | .start_jif = jiffies, |
d1f3779b | 1062 | .in_flight = ATOMIC_INIT(1), |
9e58c4da | 1063 | .done = 0, |
0e8488ad | 1064 | .flags = flags, |
d1f3779b PR |
1065 | .error = 0, |
1066 | .kref = { ATOMIC_INIT(2) }, | |
1067 | }; | |
1068 | ||
4ce49266 | 1069 | if (!get_ldev_if_state(device, D_ATTACHING)) { /* put is in drbd_bm_aio_ctx_destroy() */ |
d0180171 | 1070 | drbd_err(device, "ASSERT FAILED: get_ldev_if_state() == 1 in bm_rw()\n"); |
9e58c4da PR |
1071 | kfree(ctx); |
1072 | return -ENODEV; | |
1073 | } | |
8fe39aac PR |
1074 | /* Here D_ATTACHING is sufficient since drbd_bm_read() is called only from |
1075 | drbd_adm_attach(), after device->ldev was assigned. */ | |
9e58c4da | 1076 | |
4ce49266 | 1077 | if (0 == (ctx->flags & ~BM_AIO_READ)) |
20ceb2b2 | 1078 | WARN_ON(!(BM_LOCKED_MASK & b->bm_flags)); |
b411b363 | 1079 | |
4ce49266 LE |
1080 | spin_lock_irq(&device->resource->req_lock); |
1081 | list_add_tail(&ctx->list, &device->pending_bitmap_io); | |
1082 | spin_unlock_irq(&device->resource->req_lock); | |
1083 | ||
6850c442 | 1084 | num_pages = b->bm_number_of_pages; |
b411b363 | 1085 | |
b411b363 | 1086 | now = jiffies; |
b411b363 PR |
1087 | |
1088 | /* let the layers below us try to merge these bios... */ | |
6850c442 | 1089 | for (i = 0; i < num_pages; i++) { |
19f843aa LE |
1090 | /* ignore completely unchanged pages */ |
1091 | if (lazy_writeout_upper_idx && i == lazy_writeout_upper_idx) | |
1092 | break; | |
4ce49266 | 1093 | if (!(flags & BM_AIO_READ)) { |
45dfffeb LE |
1094 | if ((flags & BM_AIO_WRITE_HINTED) && |
1095 | !test_and_clear_bit(BM_PAGE_HINT_WRITEOUT, | |
1096 | &page_private(b->bm_pages[i]))) | |
1097 | continue; | |
fef45d29 | 1098 | |
4ce49266 | 1099 | if (!(flags & BM_AIO_WRITE_ALL_PAGES) && |
d1aa4d04 | 1100 | bm_test_page_unchanged(b->bm_pages[i])) { |
d0180171 | 1101 | dynamic_drbd_dbg(device, "skipped bm write for idx %u\n", i); |
19f843aa LE |
1102 | continue; |
1103 | } | |
1104 | /* during lazy writeout, | |
1105 | * ignore those pages not marked for lazy writeout. */ | |
1106 | if (lazy_writeout_upper_idx && | |
1107 | !bm_test_page_lazy_writeout(b->bm_pages[i])) { | |
d0180171 | 1108 | dynamic_drbd_dbg(device, "skipped bm lazy write for idx %u\n", i); |
19f843aa LE |
1109 | continue; |
1110 | } | |
1111 | } | |
d1f3779b | 1112 | atomic_inc(&ctx->in_flight); |
4ce49266 | 1113 | bm_page_io_async(ctx, i); |
19f843aa LE |
1114 | ++count; |
1115 | cond_resched(); | |
1116 | } | |
b411b363 | 1117 | |
725a97e4 | 1118 | /* |
ed15b795 | 1119 | * We initialize ctx->in_flight to one to make sure drbd_bm_endio |
9e58c4da | 1120 | * will not set ctx->done early, and decrement / test it here. If there |
725a97e4 | 1121 | * are still some bios in flight, we need to wait for them here. |
9e58c4da PR |
1122 | * If all IO is done already (or nothing had been submitted), there is |
1123 | * no need to wait. Still, we need to put the kref associated with the | |
1124 | * "in_flight reached zero, all done" event. | |
725a97e4 | 1125 | */ |
d1f3779b | 1126 | if (!atomic_dec_and_test(&ctx->in_flight)) |
b30ab791 | 1127 | wait_until_done_or_force_detached(device, device->ldev, &ctx->done); |
9e58c4da | 1128 | else |
