Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 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> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 29 | #include <linux/slab.h> |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 30 | #include <asm/kmap_types.h> |
Stephen Rothwell | f0ff135 | 2011-03-17 15:02:51 +0100 | [diff] [blame] | 31 | |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 32 | #include "drbd_int.h" |
| 33 | |
Lars Ellenberg | 95a0f10 | 2010-12-15 08:59:09 +0100 | [diff] [blame] | 34 | |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 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". |
Lars Ellenberg | 4b0715f | 2010-12-14 15:13:04 +0100 | [diff] [blame] | 41 | */ |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 42 | |
Lars Ellenberg | 4b0715f | 2010-12-14 15:13:04 +0100 | [diff] [blame] | 43 | |
| 44 | /* |
| 45 | * LIMITATIONS: |
| 46 | * We want to support >= peta byte of backend storage, while for now still using |
| 47 | * a granularity of one bit per 4KiB of storage. |
| 48 | * 1 << 50 bytes backend storage (1 PiB) |
| 49 | * 1 << (50 - 12) bits needed |
| 50 | * 38 --> we need u64 to index and count bits |
| 51 | * 1 << (38 - 3) bitmap bytes needed |
| 52 | * 35 --> we still need u64 to index and count bytes |
| 53 | * (that's 32 GiB of bitmap for 1 PiB storage) |
| 54 | * 1 << (35 - 2) 32bit longs needed |
| 55 | * 33 --> we'd even need u64 to index and count 32bit long words. |
| 56 | * 1 << (35 - 3) 64bit longs needed |
| 57 | * 32 --> we could get away with a 32bit unsigned int to index and count |
| 58 | * 64bit long words, but I rather stay with unsigned long for now. |
| 59 | * We probably should neither count nor point to bytes or long words |
| 60 | * directly, but either by bitnumber, or by page index and offset. |
| 61 | * 1 << (35 - 12) |
| 62 | * 22 --> we need that much 4KiB pages of bitmap. |
| 63 | * 1 << (22 + 3) --> on a 64bit arch, |
| 64 | * we need 32 MiB to store the array of page pointers. |
| 65 | * |
| 66 | * Because I'm lazy, and because the resulting patch was too large, too ugly |
| 67 | * and still incomplete, on 32bit we still "only" support 16 TiB (minus some), |
| 68 | * (1 << 32) bits * 4k storage. |
| 69 | * |
| 70 | |
| 71 | * bitmap storage and IO: |
| 72 | * Bitmap is stored little endian on disk, and is kept little endian in |
| 73 | * core memory. Currently we still hold the full bitmap in core as long |
| 74 | * as we are "attached" to a local disk, which at 32 GiB for 1PiB storage |
| 75 | * seems excessive. |
| 76 | * |
Bart Van Assche | 24c4830 | 2011-05-21 18:32:29 +0200 | [diff] [blame] | 77 | * We plan to reduce the amount of in-core bitmap pages by paging them in |
Lars Ellenberg | 4b0715f | 2010-12-14 15:13:04 +0100 | [diff] [blame] | 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. |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 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; |
Lars Ellenberg | 4b0715f | 2010-12-14 15:13:04 +0100 | [diff] [blame] | 96 | |
| 97 | /* see LIMITATIONS: above */ |
| 98 | |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 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; |
Thomas Gleixner | 8a03ae2 | 2010-01-29 20:39:07 +0000 | [diff] [blame] | 104 | struct mutex bm_change; /* serializes resize operations */ |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 105 | |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 106 | wait_queue_head_t bm_io_wait; /* used to serialize IO of single pages */ |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 107 | |
Lars Ellenberg | 20ceb2b | 2011-01-21 10:56:44 +0100 | [diff] [blame] | 108 | enum bm_flag bm_flags; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 109 | |
| 110 | /* debugging aid, in case we are still racy somewhere */ |
| 111 | char *bm_why; |
| 112 | struct task_struct *bm_task; |
| 113 | }; |
| 114 | |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 115 | #define bm_print_lock_info(m) __bm_print_lock_info(m, __func__) |
| 116 | static void __bm_print_lock_info(struct drbd_conf *mdev, const char *func) |
| 117 | { |
| 118 | struct drbd_bitmap *b = mdev->bitmap; |
| 119 | if (!__ratelimit(&drbd_ratelimit_state)) |
| 120 | return; |
| 121 | dev_err(DEV, "FIXME %s in %s, bitmap locked for '%s' by %s\n", |
Philipp Reisner | 392c880 | 2011-02-09 10:33:31 +0100 | [diff] [blame] | 122 | drbd_task_to_thread_name(mdev->tconn, current), |
Philipp Reisner | bed879a | 2011-02-04 14:00:37 +0100 | [diff] [blame] | 123 | func, b->bm_why ?: "?", |
Philipp Reisner | 392c880 | 2011-02-09 10:33:31 +0100 | [diff] [blame] | 124 | drbd_task_to_thread_name(mdev->tconn, b->bm_task)); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 125 | } |
| 126 | |
Lars Ellenberg | 20ceb2b | 2011-01-21 10:56:44 +0100 | [diff] [blame] | 127 | void drbd_bm_lock(struct drbd_conf *mdev, char *why, enum bm_flag flags) |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 128 | { |
| 129 | struct drbd_bitmap *b = mdev->bitmap; |
| 130 | int trylock_failed; |
| 131 | |
| 132 | if (!b) { |
| 133 | dev_err(DEV, "FIXME no bitmap in drbd_bm_lock!?\n"); |
| 134 | return; |
| 135 | } |
| 136 | |
Thomas Gleixner | 8a03ae2 | 2010-01-29 20:39:07 +0000 | [diff] [blame] | 137 | trylock_failed = !mutex_trylock(&b->bm_change); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 138 | |
| 139 | if (trylock_failed) { |
| 140 | dev_warn(DEV, "%s going to '%s' but bitmap already locked for '%s' by %s\n", |
Philipp Reisner | 392c880 | 2011-02-09 10:33:31 +0100 | [diff] [blame] | 141 | drbd_task_to_thread_name(mdev->tconn, current), |
Philipp Reisner | bed879a | 2011-02-04 14:00:37 +0100 | [diff] [blame] | 142 | why, b->bm_why ?: "?", |
Philipp Reisner | 392c880 | 2011-02-09 10:33:31 +0100 | [diff] [blame] | 143 | drbd_task_to_thread_name(mdev->tconn, b->bm_task)); |
Thomas Gleixner | 8a03ae2 | 2010-01-29 20:39:07 +0000 | [diff] [blame] | 144 | mutex_lock(&b->bm_change); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 145 | } |
Lars Ellenberg | 20ceb2b | 2011-01-21 10:56:44 +0100 | [diff] [blame] | 146 | if (BM_LOCKED_MASK & b->bm_flags) |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 147 | dev_err(DEV, "FIXME bitmap already locked in bm_lock\n"); |
Lars Ellenberg | 20ceb2b | 2011-01-21 10:56:44 +0100 | [diff] [blame] | 148 | b->bm_flags |= flags & BM_LOCKED_MASK; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 149 | |
| 150 | b->bm_why = why; |
| 151 | b->bm_task = current; |
| 152 | } |
| 153 | |
| 154 | void drbd_bm_unlock(struct drbd_conf *mdev) |
| 155 | { |
| 156 | struct drbd_bitmap *b = mdev->bitmap; |
| 157 | if (!b) { |
| 158 | dev_err(DEV, "FIXME no bitmap in drbd_bm_unlock!?\n"); |
| 159 | return; |
| 160 | } |
| 161 | |
Lars Ellenberg | 20ceb2b | 2011-01-21 10:56:44 +0100 | [diff] [blame] | 162 | if (!(BM_LOCKED_MASK & mdev->bitmap->bm_flags)) |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 163 | dev_err(DEV, "FIXME bitmap not locked in bm_unlock\n"); |
| 164 | |
Lars Ellenberg | 20ceb2b | 2011-01-21 10:56:44 +0100 | [diff] [blame] | 165 | b->bm_flags &= ~BM_LOCKED_MASK; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 166 | b->bm_why = NULL; |
| 167 | b->bm_task = NULL; |
Thomas Gleixner | 8a03ae2 | 2010-01-29 20:39:07 +0000 | [diff] [blame] | 168 | mutex_unlock(&b->bm_change); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 169 | } |
| 170 | |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 171 | /* we store some "meta" info about our pages in page->private */ |
| 172 | /* at a granularity of 4k storage per bitmap bit: |
| 173 | * one peta byte storage: 1<<50 byte, 1<<38 * 4k storage blocks |
| 174 | * 1<<38 bits, |
| 175 | * 1<<23 4k bitmap pages. |
| 176 | * Use 24 bits as page index, covers 2 peta byte storage |
| 177 | * at a granularity of 4k per bit. |
| 178 | * Used to report the failed page idx on io error from the endio handlers. |
| 179 | */ |
| 180 | #define BM_PAGE_IDX_MASK ((1UL<<24)-1) |
| 181 | /* this page is currently read in, or written back */ |
| 182 | #define BM_PAGE_IO_LOCK 31 |
| 183 | /* if there has been an IO error for this page */ |
| 184 | #define BM_PAGE_IO_ERROR 30 |
| 185 | /* this is to be able to intelligently skip disk IO, |
| 186 | * set if bits have been set since last IO. */ |
| 187 | #define BM_PAGE_NEED_WRITEOUT 29 |
| 188 | /* to mark for lazy writeout once syncer cleared all clearable bits, |
| 189 | * we if bits have been cleared since last IO. */ |
| 190 | #define BM_PAGE_LAZY_WRITEOUT 28 |
Lars Ellenberg | 45dfffe | 2011-02-21 13:21:00 +0100 | [diff] [blame] | 191 | /* pages marked with this "HINT" will be considered for writeout |
| 192 | * on activity log transactions */ |
| 193 | #define BM_PAGE_HINT_WRITEOUT 27 |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 194 | |
Bart Van Assche | 24c4830 | 2011-05-21 18:32:29 +0200 | [diff] [blame] | 195 | /* store_page_idx uses non-atomic assignment. It is only used directly after |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 196 | * allocating the page. All other bm_set_page_* and bm_clear_page_* need to |
| 197 | * use atomic bit manipulation, as set_out_of_sync (and therefore bitmap |
| 198 | * changes) may happen from various contexts, and wait_on_bit/wake_up_bit |
| 199 | * requires it all to be atomic as well. */ |
| 200 | static void bm_store_page_idx(struct page *page, unsigned long idx) |
Lars Ellenberg | 95a0f10 | 2010-12-15 08:59:09 +0100 | [diff] [blame] | 201 | { |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 202 | BUG_ON(0 != (idx & ~BM_PAGE_IDX_MASK)); |
Lars Ellenberg | f66ee69 | 2012-05-07 13:04:03 +0200 | [diff] [blame] | 203 | set_page_private(page, idx); |
Lars Ellenberg | 95a0f10 | 2010-12-15 08:59:09 +0100 | [diff] [blame] | 204 | } |
| 205 | |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 206 | static unsigned long bm_page_to_idx(struct page *page) |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 207 | { |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 208 | return page_private(page) & BM_PAGE_IDX_MASK; |
| 209 | } |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 210 | |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 211 | /* As is very unlikely that the same page is under IO from more than one |
| 212 | * context, we can get away with a bit per page and one wait queue per bitmap. |
| 213 | */ |
| 214 | static void bm_page_lock_io(struct drbd_conf *mdev, int page_nr) |
| 215 | { |
| 216 | struct drbd_bitmap *b = mdev->bitmap; |
| 217 | void *addr = &page_private(b->bm_pages[page_nr]); |
| 218 | wait_event(b->bm_io_wait, !test_and_set_bit(BM_PAGE_IO_LOCK, addr)); |
| 219 | } |
| 220 | |
| 221 | static void bm_page_unlock_io(struct drbd_conf *mdev, int page_nr) |
| 222 | { |
| 223 | struct drbd_bitmap *b = mdev->bitmap; |
