1457c8996SThomas Gleixner // SPDX-License-Identifier: GPL-2.0-only 2935fe098SMike Snitzer /* 3935fe098SMike Snitzer * bitmap.c two-level bitmap (C) Peter T. Breuer (ptb@ot.uc3m.es) 2003 4935fe098SMike Snitzer * 5935fe098SMike Snitzer * bitmap_create - sets up the bitmap structure 6935fe098SMike Snitzer * bitmap_destroy - destroys the bitmap structure 7935fe098SMike Snitzer * 8935fe098SMike Snitzer * additions, Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.: 9935fe098SMike Snitzer * - added disk storage for bitmap 10935fe098SMike Snitzer * - changes to allow various bitmap chunk sizes 11935fe098SMike Snitzer */ 12935fe098SMike Snitzer 13935fe098SMike Snitzer /* 14935fe098SMike Snitzer * Still to do: 15935fe098SMike Snitzer * 16935fe098SMike Snitzer * flush after percent set rather than just time based. (maybe both). 17935fe098SMike Snitzer */ 18935fe098SMike Snitzer 19935fe098SMike Snitzer #include <linux/blkdev.h> 20935fe098SMike Snitzer #include <linux/module.h> 21935fe098SMike Snitzer #include <linux/errno.h> 22935fe098SMike Snitzer #include <linux/slab.h> 23935fe098SMike Snitzer #include <linux/init.h> 24935fe098SMike Snitzer #include <linux/timer.h> 25935fe098SMike Snitzer #include <linux/sched.h> 26935fe098SMike Snitzer #include <linux/list.h> 27935fe098SMike Snitzer #include <linux/file.h> 28935fe098SMike Snitzer #include <linux/mount.h> 29935fe098SMike Snitzer #include <linux/buffer_head.h> 30935fe098SMike Snitzer #include <linux/seq_file.h> 31935fe098SMike Snitzer #include <trace/events/block.h> 32935fe098SMike Snitzer #include "md.h" 33935fe098SMike Snitzer #include "md-bitmap.h" 34935fe098SMike Snitzer 35935fe098SMike Snitzer static inline char *bmname(struct bitmap *bitmap) 36935fe098SMike Snitzer { 37935fe098SMike Snitzer return bitmap->mddev ? mdname(bitmap->mddev) : "mdX"; 38935fe098SMike Snitzer } 39935fe098SMike Snitzer 40935fe098SMike Snitzer /* 41935fe098SMike Snitzer * check a page and, if necessary, allocate it (or hijack it if the alloc fails) 42935fe098SMike Snitzer * 43935fe098SMike Snitzer * 1) check to see if this page is allocated, if it's not then try to alloc 44935fe098SMike Snitzer * 2) if the alloc fails, set the page's hijacked flag so we'll use the 45935fe098SMike Snitzer * page pointer directly as a counter 46935fe098SMike Snitzer * 47935fe098SMike Snitzer * if we find our page, we increment the page's refcount so that it stays 48935fe098SMike Snitzer * allocated while we're using it 49935fe098SMike Snitzer */ 50e64e4018SAndy Shevchenko static int md_bitmap_checkpage(struct bitmap_counts *bitmap, 51935fe098SMike Snitzer unsigned long page, int create, int no_hijack) 52935fe098SMike Snitzer __releases(bitmap->lock) 53935fe098SMike Snitzer __acquires(bitmap->lock) 54935fe098SMike Snitzer { 55935fe098SMike Snitzer unsigned char *mappage; 56935fe098SMike Snitzer 57301867b1SLi Nan WARN_ON_ONCE(page >= bitmap->pages); 58935fe098SMike Snitzer if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */ 59935fe098SMike Snitzer return 0; 60935fe098SMike Snitzer 61935fe098SMike Snitzer if (bitmap->bp[page].map) /* page is already allocated, just return */ 62935fe098SMike Snitzer return 0; 63935fe098SMike Snitzer 64935fe098SMike Snitzer if (!create) 65935fe098SMike Snitzer return -ENOENT; 66935fe098SMike Snitzer 67935fe098SMike Snitzer /* this page has not been allocated yet */ 68935fe098SMike Snitzer 69935fe098SMike Snitzer spin_unlock_irq(&bitmap->lock); 70935fe098SMike Snitzer /* It is possible that this is being called inside a 71935fe098SMike Snitzer * prepare_to_wait/finish_wait loop from raid5c:make_request(). 72935fe098SMike Snitzer * In general it is not permitted to sleep in that context as it 73935fe098SMike Snitzer * can cause the loop to spin freely. 74935fe098SMike Snitzer * That doesn't apply here as we can only reach this point 75935fe098SMike Snitzer * once with any loop. 76935fe098SMike Snitzer * When this function completes, either bp[page].map or 77935fe098SMike Snitzer * bp[page].hijacked. In either case, this function will 78935fe098SMike Snitzer * abort before getting to this point again. So there is 79935fe098SMike Snitzer * no risk of a free-spin, and so it is safe to assert 80935fe098SMike Snitzer * that sleeping here is allowed. 81935fe098SMike Snitzer */ 82935fe098SMike Snitzer sched_annotate_sleep(); 83935fe098SMike Snitzer mappage = kzalloc(PAGE_SIZE, GFP_NOIO); 84935fe098SMike Snitzer spin_lock_irq(&bitmap->lock); 85935fe098SMike Snitzer 86935fe098SMike Snitzer if (mappage == NULL) { 87935fe098SMike Snitzer pr_debug("md/bitmap: map page allocation failed, hijacking\n"); 88935fe098SMike Snitzer /* We don't support hijack for cluster raid */ 89935fe098SMike Snitzer if (no_hijack) 90935fe098SMike Snitzer return -ENOMEM; 91935fe098SMike Snitzer /* failed - set the hijacked flag so that we can use the 92935fe098SMike Snitzer * pointer as a counter */ 93935fe098SMike Snitzer if (!bitmap->bp[page].map) 94935fe098SMike Snitzer bitmap->bp[page].hijacked = 1; 95935fe098SMike Snitzer } else if (bitmap->bp[page].map || 96935fe098SMike Snitzer bitmap->bp[page].hijacked) { 97935fe098SMike Snitzer /* somebody beat us to getting the page */ 98935fe098SMike Snitzer kfree(mappage); 99935fe098SMike Snitzer } else { 100935fe098SMike Snitzer 101935fe098SMike Snitzer /* no page was in place and we have one, so install it */ 102935fe098SMike Snitzer 103935fe098SMike Snitzer bitmap->bp[page].map = mappage; 104935fe098SMike Snitzer bitmap->missing_pages--; 105935fe098SMike Snitzer } 106935fe098SMike Snitzer return 0; 107935fe098SMike Snitzer } 108935fe098SMike Snitzer 109935fe098SMike Snitzer /* if page is completely empty, put it back on the free list, or dealloc it */ 110935fe098SMike Snitzer /* if page was hijacked, unmark the flag so it might get alloced next time */ 111935fe098SMike Snitzer /* Note: lock should be held when calling this */ 112e64e4018SAndy Shevchenko static void md_bitmap_checkfree(struct bitmap_counts *bitmap, unsigned long page) 113935fe098SMike Snitzer { 114935fe098SMike Snitzer char *ptr; 115935fe098SMike Snitzer 116935fe098SMike Snitzer if (bitmap->bp[page].count) /* page is still busy */ 117935fe098SMike Snitzer return; 118935fe098SMike Snitzer 119935fe098SMike Snitzer /* page is no longer in use, it can be released */ 120935fe098SMike Snitzer 121935fe098SMike Snitzer if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */ 122935fe098SMike Snitzer bitmap->bp[page].hijacked = 0; 123935fe098SMike Snitzer bitmap->bp[page].map = NULL; 124935fe098SMike Snitzer } else { 125935fe098SMike Snitzer /* normal case, free the page */ 126935fe098SMike Snitzer ptr = bitmap->bp[page].map; 127935fe098SMike Snitzer bitmap->bp[page].map = NULL; 128935fe098SMike Snitzer bitmap->missing_pages++; 129935fe098SMike Snitzer kfree(ptr); 130935fe098SMike Snitzer } 131935fe098SMike Snitzer } 132935fe098SMike Snitzer 133935fe098SMike Snitzer /* 134935fe098SMike Snitzer * bitmap file handling - read and write the bitmap file and its superblock 135935fe098SMike Snitzer */ 136935fe098SMike Snitzer 137935fe098SMike Snitzer /* 138935fe098SMike Snitzer * basic page I/O operations 139935fe098SMike Snitzer */ 140935fe098SMike Snitzer 141935fe098SMike Snitzer /* IO operations when bitmap is stored near all superblocks */ 142935fe098SMike Snitzer static int read_sb_page(struct mddev *mddev, loff_t offset, 143935fe098SMike Snitzer struct page *page, 144935fe098SMike Snitzer unsigned long index, int size) 145935fe098SMike Snitzer { 146935fe098SMike Snitzer /* choose a good rdev and read the page from there */ 147935fe098SMike Snitzer 148935fe098SMike Snitzer struct md_rdev *rdev; 149935fe098SMike Snitzer sector_t target; 150935fe098SMike Snitzer 151935fe098SMike Snitzer rdev_for_each(rdev, mddev) { 152935fe098SMike Snitzer if (! test_bit(In_sync, &rdev->flags) 153935fe098SMike Snitzer || test_bit(Faulty, &rdev->flags) 154935fe098SMike Snitzer || test_bit(Bitmap_sync, &rdev->flags)) 155935fe098SMike Snitzer continue; 156935fe098SMike Snitzer 157935fe098SMike Snitzer target = offset + index * (PAGE_SIZE/512); 158935fe098SMike Snitzer 159935fe098SMike Snitzer if (sync_page_io(rdev, target, 160935fe098SMike Snitzer roundup(size, bdev_logical_block_size(rdev->bdev)), 1614ce4c73fSBart Van Assche page, REQ_OP_READ, true)) { 162935fe098SMike Snitzer page->index = index; 163935fe098SMike Snitzer return 0; 164935fe098SMike Snitzer } 165935fe098SMike Snitzer } 166935fe098SMike Snitzer return -EIO; 167935fe098SMike Snitzer } 168935fe098SMike Snitzer 169935fe098SMike Snitzer static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev) 170935fe098SMike Snitzer { 171935fe098SMike Snitzer /* Iterate the disks of an mddev, using rcu to protect access to the 172935fe098SMike Snitzer * linked list, and raising the refcount of devices we return to ensure 173935fe098SMike Snitzer * they don't disappear while in use. 174935fe098SMike Snitzer * As devices are only added or removed when raid_disk is < 0 and 175935fe098SMike Snitzer * nr_pending is 0 and In_sync is clear, the entries we return will 176935fe098SMike Snitzer * still be in the same position on the list when we re-enter 177935fe098SMike Snitzer * list_for_each_entry_continue_rcu. 178935fe098SMike Snitzer * 179935fe098SMike Snitzer * Note that if entered with 'rdev == NULL' to start at the 180935fe098SMike Snitzer * beginning, we temporarily assign 'rdev' to an address which 181935fe098SMike Snitzer * isn't really an rdev, but which can be used by 182935fe098SMike Snitzer * list_for_each_entry_continue_rcu() to find the first entry. 