4ce49266 | 1129 | kref_put(&ctx->kref, &drbd_bm_aio_ctx_destroy); |
d1f3779b | 1130 | |
c9d963a4 LE |
1131 | /* summary for global bitmap IO */ |
1132 | if (flags == 0) | |
d0180171 | 1133 | drbd_info(device, "bitmap %s of %u pages took %lu jiffies\n", |
4ce49266 | 1134 | (flags & BM_AIO_READ) ? "READ" : "WRITE", |
cdfda633 | 1135 | count, jiffies - now); |
b411b363 | 1136 | |
d1f3779b | 1137 | if (ctx->error) { |
d0180171 | 1138 | drbd_alert(device, "we had at least one MD IO ERROR during bitmap IO\n"); |
b30ab791 | 1139 | drbd_chk_io_error(device, 1, DRBD_META_IO_ERROR); |
d1f3779b | 1140 | err = -EIO; /* ctx->error ? */ |
b411b363 PR |
1141 | } |
1142 | ||
9e58c4da | 1143 | if (atomic_read(&ctx->in_flight)) |
44edfb0d | 1144 | err = -EIO; /* Disk timeout/force-detach during IO... */ |
9e58c4da | 1145 | |
b411b363 | 1146 | now = jiffies; |
4ce49266 | 1147 | if (flags & BM_AIO_READ) { |
95a0f10c | 1148 | b->bm_set = bm_count_bits(b); |
d0180171 | 1149 | drbd_info(device, "recounting of set bits took additional %lu jiffies\n", |
b411b363 PR |
1150 | jiffies - now); |
1151 | } | |
1152 | now = b->bm_set; | |
1153 | ||
4ce49266 | 1154 | if ((flags & ~BM_AIO_READ) == 0) |
d0180171 | 1155 | drbd_info(device, "%s (%lu bits) marked out-of-sync by on disk bit-map.\n", |
c9d963a4 | 1156 | ppsize(ppb, now << (BM_BLOCK_SHIFT-10)), now); |
b411b363 | 1157 | |
4ce49266 | 1158 | kref_put(&ctx->kref, &drbd_bm_aio_ctx_destroy); |
b411b363 PR |
1159 | return err; |
1160 | } | |
1161 | ||
1162 | /** | |
1163 | * drbd_bm_read() - Read the whole bitmap from its on disk location. | |
b30ab791 | 1164 | * @device: DRBD device. |
b411b363 | 1165 | */ |
b30ab791 | 1166 | int drbd_bm_read(struct drbd_device *device) __must_hold(local) |
b411b363 | 1167 | { |
4ce49266 | 1168 | return bm_rw(device, BM_AIO_READ, 0); |
b411b363 PR |
1169 | } |
1170 | ||
1171 | /** | |
1172 | * drbd_bm_write() - Write the whole bitmap to its on disk location. | |
b30ab791 | 1173 | * @device: DRBD device. |
19f843aa LE |
1174 | * |
1175 | * Will only write pages that have changed since last IO. | |
b411b363 | 1176 | */ |
b30ab791 | 1177 | int drbd_bm_write(struct drbd_device *device) __must_hold(local) |
b411b363 | 1178 | { |
4ce49266 | 1179 | return bm_rw(device, 0, 0); |
b411b363 PR |
1180 | } |
1181 | ||
d1aa4d04 PR |
1182 | /** |
1183 | * drbd_bm_write_all() - Write the whole bitmap to its on disk location. | |
b30ab791 | 1184 | * @device: DRBD device. |
d1aa4d04 PR |
1185 | * |
1186 | * Will write all pages. | |
1187 | */ | |
b30ab791 | 1188 | int drbd_bm_write_all(struct drbd_device *device) __must_hold(local) |
d1aa4d04 | 1189 | { |
4ce49266 | 1190 | return bm_rw(device, BM_AIO_WRITE_ALL_PAGES, 0); |
d1aa4d04 PR |
1191 | } |
1192 | ||
c7a58db4 LE |
1193 | /** |
1194 | * drbd_bm_write_lazy() - Write bitmap pages 0 to @upper_idx-1, if they have changed. | |
1195 | * @device: DRBD device. | |
1196 | * @upper_idx: 0: write all changed pages; +ve: page index to stop scanning for changed pages | |
1197 | */ | |
1198 | int drbd_bm_write_lazy(struct drbd_device *device, unsigned upper_idx) __must_hold(local) | |
1199 | { | |
1200 | return bm_rw(device, BM_AIO_COPY_PAGES, upper_idx); | |