| 224 | void *addr = &page_private(b->bm_pages[page_nr]); |
Lars Ellenberg | 4738fa1 | 2011-02-21 13:20:55 +0100 | [diff] [blame] | 225 | clear_bit_unlock(BM_PAGE_IO_LOCK, addr); |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 226 | wake_up(&mdev->bitmap->bm_io_wait); |
| 227 | } |
| 228 | |
| 229 | /* set _before_ submit_io, so it may be reset due to being changed |
| 230 | * while this page is in flight... will get submitted later again */ |
| 231 | static void bm_set_page_unchanged(struct page *page) |
| 232 | { |
| 233 | /* use cmpxchg? */ |
| 234 | clear_bit(BM_PAGE_NEED_WRITEOUT, &page_private(page)); |
| 235 | clear_bit(BM_PAGE_LAZY_WRITEOUT, &page_private(page)); |
| 236 | } |
| 237 | |
| 238 | static void bm_set_page_need_writeout(struct page *page) |
| 239 | { |
| 240 | set_bit(BM_PAGE_NEED_WRITEOUT, &page_private(page)); |
| 241 | } |
| 242 | |
Lars Ellenberg | 45dfffe | 2011-02-21 13:21:00 +0100 | [diff] [blame] | 243 | /** |
| 244 | * drbd_bm_mark_for_writeout() - mark a page with a "hint" to be considered for writeout |
| 245 | * @mdev: DRBD device. |
| 246 | * @page_nr: the bitmap page to mark with the "hint" flag |
| 247 | * |
| 248 | * From within an activity log transaction, we mark a few pages with these |
| 249 | * hints, then call drbd_bm_write_hinted(), which will only write out changed |
| 250 | * pages which are flagged with this mark. |
| 251 | */ |
| 252 | void drbd_bm_mark_for_writeout(struct drbd_conf *mdev, int page_nr) |
| 253 | { |
| 254 | struct page *page; |
| 255 | if (page_nr >= mdev->bitmap->bm_number_of_pages) { |
| 256 | dev_warn(DEV, "BAD: page_nr: %u, number_of_pages: %u\n", |
| 257 | page_nr, (int)mdev->bitmap->bm_number_of_pages); |
| 258 | return; |
| 259 | } |
| 260 | page = mdev->bitmap->bm_pages[page_nr]; |
| 261 | set_bit(BM_PAGE_HINT_WRITEOUT, &page_private(page)); |
| 262 | } |
| 263 | |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 264 | static int bm_test_page_unchanged(struct page *page) |
| 265 | { |
| 266 | volatile const unsigned long *addr = &page_private(page); |
| 267 | return (*addr & ((1UL<<BM_PAGE_NEED_WRITEOUT)|(1UL<<BM_PAGE_LAZY_WRITEOUT))) == 0; |
| 268 | } |
| 269 | |
| 270 | static void bm_set_page_io_err(struct page *page) |
| 271 | { |
| 272 | set_bit(BM_PAGE_IO_ERROR, &page_private(page)); |
| 273 | } |
| 274 | |
| 275 | static void bm_clear_page_io_err(struct page *page) |
| 276 | { |
| 277 | clear_bit(BM_PAGE_IO_ERROR, &page_private(page)); |
| 278 | } |
| 279 | |
| 280 | static void bm_set_page_lazy_writeout(struct page *page) |
| 281 | { |
| 282 | set_bit(BM_PAGE_LAZY_WRITEOUT, &page_private(page)); |
| 283 | } |
| 284 | |
| 285 | static int bm_test_page_lazy_writeout(struct page *page) |
| 286 | { |
| 287 | return test_bit(BM_PAGE_LAZY_WRITEOUT, &page_private(page)); |
| 288 | } |
| 289 | |
| 290 | /* on a 32bit box, this would allow for exactly (2<<38) bits. */ |
| 291 | static unsigned int bm_word_to_page_idx(struct drbd_bitmap *b, unsigned long long_nr) |
| 292 | { |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 293 | /* page_nr = (word*sizeof(long)) >> PAGE_SHIFT; */ |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 294 | unsigned int page_nr = long_nr >> (PAGE_SHIFT - LN2_BPL + 3); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 295 | BUG_ON(page_nr >= b->bm_number_of_pages); |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 296 | return page_nr; |
| 297 | } |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 298 | |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 299 | static unsigned int bm_bit_to_page_idx(struct drbd_bitmap *b, u64 bitnr) |
| 300 | { |
| 301 | /* page_nr = (bitnr/8) >> PAGE_SHIFT; */ |
| 302 | unsigned int page_nr = bitnr >> (PAGE_SHIFT + 3); |
| 303 | BUG_ON(page_nr >= b->bm_number_of_pages); |
| 304 | return page_nr; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 305 | } |
| 306 | |
Lars Ellenberg | 95a0f10 | 2010-12-15 08:59:09 +0100 | [diff] [blame] | 307 | static unsigned long *__bm_map_pidx(struct drbd_bitmap *b, unsigned int idx, const enum km_type km) |
| 308 | { |
| 309 | struct page *page = b->bm_pages[idx]; |
| 310 | return (unsigned long *) kmap_atomic(page, km); |
| 311 | } |
| 312 | |
| 313 | static unsigned long *bm_map_pidx(struct drbd_bitmap *b, unsigned int idx) |
| 314 | { |
| 315 | return __bm_map_pidx(b, idx, KM_IRQ1); |
| 316 | } |
| 317 | |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 318 | static void __bm_unmap(unsigned long *p_addr, const enum km_type km) |
| 319 | { |
| 320 | kunmap_atomic(p_addr, km); |
| 321 | }; |
| 322 | |
| 323 | static void bm_unmap(unsigned long *p_addr) |
| 324 | { |
| 325 | return __bm_unmap(p_addr, KM_IRQ1); |
| 326 | } |
| 327 | |
| 328 | /* long word offset of _bitmap_ sector */ |
| 329 | #define S2W(s) ((s)<<(BM_EXT_SHIFT-BM_BLOCK_SHIFT-LN2_BPL)) |
| 330 | /* word offset from start of bitmap to word number _in_page_ |
| 331 | * modulo longs per page |
| 332 | #define MLPP(X) ((X) % (PAGE_SIZE/sizeof(long)) |
Bart Van Assche | 24c4830 | 2011-05-21 18:32:29 +0200 | [diff] [blame] | 333 | hm, well, Philipp thinks gcc might not optimize the % into & (... - 1) |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 334 | so do it explicitly: |
| 335 | */ |
| 336 | #define MLPP(X) ((X) & ((PAGE_SIZE/sizeof(long))-1)) |
| 337 | |
| 338 | /* Long words per page */ |
| 339 | #define LWPP (PAGE_SIZE/sizeof(long)) |
| 340 | |
| 341 | /* |
| 342 | * actually most functions herein should take a struct drbd_bitmap*, not a |
| 343 | * struct drbd_conf*, but for the debug macros I like to have the mdev around |
| 344 | * to be able to report device specific. |
| 345 | */ |
| 346 | |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 347 | |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 348 | static void bm_free_pages(struct page **pages, unsigned long number) |
| 349 | { |
| 350 | unsigned long i; |
| 351 | if (!pages) |
| 352 | return; |
| 353 | |
| 354 | for (i = 0; i < number; i++) { |
| 355 | if (!pages[i]) { |
| 356 | printk(KERN_ALERT "drbd: bm_free_pages tried to free " |
| 357 | "a NULL pointer; i=%lu n=%lu\n", |
| 358 | i, number); |
| 359 | continue; |
| 360 | } |
| 361 | __free_page(pages[i]); |
| 362 | pages[i] = NULL; |
| 363 | } |
| 364 | } |
| 365 | |
| 366 | static void bm_vk_free(void *ptr, int v) |
| 367 | { |
| 368 | if (v) |
| 369 | vfree(ptr); |
| 370 | else |
| 371 | kfree(ptr); |
| 372 | } |
| 373 | |
| 374 | /* |
| 375 | * "have" and "want" are NUMBER OF PAGES. |
| 376 | */ |
| 377 | static struct page **bm_realloc_pages(struct drbd_bitmap *b, unsigned long want) |
| 378 | { |
| 379 | struct page **old_pages = b->bm_pages; |
| 380 | struct page **new_pages, *page; |
| 381 | unsigned int i, bytes, vmalloced = 0; |
| 382 | unsigned long have = b->bm_number_of_pages; |
| 383 | |
| 384 | BUG_ON(have == 0 && old_pages != NULL); |
| 385 | BUG_ON(have != 0 && old_pages == NULL); |
| 386 | |
| 387 | if (have == want) |
| 388 | return old_pages; |
| 389 | |
| 390 | /* Trying kmalloc first, falling back to vmalloc. |
Lars Ellenberg | bc891c9 | 2012-09-26 14:18:51 +0200 | [diff] [blame^] | 391 | * GFP_NOIO, as this is called while drbd IO is "suspended", |
| 392 | * and during resize or attach on diskless Primary, |
| 393 | * we must not block on IO to ourselves. |
| 394 | * Context is receiver thread or dmsetup. */ |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 395 | bytes = sizeof(struct page *)*want; |
Lars Ellenberg | bc891c9 | 2012-09-26 14:18:51 +0200 | [diff] [blame^] | 396 | new_pages = kmalloc(bytes, GFP_NOIO); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 397 | if (!new_pages) { |
Lars Ellenberg | bc891c9 | 2012-09-26 14:18:51 +0200 | [diff] [blame^] | 398 | new_pages = __vmalloc(bytes, |
| 399 | GFP_NOIO | __GFP_HIGHMEM, |
| 400 | PAGE_KERNEL); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 401 | if (!new_pages) |
| 402 | return NULL; |
| 403 | vmalloced = 1; |
| 404 | } |
| 405 | |
| 406 | memset(new_pages, 0, bytes); |
| 407 | if (want >= have) { |
| 408 | for (i = 0; i < have; i++) |
| 409 | new_pages[i] = old_pages[i]; |
| 410 | for (; i < want; i++) { |
Lars Ellenberg | bc891c9 | 2012-09-26 14:18:51 +0200 | [diff] [blame^] | 411 | page = alloc_page(GFP_NOIO | __GFP_HIGHMEM); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 412 | if (!page) { |
| 413 | bm_free_pages(new_pages + have, i - have); |
| 414 | bm_vk_free(new_pages, vmalloced); |
| 415 | return NULL; |
| 416 | } |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 417 | /* we want to know which page it is |
| 418 | * from the endio handlers */ |
| 419 | bm_store_page_idx(page, i); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 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) |
Lars Ellenberg | 20ceb2b | 2011-01-21 10:56:44 +0100 | [diff] [blame] | 431 | b->bm_flags |= BM_P_VMALLOCED; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 432 | else |
Lars Ellenberg | 20ceb2b | 2011-01-21 10:56:44 +0100 | [diff] [blame] | 433 | b->bm_flags &= ~BM_P_VMALLOCED; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 434 | |
| 435 | return new_pages; |
| 436 | } |
| 437 | |
| 438 | /* |
| 439 | * called on driver init only. TODO call when a device is created. |
| 440 | * allocates the drbd_bitmap, and stores it in mdev->bitmap. |
| 441 | */ |
| 442 | int drbd_bm_init(struct drbd_conf *mdev) |
| 443 | { |
| 444 | struct drbd_bitmap *b = mdev->bitmap; |
| 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); |
Thomas Gleixner | 8a03ae2 | 2010-01-29 20:39:07 +0000 | [diff] [blame] | 450 | mutex_init(&b->bm_change); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 451 | init_waitqueue_head(&b->bm_io_wait); |
| 452 | |
| 453 | mdev->bitmap = b; |
| 454 | |
| 455 | return 0; |
| 456 | } |
| 457 | |
| 458 | sector_t drbd_bm_capacity(struct drbd_conf *mdev) |
| 459 | { |
Andreas Gruenbacher | 841ce24 | 2010-12-15 19:31:20 +0100 | [diff] [blame] | 460 | if (!expect(mdev->bitmap)) |
| 461 | return 0; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 462 | return mdev->bitmap->bm_dev_capacity; |
| 463 | } |
| 464 | |
| 465 | /* called on driver unload. TODO: call when a device is destroyed. |
| 466 | */ |
| 467 | void drbd_bm_cleanup(struct drbd_conf *mdev) |
| 468 | { |
Andreas Gruenbacher | 841ce24 | 2010-12-15 19:31:20 +0100 | [diff] [blame] | 469 | if (!expect(mdev->bitmap)) |
| 470 | return; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 471 | bm_free_pages(mdev->bitmap->bm_pages, mdev->bitmap->bm_number_of_pages); |
Lars Ellenberg | 20ceb2b | 2011-01-21 10:56:44 +0100 | [diff] [blame] | 472 | bm_vk_free(mdev->bitmap->bm_pages, (BM_P_VMALLOCED & mdev->bitmap->bm_flags)); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 473 | kfree(mdev->bitmap); |