183935fe098SMike Snitzer */ 184935fe098SMike Snitzer rcu_read_lock(); 185935fe098SMike Snitzer if (rdev == NULL) 186935fe098SMike Snitzer /* start at the beginning */ 187935fe098SMike Snitzer rdev = list_entry(&mddev->disks, struct md_rdev, same_set); 188935fe098SMike Snitzer else { 189935fe098SMike Snitzer /* release the previous rdev and start from there. */ 190935fe098SMike Snitzer rdev_dec_pending(rdev, mddev); 191935fe098SMike Snitzer } 192935fe098SMike Snitzer list_for_each_entry_continue_rcu(rdev, &mddev->disks, same_set) { 193935fe098SMike Snitzer if (rdev->raid_disk >= 0 && 194935fe098SMike Snitzer !test_bit(Faulty, &rdev->flags)) { 195935fe098SMike Snitzer /* this is a usable devices */ 196935fe098SMike Snitzer atomic_inc(&rdev->nr_pending); 197935fe098SMike Snitzer rcu_read_unlock(); 198935fe098SMike Snitzer return rdev; 199935fe098SMike Snitzer } 200935fe098SMike Snitzer } 201935fe098SMike Snitzer rcu_read_unlock(); 202935fe098SMike Snitzer return NULL; 203935fe098SMike Snitzer } 204935fe098SMike Snitzer 2058745faa9SJon Derrick static unsigned int optimal_io_size(struct block_device *bdev, 2068745faa9SJon Derrick unsigned int last_page_size, 2078745faa9SJon Derrick unsigned int io_size) 2088745faa9SJon Derrick { 2098745faa9SJon Derrick if (bdev_io_opt(bdev) > bdev_logical_block_size(bdev)) 2108745faa9SJon Derrick return roundup(last_page_size, bdev_io_opt(bdev)); 2118745faa9SJon Derrick return io_size; 2128745faa9SJon Derrick } 2138745faa9SJon Derrick 2148745faa9SJon Derrick static unsigned int bitmap_io_size(unsigned int io_size, unsigned int opt_size, 215b1211978SJonathan Derrick loff_t start, loff_t boundary) 2168745faa9SJon Derrick { 2178745faa9SJon Derrick if (io_size != opt_size && 2188745faa9SJon Derrick start + opt_size / SECTOR_SIZE <= boundary) 2198745faa9SJon Derrick return opt_size; 2208745faa9SJon Derrick if (start + io_size / SECTOR_SIZE <= boundary) 2218745faa9SJon Derrick return io_size; 2228745faa9SJon Derrick 2238745faa9SJon Derrick /* Overflows boundary */ 2248745faa9SJon Derrick return 0; 2258745faa9SJon Derrick } 2268745faa9SJon Derrick 227328e17d8SJon Derrick static int __write_sb_page(struct md_rdev *rdev, struct bitmap *bitmap, 228328e17d8SJon Derrick struct page *page) 229935fe098SMike Snitzer { 230935fe098SMike Snitzer struct block_device *bdev; 231935fe098SMike Snitzer struct mddev *mddev = bitmap->mddev; 232935fe098SMike Snitzer struct bitmap_storage *store = &bitmap->storage; 233b1211978SJonathan Derrick loff_t sboff, offset = mddev->bitmap_info.offset; 234b1211978SJonathan Derrick sector_t ps, doff; 23510172f20SJon Derrick unsigned int size = PAGE_SIZE; 2368745faa9SJon Derrick unsigned int opt_size = PAGE_SIZE; 237935fe098SMike Snitzer 238935fe098SMike Snitzer bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev; 239935fe098SMike Snitzer if (page->index == store->file_pages - 1) { 24010172f20SJon Derrick unsigned int last_page_size = store->bytes & (PAGE_SIZE - 1); 241328e17d8SJon Derrick 242935fe098SMike Snitzer if (last_page_size == 0) 243935fe098SMike Snitzer last_page_size = PAGE_SIZE; 2448745faa9SJon Derrick size = roundup(last_page_size, bdev_logical_block_size(bdev)); 2458745faa9SJon Derrick opt_size = optimal_io_size(bdev, last_page_size, size); 246935fe098SMike Snitzer } 247328e17d8SJon Derrick 24810172f20SJon Derrick ps = page->index * PAGE_SIZE / SECTOR_SIZE; 24910172f20SJon Derrick sboff = rdev->sb_start + offset; 25010172f20SJon Derrick doff = rdev->data_offset; 25110172f20SJon Derrick 252328e17d8SJon Derrick /* Just make sure we aren't corrupting data or metadata */ 253935fe098SMike Snitzer if (mddev->external) { 254935fe098SMike Snitzer /* Bitmap could be anywhere. */ 25510172f20SJon Derrick if (sboff + ps > doff && 25610172f20SJon Derrick sboff < (doff + mddev->dev_sectors + PAGE_SIZE / SECTOR_SIZE)) 257328e17d8SJon Derrick return -EINVAL; 258935fe098SMike Snitzer } else if (offset < 0) { 259935fe098SMike Snitzer /* DATA BITMAP METADATA */ 2608745faa9SJon Derrick size = bitmap_io_size(size, opt_size, offset + ps, 0); 2618745faa9SJon Derrick if (size == 0) 262935fe098SMike Snitzer /* bitmap runs in to metadata */ 263328e17d8SJon Derrick return -EINVAL; 264328e17d8SJon Derrick 26510172f20SJon Derrick if (doff + mddev->dev_sectors > sboff) 266935fe098SMike Snitzer /* data runs in to bitmap */ 267328e17d8SJon Derrick return -EINVAL; 268935fe098SMike Snitzer } else if (rdev->sb_start < rdev->data_offset) { 269935fe098SMike Snitzer /* METADATA BITMAP DATA */ 2708745faa9SJon Derrick size = bitmap_io_size(size, opt_size, sboff + ps, doff); 2718745faa9SJon Derrick if (size == 0) 272935fe098SMike Snitzer /* bitmap runs in to data */ 273328e17d8SJon Derrick return -EINVAL; 274935fe098SMike Snitzer } else { 275935fe098SMike Snitzer /* DATA METADATA BITMAP - no problems */ 276935fe098SMike Snitzer } 277328e17d8SJon Derrick 27810172f20SJon Derrick md_super_write(mddev, rdev, sboff + ps, (int) size, page); 279328e17d8SJon Derrick return 0; 280935fe098SMike Snitzer } 281935fe098SMike Snitzer 282328e17d8SJon Derrick static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait) 283328e17d8SJon Derrick { 284328e17d8SJon Derrick struct md_rdev *rdev; 285328e17d8SJon Derrick struct mddev *mddev = bitmap->mddev; 286328e17d8SJon Derrick int ret; 287935fe098SMike Snitzer 288328e17d8SJon Derrick do { 289328e17d8SJon Derrick rdev = NULL; 290328e17d8SJon Derrick while ((rdev = next_active_rdev(rdev, mddev)) != NULL) { 291328e17d8SJon Derrick ret = __write_sb_page(rdev, bitmap, page); 292328e17d8SJon Derrick if (ret) 293328e17d8SJon Derrick return ret; 294328e17d8SJon Derrick } 295328e17d8SJon Derrick } while (wait && md_super_wait(mddev) < 0); 296328e17d8SJon Derrick 297328e17d8SJon Derrick return 0; 298935fe098SMike Snitzer } 299935fe098SMike Snitzer 300e64e4018SAndy Shevchenko static void md_bitmap_file_kick(struct bitmap *bitmap); 301935fe098SMike Snitzer /* 302935fe098SMike Snitzer * write out a page to a file 303935fe098SMike Snitzer */ 304935fe098SMike Snitzer static void write_page(struct bitmap *bitmap, struct page *page, int wait) 305935fe098SMike Snitzer { 306935fe098SMike Snitzer struct buffer_head *bh; 307935fe098SMike Snitzer 308935fe098SMike Snitzer if (bitmap->storage.file == NULL) { 309935fe098SMike Snitzer switch (write_sb_page(bitmap, page, wait)) { 310935fe098SMike Snitzer case -EINVAL: 311935fe098SMike Snitzer set_bit(BITMAP_WRITE_ERROR, &bitmap->flags); 312935fe098SMike Snitzer } 313935fe098SMike Snitzer } else { 314935fe098SMike Snitzer 315935fe098SMike Snitzer bh = page_buffers(page); 316935fe098SMike Snitzer 317935fe098SMike Snitzer while (bh && bh->b_blocknr) { 318935fe098SMike Snitzer atomic_inc(&bitmap->pending_writes); 319935fe098SMike Snitzer set_buffer_locked(bh); 320935fe098SMike Snitzer set_buffer_mapped(bh); 3211420c4a5SBart Van Assche submit_bh(REQ_OP_WRITE | REQ_SYNC, bh); 322935fe098SMike Snitzer bh = bh->b_this_page; 323935fe098SMike Snitzer } 324935fe098SMike Snitzer 325935fe098SMike Snitzer if (wait) 326935fe098SMike Snitzer wait_event(bitmap->write_wait, 327935fe098SMike Snitzer atomic_read(&bitmap->pending_writes)==0); 328935fe098SMike Snitzer } 329935fe098SMike Snitzer if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) 330e64e4018SAndy Shevchenko md_bitmap_file_kick(bitmap); 331935fe098SMike Snitzer } 332935fe098SMike Snitzer 333935fe098SMike Snitzer static void end_bitmap_write(struct buffer_head *bh, int uptodate) 334935fe098SMike Snitzer { 335935fe098SMike Snitzer struct bitmap *bitmap = bh->b_private; 336935fe098SMike Snitzer 337935fe098SMike Snitzer if (!uptodate) 338935fe098SMike Snitzer set_bit(BITMAP_WRITE_ERROR, &bitmap->flags); 339935fe098SMike Snitzer if (atomic_dec_and_test(&bitmap->pending_writes)) 340935fe098SMike Snitzer wake_up(&bitmap->write_wait); 341935fe098SMike Snitzer } 342935fe098SMike Snitzer 343935fe098SMike Snitzer static void free_buffers(struct page *page) 344935fe098SMike Snitzer { 345935fe098SMike Snitzer struct buffer_head *bh; 346935fe098SMike Snitzer 347935fe098SMike Snitzer if (!PagePrivate(page)) 348935fe098SMike Snitzer return; 349935fe098SMike Snitzer 350935fe098SMike Snitzer bh = page_buffers(page); 351935fe098SMike Snitzer while (bh) { 352935fe098SMike Snitzer struct buffer_head *next = bh->b_this_page; 353935fe098SMike Snitzer free_buffer_head(bh); 354935fe098SMike Snitzer bh = next; 355935fe098SMike Snitzer } 356db2c1d86SGuoqing Jiang detach_page_private(page); 357935fe098SMike Snitzer put_page(page); 358935fe098SMike Snitzer } 359935fe098SMike Snitzer 360935fe098SMike Snitzer /* read a page from a file. 361935fe098SMike Snitzer * We both read the page, and attach buffers to the page to record the 362935fe098SMike Snitzer * address of each block (using bmap). These addresses will be used 363935fe098SMike Snitzer * to write the block later, completely bypassing the filesystem. 364935fe098SMike Snitzer * This usage is similar to how swap files are handled, and allows us 365935fe098SMike Snitzer * to write to a file with no concerns of memory allocation failing. 