1201 | } | |
1202 | ||
0e8488ad LE |
1203 | /** |
1204 | * drbd_bm_write_copy_pages() - Write the whole bitmap to its on disk location. | |
b30ab791 | 1205 | * @device: DRBD device. |
0e8488ad LE |
1206 | * |
1207 | * Will only write pages that have changed since last IO. | |
1208 | * In contrast to drbd_bm_write(), this will copy the bitmap pages | |
1209 | * to temporary writeout pages. It is intended to trigger a full write-out | |
1210 | * while still allowing the bitmap to change, for example if a resync or online | |
1211 | * verify is aborted due to a failed peer disk, while local IO continues, or | |
1212 | * pending resync acks are still being processed. | |
1213 | */ | |
b30ab791 | 1214 | int drbd_bm_write_copy_pages(struct drbd_device *device) __must_hold(local) |
0e8488ad | 1215 | { |
4ce49266 | 1216 | return bm_rw(device, BM_AIO_COPY_PAGES, 0); |
19f843aa LE |
1217 | } |
1218 | ||
45dfffeb LE |
1219 | /** |
1220 | * drbd_bm_write_hinted() - Write bitmap pages with "hint" marks, if they have changed. | |
b30ab791 | 1221 | * @device: DRBD device. |
45dfffeb | 1222 | */ |
b30ab791 | 1223 | int drbd_bm_write_hinted(struct drbd_device *device) __must_hold(local) |
45dfffeb | 1224 | { |
4ce49266 | 1225 | return bm_rw(device, BM_AIO_WRITE_HINTED | BM_AIO_COPY_PAGES, 0); |
45dfffeb | 1226 | } |
19f843aa | 1227 | |
b411b363 PR |
1228 | /* NOTE |
1229 | * find_first_bit returns int, we return unsigned long. | |
4b0715f0 LE |
1230 | * For this to work on 32bit arch with bitnumbers > (1<<32), |
1231 | * we'd need to return u64, and get a whole lot of other places | |
1232 | * fixed where we still use unsigned long. | |
b411b363 PR |
1233 | * |
1234 | * this returns a bit number, NOT a sector! | |
1235 | */ | |
b30ab791 | 1236 | static unsigned long __bm_find_next(struct drbd_device *device, unsigned long bm_fo, |
cfd8005c | 1237 | const int find_zero_bit) |
b411b363 | 1238 | { |
b30ab791 | 1239 | struct drbd_bitmap *b = device->bitmap; |
b411b363 | 1240 | unsigned long *p_addr; |
4b0715f0 LE |
1241 | unsigned long bit_offset; |
1242 | unsigned i; | |
1243 | ||
b411b363 PR |
1244 | |
1245 | if (bm_fo > b->bm_bits) { | |
d0180171 | 1246 | drbd_err(device, "bm_fo=%lu bm_bits=%lu\n", bm_fo, b->bm_bits); |
4b0715f0 | 1247 | bm_fo = DRBD_END_OF_BITMAP; |
b411b363 PR |
1248 | } else { |
1249 | while (bm_fo < b->bm_bits) { | |
19f843aa | 1250 | /* bit offset of the first bit in the page */ |
4b0715f0 | 1251 | bit_offset = bm_fo & ~BITS_PER_PAGE_MASK; |
cfd8005c | 1252 | p_addr = __bm_map_pidx(b, bm_bit_to_page_idx(b, bm_fo)); |
b411b363 PR |
1253 | |
1254 | if (find_zero_bit) | |
7e599e6e | 1255 | i = find_next_zero_bit_le(p_addr, |
4b0715f0 | 1256 | PAGE_SIZE*8, bm_fo & BITS_PER_PAGE_MASK); |
b411b363 | 1257 | else |
7e599e6e | 1258 | i = find_next_bit_le(p_addr, |
4b0715f0 | 1259 | PAGE_SIZE*8, bm_fo & BITS_PER_PAGE_MASK); |
b411b363 | 1260 | |
cfd8005c | 1261 | __bm_unmap(p_addr); |
b411b363 | 1262 | if (i < PAGE_SIZE*8) { |
4b0715f0 LE |
1263 | bm_fo = bit_offset + i; |
1264 | if (bm_fo >= b->bm_bits) | |
b411b363 PR |
1265 | break; |
1266 | goto found; | |
1267 | } | |
1268 | bm_fo = bit_offset + PAGE_SIZE*8; | |