| 474 | mdev->bitmap = NULL; |
| 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 | */ |
Lars Ellenberg | 95a0f10 | 2010-12-15 08:59:09 +0100 | [diff] [blame] | 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) |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 485 | static int bm_clear_surplus(struct drbd_bitmap *b) |
| 486 | { |
Lars Ellenberg | 95a0f10 | 2010-12-15 08:59:09 +0100 | [diff] [blame] | 487 | unsigned long mask; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 488 | unsigned long *p_addr, *bm; |
Lars Ellenberg | 95a0f10 | 2010-12-15 08:59:09 +0100 | [diff] [blame] | 489 | int tmp; |
| 490 | int cleared = 0; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 491 | |
Lars Ellenberg | 95a0f10 | 2010-12-15 08:59:09 +0100 | [diff] [blame] | 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 | |
Lars Ellenberg | 6850c44 | 2010-12-16 00:32:38 +0100 | [diff] [blame] | 500 | p_addr = bm_map_pidx(b, b->bm_number_of_pages - 1); |
Lars Ellenberg | 95a0f10 | 2010-12-15 08:59:09 +0100 | [diff] [blame] | 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. */ |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 507 | cleared = hweight_long(*bm & ~mask); |
| 508 | *bm &= mask; |
Lars Ellenberg | 95a0f10 | 2010-12-15 08:59:09 +0100 | [diff] [blame] | 509 | bm++; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 510 | } |
| 511 | |
Lars Ellenberg | 95a0f10 | 2010-12-15 08:59:09 +0100 | [diff] [blame] | 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 */ |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 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 | { |
Lars Ellenberg | 95a0f10 | 2010-12-15 08:59:09 +0100 | [diff] [blame] | 524 | unsigned long mask; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 525 | unsigned long *p_addr, *bm; |
Lars Ellenberg | 95a0f10 | 2010-12-15 08:59:09 +0100 | [diff] [blame] | 526 | int tmp; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 527 | |
Lars Ellenberg | 95a0f10 | 2010-12-15 08:59:09 +0100 | [diff] [blame] | 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 | |
Lars Ellenberg | 6850c44 | 2010-12-16 00:32:38 +0100 | [diff] [blame] | 536 | p_addr = bm_map_pidx(b, b->bm_number_of_pages - 1); |
Lars Ellenberg | 95a0f10 | 2010-12-15 08:59:09 +0100 | [diff] [blame] | 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. */ |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 543 | *bm |= ~mask; |
Lars Ellenberg | 95a0f10 | 2010-12-15 08:59:09 +0100 | [diff] [blame] | 544 | bm++; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 545 | } |
| 546 | |
Lars Ellenberg | 95a0f10 | 2010-12-15 08:59:09 +0100 | [diff] [blame] | 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; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 551 | } |
| 552 | bm_unmap(p_addr); |
| 553 | } |
| 554 | |
Lars Ellenberg | 4b0715f | 2010-12-14 15:13:04 +0100 | [diff] [blame] | 555 | /* you better not modify the bitmap while this is running, |
| 556 | * or its results will be stale */ |
Lars Ellenberg | 95a0f10 | 2010-12-15 08:59:09 +0100 | [diff] [blame] | 557 | static unsigned long bm_count_bits(struct drbd_bitmap *b) |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 558 | { |
Lars Ellenberg | 4b0715f | 2010-12-14 15:13:04 +0100 | [diff] [blame] | 559 | unsigned long *p_addr; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 560 | unsigned long bits = 0; |
Lars Ellenberg | 4b0715f | 2010-12-14 15:13:04 +0100 | [diff] [blame] | 561 | unsigned long mask = (1UL << (b->bm_bits & BITS_PER_LONG_MASK)) -1; |
Lars Ellenberg | 6850c44 | 2010-12-16 00:32:38 +0100 | [diff] [blame] | 562 | int idx, i, last_word; |
Lars Ellenberg | 7777a8b | 2010-12-15 23:21:39 +0100 | [diff] [blame] | 563 | |
Lars Ellenberg | 4b0715f | 2010-12-14 15:13:04 +0100 | [diff] [blame] | 564 | /* all but last page */ |
Lars Ellenberg | 6850c44 | 2010-12-16 00:32:38 +0100 | [diff] [blame] | 565 | for (idx = 0; idx < b->bm_number_of_pages - 1; idx++) { |
Lars Ellenberg | 4b0715f | 2010-12-14 15:13:04 +0100 | [diff] [blame] | 566 | p_addr = __bm_map_pidx(b, idx, KM_USER0); |
| 567 | for (i = 0; i < LWPP; i++) |
| 568 | bits += hweight_long(p_addr[i]); |
Lars Ellenberg | 7777a8b | 2010-12-15 23:21:39 +0100 | [diff] [blame] | 569 | __bm_unmap(p_addr, KM_USER0); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 570 | cond_resched(); |
| 571 | } |
Lars Ellenberg | 4b0715f | 2010-12-14 15:13:04 +0100 | [diff] [blame] | 572 | /* last (or only) page */ |
| 573 | last_word = ((b->bm_bits - 1) & BITS_PER_PAGE_MASK) >> LN2_BPL; |
| 574 | p_addr = __bm_map_pidx(b, idx, KM_USER0); |
| 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; |
| 582 | __bm_unmap(p_addr, KM_USER0); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 583 | return bits; |
| 584 | } |
| 585 | |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 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; |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 590 | unsigned int idx; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 591 | size_t do_now, end; |
| 592 | |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 593 | end = offset + len; |
| 594 | |
| 595 | if (end > b->bm_words) { |
| 596 | printk(KERN_ALERT "drbd: bm_memset end > bm_words\n"); |
| 597 | return; |
| 598 | } |
| 599 | |
| 600 | while (offset < end) { |
| 601 | do_now = min_t(size_t, ALIGN(offset + 1, LWPP), end) - offset; |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 602 | idx = bm_word_to_page_idx(b, offset); |
| 603 | p_addr = bm_map_pidx(b, idx); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 604 | bm = p_addr + MLPP(offset); |
| 605 | if (bm+do_now > p_addr + LWPP) { |
| 606 | printk(KERN_ALERT "drbd: BUG BUG BUG! p_addr:%p bm:%p do_now:%d\n", |
| 607 | p_addr, bm, (int)do_now); |
Lars Ellenberg | 84e7c0f | 2010-12-16 00:37:57 +0100 | [diff] [blame] | 608 | } else |
| 609 | memset(bm, c, do_now * sizeof(long)); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 610 | bm_unmap(p_addr); |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 611 | bm_set_page_need_writeout(b->bm_pages[idx]); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 612 | offset += do_now; |
| 613 | } |
| 614 | } |
| 615 | |
| 616 | /* |
| 617 | * make sure the bitmap has enough room for the attached storage, |
| 618 | * if necessary, resize. |
| 619 | * called whenever we may have changed the device size. |
| 620 | * returns -ENOMEM if we could not allocate enough memory, 0 on success. |
| 621 | * In case this is actually a resize, we copy the old bitmap into the new one. |
| 622 | * Otherwise, the bitmap is initialized to all bits set. |
| 623 | */ |
Philipp Reisner | 02d9a94 | 2010-03-24 16:23:03 +0100 | [diff] [blame] | 624 | int drbd_bm_resize(struct drbd_conf *mdev, sector_t capacity, int set_new_bits) |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 625 | { |
| 626 | struct drbd_bitmap *b = mdev->bitmap; |
Lars Ellenberg | 6850c44 | 2010-12-16 00:32:38 +0100 | [diff] [blame] | 627 | unsigned long bits, words, owords, obits; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 628 | unsigned long want, have, onpages; /* number of pages */ |
| 629 | struct page **npages, **opages = NULL; |
| 630 | int err = 0, growing; |
| 631 | int opages_vmalloced; |
| 632 | |
Andreas Gruenbacher | 841ce24 | 2010-12-15 19:31:20 +0100 | [diff] [blame] | 633 | if (!expect(b)) |
| 634 | return -ENOMEM; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 635 | |
Lars Ellenberg | 20ceb2b | 2011-01-21 10:56:44 +0100 | [diff] [blame] | 636 | drbd_bm_lock(mdev, "resize", BM_LOCKED_MASK); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 637 | |
| 638 | dev_info(DEV, "drbd_bm_resize called with capacity == %llu\n", |
| 639 | (unsigned long long)capacity); |
| 640 | |
| 641 | if (capacity == b->bm_dev_capacity) |
| 642 | goto out; |
| 643 | |
Lars Ellenberg | 20ceb2b | 2011-01-21 10:56:44 +0100 | [diff] [blame] | 644 | opages_vmalloced = (BM_P_VMALLOCED & b->bm_flags); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 645 | |
| 646 | if (capacity == 0) { |
| 647 | spin_lock_irq(&b->bm_lock); |
| 648 | opages = b->bm_pages; |
| 649 | onpages = b->bm_number_of_pages; |
| 650 | owords = b->bm_words; |
| 651 | b->bm_pages = NULL; |
| 652 | b->bm_number_of_pages = |
| 653 | b->bm_set = |
| 654 | b->bm_bits = |
| 655 | b->bm_words = |
| 656 | b->bm_dev_capacity = 0; |
| 657 | spin_unlock_irq(&b->bm_lock); |
| 658 | bm_free_pages(opages, onpages); |
| 659 | bm_vk_free(opages, opages_vmalloced); |
| 660 | goto out; |
| 661 | } |
| 662 | bits = BM_SECT_TO_BIT(ALIGN(capacity, BM_SECT_PER_BIT)); |
| 663 | |
| 664 | /* if we would use |
| 665 | words = ALIGN(bits,BITS_PER_LONG) >> LN2_BPL; |
| 666 | a 32bit host could present the wrong number of words |
| 667 | to a 64bit host. |
| 668 | */ |
| 669 | words = ALIGN(bits, 64) >> LN2_BPL; |
| 670 | |
| 671 | if (get_ldev(mdev)) { |
Lars Ellenberg | 4b0715f | 2010-12-14 15:13:04 +0100 | [diff] [blame] | 672 | u64 bits_on_disk = ((u64)mdev->ldev->md.md_size_sect-MD_BM_OFFSET) << 12; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 673 | put_ldev(mdev); |
Lars Ellenberg | 4b0715f | 2010-12-14 15:13:04 +0100 | [diff] [blame] | 674 | if (bits > bits_on_disk) { |
| 675 | dev_info(DEV, "bits = %lu\n", bits); |
| 676 | dev_info(DEV, "bits_on_disk = %llu\n", bits_on_disk); |
| 677 | err = -ENOSPC; |
| 678 | goto out; |
| 679 | } |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 680 | } |
| 681 | |
Lars Ellenberg | 6850c44 | 2010-12-16 00:32:38 +0100 | [diff] [blame] | 682 | want = ALIGN(words*sizeof(long), PAGE_SIZE) >> PAGE_SHIFT; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 683 | have = b->bm_number_of_pages; |
| 684 | if (want == have) { |
| 685 | D_ASSERT(b->bm_pages != NULL); |
| 686 | npages = b->bm_pages; |
| 687 | } else { |
Andreas Gruenbacher | 0cf9d27 | 2010-12-07 10:43:29 +0100 | [diff] [blame] | 688 | if (drbd_insert_fault(mdev, DRBD_FAULT_BM_ALLOC)) |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 689 | npages = NULL; |
| 690 | else |
| 691 | npages = bm_realloc_pages(b, want); |
| 692 | } |
| 693 | |
| 694 | if (!npages) { |
| 695 | err = -ENOMEM; |
| 696 | goto out; |
| 697 | } |
| 698 | |
| 699 | spin_lock_irq(&b->bm_lock); |
| 700 | opages = b->bm_pages; |
| 701 | owords = b->bm_words; |
| 702 | obits = b->bm_bits; |
| 703 | |
| 704 | growing = bits > obits; |
Philipp Reisner | 5223671 | 2010-04-28 14:46:57 +0200 | [diff] [blame] | 705 | if (opages && growing && set_new_bits) |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 706 | bm_set_surplus(b); |
| 707 | |
| 708 | b->bm_pages = npages; |
| 709 | b->bm_number_of_pages = want; |
| 710 | b->bm_bits = bits; |
| 711 | b->bm_words = words; |
| 712 | b->bm_dev_capacity = capacity; |
| 713 | |
| 714 | if (growing) { |