366935fe098SMike Snitzer */ 367935fe098SMike Snitzer static int read_page(struct file *file, unsigned long index, 368935fe098SMike Snitzer struct bitmap *bitmap, 369935fe098SMike Snitzer unsigned long count, 370935fe098SMike Snitzer struct page *page) 371935fe098SMike Snitzer { 372935fe098SMike Snitzer int ret = 0; 373935fe098SMike Snitzer struct inode *inode = file_inode(file); 374935fe098SMike Snitzer struct buffer_head *bh; 37530460e1eSCarlos Maiolino sector_t block, blk_cur; 376313b825fSXianting Tian unsigned long blocksize = i_blocksize(inode); 377935fe098SMike Snitzer 378935fe098SMike Snitzer pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE, 379935fe098SMike Snitzer (unsigned long long)index << PAGE_SHIFT); 380935fe098SMike Snitzer 381313b825fSXianting Tian bh = alloc_page_buffers(page, blocksize, false); 382935fe098SMike Snitzer if (!bh) { 383935fe098SMike Snitzer ret = -ENOMEM; 384935fe098SMike Snitzer goto out; 385935fe098SMike Snitzer } 386db2c1d86SGuoqing Jiang attach_page_private(page, bh); 38730460e1eSCarlos Maiolino blk_cur = index << (PAGE_SHIFT - inode->i_blkbits); 388935fe098SMike Snitzer while (bh) { 38930460e1eSCarlos Maiolino block = blk_cur; 39030460e1eSCarlos Maiolino 391935fe098SMike Snitzer if (count == 0) 392935fe098SMike Snitzer bh->b_blocknr = 0; 393935fe098SMike Snitzer else { 39430460e1eSCarlos Maiolino ret = bmap(inode, &block); 39530460e1eSCarlos Maiolino if (ret || !block) { 396935fe098SMike Snitzer ret = -EINVAL; 39730460e1eSCarlos Maiolino bh->b_blocknr = 0; 398935fe098SMike Snitzer goto out; 399935fe098SMike Snitzer } 40030460e1eSCarlos Maiolino 40130460e1eSCarlos Maiolino bh->b_blocknr = block; 402935fe098SMike Snitzer bh->b_bdev = inode->i_sb->s_bdev; 403313b825fSXianting Tian if (count < blocksize) 404935fe098SMike Snitzer count = 0; 405935fe098SMike Snitzer else 406313b825fSXianting Tian count -= blocksize; 407935fe098SMike Snitzer 408935fe098SMike Snitzer bh->b_end_io = end_bitmap_write; 409935fe098SMike Snitzer bh->b_private = bitmap; 410935fe098SMike Snitzer atomic_inc(&bitmap->pending_writes); 411935fe098SMike Snitzer set_buffer_locked(bh); 412935fe098SMike Snitzer set_buffer_mapped(bh); 4131420c4a5SBart Van Assche submit_bh(REQ_OP_READ, bh); 414935fe098SMike Snitzer } 41530460e1eSCarlos Maiolino blk_cur++; 416935fe098SMike Snitzer bh = bh->b_this_page; 417935fe098SMike Snitzer } 418935fe098SMike Snitzer page->index = index; 419935fe098SMike Snitzer 420935fe098SMike Snitzer wait_event(bitmap->write_wait, 421935fe098SMike Snitzer atomic_read(&bitmap->pending_writes)==0); 422935fe098SMike Snitzer if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) 423935fe098SMike Snitzer ret = -EIO; 424935fe098SMike Snitzer out: 425935fe098SMike Snitzer if (ret) 426935fe098SMike Snitzer pr_err("md: bitmap read error: (%dB @ %llu): %d\n", 427935fe098SMike Snitzer (int)PAGE_SIZE, 428935fe098SMike Snitzer (unsigned long long)index << PAGE_SHIFT, 429935fe098SMike Snitzer ret); 430935fe098SMike Snitzer return ret; 431935fe098SMike Snitzer } 432935fe098SMike Snitzer 433935fe098SMike Snitzer /* 434935fe098SMike Snitzer * bitmap file superblock operations 435935fe098SMike Snitzer */ 436935fe098SMike Snitzer 437935fe098SMike Snitzer /* 438e64e4018SAndy Shevchenko * md_bitmap_wait_writes() should be called before writing any bitmap 439935fe098SMike Snitzer * blocks, to ensure previous writes, particularly from 440e64e4018SAndy Shevchenko * md_bitmap_daemon_work(), have completed. 441935fe098SMike Snitzer */ 442e64e4018SAndy Shevchenko static void md_bitmap_wait_writes(struct bitmap *bitmap) 443935fe098SMike Snitzer { 444935fe098SMike Snitzer if (bitmap->storage.file) 445935fe098SMike Snitzer wait_event(bitmap->write_wait, 446935fe098SMike Snitzer atomic_read(&bitmap->pending_writes)==0); 447935fe098SMike Snitzer else 448935fe098SMike Snitzer /* Note that we ignore the return value. The writes 449935fe098SMike Snitzer * might have failed, but that would just mean that 450935fe098SMike Snitzer * some bits which should be cleared haven't been, 451935fe098SMike Snitzer * which is safe. The relevant bitmap blocks will 452935fe098SMike Snitzer * probably get written again, but there is no great 453935fe098SMike Snitzer * loss if they aren't. 454935fe098SMike Snitzer */ 455935fe098SMike Snitzer md_super_wait(bitmap->mddev); 456935fe098SMike Snitzer } 457935fe098SMike Snitzer 458935fe098SMike Snitzer 459935fe098SMike Snitzer /* update the event counter and sync the superblock to disk */ 460e64e4018SAndy Shevchenko void md_bitmap_update_sb(struct bitmap *bitmap) 461935fe098SMike Snitzer { 462935fe098SMike Snitzer bitmap_super_t *sb; 463935fe098SMike Snitzer 464935fe098SMike Snitzer if (!bitmap || !bitmap->mddev) /* no bitmap for this array */ 465935fe098SMike Snitzer return; 466935fe098SMike Snitzer if (bitmap->mddev->bitmap_info.external) 467935fe098SMike Snitzer return; 468935fe098SMike Snitzer if (!bitmap->storage.sb_page) /* no superblock */ 469935fe098SMike Snitzer return; 470935fe098SMike Snitzer sb = kmap_atomic(bitmap->storage.sb_page); 471935fe098SMike Snitzer sb->events = cpu_to_le64(bitmap->mddev->events); 472935fe098SMike Snitzer if (bitmap->mddev->events < bitmap->events_cleared) 473935fe098SMike Snitzer /* rocking back to read-only */ 474935fe098SMike Snitzer bitmap->events_cleared = bitmap->mddev->events; 475935fe098SMike Snitzer sb->events_cleared = cpu_to_le64(bitmap->events_cleared); 47697f0eb9fSHou Tao /* 47797f0eb9fSHou Tao * clear BITMAP_WRITE_ERROR bit to protect against the case that 47897f0eb9fSHou Tao * a bitmap write error occurred but the later writes succeeded. 47997f0eb9fSHou Tao */ 48097f0eb9fSHou Tao sb->state = cpu_to_le32(bitmap->flags & ~BIT(BITMAP_WRITE_ERROR)); 481935fe098SMike Snitzer /* Just in case these have been changed via sysfs: */ 482935fe098SMike Snitzer sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ); 483935fe098SMike Snitzer sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind); 484935fe098SMike Snitzer /* This might have been changed by a reshape */ 485935fe098SMike Snitzer sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); 486935fe098SMike Snitzer sb->chunksize = cpu_to_le32(bitmap->mddev->bitmap_info.chunksize); 487935fe098SMike Snitzer sb->nodes = cpu_to_le32(bitmap->mddev->bitmap_info.nodes); 488935fe098SMike Snitzer sb->sectors_reserved = cpu_to_le32(bitmap->mddev-> 489935fe098SMike Snitzer bitmap_info.space); 490935fe098SMike Snitzer kunmap_atomic(sb); 491935fe098SMike Snitzer write_page(bitmap, bitmap->storage.sb_page, 1); 492935fe098SMike Snitzer } 493e64e4018SAndy Shevchenko EXPORT_SYMBOL(md_bitmap_update_sb); 494935fe098SMike Snitzer 495935fe098SMike Snitzer /* print out the bitmap file superblock */ 496e64e4018SAndy Shevchenko void md_bitmap_print_sb(struct bitmap *bitmap) 497935fe098SMike Snitzer { 498935fe098SMike Snitzer bitmap_super_t *sb; 499935fe098SMike Snitzer 500935fe098SMike Snitzer if (!bitmap || !bitmap->storage.sb_page) 501935fe098SMike Snitzer return; 502935fe098SMike Snitzer sb = kmap_atomic(bitmap->storage.sb_page); 503935fe098SMike Snitzer pr_debug("%s: bitmap file superblock:\n", bmname(bitmap)); 504935fe098SMike Snitzer pr_debug(" magic: %08x\n", le32_to_cpu(sb->magic)); 50545552111SFlorian-Ewald Mueller pr_debug(" version: %u\n", le32_to_cpu(sb->version)); 506935fe098SMike Snitzer pr_debug(" uuid: %08x.%08x.%08x.%08x\n", 507c35403f8SChristoph Hellwig le32_to_cpu(*(__le32 *)(sb->uuid+0)), 508c35403f8SChristoph Hellwig le32_to_cpu(*(__le32 *)(sb->uuid+4)), 509c35403f8SChristoph Hellwig le32_to_cpu(*(__le32 *)(sb->uuid+8)), 510c35403f8SChristoph Hellwig le32_to_cpu(*(__le32 *)(sb->uuid+12))); 511935fe098SMike Snitzer pr_debug(" events: %llu\n", 512935fe098SMike Snitzer (unsigned long long) le64_to_cpu(sb->events)); 513935fe098SMike Snitzer pr_debug("events cleared: %llu\n", 514935fe098SMike Snitzer (unsigned long long) le64_to_cpu(sb->events_cleared)); 515935fe098SMike Snitzer pr_debug(" state: %08x\n", le32_to_cpu(sb->state)); 51645552111SFlorian-Ewald Mueller pr_debug(" chunksize: %u B\n", le32_to_cpu(sb->chunksize)); 51745552111SFlorian-Ewald Mueller pr_debug(" daemon sleep: %us\n", le32_to_cpu(sb->daemon_sleep)); 518935fe098SMike Snitzer pr_debug(" sync size: %llu KB\n", 519935fe098SMike Snitzer (unsigned long long)le64_to_cpu(sb->sync_size)/2); 52045552111SFlorian-Ewald Mueller pr_debug("max write behind: %u\n", le32_to_cpu(sb->write_behind)); 521935fe098SMike Snitzer kunmap_atomic(sb); 522935fe098SMike Snitzer } 523935fe098SMike Snitzer 524935fe098SMike Snitzer /* 525935fe098SMike Snitzer * bitmap_new_disk_sb 526935fe098SMike Snitzer * @bitmap 527935fe098SMike Snitzer * 528935fe098SMike Snitzer * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb 529935fe098SMike Snitzer * reads and verifies the on-disk bitmap superblock and populates bitmap_info. 530935fe098SMike Snitzer * This function verifies 'bitmap_info' and populates the on-disk bitmap 531935fe098SMike Snitzer * structure, which is to be written to disk. 532935fe098SMike Snitzer * 533935fe098SMike Snitzer * Returns: 0 on success, -Exxx on error 534935fe098SMike Snitzer */ 535e64e4018SAndy Shevchenko static int md_bitmap_new_disk_sb(struct bitmap *bitmap) 536935fe098SMike Snitzer { 537935fe098SMike Snitzer bitmap_super_t *sb; 538935fe098SMike Snitzer unsigned long chunksize, daemon_sleep, write_behind; 539935fe098SMike Snitzer 540935fe098SMike Snitzer bitmap->storage.sb_page = alloc_page(GFP_KERNEL | __GFP_ZERO); 541935fe098SMike Snitzer if (bitmap->storage.sb_page == NULL) 542935fe098SMike Snitzer return -ENOMEM; 543935fe098SMike Snitzer bitmap->storage.sb_page->index = 0; 544935fe098SMike Snitzer 545935fe098SMike Snitzer sb = kmap_atomic(bitmap->storage.sb_page); 546935fe098SMike Snitzer 547935fe098SMike Snitzer sb->magic = cpu_to_le32(BITMAP_MAGIC); 548935fe098SMike Snitzer sb->version = cpu_to_le32(BITMAP_MAJOR_HI); 549935fe098SMike Snitzer 550935fe098SMike Snitzer chunksize = bitmap->mddev->bitmap_info.chunksize; 551935fe098SMike Snitzer BUG_ON(!chunksize); 552935fe098SMike Snitzer if (!is_power_of_2(chunksize)) { 553935fe098SMike Snitzer kunmap_atomic(sb); 554935fe098SMike Snitzer pr_warn("bitmap chunksize not a power of 2\n"); 555935fe098SMike Snitzer return -EINVAL; 556935fe098SMike Snitzer } 557935fe098SMike Snitzer sb->chunksize = cpu_to_le32(chunksize); 558935fe098SMike Snitzer 559935fe098SMike Snitzer daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep; 560935fe098SMike Snitzer if (!daemon_sleep || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) { 561935fe098SMike Snitzer pr_debug("Choosing daemon_sleep default (5 sec)\n"); 562935fe098SMike Snitzer daemon_sleep = 5 * HZ; 563935fe098SMike Snitzer } 564935fe098SMike Snitzer sb->daemon_sleep = cpu_to_le32(daemon_sleep); 565935fe098SMike Snitzer bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep; 566935fe098SMike Snitzer 567935fe098SMike Snitzer /* 568935fe098SMike Snitzer * FIXME: write_behind for RAID1. If not specified, what 569935fe098SMike Snitzer * is a good choice? We choose COUNTER_MAX / 2 arbitrarily. 