1269 | } | |
4b0715f0 | 1270 | bm_fo = DRBD_END_OF_BITMAP; |
b411b363 PR |
1271 | } |
1272 | found: | |
4b0715f0 | 1273 | return bm_fo; |
b411b363 PR |
1274 | } |
1275 | ||
b30ab791 | 1276 | static unsigned long bm_find_next(struct drbd_device *device, |
b411b363 PR |
1277 | unsigned long bm_fo, const int find_zero_bit) |
1278 | { | |
b30ab791 | 1279 | struct drbd_bitmap *b = device->bitmap; |
4b0715f0 | 1280 | unsigned long i = DRBD_END_OF_BITMAP; |
b411b363 | 1281 | |
841ce241 AG |
1282 | if (!expect(b)) |
1283 | return i; | |
1284 | if (!expect(b->bm_pages)) | |
1285 | return i; | |
b411b363 PR |
1286 | |
1287 | spin_lock_irq(&b->bm_lock); | |
20ceb2b2 | 1288 | if (BM_DONT_TEST & b->bm_flags) |
b30ab791 | 1289 | bm_print_lock_info(device); |
b411b363 | 1290 | |
b30ab791 | 1291 | i = __bm_find_next(device, bm_fo, find_zero_bit); |
b411b363 PR |
1292 | |
1293 | spin_unlock_irq(&b->bm_lock); | |
1294 | return i; | |
1295 | } | |
1296 | ||
b30ab791 | 1297 | unsigned long drbd_bm_find_next(struct drbd_device *device, unsigned long bm_fo) |
b411b363 | 1298 | { |
b30ab791 | 1299 | return bm_find_next(device, bm_fo, 0); |
b411b363 PR |
1300 | } |
1301 | ||
1302 | #if 0 | |
1303 | /* not yet needed for anything. */ | |
b30ab791 | 1304 | unsigned long drbd_bm_find_next_zero(struct drbd_device *device, unsigned long bm_fo) |
b411b363 | 1305 | { |
b30ab791 | 1306 | return bm_find_next(device, bm_fo, 1); |
b411b363 PR |
1307 | } |
1308 | #endif | |
1309 | ||
1310 | /* does not spin_lock_irqsave. | |
1311 | * you must take drbd_bm_lock() first */ | |
b30ab791 | 1312 | unsigned long _drbd_bm_find_next(struct drbd_device *device, unsigned long bm_fo) |
b411b363 | 1313 | { |
b30ab791 AG |
1314 | /* WARN_ON(!(BM_DONT_SET & device->b->bm_flags)); */ |
1315 | return __bm_find_next(device, bm_fo, 0); | |
b411b363 PR |
1316 | } |
1317 | ||
b30ab791 | 1318 | unsigned long _drbd_bm_find_next_zero(struct drbd_device *device, unsigned long bm_fo) |
b411b363 | 1319 | { |
b30ab791 AG |
1320 | /* WARN_ON(!(BM_DONT_SET & device->b->bm_flags)); */ |
1321 | return __bm_find_next(device, bm_fo, 1); | |
b411b363 PR |
1322 | } |
1323 | ||
1324 | /* returns number of bits actually changed. | |
1325 | * for val != 0, we change 0 -> 1, return code positive | |
1326 | * for val == 0, we change 1 -> 0, return code negative | |
1327 | * wants bitnr, not sector. | |
1328 | * expected to be called for only a few bits (e - s about BITS_PER_LONG). | |
1329 | * Must hold bitmap lock already. */ | |
b30ab791 | 1330 | static int __bm_change_bits_to(struct drbd_device *device, const unsigned long s, |
829c6087 | 1331 | unsigned long e, int val) |
b411b363 | 1332 | { |
b30ab791 | 1333 | struct drbd_bitmap *b = device->bitmap; |
b411b363 PR |
1334 | unsigned long *p_addr = NULL; |
1335 | unsigned long bitnr; | |
19f843aa | 1336 | unsigned int last_page_nr = -1U; |
b411b363 | 1337 | int c = 0; |
19f843aa | 1338 | int changed_total = 0; |
b411b363 PR |
1339 | |
1340 | if (e >= b->bm_bits) { | |
d0180171 | 1341 | drbd_err(device, "ASSERT FAILED: bit_s=%lu bit_e=%lu bm_bits=%lu\n", |
b411b363 PR |
1342 | s, e, b->bm_bits); |
1343 | e = b->bm_bits ? b->bm_bits -1 : 0; | |
1344 | } | |