Philipp Reisner | 02d9a94 | 2010-03-24 16:23:03 +0100 | [diff] [blame] | 715 | if (set_new_bits) { |
| 716 | bm_memset(b, owords, 0xff, words-owords); |
| 717 | b->bm_set += bits - obits; |
| 718 | } else |
| 719 | bm_memset(b, owords, 0x00, words-owords); |
| 720 | |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 721 | } |
| 722 | |
| 723 | if (want < have) { |
| 724 | /* implicit: (opages != NULL) && (opages != npages) */ |
| 725 | bm_free_pages(opages + want, have - want); |
| 726 | } |
| 727 | |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 728 | (void)bm_clear_surplus(b); |
| 729 | |
| 730 | spin_unlock_irq(&b->bm_lock); |
| 731 | if (opages != npages) |
| 732 | bm_vk_free(opages, opages_vmalloced); |
| 733 | if (!growing) |
| 734 | b->bm_set = bm_count_bits(b); |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 735 | dev_info(DEV, "resync bitmap: bits=%lu words=%lu pages=%lu\n", bits, words, want); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 736 | |
| 737 | out: |
| 738 | drbd_bm_unlock(mdev); |
| 739 | return err; |
| 740 | } |
| 741 | |
| 742 | /* inherently racy: |
| 743 | * if not protected by other means, return value may be out of date when |
| 744 | * leaving this function... |
| 745 | * we still need to lock it, since it is important that this returns |
| 746 | * bm_set == 0 precisely. |
| 747 | * |
| 748 | * maybe bm_set should be atomic_t ? |
| 749 | */ |
Philipp Reisner | 0778286 | 2010-08-31 12:00:50 +0200 | [diff] [blame] | 750 | unsigned long _drbd_bm_total_weight(struct drbd_conf *mdev) |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 751 | { |
| 752 | struct drbd_bitmap *b = mdev->bitmap; |
| 753 | unsigned long s; |
| 754 | unsigned long flags; |
| 755 | |
Andreas Gruenbacher | 841ce24 | 2010-12-15 19:31:20 +0100 | [diff] [blame] | 756 | if (!expect(b)) |
| 757 | return 0; |
| 758 | if (!expect(b->bm_pages)) |
| 759 | return 0; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 760 | |
| 761 | spin_lock_irqsave(&b->bm_lock, flags); |
| 762 | s = b->bm_set; |
| 763 | spin_unlock_irqrestore(&b->bm_lock, flags); |
| 764 | |
| 765 | return s; |
| 766 | } |
| 767 | |
| 768 | unsigned long drbd_bm_total_weight(struct drbd_conf *mdev) |
| 769 | { |
| 770 | unsigned long s; |
| 771 | /* if I don't have a disk, I don't know about out-of-sync status */ |
| 772 | if (!get_ldev_if_state(mdev, D_NEGOTIATING)) |
| 773 | return 0; |
| 774 | s = _drbd_bm_total_weight(mdev); |
| 775 | put_ldev(mdev); |
| 776 | return s; |
| 777 | } |
| 778 | |
| 779 | size_t drbd_bm_words(struct drbd_conf *mdev) |
| 780 | { |
| 781 | struct drbd_bitmap *b = mdev->bitmap; |
Andreas Gruenbacher | 841ce24 | 2010-12-15 19:31:20 +0100 | [diff] [blame] | 782 | if (!expect(b)) |
| 783 | return 0; |
| 784 | if (!expect(b->bm_pages)) |
| 785 | return 0; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 786 | |
| 787 | return b->bm_words; |
| 788 | } |
| 789 | |
| 790 | unsigned long drbd_bm_bits(struct drbd_conf *mdev) |
| 791 | { |
| 792 | struct drbd_bitmap *b = mdev->bitmap; |
Andreas Gruenbacher | 841ce24 | 2010-12-15 19:31:20 +0100 | [diff] [blame] | 793 | if (!expect(b)) |
| 794 | return 0; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 795 | |
| 796 | return b->bm_bits; |
| 797 | } |
| 798 | |
| 799 | /* merge number words from buffer into the bitmap starting at offset. |
| 800 | * buffer[i] is expected to be little endian unsigned long. |
| 801 | * bitmap must be locked by drbd_bm_lock. |
| 802 | * currently only used from receive_bitmap. |
| 803 | */ |
| 804 | void drbd_bm_merge_lel(struct drbd_conf *mdev, size_t offset, size_t number, |
| 805 | unsigned long *buffer) |
| 806 | { |
| 807 | struct drbd_bitmap *b = mdev->bitmap; |
| 808 | unsigned long *p_addr, *bm; |
| 809 | unsigned long word, bits; |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 810 | unsigned int idx; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 811 | size_t end, do_now; |
| 812 | |
| 813 | end = offset + number; |
| 814 | |
Andreas Gruenbacher | 841ce24 | 2010-12-15 19:31:20 +0100 | [diff] [blame] | 815 | if (!expect(b)) |
| 816 | return; |
| 817 | if (!expect(b->bm_pages)) |
| 818 | return; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 819 | if (number == 0) |
| 820 | return; |
| 821 | WARN_ON(offset >= b->bm_words); |
| 822 | WARN_ON(end > b->bm_words); |
| 823 | |
| 824 | spin_lock_irq(&b->bm_lock); |
| 825 | while (offset < end) { |
| 826 | do_now = min_t(size_t, ALIGN(offset+1, LWPP), end) - offset; |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 827 | idx = bm_word_to_page_idx(b, offset); |
| 828 | p_addr = bm_map_pidx(b, idx); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 829 | bm = p_addr + MLPP(offset); |
| 830 | offset += do_now; |
| 831 | while (do_now--) { |
| 832 | bits = hweight_long(*bm); |
Lars Ellenberg | 95a0f10 | 2010-12-15 08:59:09 +0100 | [diff] [blame] | 833 | word = *bm | *buffer++; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 834 | *bm++ = word; |
| 835 | b->bm_set += hweight_long(word) - bits; |
| 836 | } |
| 837 | bm_unmap(p_addr); |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 838 | bm_set_page_need_writeout(b->bm_pages[idx]); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 839 | } |
| 840 | /* with 32bit <-> 64bit cross-platform connect |
| 841 | * this is only correct for current usage, |
| 842 | * where we _know_ that we are 64 bit aligned, |
| 843 | * and know that this function is used in this way, too... |
| 844 | */ |
| 845 | if (end == b->bm_words) |
| 846 | b->bm_set -= bm_clear_surplus(b); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 847 | spin_unlock_irq(&b->bm_lock); |
| 848 | } |
| 849 | |
| 850 | /* copy number words from the bitmap starting at offset into the buffer. |
| 851 | * buffer[i] will be little endian unsigned long. |
| 852 | */ |
| 853 | void drbd_bm_get_lel(struct drbd_conf *mdev, size_t offset, size_t number, |
| 854 | unsigned long *buffer) |
| 855 | { |
| 856 | struct drbd_bitmap *b = mdev->bitmap; |
| 857 | unsigned long *p_addr, *bm; |
| 858 | size_t end, do_now; |
| 859 | |
| 860 | end = offset + number; |
| 861 | |
Andreas Gruenbacher | 841ce24 | 2010-12-15 19:31:20 +0100 | [diff] [blame] | 862 | if (!expect(b)) |
| 863 | return; |
| 864 | if (!expect(b->bm_pages)) |
| 865 | return; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 866 | |
| 867 | spin_lock_irq(&b->bm_lock); |
| 868 | if ((offset >= b->bm_words) || |
| 869 | (end > b->bm_words) || |
| 870 | (number <= 0)) |
| 871 | dev_err(DEV, "offset=%lu number=%lu bm_words=%lu\n", |
| 872 | (unsigned long) offset, |
| 873 | (unsigned long) number, |
| 874 | (unsigned long) b->bm_words); |
| 875 | else { |
| 876 | while (offset < end) { |
| 877 | do_now = min_t(size_t, ALIGN(offset+1, LWPP), end) - offset; |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 878 | p_addr = bm_map_pidx(b, bm_word_to_page_idx(b, offset)); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 879 | bm = p_addr + MLPP(offset); |
| 880 | offset += do_now; |
| 881 | while (do_now--) |
Lars Ellenberg | 95a0f10 | 2010-12-15 08:59:09 +0100 | [diff] [blame] | 882 | *buffer++ = *bm++; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 883 | bm_unmap(p_addr); |
| 884 | } |
| 885 | } |
| 886 | spin_unlock_irq(&b->bm_lock); |
| 887 | } |
| 888 | |
| 889 | /* set all bits in the bitmap */ |
| 890 | void drbd_bm_set_all(struct drbd_conf *mdev) |
| 891 | { |
| 892 | struct drbd_bitmap *b = mdev->bitmap; |
Andreas Gruenbacher | 841ce24 | 2010-12-15 19:31:20 +0100 | [diff] [blame] | 893 | if (!expect(b)) |
| 894 | return; |
| 895 | if (!expect(b->bm_pages)) |
| 896 | return; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 897 | |
| 898 | spin_lock_irq(&b->bm_lock); |
| 899 | bm_memset(b, 0, 0xff, b->bm_words); |
| 900 | (void)bm_clear_surplus(b); |
| 901 | b->bm_set = b->bm_bits; |
| 902 | spin_unlock_irq(&b->bm_lock); |
| 903 | } |
| 904 | |
| 905 | /* clear all bits in the bitmap */ |
| 906 | void drbd_bm_clear_all(struct drbd_conf *mdev) |
| 907 | { |
| 908 | struct drbd_bitmap *b = mdev->bitmap; |
Andreas Gruenbacher | 841ce24 | 2010-12-15 19:31:20 +0100 | [diff] [blame] | 909 | if (!expect(b)) |
| 910 | return; |
| 911 | if (!expect(b->bm_pages)) |
| 912 | return; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 913 | |
| 914 | spin_lock_irq(&b->bm_lock); |
| 915 | bm_memset(b, 0, 0, b->bm_words); |
| 916 | b->bm_set = 0; |
| 917 | spin_unlock_irq(&b->bm_lock); |
| 918 | } |
| 919 | |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 920 | struct bm_aio_ctx { |
| 921 | struct drbd_conf *mdev; |
| 922 | atomic_t in_flight; |
Philipp Reisner | cdfda63 | 2011-07-05 15:38:59 +0200 | [diff] [blame] | 923 | unsigned int done; |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 924 | unsigned flags; |
| 925 | #define BM_AIO_COPY_PAGES 1 |
Lars Ellenberg | 45dfffe | 2011-02-21 13:21:00 +0100 | [diff] [blame] | 926 | #define BM_AIO_WRITE_HINTED 2 |
Philipp Reisner | fef45d2 | 2012-08-14 11:46:59 +0200 | [diff] [blame] | 927 | #define BM_WRITE_ALL_PAGES 4 |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 928 | int error; |
Philipp Reisner | cdfda63 | 2011-07-05 15:38:59 +0200 | [diff] [blame] | 929 | struct kref kref; |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 930 | }; |
| 931 | |
Philipp Reisner | cdfda63 | 2011-07-05 15:38:59 +0200 | [diff] [blame] | 932 | static void bm_aio_ctx_destroy(struct kref *kref) |
| 933 | { |
| 934 | struct bm_aio_ctx *ctx = container_of(kref, struct bm_aio_ctx, kref); |
| 935 | |
| 936 | put_ldev(ctx->mdev); |
| 937 | kfree(ctx); |
| 938 | } |
| 939 | |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 940 | /* bv_page may be a copy, or may be the original */ |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 941 | static void bm_async_io_complete(struct bio *bio, int error) |
| 942 | { |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 943 | struct bm_aio_ctx *ctx = bio->bi_private; |
| 944 | struct drbd_conf *mdev = ctx->mdev; |
| 945 | struct drbd_bitmap *b = mdev->bitmap; |
| 946 | unsigned int idx = bm_page_to_idx(bio->bi_io_vec[0].bv_page); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 947 | int uptodate = bio_flagged(bio, BIO_UPTODATE); |
| 948 | |
| 949 | |
| 950 | /* strange behavior of some lower level drivers... |
| 951 | * fail the request by clearing the uptodate flag, |
| 952 | * but do not return any error?! |
| 953 | * do we want to WARN() on this? */ |
| 954 | if (!error && !uptodate) |
| 955 | error = -EIO; |
| 956 | |
Lars Ellenberg | 7648cdf | 2010-12-17 23:58:41 +0100 | [diff] [blame] | 957 | if ((ctx->flags & BM_AIO_COPY_PAGES) == 0 && |
| 958 | !bm_test_page_unchanged(b->bm_pages[idx])) |