570935fe098SMike Snitzer */ 571935fe098SMike Snitzer write_behind = bitmap->mddev->bitmap_info.max_write_behind; 572935fe098SMike Snitzer if (write_behind > COUNTER_MAX) 573935fe098SMike Snitzer write_behind = COUNTER_MAX / 2; 574935fe098SMike Snitzer sb->write_behind = cpu_to_le32(write_behind); 575935fe098SMike Snitzer bitmap->mddev->bitmap_info.max_write_behind = write_behind; 576935fe098SMike Snitzer 577935fe098SMike Snitzer /* keep the array size field of the bitmap superblock up to date */ 578935fe098SMike Snitzer sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); 579935fe098SMike Snitzer 580935fe098SMike Snitzer memcpy(sb->uuid, bitmap->mddev->uuid, 16); 581935fe098SMike Snitzer 582935fe098SMike Snitzer set_bit(BITMAP_STALE, &bitmap->flags); 583935fe098SMike Snitzer sb->state = cpu_to_le32(bitmap->flags); 584935fe098SMike Snitzer bitmap->events_cleared = bitmap->mddev->events; 585935fe098SMike Snitzer sb->events_cleared = cpu_to_le64(bitmap->mddev->events); 586935fe098SMike Snitzer bitmap->mddev->bitmap_info.nodes = 0; 587935fe098SMike Snitzer 588935fe098SMike Snitzer kunmap_atomic(sb); 589935fe098SMike Snitzer 590935fe098SMike Snitzer return 0; 591935fe098SMike Snitzer } 592935fe098SMike Snitzer 593935fe098SMike Snitzer /* read the superblock from the bitmap file and initialize some bitmap fields */ 594e64e4018SAndy Shevchenko static int md_bitmap_read_sb(struct bitmap *bitmap) 595935fe098SMike Snitzer { 596935fe098SMike Snitzer char *reason = NULL; 597935fe098SMike Snitzer bitmap_super_t *sb; 598935fe098SMike Snitzer unsigned long chunksize, daemon_sleep, write_behind; 599935fe098SMike Snitzer unsigned long long events; 600935fe098SMike Snitzer int nodes = 0; 601935fe098SMike Snitzer unsigned long sectors_reserved = 0; 602935fe098SMike Snitzer int err = -EINVAL; 603935fe098SMike Snitzer struct page *sb_page; 604935fe098SMike Snitzer loff_t offset = bitmap->mddev->bitmap_info.offset; 605935fe098SMike Snitzer 606935fe098SMike Snitzer if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) { 607935fe098SMike Snitzer chunksize = 128 * 1024 * 1024; 608935fe098SMike Snitzer daemon_sleep = 5 * HZ; 609935fe098SMike Snitzer write_behind = 0; 610935fe098SMike Snitzer set_bit(BITMAP_STALE, &bitmap->flags); 611935fe098SMike Snitzer err = 0; 612935fe098SMike Snitzer goto out_no_sb; 613935fe098SMike Snitzer } 614935fe098SMike Snitzer /* page 0 is the superblock, read it... */ 615935fe098SMike Snitzer sb_page = alloc_page(GFP_KERNEL); 616935fe098SMike Snitzer if (!sb_page) 617935fe098SMike Snitzer return -ENOMEM; 618935fe098SMike Snitzer bitmap->storage.sb_page = sb_page; 619935fe098SMike Snitzer 620935fe098SMike Snitzer re_read: 621935fe098SMike Snitzer /* If cluster_slot is set, the cluster is setup */ 622935fe098SMike Snitzer if (bitmap->cluster_slot >= 0) { 623935fe098SMike Snitzer sector_t bm_blocks = bitmap->mddev->resync_max_sectors; 624935fe098SMike Snitzer 625a913096dSZhao Heming bm_blocks = DIV_ROUND_UP_SECTOR_T(bm_blocks, 626a913096dSZhao Heming (bitmap->mddev->bitmap_info.chunksize >> 9)); 627935fe098SMike Snitzer /* bits to bytes */ 628935fe098SMike Snitzer bm_blocks = ((bm_blocks+7) >> 3) + sizeof(bitmap_super_t); 629935fe098SMike Snitzer /* to 4k blocks */ 630935fe098SMike Snitzer bm_blocks = DIV_ROUND_UP_SECTOR_T(bm_blocks, 4096); 631935fe098SMike Snitzer offset = bitmap->mddev->bitmap_info.offset + (bitmap->cluster_slot * (bm_blocks << 3)); 632935fe098SMike Snitzer pr_debug("%s:%d bm slot: %d offset: %llu\n", __func__, __LINE__, 633935fe098SMike Snitzer bitmap->cluster_slot, offset); 634935fe098SMike Snitzer } 635935fe098SMike Snitzer 636935fe098SMike Snitzer if (bitmap->storage.file) { 637935fe098SMike Snitzer loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host); 638935fe098SMike Snitzer int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize; 639935fe098SMike Snitzer 640935fe098SMike Snitzer err = read_page(bitmap->storage.file, 0, 641935fe098SMike Snitzer bitmap, bytes, sb_page); 642935fe098SMike Snitzer } else { 643935fe098SMike Snitzer err = read_sb_page(bitmap->mddev, 644935fe098SMike Snitzer offset, 645935fe098SMike Snitzer sb_page, 646935fe098SMike Snitzer 0, sizeof(bitmap_super_t)); 647935fe098SMike Snitzer } 648935fe098SMike Snitzer if (err) 649935fe098SMike Snitzer return err; 650935fe098SMike Snitzer 651935fe098SMike Snitzer err = -EINVAL; 652935fe098SMike Snitzer sb = kmap_atomic(sb_page); 653935fe098SMike Snitzer 654935fe098SMike Snitzer chunksize = le32_to_cpu(sb->chunksize); 655935fe098SMike Snitzer daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ; 656935fe098SMike Snitzer write_behind = le32_to_cpu(sb->write_behind); 657935fe098SMike Snitzer sectors_reserved = le32_to_cpu(sb->sectors_reserved); 658935fe098SMike Snitzer 659935fe098SMike Snitzer /* verify that the bitmap-specific fields are valid */ 660935fe098SMike Snitzer if (sb->magic != cpu_to_le32(BITMAP_MAGIC)) 661935fe098SMike Snitzer reason = "bad magic"; 662935fe098SMike Snitzer else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO || 663935fe098SMike Snitzer le32_to_cpu(sb->version) > BITMAP_MAJOR_CLUSTERED) 664935fe098SMike Snitzer reason = "unrecognized superblock version"; 665935fe098SMike Snitzer else if (chunksize < 512) 666935fe098SMike Snitzer reason = "bitmap chunksize too small"; 667935fe098SMike Snitzer else if (!is_power_of_2(chunksize)) 668935fe098SMike Snitzer reason = "bitmap chunksize not a power of 2"; 669935fe098SMike Snitzer else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT) 670935fe098SMike Snitzer reason = "daemon sleep period out of range"; 671935fe098SMike Snitzer else if (write_behind > COUNTER_MAX) 672935fe098SMike Snitzer reason = "write-behind limit out of range (0 - 16383)"; 673935fe098SMike Snitzer if (reason) { 674935fe098SMike Snitzer pr_warn("%s: invalid bitmap file superblock: %s\n", 675935fe098SMike Snitzer bmname(bitmap), reason); 676935fe098SMike Snitzer goto out; 677935fe098SMike Snitzer } 678935fe098SMike Snitzer 679e68cb83aSHeming Zhao /* 680e68cb83aSHeming Zhao * Setup nodes/clustername only if bitmap version is 681e68cb83aSHeming Zhao * cluster-compatible 682e68cb83aSHeming Zhao */ 683e68cb83aSHeming Zhao if (sb->version == cpu_to_le32(BITMAP_MAJOR_CLUSTERED)) { 684e68cb83aSHeming Zhao nodes = le32_to_cpu(sb->nodes); 68592d9aac9SHeming Zhao strscpy(bitmap->mddev->bitmap_info.cluster_name, 686e68cb83aSHeming Zhao sb->cluster_name, 64); 687e68cb83aSHeming Zhao } 688e68cb83aSHeming Zhao 689935fe098SMike Snitzer /* keep the array size field of the bitmap superblock up to date */ 690935fe098SMike Snitzer sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); 691935fe098SMike Snitzer 692935fe098SMike Snitzer if (bitmap->mddev->persistent) { 693935fe098SMike Snitzer /* 694935fe098SMike Snitzer * We have a persistent array superblock, so compare the 695935fe098SMike Snitzer * bitmap's UUID and event counter to the mddev's 696935fe098SMike Snitzer */ 697935fe098SMike Snitzer if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) { 698935fe098SMike Snitzer pr_warn("%s: bitmap superblock UUID mismatch\n", 699935fe098SMike Snitzer bmname(bitmap)); 700935fe098SMike Snitzer goto out; 701935fe098SMike Snitzer } 702935fe098SMike Snitzer events = le64_to_cpu(sb->events); 703935fe098SMike Snitzer if (!nodes && (events < bitmap->mddev->events)) { 704935fe098SMike Snitzer pr_warn("%s: bitmap file is out of date (%llu < %llu) -- forcing full recovery\n", 705935fe098SMike Snitzer bmname(bitmap), events, 706935fe098SMike Snitzer (unsigned long long) bitmap->mddev->events); 707935fe098SMike Snitzer set_bit(BITMAP_STALE, &bitmap->flags); 708935fe098SMike Snitzer } 709935fe098SMike Snitzer } 710935fe098SMike Snitzer 711935fe098SMike Snitzer /* assign fields using values from superblock */ 712935fe098SMike Snitzer bitmap->flags |= le32_to_cpu(sb->state); 713935fe098SMike Snitzer if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN) 714935fe098SMike Snitzer set_bit(BITMAP_HOSTENDIAN, &bitmap->flags); 715935fe098SMike Snitzer bitmap->events_cleared = le64_to_cpu(sb->events_cleared); 716935fe098SMike Snitzer err = 0; 717935fe098SMike Snitzer 718935fe098SMike Snitzer out: 719935fe098SMike Snitzer kunmap_atomic(sb); 720e68cb83aSHeming Zhao if (err == 0 && nodes && (bitmap->cluster_slot < 0)) { 721935fe098SMike Snitzer /* Assigning chunksize is required for "re_read" */ 722935fe098SMike Snitzer bitmap->mddev->bitmap_info.chunksize = chunksize; 723935fe098SMike Snitzer err = md_setup_cluster(bitmap->mddev, nodes); 724935fe098SMike Snitzer if (err) { 725935fe098SMike Snitzer pr_warn("%s: Could not setup cluster service (%d)\n", 726935fe098SMike Snitzer bmname(bitmap), err); 727935fe098SMike Snitzer goto out_no_sb; 728935fe098SMike Snitzer } 729935fe098SMike Snitzer bitmap->cluster_slot = md_cluster_ops->slot_number(bitmap->mddev); 730935fe098SMike Snitzer goto re_read; 731935fe098SMike Snitzer } 732935fe098SMike Snitzer 733935fe098SMike Snitzer out_no_sb: 734e68cb83aSHeming Zhao if (err == 0) { 735935fe098SMike Snitzer if (test_bit(BITMAP_STALE, &bitmap->flags)) 736935fe098SMike Snitzer bitmap->events_cleared = bitmap->mddev->events; 737935fe098SMike Snitzer bitmap->mddev->bitmap_info.chunksize = chunksize; 738935fe098SMike Snitzer bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep; 739935fe098SMike Snitzer bitmap->mddev->bitmap_info.max_write_behind = write_behind; 740935fe098SMike Snitzer bitmap->mddev->bitmap_info.nodes = nodes; 741935fe098SMike