1345 | for (bitnr = s; bitnr <= e; bitnr++) { | |
19f843aa | 1346 | unsigned int page_nr = bm_bit_to_page_idx(b, bitnr); |
b411b363 PR |
1347 | if (page_nr != last_page_nr) { |
1348 | if (p_addr) | |
cfd8005c | 1349 | __bm_unmap(p_addr); |
19f843aa LE |
1350 | if (c < 0) |
1351 | bm_set_page_lazy_writeout(b->bm_pages[last_page_nr]); | |
1352 | else if (c > 0) | |
1353 | bm_set_page_need_writeout(b->bm_pages[last_page_nr]); | |
1354 | changed_total += c; | |
1355 | c = 0; | |
cfd8005c | 1356 | p_addr = __bm_map_pidx(b, page_nr); |
b411b363 PR |
1357 | last_page_nr = page_nr; |
1358 | } | |
1359 | if (val) | |
7e599e6e | 1360 | c += (0 == __test_and_set_bit_le(bitnr & BITS_PER_PAGE_MASK, p_addr)); |
b411b363 | 1361 | else |
7e599e6e | 1362 | c -= (0 != __test_and_clear_bit_le(bitnr & BITS_PER_PAGE_MASK, p_addr)); |
b411b363 PR |
1363 | } |
1364 | if (p_addr) | |
cfd8005c | 1365 | __bm_unmap(p_addr); |
19f843aa LE |
1366 | if (c < 0) |
1367 | bm_set_page_lazy_writeout(b->bm_pages[last_page_nr]); | |
1368 | else if (c > 0) | |
1369 | bm_set_page_need_writeout(b->bm_pages[last_page_nr]); | |
1370 | changed_total += c; | |
1371 | b->bm_set += changed_total; | |
1372 | return changed_total; | |
b411b363 PR |
1373 | } |
1374 | ||
1375 | /* returns number of bits actually changed. | |
1376 | * for val != 0, we change 0 -> 1, return code positive | |
1377 | * for val == 0, we change 1 -> 0, return code negative | |
1378 | * wants bitnr, not sector */ | |
b30ab791 | 1379 | static int bm_change_bits_to(struct drbd_device *device, const unsigned long s, |
b411b363 PR |
1380 | const unsigned long e, int val) |
1381 | { | |
1382 | unsigned long flags; | |
b30ab791 | 1383 | struct drbd_bitmap *b = device->bitmap; |
b411b363 PR |
1384 | int c = 0; |
1385 | ||
841ce241 AG |
1386 | if (!expect(b)) |
1387 | return 1; | |
1388 | if (!expect(b->bm_pages)) | |
1389 | return 0; | |
b411b363 PR |
1390 | |
1391 | spin_lock_irqsave(&b->bm_lock, flags); | |
20ceb2b2 | 1392 | if ((val ? BM_DONT_SET : BM_DONT_CLEAR) & b->bm_flags) |
b30ab791 | 1393 | bm_print_lock_info(device); |
b411b363 | 1394 | |
b30ab791 | 1395 | c = __bm_change_bits_to(device, s, e, val); |
b411b363 PR |
1396 | |
1397 | spin_unlock_irqrestore(&b->bm_lock, flags); | |
1398 | return c; | |
1399 | } | |
1400 | ||
1401 | /* returns number of bits changed 0 -> 1 */ | |
b30ab791 | 1402 | int drbd_bm_set_bits(struct drbd_device *device, const unsigned long s, const unsigned long e) |
b411b363 | 1403 | { |
b30ab791 | 1404 | return bm_change_bits_to(device, s, e, 1); |
b411b363 PR |
1405 | } |
1406 | ||
1407 | /* returns number of bits changed 1 -> 0 */ | |
b30ab791 | 1408 | int drbd_bm_clear_bits(struct drbd_device *device, const unsigned long s, const unsigned long e) |
b411b363 | 1409 | { |
b30ab791 | 1410 | return -bm_change_bits_to(device, s, e, 0); |
b411b363 PR |
1411 | } |
1412 | ||
1413 | /* sets all bits in full words, | |
1414 | * from first_word up to, but not including, last_word */ | |
1415 | static inline void bm_set_full_words_within_one_page(struct drbd_bitmap *b, | |
1416 | int page_nr, int first_word, int last_word) | |
1417 | { | |
1418 | int i; | |
1419 | int bits; | |