| 959 | dev_warn(DEV, "bitmap page idx %u changed during IO!\n", idx); |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 960 | |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 961 | if (error) { |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 962 | /* ctx error will hold the completed-last non-zero error code, |
| 963 | * in case error codes differ. */ |
| 964 | ctx->error = error; |
| 965 | bm_set_page_io_err(b->bm_pages[idx]); |
| 966 | /* Not identical to on disk version of it. |
| 967 | * Is BM_PAGE_IO_ERROR enough? */ |
| 968 | if (__ratelimit(&drbd_ratelimit_state)) |
| 969 | dev_err(DEV, "IO ERROR %d on bitmap page idx %u\n", |
| 970 | error, idx); |
| 971 | } else { |
| 972 | bm_clear_page_io_err(b->bm_pages[idx]); |
| 973 | dynamic_dev_dbg(DEV, "bitmap page idx %u completed\n", idx); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 974 | } |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 975 | |
| 976 | bm_page_unlock_io(mdev, idx); |
| 977 | |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 978 | if (ctx->flags & BM_AIO_COPY_PAGES) |
Lars Ellenberg | 9db4e77 | 2011-02-23 15:38:47 +0100 | [diff] [blame] | 979 | mempool_free(bio->bi_io_vec[0].bv_page, drbd_md_io_page_pool); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 980 | |
| 981 | bio_put(bio); |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 982 | |
Philipp Reisner | cdfda63 | 2011-07-05 15:38:59 +0200 | [diff] [blame] | 983 | if (atomic_dec_and_test(&ctx->in_flight)) { |
| 984 | ctx->done = 1; |
| 985 | wake_up(&mdev->misc_wait); |
| 986 | kref_put(&ctx->kref, &bm_aio_ctx_destroy); |
| 987 | } |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 988 | } |
| 989 | |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 990 | static void bm_page_io_async(struct bm_aio_ctx *ctx, int page_nr, int rw) __must_hold(local) |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 991 | { |
Philipp Reisner | cdfda63 | 2011-07-05 15:38:59 +0200 | [diff] [blame] | 992 | struct bio *bio = bio_alloc_drbd(GFP_NOIO); |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 993 | struct drbd_conf *mdev = ctx->mdev; |
| 994 | struct drbd_bitmap *b = mdev->bitmap; |
| 995 | struct page *page; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 996 | unsigned int len; |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 997 | |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 998 | sector_t on_disk_sector = |
| 999 | mdev->ldev->md.md_offset + mdev->ldev->md.bm_offset; |
| 1000 | on_disk_sector += ((sector_t)page_nr) << (PAGE_SHIFT-9); |
| 1001 | |
| 1002 | /* this might happen with very small |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1003 | * flexible external meta data device, |
| 1004 | * or with PAGE_SIZE > 4k */ |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1005 | len = min_t(unsigned int, PAGE_SIZE, |
| 1006 | (drbd_md_last_sector(mdev->ldev) - on_disk_sector + 1)<<9); |
| 1007 | |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1008 | /* serialize IO on this page */ |
| 1009 | bm_page_lock_io(mdev, page_nr); |
| 1010 | /* before memcpy and submit, |
| 1011 | * so it can be redirtied any time */ |
| 1012 | bm_set_page_unchanged(b->bm_pages[page_nr]); |
| 1013 | |
| 1014 | if (ctx->flags & BM_AIO_COPY_PAGES) { |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1015 | void *src, *dest; |
Lars Ellenberg | 9db4e77 | 2011-02-23 15:38:47 +0100 | [diff] [blame] | 1016 | page = mempool_alloc(drbd_md_io_page_pool, __GFP_HIGHMEM|__GFP_WAIT); |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1017 | dest = kmap_atomic(page, KM_USER0); |
| 1018 | src = kmap_atomic(b->bm_pages[page_nr], KM_USER1); |
| 1019 | memcpy(dest, src, PAGE_SIZE); |
| 1020 | kunmap_atomic(src, KM_USER1); |
| 1021 | kunmap_atomic(dest, KM_USER0); |
| 1022 | bm_store_page_idx(page, page_nr); |
| 1023 | } else |
| 1024 | page = b->bm_pages[page_nr]; |
| 1025 | |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1026 | bio->bi_bdev = mdev->ldev->md_bdev; |
| 1027 | bio->bi_sector = on_disk_sector; |
Lars Ellenberg | 9db4e77 | 2011-02-23 15:38:47 +0100 | [diff] [blame] | 1028 | /* bio_add_page of a single page to an empty bio will always succeed, |
| 1029 | * according to api. Do we want to assert that? */ |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1030 | bio_add_page(bio, page, len, 0); |
| 1031 | bio->bi_private = ctx; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1032 | bio->bi_end_io = bm_async_io_complete; |
| 1033 | |
Andreas Gruenbacher | 0cf9d27 | 2010-12-07 10:43:29 +0100 | [diff] [blame] | 1034 | if (drbd_insert_fault(mdev, (rw & WRITE) ? DRBD_FAULT_MD_WR : DRBD_FAULT_MD_RD)) { |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1035 | bio->bi_rw |= rw; |
| 1036 | bio_endio(bio, -EIO); |
| 1037 | } else { |
| 1038 | submit_bio(rw, bio); |
Lars Ellenberg | 5a8b424 | 2011-06-14 14:18:23 +0200 | [diff] [blame] | 1039 | /* this should not count as user activity and cause the |
| 1040 | * resync to throttle -- see drbd_rs_should_slow_down(). */ |
| 1041 | atomic_add(len >> 9, &mdev->rs_sect_ev); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1042 | } |
| 1043 | } |
| 1044 | |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1045 | /* |
| 1046 | * bm_rw: read/write the whole bitmap from/to its on disk location. |
| 1047 | */ |
Lars Ellenberg | 45dfffe | 2011-02-21 13:21:00 +0100 | [diff] [blame] | 1048 | static int bm_rw(struct drbd_conf *mdev, int rw, unsigned flags, unsigned lazy_writeout_upper_idx) __must_hold(local) |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1049 | { |
Philipp Reisner | cdfda63 | 2011-07-05 15:38:59 +0200 | [diff] [blame] | 1050 | struct bm_aio_ctx *ctx; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1051 | struct drbd_bitmap *b = mdev->bitmap; |
Lars Ellenberg | 6850c44 | 2010-12-16 00:32:38 +0100 | [diff] [blame] | 1052 | int num_pages, i, count = 0; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1053 | unsigned long now; |
| 1054 | char ppb[10]; |
| 1055 | int err = 0; |
| 1056 | |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1057 | /* |
| 1058 | * We are protected against bitmap disappearing/resizing by holding an |
| 1059 | * ldev reference (caller must have called get_ldev()). |
| 1060 | * For read/write, we are protected against changes to the bitmap by |
| 1061 | * the bitmap lock (see drbd_bitmap_io). |
| 1062 | * For lazy writeout, we don't care for ongoing changes to the bitmap, |
| 1063 | * as we submit copies of pages anyways. |
| 1064 | */ |
Philipp Reisner | cdfda63 | 2011-07-05 15:38:59 +0200 | [diff] [blame] | 1065 | |
| 1066 | ctx = kmalloc(sizeof(struct bm_aio_ctx), GFP_NOIO); |
| 1067 | if (!ctx) |
| 1068 | return -ENOMEM; |
| 1069 | |
| 1070 | *ctx = (struct bm_aio_ctx) { |
| 1071 | .mdev = mdev, |
| 1072 | .in_flight = ATOMIC_INIT(1), |
| 1073 | .done = 0, |
| 1074 | .flags = flags, |
| 1075 | .error = 0, |
| 1076 | .kref = { ATOMIC_INIT(2) }, |
| 1077 | }; |
| 1078 | |
| 1079 | if (!get_ldev_if_state(mdev, D_ATTACHING)) { /* put is in bm_aio_ctx_destroy() */ |
| 1080 | dev_err(DEV, "ASSERT FAILED: get_ldev_if_state() == 1 in bm_rw()\n"); |
Lars Ellenberg | 9dab384 | 2012-03-26 16:22:00 +0200 | [diff] [blame] | 1081 | kfree(ctx); |
| 1082 | return -ENODEV; |
Philipp Reisner | cdfda63 | 2011-07-05 15:38:59 +0200 | [diff] [blame] | 1083 | } |
| 1084 | |
| 1085 | if (!ctx->flags) |
Lars Ellenberg | 20ceb2b | 2011-01-21 10:56:44 +0100 | [diff] [blame] | 1086 | WARN_ON(!(BM_LOCKED_MASK & b->bm_flags)); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1087 | |
Lars Ellenberg | 6850c44 | 2010-12-16 00:32:38 +0100 | [diff] [blame] | 1088 | num_pages = b->bm_number_of_pages; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1089 | |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1090 | now = jiffies; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1091 | |
| 1092 | /* let the layers below us try to merge these bios... */ |
Lars Ellenberg | 6850c44 | 2010-12-16 00:32:38 +0100 | [diff] [blame] | 1093 | for (i = 0; i < num_pages; i++) { |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1094 | /* ignore completely unchanged pages */ |
| 1095 | if (lazy_writeout_upper_idx && i == lazy_writeout_upper_idx) |
| 1096 | break; |
| 1097 | if (rw & WRITE) { |
Lars Ellenberg | 45dfffe | 2011-02-21 13:21:00 +0100 | [diff] [blame] | 1098 | if ((flags & BM_AIO_WRITE_HINTED) && |
| 1099 | !test_and_clear_bit(BM_PAGE_HINT_WRITEOUT, |
| 1100 | &page_private(b->bm_pages[i]))) |
| 1101 | continue; |
Philipp Reisner | fef45d2 | 2012-08-14 11:46:59 +0200 | [diff] [blame] | 1102 | |
| 1103 | if (!(flags & BM_WRITE_ALL_PAGES) && |
| 1104 | bm_test_page_unchanged(b->bm_pages[i])) { |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1105 | dynamic_dev_dbg(DEV, "skipped bm write for idx %u\n", i); |
| 1106 | continue; |
| 1107 | } |
| 1108 | /* during lazy writeout, |
| 1109 | * ignore those pages not marked for lazy writeout. */ |
| 1110 | if (lazy_writeout_upper_idx && |
| 1111 | !bm_test_page_lazy_writeout(b->bm_pages[i])) { |
| 1112 | dynamic_dev_dbg(DEV, "skipped bm lazy write for idx %u\n", i); |
| 1113 | continue; |
| 1114 | } |
| 1115 | } |
Philipp Reisner | cdfda63 | 2011-07-05 15:38:59 +0200 | [diff] [blame] | 1116 | atomic_inc(&ctx->in_flight); |
| 1117 | bm_page_io_async(ctx, i, rw); |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1118 | ++count; |
| 1119 | cond_resched(); |
| 1120 | } |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1121 | |
Lars Ellenberg | 725a97e | 2010-12-19 11:29:55 +0100 | [diff] [blame] | 1122 | /* |
Philipp Reisner | cdfda63 | 2011-07-05 15:38:59 +0200 | [diff] [blame] | 1123 | * We initialize ctx->in_flight to one to make sure bm_async_io_complete |
| 1124 | * will not set ctx->done early, and decrement / test it here. If there |
Lars Ellenberg | 725a97e | 2010-12-19 11:29:55 +0100 | [diff] [blame] | 1125 | * are still some bios in flight, we need to wait for them here. |
Philipp Reisner | cdfda63 | 2011-07-05 15:38:59 +0200 | [diff] [blame] | 1126 | * If all IO is done already (or nothing had been submitted), there is |
| 1127 | * no need to wait. Still, we need to put the kref associated with the |
| 1128 | * "in_flight reached zero, all done" event. |
Lars Ellenberg | 725a97e | 2010-12-19 11:29:55 +0100 | [diff] [blame] | 1129 | */ |
Philipp Reisner | cdfda63 | 2011-07-05 15:38:59 +0200 | [diff] [blame] | 1130 | if (!atomic_dec_and_test(&ctx->in_flight)) |
Philipp Reisner | 32db80f | 2012-02-22 11:51:57 +0100 | [diff] [blame] | 1131 | wait_until_done_or_disk_failure(mdev, mdev->ldev, &ctx->done); |
Philipp Reisner | cdfda63 | 2011-07-05 15:38:59 +0200 | [diff] [blame] | 1132 | else |
| 1133 | kref_put(&ctx->kref, &bm_aio_ctx_destroy); |
Lars Ellenberg | c9d963a | 2011-02-21 15:10:23 +0100 | [diff] [blame] | 1134 | |
| 1135 | /* summary for global bitmap IO */ |