Snitzer if (bitmap->mddev->bitmap_info.space == 0 || 742935fe098SMike Snitzer bitmap->mddev->bitmap_info.space > sectors_reserved) 743935fe098SMike Snitzer bitmap->mddev->bitmap_info.space = sectors_reserved; 744e68cb83aSHeming Zhao } else { 745e64e4018SAndy Shevchenko md_bitmap_print_sb(bitmap); 746935fe098SMike Snitzer if (bitmap->cluster_slot < 0) 747935fe098SMike Snitzer md_cluster_stop(bitmap->mddev); 748935fe098SMike Snitzer } 749935fe098SMike Snitzer return err; 750935fe098SMike Snitzer } 751935fe098SMike Snitzer 752935fe098SMike Snitzer /* 753935fe098SMike Snitzer * general bitmap file operations 754935fe098SMike Snitzer */ 755935fe098SMike Snitzer 756935fe098SMike Snitzer /* 757935fe098SMike Snitzer * on-disk bitmap: 758935fe098SMike Snitzer * 759935fe098SMike Snitzer * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap 760935fe098SMike Snitzer * file a page at a time. There's a superblock at the start of the file. 761935fe098SMike Snitzer */ 762935fe098SMike Snitzer /* calculate the index of the page that contains this bit */ 763935fe098SMike Snitzer static inline unsigned long file_page_index(struct bitmap_storage *store, 764935fe098SMike Snitzer unsigned long chunk) 765935fe098SMike Snitzer { 766935fe098SMike Snitzer if (store->sb_page) 767935fe098SMike Snitzer chunk += sizeof(bitmap_super_t) << 3; 768935fe098SMike Snitzer return chunk >> PAGE_BIT_SHIFT; 769935fe098SMike Snitzer } 770935fe098SMike Snitzer 771935fe098SMike Snitzer /* calculate the (bit) offset of this bit within a page */ 772935fe098SMike Snitzer static inline unsigned long file_page_offset(struct bitmap_storage *store, 773935fe098SMike Snitzer unsigned long chunk) 774935fe098SMike Snitzer { 775935fe098SMike Snitzer if (store->sb_page) 776935fe098SMike Snitzer chunk += sizeof(bitmap_super_t) << 3; 777935fe098SMike Snitzer return chunk & (PAGE_BITS - 1); 778935fe098SMike Snitzer } 779935fe098SMike Snitzer 780935fe098SMike Snitzer /* 781935fe098SMike Snitzer * return a pointer to the page in the filemap that contains the given bit 782935fe098SMike Snitzer * 783935fe098SMike Snitzer */ 784935fe098SMike Snitzer static inline struct page *filemap_get_page(struct bitmap_storage *store, 785935fe098SMike Snitzer unsigned long chunk) 786935fe098SMike Snitzer { 787935fe098SMike Snitzer if (file_page_index(store, chunk) >= store->file_pages) 788935fe098SMike Snitzer return NULL; 789935fe098SMike Snitzer return store->filemap[file_page_index(store, chunk)]; 790935fe098SMike Snitzer } 791935fe098SMike Snitzer 792e64e4018SAndy Shevchenko static int md_bitmap_storage_alloc(struct bitmap_storage *store, 793935fe098SMike Snitzer unsigned long chunks, int with_super, 794935fe098SMike Snitzer int slot_number) 795935fe098SMike Snitzer { 796935fe098SMike Snitzer int pnum, offset = 0; 797935fe098SMike Snitzer unsigned long num_pages; 798935fe098SMike Snitzer unsigned long bytes; 799935fe098SMike Snitzer 800935fe098SMike Snitzer bytes = DIV_ROUND_UP(chunks, 8); 801935fe098SMike Snitzer if (with_super) 802935fe098SMike Snitzer bytes += sizeof(bitmap_super_t); 803935fe098SMike Snitzer 804935fe098SMike Snitzer num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE); 805935fe098SMike Snitzer offset = slot_number * num_pages; 806935fe098SMike Snitzer 8076da2ec56SKees Cook store->filemap = kmalloc_array(num_pages, sizeof(struct page *), 8086da2ec56SKees Cook GFP_KERNEL); 809935fe098SMike Snitzer if (!store->filemap) 810935fe098SMike Snitzer return -ENOMEM; 811935fe098SMike Snitzer 812935fe098SMike Snitzer if (with_super && !store->sb_page) { 813935fe098SMike Snitzer store->sb_page = alloc_page(GFP_KERNEL|__GFP_ZERO); 814935fe098SMike Snitzer if (store->sb_page == NULL) 815935fe098SMike Snitzer return -ENOMEM; 816935fe098SMike Snitzer } 817935fe098SMike Snitzer 818935fe098SMike Snitzer pnum = 0; 819935fe098SMike Snitzer if (store->sb_page) { 820935fe098SMike Snitzer store->filemap[0] = store->sb_page; 821935fe098SMike Snitzer pnum = 1; 822935fe098SMike Snitzer store->sb_page->index = offset; 823935fe098SMike Snitzer } 824935fe098SMike Snitzer 825935fe098SMike Snitzer for ( ; pnum < num_pages; pnum++) { 826935fe098SMike Snitzer store->filemap[pnum] = alloc_page(GFP_KERNEL|__GFP_ZERO); 827935fe098SMike Snitzer if (!store->filemap[pnum]) { 828935fe098SMike Snitzer store->file_pages = pnum; 829935fe098SMike Snitzer return -ENOMEM; 830935fe098SMike Snitzer } 831935fe098SMike Snitzer store->filemap[pnum]->index = pnum + offset; 832935fe098SMike Snitzer } 833935fe098SMike Snitzer store->file_pages = pnum; 834935fe098SMike Snitzer 835935fe098SMike Snitzer /* We need 4 bits per page, rounded up to a multiple 836935fe098SMike Snitzer * of sizeof(unsigned long) */ 837935fe098SMike Snitzer store->filemap_attr = kzalloc( 838935fe098SMike Snitzer roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)), 839935fe098SMike Snitzer GFP_KERNEL); 840935fe098SMike Snitzer if (!store->filemap_attr) 841935fe098SMike Snitzer return -ENOMEM; 842935fe098SMike Snitzer 843935fe098SMike Snitzer store->bytes = bytes; 844935fe098SMike Snitzer 845935fe098SMike Snitzer return 0; 846935fe098SMike Snitzer } 847935fe098SMike Snitzer 848e64e4018SAndy Shevchenko static void md_bitmap_file_unmap(struct bitmap_storage *store) 849935fe098SMike Snitzer { 850935fe098SMike Snitzer struct page **map, *sb_page; 851935fe098SMike Snitzer int pages; 852935fe098SMike Snitzer struct file *file; 853935fe098SMike Snitzer 854935fe098SMike Snitzer file = store->file; 855935fe098SMike Snitzer map = store->filemap; 856935fe098SMike Snitzer pages = store->file_pages; 857935fe098SMike Snitzer sb_page = store->sb_page; 858935fe098SMike Snitzer 859935fe098SMike Snitzer while (pages--) 860935fe098SMike Snitzer if (map[pages] != sb_page) /* 0 is sb_page, release it below */ 861935fe098SMike Snitzer free_buffers(map[pages]); 862935fe098SMike Snitzer kfree(map); 863935fe098SMike Snitzer kfree(store->filemap_attr); 864935fe098SMike Snitzer 865935fe098SMike Snitzer if (sb_page) 866935fe098SMike Snitzer free_buffers(sb_page); 867935fe098SMike Snitzer 868935fe098SMike Snitzer if (file) { 869935fe098SMike Snitzer struct inode *inode = file_inode(file); 870935fe098SMike Snitzer invalidate_mapping_pages(inode->i_mapping, 0, -1); 871935fe098SMike Snitzer fput(file); 872935fe098SMike Snitzer } 873935fe098SMike Snitzer } 874935fe098SMike Snitzer 875935fe098SMike Snitzer /* 876935fe098SMike Snitzer * bitmap_file_kick - if an error occurs while manipulating the bitmap file 877935fe098SMike Snitzer * then it is no longer reliable, so we stop using it and we mark the file 878935fe098SMike Snitzer * as failed in the superblock 879935fe098SMike Snitzer */ 880e64e4018SAndy Shevchenko static void md_bitmap_file_kick(struct bitmap *bitmap) 881935fe098SMike Snitzer { 882935fe098SMike Snitzer char *path, *ptr = NULL; 883935fe098SMike Snitzer 884935fe098SMike Snitzer if (!test_and_set_bit(BITMAP_STALE, &bitmap->flags)) { 885e64e4018SAndy Shevchenko md_bitmap_update_sb(bitmap); 886935fe098SMike Snitzer 887935fe098SMike Snitzer if (bitmap->storage.file) { 888935fe098SMike Snitzer path = kmalloc(PAGE_SIZE, GFP_KERNEL); 889935fe098SMike Snitzer if (path) 890935fe098SMike Snitzer ptr = file_path(bitmap->storage.file, 891935fe098SMike Snitzer path, PAGE_SIZE); 892935fe098SMike Snitzer 893935fe098SMike Snitzer pr_warn("%s: kicking failed bitmap file %s from array!\n", 894935fe098SMike Snitzer bmname(bitmap), IS_ERR(ptr) ? "" : ptr); 895935fe098SMike Snitzer 896935fe098SMike Snitzer kfree(path); 897935fe098SMike Snitzer } else 898935fe098SMike Snitzer pr_warn("%s: disabling internal bitmap due to errors\n", 899935fe098SMike Snitzer bmname(bitmap)); 900935fe098SMike Snitzer } 901935fe098SMike Snitzer } 902935fe098SMike Snitzer 903935fe098SMike Snitzer enum bitmap_page_attr { 904935fe098SMike Snitzer BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */ 905935fe098SMike Snitzer BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned. 906935fe098SMike Snitzer * i.e. counter is 1 or 2. */ 907935fe098SMike Snitzer BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */ 908935fe098SMike Snitzer }; 909935fe098SMike Snitzer 910935fe098SMike Snitzer static inline void set_page_attr(struct bitmap *bitmap, int pnum, 911935fe098SMike Snitzer enum bitmap_page_attr attr) 912935fe098SMike Snitzer { 913935fe098SMike Snitzer set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); 914935fe098SMike Snitzer } 915935fe098SMike Snitzer 916935fe098SMike Snitzer static inline void clear_page_attr(struct bitmap *bitmap, int pnum, 917935fe098SMike Snitzer enum bitmap_page_attr attr) 918935fe098SMike Snitzer { 919935fe098SMike Snitzer clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); 920935fe098SMike Snitzer } 921935fe098SMike Snitzer 922935fe098SMike Snitzer static inline int test_page_attr(struct bitmap *bitmap, int pnum, 923935fe098SMike Snitzer enum bitmap_page_attr attr) 924935fe098SMike Snitzer { 925935fe098SMike Snitzer return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); 926935fe098SMike Snitzer } 927935fe098SMike Snitzer 928935fe098SMike Snitzer static inline int test_and_clear_page_attr(struct bitmap *bitmap, int pnum, 929935fe098SMike Snitzer enum bitmap_page_attr attr) 930935fe098SMike Snitzer { 931935fe098SMike Snitzer return test_and_clear_bit((pnum<<2) + attr, 932935fe098SMike Snitzer bitmap->storage.filemap_attr); 933935fe098SMike Snitzer } 934935fe098SMike Snitzer /* 935935fe098SMike Snitzer * bitmap_file_set_bit -- called before performing a write to the md device 936935fe098SMike Snitzer * to set (and eventually sync) a particular bit in the bitmap file 937935fe098SMike Snitzer * 938935fe098SMike Snitzer * we set the bit immediately, then we record the page number so that 939935fe098SMike Snitzer * when an unplug occurs, we