22d81140 | 1420 | int changed = 0; |
cfd8005c | 1421 | unsigned long *paddr = kmap_atomic(b->bm_pages[page_nr]); |
b411b363 PR |
1422 | for (i = first_word; i < last_word; i++) { |
1423 | bits = hweight_long(paddr[i]); | |
1424 | paddr[i] = ~0UL; | |
22d81140 | 1425 | changed += BITS_PER_LONG - bits; |
b411b363 | 1426 | } |
cfd8005c | 1427 | kunmap_atomic(paddr); |
22d81140 LE |
1428 | if (changed) { |
1429 | /* We only need lazy writeout, the information is still in the | |
1430 | * remote bitmap as well, and is reconstructed during the next | |
1431 | * bitmap exchange, if lost locally due to a crash. */ | |
1432 | bm_set_page_lazy_writeout(b->bm_pages[page_nr]); | |
1433 | b->bm_set += changed; | |
1434 | } | |
b411b363 PR |
1435 | } |
1436 | ||
829c6087 LE |
1437 | /* Same thing as drbd_bm_set_bits, |
1438 | * but more efficient for a large bit range. | |
b411b363 PR |
1439 | * You must first drbd_bm_lock(). |
1440 | * Can be called to set the whole bitmap in one go. | |
1441 | * Sets bits from s to e _inclusive_. */ | |
b30ab791 | 1442 | void _drbd_bm_set_bits(struct drbd_device *device, const unsigned long s, const unsigned long e) |
b411b363 PR |
1443 | { |
1444 | /* First set_bit from the first bit (s) | |
1445 | * up to the next long boundary (sl), | |
1446 | * then assign full words up to the last long boundary (el), | |
1447 | * then set_bit up to and including the last bit (e). | |
1448 | * | |
1449 | * Do not use memset, because we must account for changes, | |
1450 | * so we need to loop over the words with hweight() anyways. | |
1451 | */ | |
b30ab791 | 1452 | struct drbd_bitmap *b = device->bitmap; |
b411b363 PR |
1453 | unsigned long sl = ALIGN(s,BITS_PER_LONG); |
1454 | unsigned long el = (e+1) & ~((unsigned long)BITS_PER_LONG-1); | |
1455 | int first_page; | |
1456 | int last_page; | |
1457 | int page_nr; | |
1458 | int first_word; | |
1459 | int last_word; | |
1460 | ||
1461 | if (e - s <= 3*BITS_PER_LONG) { | |
1462 | /* don't bother; el and sl may even be wrong. */ | |
829c6087 | 1463 | spin_lock_irq(&b->bm_lock); |
b30ab791 | 1464 | __bm_change_bits_to(device, s, e, 1); |
829c6087 | 1465 | spin_unlock_irq(&b->bm_lock); |
b411b363 PR |
1466 | return; |
1467 | } | |
1468 | ||
1469 | /* difference is large enough that we can trust sl and el */ | |
1470 | ||
829c6087 LE |
1471 | spin_lock_irq(&b->bm_lock); |
1472 | ||
b411b363 PR |
1473 | /* bits filling the current long */ |
1474 | if (sl) | |
b30ab791 | 1475 | __bm_change_bits_to(device, s, sl-1, 1); |
b411b363 PR |
1476 | |
1477 | first_page = sl >> (3 + PAGE_SHIFT); | |
1478 | last_page = el >> (3 + PAGE_SHIFT); | |
1479 | ||
1480 | /* MLPP: modulo longs per page */ | |
1481 | /* LWPP: long words per page */ | |
1482 | first_word = MLPP(sl >> LN2_BPL); | |
1483 | last_word = LWPP; | |
1484 | ||
1485 | /* first and full pages, unless first page == last page */ | |
1486 | for (page_nr = first_page; page_nr < last_page; page_nr++) { | |
b30ab791 | 1487 | bm_set_full_words_within_one_page(device->bitmap, page_nr, first_word, last_word); |
8ccee20e LE |
1488 | spin_unlock_irq(&b->bm_lock); |
1489 | cond_resched(); | |
b411b363 | 1490 | first_word = 0; |