| 1136 | if (flags == 0) |
| 1137 | dev_info(DEV, "bitmap %s of %u pages took %lu jiffies\n", |
Philipp Reisner | cdfda63 | 2011-07-05 15:38:59 +0200 | [diff] [blame] | 1138 | rw == WRITE ? "WRITE" : "READ", |
| 1139 | count, jiffies - now); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1140 | |
Philipp Reisner | cdfda63 | 2011-07-05 15:38:59 +0200 | [diff] [blame] | 1141 | if (ctx->error) { |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1142 | dev_alert(DEV, "we had at least one MD IO ERROR during bitmap IO\n"); |
Lars Ellenberg | 0c84966 | 2012-07-30 09:07:28 +0200 | [diff] [blame] | 1143 | drbd_chk_io_error(mdev, 1, DRBD_META_IO_ERROR); |
Philipp Reisner | cdfda63 | 2011-07-05 15:38:59 +0200 | [diff] [blame] | 1144 | err = -EIO; /* ctx->error ? */ |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1145 | } |
| 1146 | |
Philipp Reisner | cdfda63 | 2011-07-05 15:38:59 +0200 | [diff] [blame] | 1147 | if (atomic_read(&ctx->in_flight)) |
| 1148 | err = -EIO; /* Disk failed during IO... */ |
| 1149 | |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1150 | now = jiffies; |
| 1151 | if (rw == WRITE) { |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1152 | drbd_md_flush(mdev); |
| 1153 | } else /* rw == READ */ { |
Lars Ellenberg | 95a0f10 | 2010-12-15 08:59:09 +0100 | [diff] [blame] | 1154 | b->bm_set = bm_count_bits(b); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1155 | dev_info(DEV, "recounting of set bits took additional %lu jiffies\n", |
| 1156 | jiffies - now); |
| 1157 | } |
| 1158 | now = b->bm_set; |
| 1159 | |
Lars Ellenberg | c9d963a | 2011-02-21 15:10:23 +0100 | [diff] [blame] | 1160 | if (flags == 0) |
| 1161 | dev_info(DEV, "%s (%lu bits) marked out-of-sync by on disk bit-map.\n", |
| 1162 | ppsize(ppb, now << (BM_BLOCK_SHIFT-10)), now); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1163 | |
Philipp Reisner | cdfda63 | 2011-07-05 15:38:59 +0200 | [diff] [blame] | 1164 | kref_put(&ctx->kref, &bm_aio_ctx_destroy); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1165 | return err; |
| 1166 | } |
| 1167 | |
| 1168 | /** |
| 1169 | * drbd_bm_read() - Read the whole bitmap from its on disk location. |
| 1170 | * @mdev: DRBD device. |
| 1171 | */ |
| 1172 | int drbd_bm_read(struct drbd_conf *mdev) __must_hold(local) |
| 1173 | { |
Lars Ellenberg | 45dfffe | 2011-02-21 13:21:00 +0100 | [diff] [blame] | 1174 | return bm_rw(mdev, READ, 0, 0); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1175 | } |
| 1176 | |
| 1177 | /** |
| 1178 | * drbd_bm_write() - Write the whole bitmap to its on disk location. |
| 1179 | * @mdev: DRBD device. |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1180 | * |
| 1181 | * Will only write pages that have changed since last IO. |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1182 | */ |
| 1183 | int drbd_bm_write(struct drbd_conf *mdev) __must_hold(local) |
| 1184 | { |
Lars Ellenberg | 45dfffe | 2011-02-21 13:21:00 +0100 | [diff] [blame] | 1185 | return bm_rw(mdev, WRITE, 0, 0); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1186 | } |
| 1187 | |
| 1188 | /** |
Philipp Reisner | fef45d2 | 2012-08-14 11:46:59 +0200 | [diff] [blame] | 1189 | * drbd_bm_write_all() - Write the whole bitmap to its on disk location. |
| 1190 | * @mdev: DRBD device. |
| 1191 | * |
| 1192 | * Will write all pages. |
| 1193 | */ |
| 1194 | int drbd_bm_write_all(struct drbd_conf *mdev) __must_hold(local) |
| 1195 | { |
| 1196 | return bm_rw(mdev, WRITE, BM_WRITE_ALL_PAGES, 0); |
| 1197 | } |
| 1198 | |
| 1199 | /** |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1200 | * drbd_bm_lazy_write_out() - Write bitmap pages 0 to @upper_idx-1, if they have changed. |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1201 | * @mdev: DRBD device. |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1202 | * @upper_idx: 0: write all changed pages; +ve: page index to stop scanning for changed pages |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1203 | */ |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1204 | int drbd_bm_write_lazy(struct drbd_conf *mdev, unsigned upper_idx) __must_hold(local) |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1205 | { |
Lars Ellenberg | 45dfffe | 2011-02-21 13:21:00 +0100 | [diff] [blame] | 1206 | return bm_rw(mdev, WRITE, BM_AIO_COPY_PAGES, upper_idx); |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1207 | } |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1208 | |
Lars Ellenberg | 45dfffe | 2011-02-21 13:21:00 +0100 | [diff] [blame] | 1209 | /** |
Lars Ellenberg | a220d29 | 2012-05-07 12:07:18 +0200 | [diff] [blame] | 1210 | * drbd_bm_write_copy_pages() - Write the whole bitmap to its on disk location. |
| 1211 | * @mdev: DRBD device. |
| 1212 | * |
| 1213 | * Will only write pages that have changed since last IO. |
| 1214 | * In contrast to drbd_bm_write(), this will copy the bitmap pages |
| 1215 | * to temporary writeout pages. It is intended to trigger a full write-out |
| 1216 | * while still allowing the bitmap to change, for example if a resync or online |
| 1217 | * verify is aborted due to a failed peer disk, while local IO continues, or |
| 1218 | * pending resync acks are still being processed. |
| 1219 | */ |
| 1220 | int drbd_bm_write_copy_pages(struct drbd_conf *mdev) __must_hold(local) |
| 1221 | { |
| 1222 | return bm_rw(mdev, WRITE, BM_AIO_COPY_PAGES, 0); |
| 1223 | } |
| 1224 | |
| 1225 | /** |
Lars Ellenberg | 45dfffe | 2011-02-21 13:21:00 +0100 | [diff] [blame] | 1226 | * drbd_bm_write_hinted() - Write bitmap pages with "hint" marks, if they have changed. |
| 1227 | * @mdev: DRBD device. |
| 1228 | */ |
| 1229 | int drbd_bm_write_hinted(struct drbd_conf *mdev) __must_hold(local) |
| 1230 | { |
| 1231 | return bm_rw(mdev, WRITE, BM_AIO_WRITE_HINTED | BM_AIO_COPY_PAGES, 0); |
| 1232 | } |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1233 | |
| 1234 | /** |
Lars Ellenberg | 45dfffe | 2011-02-21 13:21:00 +0100 | [diff] [blame] | 1235 | * drbd_bm_write_page() - Writes a PAGE_SIZE aligned piece of bitmap |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1236 | * @mdev: DRBD device. |
| 1237 | * @idx: bitmap page index |
| 1238 | * |
Lars Ellenberg | 4b0715f | 2010-12-14 15:13:04 +0100 | [diff] [blame] | 1239 | * We don't want to special case on logical_block_size of the backend device, |
| 1240 | * so we submit PAGE_SIZE aligned pieces. |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1241 | * Note that on "most" systems, PAGE_SIZE is 4k. |
Lars Ellenberg | 4b0715f | 2010-12-14 15:13:04 +0100 | [diff] [blame] | 1242 | * |
| 1243 | * In case this becomes an issue on systems with larger PAGE_SIZE, |
| 1244 | * we may want to change this again to write 4k aligned 4k pieces. |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1245 | */ |
| 1246 | int drbd_bm_write_page(struct drbd_conf *mdev, unsigned int idx) __must_hold(local) |
| 1247 | { |
Philipp Reisner | cdfda63 | 2011-07-05 15:38:59 +0200 | [diff] [blame] | 1248 | struct bm_aio_ctx *ctx; |
| 1249 | int err; |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1250 | |
| 1251 | if (bm_test_page_unchanged(mdev->bitmap->bm_pages[idx])) { |
Lars Ellenberg | 7648cdf | 2010-12-17 23:58:41 +0100 | [diff] [blame] | 1252 | dynamic_dev_dbg(DEV, "skipped bm page write for idx %u\n", idx); |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1253 | return 0; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1254 | } |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1255 | |
Philipp Reisner | cdfda63 | 2011-07-05 15:38:59 +0200 | [diff] [blame] | 1256 | ctx = kmalloc(sizeof(struct bm_aio_ctx), GFP_NOIO); |
| 1257 | if (!ctx) |
| 1258 | return -ENOMEM; |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1259 | |
Philipp Reisner | cdfda63 | 2011-07-05 15:38:59 +0200 | [diff] [blame] | 1260 | *ctx = (struct bm_aio_ctx) { |
| 1261 | .mdev = mdev, |
| 1262 | .in_flight = ATOMIC_INIT(1), |
| 1263 | .done = 0, |
| 1264 | .flags = BM_AIO_COPY_PAGES, |
| 1265 | .error = 0, |
| 1266 | .kref = { ATOMIC_INIT(2) }, |
| 1267 | }; |
| 1268 | |
| 1269 | if (!get_ldev_if_state(mdev, D_ATTACHING)) { /* put is in bm_aio_ctx_destroy() */ |
| 1270 | dev_err(DEV, "ASSERT FAILED: get_ldev_if_state() == 1 in drbd_bm_write_page()\n"); |
Lars Ellenberg | 9dab384 | 2012-03-26 16:22:00 +0200 | [diff] [blame] | 1271 | kfree(ctx); |
| 1272 | return -ENODEV; |
Philipp Reisner | cdfda63 | 2011-07-05 15:38:59 +0200 | [diff] [blame] | 1273 | } |
| 1274 | |
| 1275 | bm_page_io_async(ctx, idx, WRITE_SYNC); |
Philipp Reisner | 32db80f | 2012-02-22 11:51:57 +0100 | [diff] [blame] | 1276 | wait_until_done_or_disk_failure(mdev, mdev->ldev, &ctx->done); |
Philipp Reisner | cdfda63 | 2011-07-05 15:38:59 +0200 | [diff] [blame] | 1277 | |
| 1278 | if (ctx->error) |
Lars Ellenberg | 0c84966 | 2012-07-30 09:07:28 +0200 | [diff] [blame] | 1279 | drbd_chk_io_error(mdev, 1, DRBD_META_IO_ERROR); |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1280 | /* that should force detach, so the in memory bitmap will be |
| 1281 | * gone in a moment as well. */ |
| 1282 | |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1283 | mdev->bm_writ_cnt++; |
Philipp Reisner | cdfda63 | 2011-07-05 15:38:59 +0200 | [diff] [blame] | 1284 | err = atomic_read(&ctx->in_flight) ? -EIO : ctx->error; |
Philipp Reisner | cdfda63 | 2011-07-05 15:38:59 +0200 | [diff] [blame] | 1285 | kref_put(&ctx->kref, &bm_aio_ctx_destroy); |
| 1286 | return err; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1287 | } |
| 1288 | |
| 1289 | /* NOTE |
| 1290 | * find_first_bit returns int, we return unsigned long. |
Lars Ellenberg | 4b0715f | 2010-12-14 15:13:04 +0100 | [diff] [blame] | 1291 | * For this to work on 32bit arch with bitnumbers > (1<<32), |
| 1292 | * we'd need to return u64, and get a whole lot of other places |
| 1293 | * fixed where we still use unsigned long. |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1294 | * |
| 1295 | * this returns a bit number, NOT a sector! |
| 1296 | */ |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1297 | static unsigned long __bm_find_next(struct drbd_conf *mdev, unsigned long bm_fo, |
| 1298 | const int find_zero_bit, const enum km_type km) |
| 1299 | { |
| 1300 | struct drbd_bitmap *b = mdev->bitmap; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1301 | unsigned long *p_addr; |
Lars Ellenberg | 4b0715f | 2010-12-14 15:13:04 +0100 | [diff] [blame] | 1302 | unsigned long bit_offset; |
| 1303 | unsigned i; |
| 1304 | |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1305 | |
| 1306 | if (bm_fo > b->bm_bits) { |
| 1307 | dev_err(DEV, "bm_fo=%lu bm_bits=%lu\n", bm_fo, b->bm_bits); |
Lars Ellenberg | 4b0715f | 2010-12-14 15:13:04 +0100 | [diff] [blame] | 1308 | bm_fo = DRBD_END_OF_BITMAP; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1309 | } else { |