can flush the dirty pages out to disk 940935fe098SMike Snitzer */ 941e64e4018SAndy Shevchenko static void md_bitmap_file_set_bit(struct bitmap *bitmap, sector_t block) 942935fe098SMike Snitzer { 943935fe098SMike Snitzer unsigned long bit; 944935fe098SMike Snitzer struct page *page; 945935fe098SMike Snitzer void *kaddr; 946935fe098SMike Snitzer unsigned long chunk = block >> bitmap->counts.chunkshift; 947935fe098SMike Snitzer struct bitmap_storage *store = &bitmap->storage; 948935fe098SMike Snitzer unsigned long node_offset = 0; 949935fe098SMike Snitzer 950935fe098SMike Snitzer if (mddev_is_clustered(bitmap->mddev)) 951935fe098SMike Snitzer node_offset = bitmap->cluster_slot * store->file_pages; 952935fe098SMike Snitzer 953935fe098SMike Snitzer page = filemap_get_page(&bitmap->storage, chunk); 954935fe098SMike Snitzer if (!page) 955935fe098SMike Snitzer return; 956935fe098SMike Snitzer bit = file_page_offset(&bitmap->storage, chunk); 957935fe098SMike Snitzer 958935fe098SMike Snitzer /* set the bit */ 959935fe098SMike Snitzer kaddr = kmap_atomic(page); 960935fe098SMike Snitzer if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) 961935fe098SMike Snitzer set_bit(bit, kaddr); 962935fe098SMike Snitzer else 963935fe098SMike Snitzer set_bit_le(bit, kaddr); 964935fe098SMike Snitzer kunmap_atomic(kaddr); 965935fe098SMike Snitzer pr_debug("set file bit %lu page %lu\n", bit, page->index); 966935fe098SMike Snitzer /* record page number so it gets flushed to disk when unplug occurs */ 967935fe098SMike Snitzer set_page_attr(bitmap, page->index - node_offset, BITMAP_PAGE_DIRTY); 968935fe098SMike Snitzer } 969935fe098SMike Snitzer 970e64e4018SAndy Shevchenko static void md_bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block) 971935fe098SMike Snitzer { 972935fe098SMike Snitzer unsigned long bit; 973935fe098SMike Snitzer struct page *page; 974935fe098SMike Snitzer void *paddr; 975935fe098SMike Snitzer unsigned long chunk = block >> bitmap->counts.chunkshift; 976935fe098SMike Snitzer struct bitmap_storage *store = &bitmap->storage; 977935fe098SMike Snitzer unsigned long node_offset = 0; 978935fe098SMike Snitzer 979935fe098SMike Snitzer if (mddev_is_clustered(bitmap->mddev)) 980935fe098SMike Snitzer node_offset = bitmap->cluster_slot * store->file_pages; 981935fe098SMike Snitzer 982935fe098SMike Snitzer page = filemap_get_page(&bitmap->storage, chunk); 983935fe098SMike Snitzer if (!page) 984935fe098SMike Snitzer return; 985935fe098SMike Snitzer bit = file_page_offset(&bitmap->storage, chunk); 986935fe098SMike Snitzer paddr = kmap_atomic(page); 987935fe098SMike Snitzer if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) 988935fe098SMike Snitzer clear_bit(bit, paddr); 989935fe098SMike Snitzer else 990935fe098SMike Snitzer clear_bit_le(bit, paddr); 991935fe098SMike Snitzer kunmap_atomic(paddr); 992935fe098SMike Snitzer if (!test_page_attr(bitmap, page->index - node_offset, BITMAP_PAGE_NEEDWRITE)) { 993935fe098SMike Snitzer set_page_attr(bitmap, page->index - node_offset, BITMAP_PAGE_PENDING); 994935fe098SMike Snitzer bitmap->allclean = 0; 995935fe098SMike Snitzer } 996935fe098SMike Snitzer } 997935fe098SMike Snitzer 998e64e4018SAndy Shevchenko static int md_bitmap_file_test_bit(struct bitmap *bitmap, sector_t block) 999935fe098SMike Snitzer { 1000935fe098SMike Snitzer unsigned long bit; 1001935fe098SMike Snitzer struct page *page; 1002935fe098SMike Snitzer void *paddr; 1003935fe098SMike Snitzer unsigned long chunk = block >> bitmap->counts.chunkshift; 1004935fe098SMike Snitzer int set = 0; 1005935fe098SMike Snitzer 1006935fe098SMike Snitzer page = filemap_get_page(&bitmap->storage, chunk); 1007935fe098SMike Snitzer if (!page) 1008935fe098SMike Snitzer return -EINVAL; 1009935fe098SMike Snitzer bit = file_page_offset(&bitmap->storage, chunk); 1010935fe098SMike Snitzer paddr = kmap_atomic(page); 1011935fe098SMike Snitzer if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) 1012935fe098SMike Snitzer set = test_bit(bit, paddr); 1013935fe098SMike Snitzer else 1014935fe098SMike Snitzer set = test_bit_le(bit, paddr); 1015935fe098SMike Snitzer kunmap_atomic(paddr); 1016935fe098SMike Snitzer return set; 1017935fe098SMike Snitzer } 1018935fe098SMike Snitzer 1019935fe098SMike Snitzer 1020935fe098SMike Snitzer /* this gets called when the md device is ready to unplug its underlying 1021935fe098SMike Snitzer * (slave) device queues -- before we let any writes go down, we need to 1022935fe098SMike Snitzer * sync the dirty pages of the bitmap file to disk */ 1023e64e4018SAndy Shevchenko void md_bitmap_unplug(struct bitmap *bitmap) 1024935fe098SMike Snitzer { 1025935fe098SMike Snitzer unsigned long i; 1026935fe098SMike Snitzer int dirty, need_write; 1027935fe098SMike Snitzer int writing = 0; 1028935fe098SMike Snitzer 1029935fe098SMike Snitzer if (!bitmap || !bitmap->storage.filemap || 1030935fe098SMike Snitzer test_bit(BITMAP_STALE, &bitmap->flags)) 1031935fe098SMike Snitzer return; 1032935fe098SMike Snitzer 1033935fe098SMike Snitzer /* look at each page to see if there are any set bits that need to be 1034935fe098SMike Snitzer * flushed out to disk */ 1035935fe098SMike Snitzer for (i = 0; i < bitmap->storage.file_pages; i++) { 1036935fe098SMike Snitzer dirty = test_and_clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY); 1037935fe098SMike Snitzer need_write = test_and_clear_page_attr(bitmap, i, 1038935fe098SMike Snitzer BITMAP_PAGE_NEEDWRITE); 1039935fe098SMike Snitzer if (dirty || need_write) { 1040935fe098SMike Snitzer if (!writing) { 1041e64e4018SAndy Shevchenko md_bitmap_wait_writes(bitmap); 1042935fe098SMike Snitzer if (bitmap->mddev->queue) 1043935fe098SMike Snitzer blk_add_trace_msg(bitmap->mddev->queue, 1044935fe098SMike Snitzer "md bitmap_unplug"); 1045935fe098SMike Snitzer } 1046935fe098SMike Snitzer clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING); 1047935fe098SMike Snitzer write_page(bitmap, bitmap->storage.filemap[i], 0); 1048935fe098SMike Snitzer writing = 1; 1049935fe098SMike Snitzer } 1050935fe098SMike Snitzer } 1051935fe098SMike Snitzer if (writing) 1052e64e4018SAndy Shevchenko md_bitmap_wait_writes(bitmap); 1053935fe098SMike Snitzer 1054935fe098SMike Snitzer if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) 1055e64e4018SAndy Shevchenko md_bitmap_file_kick(bitmap); 1056935fe098SMike Snitzer } 1057e64e4018SAndy Shevchenko EXPORT_SYMBOL(md_bitmap_unplug); 1058935fe098SMike Snitzer 1059e64e4018SAndy Shevchenko static void md_bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed); 1060935fe098SMike Snitzer /* * bitmap_init_from_disk -- called at bitmap_create time to initialize 1061935fe098SMike Snitzer * the in-memory bitmap from the on-disk bitmap -- also, sets up the 1062935fe098SMike Snitzer * memory mapping of the bitmap file 1063935fe098SMike Snitzer * Special cases: 1064935fe098SMike Snitzer * if there's no bitmap file, or if the bitmap file had been 1065935fe098SMike Snitzer * previously kicked from the array, we mark all the bits as 1066935fe098SMike Snitzer * 1's in order to cause a full resync. 1067935fe098SMike Snitzer * 1068935fe098SMike Snitzer * We ignore all bits for sectors that end earlier than 'start'. 1069935fe098SMike Snitzer * This is used when reading an out-of-date bitmap... 1070935fe098SMike Snitzer */ 1071e64e4018SAndy Shevchenko static int md_bitmap_init_from_disk(struct bitmap *bitmap, sector_t start) 1072935fe098SMike Snitzer { 1073935fe098SMike Snitzer unsigned long i, chunks, index, oldindex, bit, node_offset = 0; 1074935fe098SMike Snitzer struct page *page = NULL; 1075935fe098SMike Snitzer unsigned long bit_cnt = 0; 1076935fe098SMike Snitzer struct file *file; 1077935fe098SMike Snitzer unsigned long offset; 1078935fe098SMike Snitzer int outofdate; 1079935fe098SMike Snitzer int ret = -ENOSPC; 1080935fe098SMike Snitzer void *paddr; 1081935fe098SMike Snitzer struct bitmap_storage *store = &bitmap->storage; 1082935fe098SMike Snitzer 1083935fe098SMike Snitzer chunks = bitmap->counts.chunks; 1084935fe098SMike Snitzer file = store->file; 1085935fe098SMike Snitzer 1086935fe098SMike Snitzer if (!file && !bitmap->mddev->bitmap_info.offset) { 1087935fe098SMike Snitzer /* No permanent bitmap - fill with '1s'. */ 1088935fe098SMike Snitzer store->filemap = NULL; 1089935fe098SMike Snitzer store->file_pages = 0; 1090935fe098SMike Snitzer for (i = 0; i < chunks ; i++) { 1091935fe098SMike Snitzer /* if the disk bit is set, set the memory bit */ 1092935fe098SMike Snitzer int needed = ((sector_t)(i+1) << (bitmap->counts.chunkshift) 1093935fe098SMike Snitzer >= start); 1094e64e4018SAndy Shevchenko md_bitmap_set_memory_bits(bitmap, 1095935fe098SMike Snitzer (sector_t)i << bitmap->counts.chunkshift, 1096935fe098SMike Snitzer needed); 1097935fe098SMike Snitzer } 1098935fe098SMike Snitzer return 0; 1099935fe098SMike Snitzer } 1100935fe098SMike Snitzer 1101935fe098SMike Snitzer outofdate = test_bit(BITMAP_STALE, &bitmap->flags); 1102935fe098SMike Snitzer if (outofdate) 1103935fe098SMike Snitzer pr_warn("%s: bitmap file is out of date, doing full recovery\n", bmname(bitmap)); 1104935fe098SMike Snitzer 1105935fe098SMike Snitzer if (file && i_size_read(file->f_mapping->host) < store->bytes) { 1106935fe098SMike Snitzer pr_warn("%s: bitmap file too short %lu < %lu\n", 1107935fe098SMike Snitzer bmname(bitmap), 1108935fe098SMike Snitzer (unsigned long) i_size_read(file->f_mapping->host), 1109935fe098SMike Snitzer store->bytes); 1110935fe098SMike Snitzer goto err; 1111935fe098SMike Snitzer } 1112935fe098SMike Snitzer 1113935fe098SMike Snitzer oldindex = ~0L; 1114935fe098SMike Snitzer offset = 0; 1115935fe098SMike Snitzer if (!bitmap->mddev->bitmap_info.external) 1116935fe098SMike Snitzer offset = sizeof(bitmap_super_t); 1117935fe098SMike Snitzer 1118935fe098SMike Snitzer if (mddev_is_clustered(bitmap->mddev)) 1119935fe098SMike Snitzer node_offset = bitmap->cluster_slot * (DIV_ROUND_UP(store->bytes, PAGE_SIZE)); 1120935fe098SMike Snitzer 1121935fe098SMike Snitzer for (i = 0; i < chunks; i++) { 1122935fe098SMike Snitzer int b; 1123935fe098SMike Snitzer index = file_page_index(&bitmap->storage, i); 1124935fe098SMike Snitzer bit = file_page_offset(&bitmap->storage, i); 1125935fe098SMike Snitzer if (index != oldindex) { /* this is a new page, read it in */ 1126935fe098SMike Snitzer int count; 1127935fe098SMike Snitzer /* unmap the old page, we're done with it */ 1128935fe098SMike Snitzer if (index == store->file_pages-1) 1129935fe098SMike Snitzer count = store->bytes - index * PAGE_SIZE; 1130935fe098SMike Snitzer else 1131935fe098SMike Snitzer count = PAGE_SIZE; 1132935fe098SMike Snitzer page = store->filemap[index]; 1133935fe098SMike Snitzer if (file) 1134935fe098SMike Snitzer ret = read_page(file, index, bitmap, 1135935fe098SMike Snitzer count, page); 1136935fe098SMike Snitzer else 1137935fe098SMike Snitzer ret = read_sb_page( 1138935fe098SMike Snitzer bitmap->mddev, 1139935fe098SMike Snitzer bitmap->mddev->bitmap_info.offset, 1140935fe098SMike Snitzer page, 1141935fe098SMike Snitzer index + node_offset, count); 1142935fe098SMike Snitzer 1143935fe098SMike Snitzer if (ret) 1144935fe098SMike Snitzer goto err; 1145935fe098SMike Snitzer 1146935fe098SMike Snitzer oldindex = index; 1147935fe098SMike Snitzer 1148935fe098SMike Snitzer if (outofdate) { 1149935fe098SMike Snitzer /* 1150935fe098SMike Snitzer * if bitmap is out of date, dirty the 1151935fe098SMike Snitzer * whole page and write it out 1152935fe098SMike Snitzer */ 1153935fe098SMike Snitzer paddr = kmap_atomic(page); 1154935fe098SMike Snitzer memset(paddr + offset, 0xff, 1155935fe098SMike Snitzer PAGE_SIZE - offset); 1156935fe098SMike Snitzer kunmap_atomic(paddr); 1157935fe098SMike Snitzer write_page(bitmap, page, 1); 1158935fe098SMike Snitzer 1159935fe098SMike Snitzer ret = -EIO; 1160935fe098SMike Snitzer if (test_bit(BITMAP_WRITE_ERROR, 1161935fe098SMike Snitzer &bitmap->flags)) 1162935fe098SMike Snitzer goto err; 1163935fe098SMike Snitzer } 1164935fe098SMike Snitzer } 1165935fe098SMike Snitzer paddr = kmap_atomic(page); 1166935fe098SMike Snitzer if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) 1167935fe098SMike Snitzer b = test_bit(bit, paddr); 1168935fe098SMike Snitzer else 1169935fe098SMike Snitzer b = test_bit_le(bit, paddr); 1170935fe098SMike Snitzer kunmap_atomic(paddr); 1171935fe098SMike Snitzer if (b) { 1172935fe098SMike Snitzer /* if the disk bit is set, set the memory bit */ 1173935fe098SMike Snitzer int needed = ((sector_t)(i+1) << bitmap->counts.chunkshift 1174935fe098SMike Snitzer >= start); 1175e64e4018SAndy Shevchenko md_bitmap_set_memory_bits(bitmap, 1176935fe098SMike Snitzer (sector_t)i << bitmap->counts.chunkshift, 1177935fe098SMike Snitzer needed); 1178935fe098SMike Snitzer bit_cnt++; 1179935fe098SMike Snitzer } 1180935fe098SMike Snitzer offset = 0; 1181935fe098SMike Snitzer } 1182935fe098SMike Snitzer 1183935fe098SMike Snitzer pr_debug("%s: bitmap initialized from disk: read %lu pages, set %lu of %lu bits\n", 1184935fe098SMike Snitzer bmname(bitmap), store->file_pages, 1185935fe098SMike Snitzer bit_cnt, chunks); 1186935fe098SMike Snitzer 1187935fe098SMike Snitzer return 0; 1188935fe098SMike Snitzer 1189935fe098SMike Snitzer err: 1190935fe098SMike Snitzer pr_warn("%s: bitmap initialisation failed: %d\n", 1191935fe098SMike Snitzer bmname(bitmap), ret); 1192935fe098SMike Snitzer return ret; 1193935fe098SMike Snitzer } 1194935fe098SMike Snitzer 1195e64e4018SAndy Shevchenko void md_bitmap_write_all(struct bitmap *bitmap) 1196935fe098SMike Snitzer { 1197935fe098SMike Snitzer /* We don't actually write all bitmap blocks here, 1198935fe098SMike Snitzer * just flag them as needing to be written 1199935fe098SMike Snitzer */ 1200935fe098SMike Snitzer int i; 1201935fe098SMike Snitzer 1202935fe098SMike Snitzer if (!bitmap || !bitmap->storage.filemap) 1203935fe098SMike Snitzer return; 1204935fe098SMike Snitzer if (bitmap->storage.file) 1205935fe098SMike Snitzer /* Only one copy, so nothing needed */ 1206935fe098SMike Snitzer return; 1207935fe098SMike Snitzer 1208935fe098SMike Snitzer for (i = 0; i < bitmap->storage.file_pages; i++) 1209935fe098SMike Snitzer set_page_attr(bitmap, i, 1210935fe098SMike Snitzer BITMAP_PAGE_NEEDWRITE); 1211935fe098SMike Snitzer bitmap->allclean = 0; 1212935fe098SMike Snitzer } 1213935fe098SMike Snitzer 1214e64e4018SAndy Shevchenko static void md_bitmap_count_page(struct bitmap_counts *bitmap, 1215935fe098SMike Snitzer sector_t offset, int inc) 1216935fe098SMike Snitzer { 1217935fe098SMike Snitzer sector_t chunk = offset >> bitmap->chunkshift; 1218935fe098SMike Snitzer unsigned long page = chunk >> PAGE_COUNTER_SHIFT; 1219935fe098SMike Snitzer bitmap->bp[page].count += inc; 1220e64e4018SAndy Shevchenko md_bitmap_checkfree(bitmap, page); 1221935fe098SMike Snitzer } 1222935fe098SMike Snitzer 1223e64e4018SAndy Shevchenko static void md_bitmap_set_pending(struct bitmap_counts *bitmap, sector_t offset) 1224935fe098SMike Snitzer { 1225935fe098SMike Snitzer sector_t chunk = offset >> bitmap->chunkshift; 1226935fe098SMike Snitzer unsigned long page = chunk >> PAGE_COUNTER_SHIFT; 1227935fe098SMike Snitzer struct bitmap_page *bp = &bitmap->bp[page]; 1228935fe098SMike Snitzer 1229935fe098SMike Snitzer if (!bp->pending) 1230935fe098SMike Snitzer bp->pending = 1; 1231935fe098SMike Snitzer } 1232935fe098SMike Snitzer 1233e64e4018SAndy Shevchenko static bitmap_counter_t *md_bitmap_get_counter(struct bitmap_counts *bitmap, 1234935fe098SMike Snitzer sector_t offset, sector_t *blocks, 1235935fe098SMike Snitzer int create); 1236935fe098SMike Snitzer 1237935fe098SMike Snitzer /* 1238935fe098SMike Snitzer * bitmap daemon -- periodically wakes up to clean bits and flush pages 1239935fe098SMike Snitzer * out to disk 1240935fe098SMike Snitzer */ 1241935fe098SMike Snitzer 1242e64e4018SAndy Shevchenko void md_bitmap_daemon_work(struct mddev *mddev) 1243935fe098SMike Snitzer { 1244935fe098SMike Snitzer struct bitmap *bitmap; 1245935fe098SMike Snitzer unsigned long j; 1246935fe098SMike Snitzer unsigned long nextpage; 1247935fe098SMike Snitzer sector_t blocks; 1248935fe098SMike Snitzer struct bitmap_counts *counts; 1249935fe098SMike Snitzer 1250935fe098SMike Snitzer /* Use a mutex to guard daemon_work against 1251935fe098SMike Snitzer * bitmap_destroy. 1252935fe098SMike Snitzer */ 1253935fe098SMike Snitzer mutex_lock(&mddev->bitmap_info.mutex); 1254935fe098SMike Snitzer bitmap = mddev->bitmap; 1255935fe098SMike Snitzer if (bitmap == NULL) { 1256935fe098SMike Snitzer mutex_unlock(&mddev->bitmap_info.mutex); 1257935fe098SMike Snitzer return; 1258935fe098SMike Snitzer } 1259935fe098SMike Snitzer if (time_before(jiffies, bitmap->daemon_lastrun 1260935fe098SMike Snitzer + mddev->bitmap_info.daemon_sleep)) 1261935fe098SMike Snitzer goto done; 1262935fe098SMike Snitzer 1263935fe098SMike Snitzer bitmap->daemon_lastrun = jiffies; 1264935fe098SMike Snitzer if (bitmap->allclean) { 1265935fe098SMike Snitzer mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; 1266935fe098SMike Snitzer goto done; 1267935fe098SMike Snitzer } 1268935fe098SMike Snitzer bitmap->allclean = 1; 1269935fe098SMike Snitzer 1270935fe098SMike Snitzer if (bitmap->mddev->queue) 1271935fe098SMike Snitzer blk_add_trace_msg(bitmap->mddev->queue, 1272935fe098SMike Snitzer "md bitmap_daemon_work"); 1273935fe098SMike Snitzer 1274935fe098SMike Snitzer /* Any file-page which is PENDING now needs to be written. 1275935fe098SMike Snitzer * So set NEEDWRITE now, then after we make any last-minute changes 1276935fe098SMike Snitzer * we will write it. 1277935fe098SMike Snitzer */ 1278935fe098SMike Snitzer for (j = 0; j < bitmap->storage.file_pages; j++) 1279935fe098SMike Snitzer if (test_and_clear_page_attr(bitmap, j, 1280935fe098SMike Snitzer BITMAP_PAGE_PENDING)) 1281935fe098SMike Snitzer set_page_attr(bitmap, j, 1282935fe098SMike Snitzer BITMAP_PAGE_NEEDWRITE); 1283935fe098SMike Snitzer 1284935fe098SMike Snitzer if (bitmap->need_sync && 1285935fe098SMike Snitzer mddev->bitmap_info.external == 0) { 1286935fe098SMike Snitzer /* Arrange for superblock update as well as 1287935fe098SMike Snitzer * other changes */ 1288935fe098SMike Snitzer bitmap_super_t *sb; 1289935fe098SMike Snitzer bitmap->need_sync = 0; 1290935fe098SMike Snitzer if (bitmap->storage.filemap) { 1291935fe098SMike Snitzer sb = kmap_atomic(bitmap->storage.sb_page); 1292935fe098SMike Snitzer sb->events_cleared = 1293935fe098SMike Snitzer cpu_to_le64(bitmap->events_cleared); 1294935fe098SMike Snitzer kunmap_atomic(sb); 1295935fe098SMike Snitzer set_page_attr(bitmap, 0, 1296935fe098SMike Snitzer BITMAP_PAGE_NEEDWRITE); 1297935fe098SMike Snitzer } 1298935fe098SMike Snitzer } 1299935fe098SMike Snitzer /* Now look at the bitmap counters and if any are '2' or '1', 1300935fe098SMike Snitzer * decrement and handle accordingly. 