8ccee20e | 1491 | spin_lock_irq(&b->bm_lock); |
b411b363 | 1492 | } |
b411b363 PR |
1493 | /* last page (respectively only page, for first page == last page) */ |
1494 | last_word = MLPP(el >> LN2_BPL); | |
4eccc579 LE |
1495 | |
1496 | /* consider bitmap->bm_bits = 32768, bitmap->bm_number_of_pages = 1. (or multiples). | |
1497 | * ==> e = 32767, el = 32768, last_page = 2, | |
1498 | * and now last_word = 0. | |
1499 | * We do not want to touch last_page in this case, | |
1500 | * as we did not allocate it, it is not present in bitmap->bm_pages. | |
1501 | */ | |
1502 | if (last_word) | |
b30ab791 | 1503 | bm_set_full_words_within_one_page(device->bitmap, last_page, first_word, last_word); |
b411b363 PR |
1504 | |
1505 | /* possibly trailing bits. | |
1506 | * example: (e & 63) == 63, el will be e+1. | |
1507 | * if that even was the very last bit, | |
1508 | * it would trigger an assert in __bm_change_bits_to() | |
1509 | */ | |
1510 | if (el <= e) | |
b30ab791 | 1511 | __bm_change_bits_to(device, el, e, 1); |
829c6087 | 1512 | spin_unlock_irq(&b->bm_lock); |
b411b363 PR |
1513 | } |
1514 | ||
1515 | /* returns bit state | |
1516 | * wants bitnr, NOT sector. | |
1517 | * inherently racy... area needs to be locked by means of {al,rs}_lru | |
1518 | * 1 ... bit set | |
1519 | * 0 ... bit not set | |
1520 | * -1 ... first out of bounds access, stop testing for bits! | |
1521 | */ | |
b30ab791 | 1522 | int drbd_bm_test_bit(struct drbd_device *device, const unsigned long bitnr) |
b411b363 PR |
1523 | { |
1524 | unsigned long flags; | |
b30ab791 | 1525 | struct drbd_bitmap *b = device->bitmap; |
b411b363 PR |
1526 | unsigned long *p_addr; |
1527 | int i; | |
1528 | ||
841ce241 AG |
1529 | if (!expect(b)) |
1530 | return 0; | |
1531 | if (!expect(b->bm_pages)) | |
1532 | return 0; | |
b411b363 PR |
1533 | |
1534 | spin_lock_irqsave(&b->bm_lock, flags); | |
20ceb2b2 | 1535 | if (BM_DONT_TEST & b->bm_flags) |
b30ab791 | 1536 | bm_print_lock_info(device); |
b411b363 | 1537 | if (bitnr < b->bm_bits) { |
19f843aa | 1538 | p_addr = bm_map_pidx(b, bm_bit_to_page_idx(b, bitnr)); |
7e599e6e | 1539 | i = test_bit_le(bitnr & BITS_PER_PAGE_MASK, p_addr) ? 1 : 0; |
b411b363 PR |
1540 | bm_unmap(p_addr); |
1541 | } else if (bitnr == b->bm_bits) { | |
1542 | i = -1; | |
1543 | } else { /* (bitnr > b->bm_bits) */ | |
d0180171 | 1544 | drbd_err(device, "bitnr=%lu > bm_bits=%lu\n", bitnr, b->bm_bits); |
b411b363 PR |
1545 | i = 0; |
1546 | } | |
1547 | ||
1548 | spin_unlock_irqrestore(&b->bm_lock, flags); | |
1549 | return i; | |
1550 | } | |
1551 | ||
1552 | /* returns number of bits set in the range [s, e] */ | |
b30ab791 | 1553 | int drbd_bm_count_bits(struct drbd_device *device, const unsigned long s, const unsigned long e) |
b411b363 PR |
1554 | { |
1555 | unsigned long flags; | |
b30ab791 | 1556 | struct drbd_bitmap *b = device->bitmap; |
19f843aa | 1557 | unsigned long *p_addr = NULL; |
b411b363 | 1558 | unsigned long bitnr; |
19f843aa | 1559 | unsigned int page_nr = -1U; |
b411b363 | 1560 | int c = 0; |
b411b363 PR |
1561 | |
1562 | /* If this is called without a bitmap, that is a bug. But just to be | |
1563 | * robust in case we screwed up elsewhere, in that case pretend there | |