| 1310 | while (bm_fo < b->bm_bits) { |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1311 | /* bit offset of the first bit in the page */ |
Lars Ellenberg | 4b0715f | 2010-12-14 15:13:04 +0100 | [diff] [blame] | 1312 | bit_offset = bm_fo & ~BITS_PER_PAGE_MASK; |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1313 | p_addr = __bm_map_pidx(b, bm_bit_to_page_idx(b, bm_fo), km); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1314 | |
| 1315 | if (find_zero_bit) |
Linus Torvalds | 7e599e6 | 2011-03-28 07:42:58 -0700 | [diff] [blame] | 1316 | i = find_next_zero_bit_le(p_addr, |
Lars Ellenberg | 4b0715f | 2010-12-14 15:13:04 +0100 | [diff] [blame] | 1317 | PAGE_SIZE*8, bm_fo & BITS_PER_PAGE_MASK); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1318 | else |
Linus Torvalds | 7e599e6 | 2011-03-28 07:42:58 -0700 | [diff] [blame] | 1319 | i = find_next_bit_le(p_addr, |
Lars Ellenberg | 4b0715f | 2010-12-14 15:13:04 +0100 | [diff] [blame] | 1320 | PAGE_SIZE*8, bm_fo & BITS_PER_PAGE_MASK); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1321 | |
| 1322 | __bm_unmap(p_addr, km); |
| 1323 | if (i < PAGE_SIZE*8) { |
Lars Ellenberg | 4b0715f | 2010-12-14 15:13:04 +0100 | [diff] [blame] | 1324 | bm_fo = bit_offset + i; |
| 1325 | if (bm_fo >= b->bm_bits) |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1326 | break; |
| 1327 | goto found; |
| 1328 | } |
| 1329 | bm_fo = bit_offset + PAGE_SIZE*8; |
| 1330 | } |
Lars Ellenberg | 4b0715f | 2010-12-14 15:13:04 +0100 | [diff] [blame] | 1331 | bm_fo = DRBD_END_OF_BITMAP; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1332 | } |
| 1333 | found: |
Lars Ellenberg | 4b0715f | 2010-12-14 15:13:04 +0100 | [diff] [blame] | 1334 | return bm_fo; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1335 | } |
| 1336 | |
| 1337 | static unsigned long bm_find_next(struct drbd_conf *mdev, |
| 1338 | unsigned long bm_fo, const int find_zero_bit) |
| 1339 | { |
| 1340 | struct drbd_bitmap *b = mdev->bitmap; |
Lars Ellenberg | 4b0715f | 2010-12-14 15:13:04 +0100 | [diff] [blame] | 1341 | unsigned long i = DRBD_END_OF_BITMAP; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1342 | |
Andreas Gruenbacher | 841ce24 | 2010-12-15 19:31:20 +0100 | [diff] [blame] | 1343 | if (!expect(b)) |
| 1344 | return i; |
| 1345 | if (!expect(b->bm_pages)) |
| 1346 | return i; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1347 | |
| 1348 | spin_lock_irq(&b->bm_lock); |
Lars Ellenberg | 20ceb2b | 2011-01-21 10:56:44 +0100 | [diff] [blame] | 1349 | if (BM_DONT_TEST & b->bm_flags) |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1350 | bm_print_lock_info(mdev); |
| 1351 | |
| 1352 | i = __bm_find_next(mdev, bm_fo, find_zero_bit, KM_IRQ1); |
| 1353 | |
| 1354 | spin_unlock_irq(&b->bm_lock); |
| 1355 | return i; |
| 1356 | } |
| 1357 | |
| 1358 | unsigned long drbd_bm_find_next(struct drbd_conf *mdev, unsigned long bm_fo) |
| 1359 | { |
| 1360 | return bm_find_next(mdev, bm_fo, 0); |
| 1361 | } |
| 1362 | |
| 1363 | #if 0 |
| 1364 | /* not yet needed for anything. */ |
| 1365 | unsigned long drbd_bm_find_next_zero(struct drbd_conf *mdev, unsigned long bm_fo) |
| 1366 | { |
| 1367 | return bm_find_next(mdev, bm_fo, 1); |
| 1368 | } |
| 1369 | #endif |
| 1370 | |
| 1371 | /* does not spin_lock_irqsave. |
| 1372 | * you must take drbd_bm_lock() first */ |
| 1373 | unsigned long _drbd_bm_find_next(struct drbd_conf *mdev, unsigned long bm_fo) |
| 1374 | { |
Lars Ellenberg | 20ceb2b | 2011-01-21 10:56:44 +0100 | [diff] [blame] | 1375 | /* WARN_ON(!(BM_DONT_SET & mdev->b->bm_flags)); */ |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1376 | return __bm_find_next(mdev, bm_fo, 0, KM_USER1); |
| 1377 | } |
| 1378 | |
| 1379 | unsigned long _drbd_bm_find_next_zero(struct drbd_conf *mdev, unsigned long bm_fo) |
| 1380 | { |
Lars Ellenberg | 20ceb2b | 2011-01-21 10:56:44 +0100 | [diff] [blame] | 1381 | /* WARN_ON(!(BM_DONT_SET & mdev->b->bm_flags)); */ |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1382 | return __bm_find_next(mdev, bm_fo, 1, KM_USER1); |
| 1383 | } |
| 1384 | |
| 1385 | /* returns number of bits actually changed. |
| 1386 | * for val != 0, we change 0 -> 1, return code positive |
| 1387 | * for val == 0, we change 1 -> 0, return code negative |
| 1388 | * wants bitnr, not sector. |
| 1389 | * expected to be called for only a few bits (e - s about BITS_PER_LONG). |
| 1390 | * Must hold bitmap lock already. */ |
Philipp Reisner | b4ee79d | 2010-04-01 09:57:40 +0200 | [diff] [blame] | 1391 | static int __bm_change_bits_to(struct drbd_conf *mdev, const unsigned long s, |
Lars Ellenberg | 829c608 | 2011-06-03 21:18:13 +0200 | [diff] [blame] | 1392 | unsigned long e, int val) |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1393 | { |
| 1394 | struct drbd_bitmap *b = mdev->bitmap; |
| 1395 | unsigned long *p_addr = NULL; |
| 1396 | unsigned long bitnr; |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1397 | unsigned int last_page_nr = -1U; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1398 | int c = 0; |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1399 | int changed_total = 0; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1400 | |
| 1401 | if (e >= b->bm_bits) { |
| 1402 | dev_err(DEV, "ASSERT FAILED: bit_s=%lu bit_e=%lu bm_bits=%lu\n", |
| 1403 | s, e, b->bm_bits); |
| 1404 | e = b->bm_bits ? b->bm_bits -1 : 0; |
| 1405 | } |
| 1406 | for (bitnr = s; bitnr <= e; bitnr++) { |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1407 | unsigned int page_nr = bm_bit_to_page_idx(b, bitnr); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1408 | if (page_nr != last_page_nr) { |
| 1409 | if (p_addr) |
Lars Ellenberg | 829c608 | 2011-06-03 21:18:13 +0200 | [diff] [blame] | 1410 | __bm_unmap(p_addr, KM_IRQ1); |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1411 | if (c < 0) |
| 1412 | bm_set_page_lazy_writeout(b->bm_pages[last_page_nr]); |
| 1413 | else if (c > 0) |
| 1414 | bm_set_page_need_writeout(b->bm_pages[last_page_nr]); |
| 1415 | changed_total += c; |
| 1416 | c = 0; |
Lars Ellenberg | 829c608 | 2011-06-03 21:18:13 +0200 | [diff] [blame] | 1417 | p_addr = __bm_map_pidx(b, page_nr, KM_IRQ1); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1418 | last_page_nr = page_nr; |
| 1419 | } |
| 1420 | if (val) |
Linus Torvalds | 7e599e6 | 2011-03-28 07:42:58 -0700 | [diff] [blame] | 1421 | c += (0 == __test_and_set_bit_le(bitnr & BITS_PER_PAGE_MASK, p_addr)); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1422 | else |
Linus Torvalds | 7e599e6 | 2011-03-28 07:42:58 -0700 | [diff] [blame] | 1423 | c -= (0 != __test_and_clear_bit_le(bitnr & BITS_PER_PAGE_MASK, p_addr)); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1424 | } |
| 1425 | if (p_addr) |
Lars Ellenberg | 829c608 | 2011-06-03 21:18:13 +0200 | [diff] [blame] | 1426 | __bm_unmap(p_addr, KM_IRQ1); |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1427 | if (c < 0) |
| 1428 | bm_set_page_lazy_writeout(b->bm_pages[last_page_nr]); |
| 1429 | else if (c > 0) |
| 1430 | bm_set_page_need_writeout(b->bm_pages[last_page_nr]); |
| 1431 | changed_total += c; |
| 1432 | b->bm_set += changed_total; |
| 1433 | return changed_total; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1434 | } |
| 1435 | |
| 1436 | /* returns number of bits actually changed. |
| 1437 | * for val != 0, we change 0 -> 1, return code positive |
| 1438 | * for val == 0, we change 1 -> 0, return code negative |
| 1439 | * wants bitnr, not sector */ |
Philipp Reisner | b4ee79d | 2010-04-01 09:57:40 +0200 | [diff] [blame] | 1440 | static int bm_change_bits_to(struct drbd_conf *mdev, const unsigned long s, |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1441 | const unsigned long e, int val) |
| 1442 | { |
| 1443 | unsigned long flags; |
| 1444 | struct drbd_bitmap *b = mdev->bitmap; |
| 1445 | int c = 0; |
| 1446 | |
Andreas Gruenbacher | 841ce24 | 2010-12-15 19:31:20 +0100 | [diff] [blame] | 1447 | if (!expect(b)) |
| 1448 | return 1; |
| 1449 | if (!expect(b->bm_pages)) |
| 1450 | return 0; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1451 | |
| 1452 | spin_lock_irqsave(&b->bm_lock, flags); |
Lars Ellenberg | 20ceb2b | 2011-01-21 10:56:44 +0100 | [diff] [blame] | 1453 | if ((val ? BM_DONT_SET : BM_DONT_CLEAR) & b->bm_flags) |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1454 | bm_print_lock_info(mdev); |
| 1455 | |
Lars Ellenberg | 829c608 | 2011-06-03 21:18:13 +0200 | [diff] [blame] | 1456 | c = __bm_change_bits_to(mdev, s, e, val); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1457 | |
| 1458 | spin_unlock_irqrestore(&b->bm_lock, flags); |
| 1459 | return c; |
| 1460 | } |
| 1461 | |
| 1462 | /* returns number of bits changed 0 -> 1 */ |
| 1463 | int drbd_bm_set_bits(struct drbd_conf *mdev, const unsigned long s, const unsigned long e) |
| 1464 | { |
| 1465 | return bm_change_bits_to(mdev, s, e, 1); |
| 1466 | } |
| 1467 | |
| 1468 | /* returns number of bits changed 1 -> 0 */ |
| 1469 | int drbd_bm_clear_bits(struct drbd_conf *mdev, const unsigned long s, const unsigned long e) |
| 1470 | { |
| 1471 | return -bm_change_bits_to(mdev, s, e, 0); |
| 1472 | } |
| 1473 | |
| 1474 | /* sets all bits in full words, |
| 1475 | * from first_word up to, but not including, last_word */ |
| 1476 | static inline void bm_set_full_words_within_one_page(struct drbd_bitmap *b, |
| 1477 | int page_nr, int first_word, int last_word) |
| 1478 | { |
| 1479 | int i; |
| 1480 | int bits; |
Lars Ellenberg | 22d8114 | 2011-09-26 09:58:04 +0200 | [diff] [blame] | 1481 | int changed = 0; |
Lars Ellenberg | 829c608 | 2011-06-03 21:18:13 +0200 | [diff] [blame] | 1482 | unsigned long *paddr = kmap_atomic(b->bm_pages[page_nr], KM_IRQ1); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1483 | for (i = first_word; i < last_word; i++) { |
| 1484 | bits = hweight_long(paddr[i]); |
| 1485 | paddr[i] = ~0UL; |
Lars Ellenberg | 22d8114 | 2011-09-26 09:58:04 +0200 | [diff] [blame] | 1486 | changed += BITS_PER_LONG - bits; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1487 | } |
Lars Ellenberg | 829c608 | 2011-06-03 21:18:13 +0200 | [diff] [blame] | 1488 | kunmap_atomic(paddr, KM_IRQ1); |
Lars Ellenberg | 22d8114 | 2011-09-26 09:58:04 +0200 | [diff] [blame] | 1489 | if (changed) { |
| 1490 | /* We only need lazy writeout, the information is still in the |
| 1491 | * remote bitmap as well, and is reconstructed during the next |
| 1492 | * bitmap exchange, if lost locally due to a crash. */ |
| 1493 | bm_set_page_lazy_writeout(b->bm_pages[page_nr]); |
| 1494 | b->bm_set += changed; |
| 1495 | } |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1496 | } |
| 1497 | |
Lars Ellenberg | 829c608 | 2011-06-03 21:18:13 +0200 | [diff] [blame] | 1498 | /* Same thing as drbd_bm_set_bits, |
| 1499 | * but more efficient for a large bit range. |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1500 | * You must first drbd_bm_lock(). |
| 1501 | * Can be called to set the whole bitmap in one go. |
| 1502 | * Sets bits from s to e _inclusive_. */ |
| 1503 | void _drbd_bm_set_bits(struct drbd_conf *mdev, const unsigned long s, const unsigned long e) |
| 1504 | { |
| 1505 | /* First set_bit from the first bit (s) |
| 1506 | * up to the next long boundary (sl), |
| 1507 | * then assign full words up to the last long boundary (el), |
| 1508 | * then set_bit up to and including the last bit (e). |
| 1509 | * |
| 1510 | * Do not use memset, because we must account for changes, |
| 1511 | * so we need to loop over the words with hweight() anyways. |
| 1512 | */ |
Lars Ellenberg | 829c608 | 2011-06-03 21:18:13 +0200 | [diff] [blame] | 1513 | struct drbd_bitmap *b = mdev->bitmap; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1514 | unsigned long sl = ALIGN(s,BITS_PER_LONG); |
| 1515 | unsigned long el = (e+1) & ~((unsigned long)BITS_PER_LONG-1); |
| 1516 | int first_page; |
| 1517 | int last_page; |
| 1518 | int page_nr; |
| 1519 | int first_word; |
| 1520 | int last_word; |
| 1521 | |
| 1522 | if (e - s <= 3*BITS_PER_LONG) { |
| 1523 | /* don't bother; el and sl may even be wrong. */ |
Lars Ellenberg | 829c608 | 2011-06-03 21:18:13 +0200 | [diff] [blame] | 1524 | spin_lock_irq(&b->bm_lock); |
| 1525 | __bm_change_bits_to(mdev, s, e, 1); |
| 1526 | spin_unlock_irq(&b->bm_lock); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1527 | return; |
| 1528 | } |
| 1529 | |
| 1530 | /* difference is large enough that we can trust sl and el */ |
| 1531 | |
Lars Ellenberg | 829c608 | 2011-06-03 21:18:13 +0200 | [diff] [blame] | 1532 | spin_lock_irq(&b->bm_lock); |
| 1533 | |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1534 | /* bits filling the current long */ |
| 1535 | if (sl) |
Lars Ellenberg | 829c608 | 2011-06-03 21:18:13 +0200 | [diff] [blame] | 1536 | __bm_change_bits_to(mdev, s, sl-1, 1); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1537 | |
| 1538 | first_page = sl >> (3 + PAGE_SHIFT); |
| 1539 | last_page = el >> (3 + PAGE_SHIFT); |
| 1540 | |
| 1541 | /* MLPP: modulo longs per page */ |
| 1542 | /* LWPP: long words per page */ |
| 1543 | first_word = MLPP(sl >> LN2_BPL); |
| 1544 | last_word = LWPP; |
| 1545 | |
| 1546 | /* first and full pages, unless first page == last page */ |
| 1547 | for (page_nr = first_page; page_nr < last_page; page_nr++) { |
| 1548 | bm_set_full_words_within_one_page(mdev->bitmap, page_nr, first_word, last_word); |
Lars Ellenberg | 8ccee20 | 2011-06-06 11:31:42 +0200 | [diff] [blame] | 1549 | spin_unlock_irq(&b->bm_lock); |
| 1550 | cond_resched(); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1551 | first_word = 0; |
Lars Ellenberg | 8ccee20 | 2011-06-06 11:31:42 +0200 | [diff] [blame] | 1552 | spin_lock_irq(&b->bm_lock); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1553 | } |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1554 | /* last page (respectively only page, for first page == last page) */ |
| 1555 | last_word = MLPP(el >> LN2_BPL); |
Lars Ellenberg | 1b6f197 | 2012-06-08 15:06:39 +0200 | [diff] [blame] | 1556 | |
| 1557 | /* consider bitmap->bm_bits = 32768, bitmap->bm_number_of_pages = 1. (or multiples). |
| 1558 | * ==> e = 32767, el = 32768, last_page = 2, |
| 1559 | * and now last_word = 0. |
| 1560 | * We do not want to touch last_page in this case, |
| 1561 | * as we did not allocate it, it is not present in bitmap->bm_pages. |
| 1562 | */ |
| 1563 | if (last_word) |
| 1564 | bm_set_full_words_within_one_page(mdev->bitmap, last_page, first_word, last_word); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1565 | |
| 1566 | /* possibly trailing bits. |
| 1567 | * example: (e & 63) == 63, el will be e+1. |
| 1568 | * if that even was the very last bit, |
| 1569 | * it would trigger an assert in __bm_change_bits_to() |
| 1570 | */ |
| 1571 | if (el <= e) |
Lars Ellenberg | 829c608 | 2011-06-03 21:18:13 +0200 | [diff] [blame] | 1572 | __bm_change_bits_to(mdev, el, e, 1); |
| 1573 | spin_unlock_irq(&b->bm_lock); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1574 | } |
| 1575 | |
| 1576 | /* returns bit state |
| 1577 | * wants bitnr, NOT sector. |
| 1578 | * inherently racy... area needs to be locked by means of {al,rs}_lru |
| 1579 | * 1 ... bit set |
| 1580 | * 0 ... bit not set |
| 1581 | * -1 ... first out of bounds access, stop testing for bits! |
| 1582 | */ |
| 1583 | int drbd_bm_test_bit(struct drbd_conf *mdev, const unsigned long bitnr) |
| 1584 | { |
| 1585 | unsigned long flags; |
| 1586 | struct drbd_bitmap *b = mdev->bitmap; |
| 1587 | unsigned long *p_addr; |
| 1588 | int i; |
| 1589 | |
Andreas Gruenbacher | 841ce24 | 2010-12-15 19:31:20 +0100 | [diff] [blame] | 1590 | if (!expect(b)) |
| 1591 | return 0; |
| 1592 | if (!expect(b->bm_pages)) |
| 1593 | return 0; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1594 | |
| 1595 | spin_lock_irqsave(&b->bm_lock, flags); |
Lars Ellenberg | 20ceb2b | 2011-01-21 10:56:44 +0100 | [diff] [blame] | 1596 | if (BM_DONT_TEST & b->bm_flags) |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1597 | bm_print_lock_info(mdev); |
| 1598 | if (bitnr < b->bm_bits) { |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1599 | p_addr = bm_map_pidx(b, bm_bit_to_page_idx(b, bitnr)); |
Linus Torvalds | 7e599e6 | 2011-03-28 07:42:58 -0700 | [diff] [blame] | 1600 | i = test_bit_le(bitnr & BITS_PER_PAGE_MASK, p_addr) ? 1 : 0; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1601 | bm_unmap(p_addr); |
| 1602 | } else if (bitnr == b->bm_bits) { |
| 1603 | i = -1; |
| 1604 | } else { /* (bitnr > b->bm_bits) */ |
| 1605 | dev_err(DEV, "bitnr=%lu > bm_bits=%lu\n", bitnr, b->bm_bits); |
| 1606 | i = 0; |
| 1607 | } |
| 1608 | |
| 1609 | spin_unlock_irqrestore(&b->bm_lock, flags); |
| 1610 | return i; |
| 1611 | } |
| 1612 | |
| 1613 | /* returns number of bits set in the range [s, e] */ |
| 1614 | int drbd_bm_count_bits(struct drbd_conf *mdev, const unsigned long s, const unsigned long e) |
| 1615 | { |
| 1616 | unsigned long flags; |
| 1617 | struct drbd_bitmap *b = mdev->bitmap; |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1618 | unsigned long *p_addr = NULL; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1619 | unsigned long bitnr; |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1620 | unsigned int page_nr = -1U; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1621 | int c = 0; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1622 | |
| 1623 | /* If this is called without a bitmap, that is a bug. But just to be |
| 1624 | * robust in case we screwed up elsewhere, in that case pretend there |
| 1625 | * was one dirty bit in the requested area, so we won't try to do a |
| 1626 | * local read there (no bitmap probably implies no disk) */ |
Andreas Gruenbacher | 841ce24 | 2010-12-15 19:31:20 +0100 | [diff] [blame] | 1627 | if (!expect(b)) |
| 1628 | return 1; |
| 1629 | if (!expect(b->bm_pages)) |
| 1630 | return 1; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1631 | |
| 1632 | spin_lock_irqsave(&b->bm_lock, flags); |
Lars Ellenberg | 20ceb2b | 2011-01-21 10:56:44 +0100 | [diff] [blame] | 1633 | if (BM_DONT_TEST & b->bm_flags) |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1634 | bm_print_lock_info(mdev); |
| 1635 | for (bitnr = s; bitnr <= e; bitnr++) { |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1636 | unsigned int idx = bm_bit_to_page_idx(b, bitnr); |
| 1637 | if (page_nr != idx) { |
| 1638 | page_nr = idx; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1639 | if (p_addr) |
| 1640 | bm_unmap(p_addr); |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1641 | p_addr = bm_map_pidx(b, idx); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1642 | } |
Andreas Gruenbacher | 841ce24 | 2010-12-15 19:31:20 +0100 | [diff] [blame] | 1643 | if (expect(bitnr < b->bm_bits)) |
Linus Torvalds | 7e599e6 | 2011-03-28 07:42:58 -0700 | [diff] [blame] | 1644 | c += (0 != test_bit_le(bitnr - (page_nr << (PAGE_SHIFT+3)), p_addr)); |
Andreas Gruenbacher | 841ce24 | 2010-12-15 19:31:20 +0100 | [diff] [blame] | 1645 | else |
| 1646 | dev_err(DEV, "bitnr=%lu bm_bits=%lu\n", bitnr, b->bm_bits); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1647 | } |
| 1648 | if (p_addr) |
| 1649 | bm_unmap(p_addr); |
| 1650 | spin_unlock_irqrestore(&b->bm_lock, flags); |
| 1651 | return c; |
| 1652 | } |
| 1653 | |
| 1654 | |
| 1655 | /* inherently racy... |
| 1656 | * return value may be already out-of-date when this function returns. |
| 1657 | * but the general usage is that this is only use during a cstate when bits are |
| 1658 | * only cleared, not set, and typically only care for the case when the return |
| 1659 | * value is zero, or we already "locked" this "bitmap extent" by other means. |
| 1660 | * |
| 1661 | * enr is bm-extent number, since we chose to name one sector (512 bytes) |
| 1662 | * worth of the bitmap a "bitmap extent". |
| 1663 | * |
| 1664 | * TODO |
| 1665 | * I think since we use it like a reference count, we should use the real |
| 1666 | * reference count of some bitmap extent element from some lru instead... |
| 1667 | * |
| 1668 | */ |
| 1669 | int drbd_bm_e_weight(struct drbd_conf *mdev, unsigned long enr) |
| 1670 | { |
| 1671 | struct drbd_bitmap *b = mdev->bitmap; |
| 1672 | int count, s, e; |
| 1673 | unsigned long flags; |
| 1674 | unsigned long *p_addr, *bm; |
| 1675 | |
Andreas Gruenbacher | 841ce24 | 2010-12-15 19:31:20 +0100 | [diff] [blame] | 1676 | if (!expect(b)) |
| 1677 | return 0; |
| 1678 | if (!expect(b->bm_pages)) |
| 1679 | return 0; |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1680 | |
| 1681 | spin_lock_irqsave(&b->bm_lock, flags); |
Lars Ellenberg | 20ceb2b | 2011-01-21 10:56:44 +0100 | [diff] [blame] | 1682 | if (BM_DONT_TEST & b->bm_flags) |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1683 | bm_print_lock_info(mdev); |
| 1684 | |
| 1685 | s = S2W(enr); |
| 1686 | e = min((size_t)S2W(enr+1), b->bm_words); |
| 1687 | count = 0; |
| 1688 | if (s < b->bm_words) { |
| 1689 | int n = e-s; |
Lars Ellenberg | 19f843a | 2010-12-15 08:59:11 +0100 | [diff] [blame] | 1690 | p_addr = bm_map_pidx(b, bm_word_to_page_idx(b, s)); |
Philipp Reisner | b411b36 | 2009-09-25 16:07:19 -0700 | [diff] [blame] | 1691 | bm = p_addr + MLPP(s); |
| 1692 | while (n--) |
| 1693 | count += hweight_long(*bm++); |
| 1694 | bm_unmap(p_addr); |
| 1695 | } else { |
| 1696 | dev_err(DEV, "start offset (%d) too large in drbd_bm_e_weight\n", s); |
| 1697 | } |
| 1698 | spin_unlock_irqrestore(&b->bm_lock, flags); |
| 1699 | return count; |
| 1700 | } |