1564 | * was one dirty bit in the requested area, so we won't try to do a | |
1565 | * local read there (no bitmap probably implies no disk) */ | |
841ce241 AG |
1566 | if (!expect(b)) |
1567 | return 1; | |
1568 | if (!expect(b->bm_pages)) | |
1569 | return 1; | |
b411b363 PR |
1570 | |
1571 | spin_lock_irqsave(&b->bm_lock, flags); | |
20ceb2b2 | 1572 | if (BM_DONT_TEST & b->bm_flags) |
b30ab791 | 1573 | bm_print_lock_info(device); |
b411b363 | 1574 | for (bitnr = s; bitnr <= e; bitnr++) { |
19f843aa LE |
1575 | unsigned int idx = bm_bit_to_page_idx(b, bitnr); |
1576 | if (page_nr != idx) { | |
1577 | page_nr = idx; | |
b411b363 PR |
1578 | if (p_addr) |
1579 | bm_unmap(p_addr); | |
19f843aa | 1580 | p_addr = bm_map_pidx(b, idx); |
b411b363 | 1581 | } |
841ce241 | 1582 | if (expect(bitnr < b->bm_bits)) |
7e599e6e | 1583 | c += (0 != test_bit_le(bitnr - (page_nr << (PAGE_SHIFT+3)), p_addr)); |
841ce241 | 1584 | else |
d0180171 | 1585 | drbd_err(device, "bitnr=%lu bm_bits=%lu\n", bitnr, b->bm_bits); |
b411b363 PR |
1586 | } |
1587 | if (p_addr) | |
1588 | bm_unmap(p_addr); | |
1589 | spin_unlock_irqrestore(&b->bm_lock, flags); | |
1590 | return c; | |
1591 | } | |
1592 | ||
1593 | ||
1594 | /* inherently racy... | |
1595 | * return value may be already out-of-date when this function returns. | |
1596 | * but the general usage is that this is only use during a cstate when bits are | |
1597 | * only cleared, not set, and typically only care for the case when the return | |
1598 | * value is zero, or we already "locked" this "bitmap extent" by other means. | |
1599 | * | |
1600 | * enr is bm-extent number, since we chose to name one sector (512 bytes) | |
1601 | * worth of the bitmap a "bitmap extent". | |
1602 | * | |
1603 | * TODO | |
1604 | * I think since we use it like a reference count, we should use the real | |
1605 | * reference count of some bitmap extent element from some lru instead... | |
1606 | * | |
1607 | */ | |
b30ab791 | 1608 | int drbd_bm_e_weight(struct drbd_device *device, unsigned long enr) |
b411b363 | 1609 | { |
b30ab791 | 1610 | struct drbd_bitmap *b = device->bitmap; |
b411b363 PR |
1611 | int count, s, e; |
1612 | unsigned long flags; | |
1613 | unsigned long *p_addr, *bm; | |
1614 | ||
841ce241 AG |
1615 | if (!expect(b)) |
1616 | return 0; | |
1617 | if (!expect(b->bm_pages)) | |
1618 | return 0; | |
b411b363 PR |
1619 | |
1620 | spin_lock_irqsave(&b->bm_lock, flags); | |
20ceb2b2 | 1621 | if (BM_DONT_TEST & b->bm_flags) |
b30ab791 | 1622 | bm_print_lock_info(device); |
b411b363 PR |
1623 | |
1624 | s = S2W(enr); | |
1625 | e = min((size_t)S2W(enr+1), b->bm_words); | |
1626 | count = 0; | |
1627 | if (s < b->bm_words) { | |
1628 | int n = e-s; | |
19f843aa | 1629 | p_addr = bm_map_pidx(b, bm_word_to_page_idx(b, s)); |
b411b363 PR |
1630 | bm = p_addr + MLPP(s); |
1631 | while (n--) | |
1632 | count += hweight_long(*bm++); | |
1633 | bm_unmap(p_addr); | |
1634 | } else { | |
d0180171 | 1635 | drbd_err(device, "start offset (%d) too large in drbd_bm_e_weight\n", s); |
b411b363 PR |
1636 | } |
1637 | spin_unlock_irqrestore(&b->bm_lock, flags); | |
1638 | return count; | |
1639 | } |