1301935fe098SMike Snitzer */ 1302935fe098SMike Snitzer counts = &bitmap->counts; 1303935fe098SMike Snitzer spin_lock_irq(&counts->lock); 1304935fe098SMike Snitzer nextpage = 0; 1305935fe098SMike Snitzer for (j = 0; j < counts->chunks; j++) { 1306935fe098SMike Snitzer bitmap_counter_t *bmc; 1307935fe098SMike Snitzer sector_t block = (sector_t)j << counts->chunkshift; 1308935fe098SMike Snitzer 1309935fe098SMike Snitzer if (j == nextpage) { 1310935fe098SMike Snitzer nextpage += PAGE_COUNTER_RATIO; 1311935fe098SMike Snitzer if (!counts->bp[j >> PAGE_COUNTER_SHIFT].pending) { 1312935fe098SMike Snitzer j |= PAGE_COUNTER_MASK; 1313935fe098SMike Snitzer continue; 1314935fe098SMike Snitzer } 1315935fe098SMike Snitzer counts->bp[j >> PAGE_COUNTER_SHIFT].pending = 0; 1316935fe098SMike Snitzer } 1317935fe098SMike Snitzer 1318e64e4018SAndy Shevchenko bmc = md_bitmap_get_counter(counts, block, &blocks, 0); 1319935fe098SMike Snitzer if (!bmc) { 1320935fe098SMike Snitzer j |= PAGE_COUNTER_MASK; 1321935fe098SMike Snitzer continue; 1322935fe098SMike Snitzer } 1323935fe098SMike Snitzer if (*bmc == 1 && !bitmap->need_sync) { 1324935fe098SMike Snitzer /* We can clear the bit */ 1325935fe098SMike Snitzer *bmc = 0; 1326e64e4018SAndy Shevchenko md_bitmap_count_page(counts, block, -1); 1327e64e4018SAndy Shevchenko md_bitmap_file_clear_bit(bitmap, block); 1328935fe098SMike Snitzer } else if (*bmc && *bmc <= 2) { 1329935fe098SMike Snitzer *bmc = 1; 1330e64e4018SAndy Shevchenko md_bitmap_set_pending(counts, block); 1331935fe098SMike Snitzer bitmap->allclean = 0; 1332935fe098SMike Snitzer } 1333935fe098SMike Snitzer } 1334935fe098SMike Snitzer spin_unlock_irq(&counts->lock); 1335935fe098SMike Snitzer 1336e64e4018SAndy Shevchenko md_bitmap_wait_writes(bitmap); 1337935fe098SMike Snitzer /* Now start writeout on any page in NEEDWRITE that isn't DIRTY. 1338935fe098SMike Snitzer * DIRTY pages need to be written by bitmap_unplug so it can wait 1339935fe098SMike Snitzer * for them. 1340935fe098SMike Snitzer * If we find any DIRTY page we stop there and let bitmap_unplug 1341935fe098SMike Snitzer * handle all the rest. This is important in the case where 1342935fe098SMike Snitzer * the first blocking holds the superblock and it has been updated. 1343935fe098SMike Snitzer * We mustn't write any other blocks before the superblock. 1344935fe098SMike Snitzer */ 1345935fe098SMike Snitzer for (j = 0; 1346935fe098SMike Snitzer j < bitmap->storage.file_pages 1347935fe098SMike Snitzer && !test_bit(BITMAP_STALE, &bitmap->flags); 1348935fe098SMike Snitzer j++) { 1349935fe098SMike Snitzer if (test_page_attr(bitmap, j, 1350935fe098SMike Snitzer BITMAP_PAGE_DIRTY)) 1351935fe098SMike Snitzer /* bitmap_unplug will handle the rest */ 1352935fe098SMike Snitzer break; 135355180498SZhiqiang Liu if (bitmap->storage.filemap && 135455180498SZhiqiang Liu test_and_clear_page_attr(bitmap, j, 1355935fe098SMike Snitzer BITMAP_PAGE_NEEDWRITE)) { 1356935fe098SMike Snitzer write_page(bitmap, bitmap->storage.filemap[j], 0); 1357935fe098SMike Snitzer } 1358935fe098SMike Snitzer } 1359935fe098SMike Snitzer 1360935fe098SMike Snitzer done: 1361935fe098SMike Snitzer if (bitmap->allclean == 0) 1362935fe098SMike Snitzer mddev->thread->timeout = 1363935fe098SMike Snitzer mddev->bitmap_info.daemon_sleep; 1364935fe098SMike Snitzer mutex_unlock(&mddev->bitmap_info.mutex); 1365935fe098SMike Snitzer } 1366935fe098SMike Snitzer 1367e64e4018SAndy Shevchenko static bitmap_counter_t *md_bitmap_get_counter(struct bitmap_counts *bitmap, 1368935fe098SMike Snitzer sector_t offset, sector_t *blocks, 1369935fe098SMike Snitzer int create) 1370935fe098SMike Snitzer __releases(bitmap->lock) 1371935fe098SMike Snitzer __acquires(bitmap->lock) 1372935fe098SMike Snitzer { 1373935fe098SMike Snitzer /* If 'create', we might release the lock and reclaim it. 1374935fe098SMike Snitzer * The lock must have been taken with interrupts enabled. 1375935fe098SMike Snitzer * If !create, we don't release the lock. 1376935fe098SMike Snitzer */ 1377935fe098SMike Snitzer sector_t chunk = offset >> bitmap->chunkshift; 1378935fe098SMike Snitzer unsigned long page = chunk >> PAGE_COUNTER_SHIFT; 1379935fe098SMike Snitzer unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT; 1380935fe098SMike Snitzer sector_t csize; 1381935fe098SMike Snitzer int err; 1382935fe098SMike Snitzer 1383301867b1SLi Nan if (page >= bitmap->pages) { 1384301867b1SLi Nan /* 1385301867b1SLi Nan * This can happen if bitmap_start_sync goes beyond 1386301867b1SLi Nan * End-of-device while looking for a whole page or 1387301867b1SLi Nan * user set a huge number to sysfs bitmap_set_bits. 1388301867b1SLi Nan */ 1389301867b1SLi Nan return NULL; 1390301867b1SLi Nan } 1391e64e4018SAndy Shevchenko err = md_bitmap_checkpage(bitmap, page, create, 0); 1392935fe098SMike Snitzer 1393935fe098SMike Snitzer if (bitmap->bp[page].hijacked || 1394935fe098SMike Snitzer bitmap->bp[page].map == NULL) 1395935fe098SMike Snitzer csize = ((sector_t)1) << (bitmap->chunkshift + 1396d837f727SZhao Heming PAGE_COUNTER_SHIFT); 1397935fe098SMike Snitzer else 1398935fe098SMike Snitzer csize = ((sector_t)1) << bitmap->chunkshift; 1399935fe098SMike Snitzer *blocks = csize - (offset & (csize - 1)); 1400935fe098SMike Snitzer 1401935fe098SMike Snitzer if (err < 0) 1402935fe098SMike Snitzer return NULL; 1403935fe098SMike Snitzer 1404935fe098SMike Snitzer /* now locked ... */ 1405935fe098SMike Snitzer 1406935fe098SMike Snitzer if (bitmap->bp[page].hijacked) { /* hijacked pointer */ 1407935fe098SMike Snitzer /* should we use the first or second counter field 1408935fe098SMike Snitzer * of the hijacked pointer? */ 1409935fe098SMike Snitzer int hi = (pageoff > PAGE_COUNTER_MASK); 1410935fe098SMike Snitzer return &((bitmap_counter_t *) 1411935fe098SMike Snitzer &bitmap->bp[page].map)[hi]; 1412935fe098SMike Snitzer } else /* page is allocated */ 1413935fe098SMike Snitzer return (bitmap_counter_t *) 1414935fe098SMike Snitzer &(bitmap->bp[page].map[pageoff]); 1415935fe098SMike Snitzer } 1416935fe098SMike Snitzer 1417e64e4018SAndy Shevchenko int md_bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind) 1418935fe098SMike Snitzer { 1419935fe098SMike Snitzer if (!bitmap) 1420935fe098SMike Snitzer return 0; 1421935fe098SMike Snitzer 1422935fe098SMike Snitzer if (behind) { 1423935fe098SMike Snitzer int bw; 1424935fe098SMike Snitzer atomic_inc(&bitmap->behind_writes); 1425935fe098SMike Snitzer bw = atomic_read(&bitmap->behind_writes); 1426935fe098SMike Snitzer if (bw > bitmap->behind_writes_used) 1427935fe098SMike Snitzer bitmap->behind_writes_used = bw; 1428935fe098SMike Snitzer 1429935fe098SMike Snitzer pr_debug("inc write-behind count %d/%lu\n", 1430935fe098SMike Snitzer bw, bitmap->mddev->bitmap_info.max_write_behind); 1431935fe098SMike Snitzer } 1432935fe098SMike Snitzer 1433935fe098SMike Snitzer while (sectors) { 1434935fe098SMike Snitzer sector_t blocks; 1435935fe098SMike Snitzer bitmap_counter_t *bmc; 1436935fe098SMike Snitzer 1437935fe098SMike Snitzer spin_lock_irq(&bitmap->counts.lock); 1438e64e4018SAndy Shevchenko bmc = md_bitmap_get_counter(&bitmap->counts, offset, &blocks, 1); 1439935fe098SMike Snitzer if (!bmc) { 1440935fe098SMike Snitzer spin_unlock_irq(&bitmap->counts.lock); 1441935fe098SMike Snitzer return 0; 1442935fe098SMike Snitzer } 1443935fe098SMike Snitzer 1444935fe098SMike Snitzer if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) { 1445935fe098SMike Snitzer DEFINE_WAIT(__wait); 1446935fe098SMike Snitzer /* note that it is safe to do the prepare_to_wait 1447935fe098SMike Snitzer * after the test as long as we do it before dropping 1448935fe098SMike Snitzer * the spinlock. 1449935fe098SMike Snitzer */ 1450935fe098SMike Snitzer prepare_to_wait(&bitmap->overflow_wait, &__wait, 1451935fe098SMike Snitzer TASK_UNINTERRUPTIBLE); 1452935fe098SMike Snitzer spin_unlock_irq(&bitmap->counts.lock); 1453935fe098SMike Snitzer schedule(); 1454935fe098SMike Snitzer finish_wait(&bitmap->overflow_wait, &__wait); 1455935fe098SMike Snitzer continue; 1456935fe098SMike Snitzer } 1457935fe098SMike Snitzer 1458935fe098SMike Snitzer switch (*bmc) { 1459935fe098SMike Snitzer case 0: 1460e64e4018SAndy Shevchenko md_bitmap_file_set_bit(bitmap, offset); 1461e64e4018SAndy Shevchenko md_bitmap_count_page(&bitmap->counts, offset, 1); 1462df561f66SGustavo A. R. Silva fallthrough; 1463935fe098SMike Snitzer case 1: 1464935fe098SMike Snitzer *bmc = 2; 1465935fe098SMike Snitzer } 1466935fe098SMike Snitzer 1467935fe098SMike Snitzer (*bmc)++; 1468935fe098SMike Snitzer 1469935fe098SMike Snitzer spin_unlock_irq(&bitmap->counts.lock); 1470935fe098SMike Snitzer 1471935fe098SMike Snitzer offset += blocks; 1472935fe098SMike Snitzer if (sectors > blocks) 1473935fe098SMike Snitzer sectors -= blocks; 1474935fe098SMike Snitzer else 1475935fe098SMike Snitzer sectors = 0; 1476935fe098SMike Snitzer } 1477935fe098SMike Snitzer return 0; 1478935fe098SMike Snitzer } 1479e64e4018SAndy Shevchenko EXPORT_SYMBOL(md_bitmap_startwrite); 1480935fe098SMike Snitzer 1481e64e4018SAndy Shevchenko void md_bitmap_endwrite(struct bitmap *bitmap, sector_t offset, 1482e64e4018SAndy Shevchenko unsigned long sectors, int success, int behind) 1483935fe098SMike Snitzer { 1484935fe098SMike Snitzer if (!bitmap) 1485935fe098SMike Snitzer return; 1486935fe098SMike Snitzer if (behind) { 1487935fe098SMike Snitzer if (atomic_dec_and_test(&bitmap->behind_writes)) 1488935fe098SMike Snitzer wake_up(&bitmap->behind_wait); 1489935fe098SMike Snitzer pr_debug("dec write-behind count %d/%lu\n", 1490935fe098SMike Snitzer atomic_read(&bitmap->behind_writes), 1491935fe098SMike Snitzer bitmap->mddev->bitmap_info.max_write_behind); 1492935fe098SMike Snitzer } 1493935fe098SMike Snitzer 1494935fe098SMike Snitzer while (sectors) { 1495935fe098SMike Snitzer sector_t blocks; 1496935fe098SMike Snitzer unsigned long flags; 1497935fe098SMike Snitzer bitmap_counter_t *bmc; 1498935fe098SMike Snitzer 1499935fe098SMike Snitzer spin_lock_irqsave(&bitmap->counts.lock, flags); 1500e64e4018SAndy Shevchenko bmc = md_bitmap_get_counter(&bitmap->counts, offset, &blocks, 0); 1501935fe098SMike Snitzer if (!bmc) { 1502935fe098SMike Snitzer spin_unlock_irqrestore(&bitmap->counts.lock, flags); 1503935fe098SMike Snitzer return; 1504935fe098SMike Snitzer } 1505935fe098SMike Snitzer 1506935fe098SMike Snitzer if (success && !bitmap->mddev->degraded && 1507935fe098SMike Snitzer bitmap->events_cleared < bitmap->mddev->events) { 1508935fe098SMike Snitzer bitmap->events_cleared = bitmap->mddev->events; 1509935fe098SMike Snitzer bitmap->need_sync = 1; 1510935fe098SMike Snitzer sysfs_notify_dirent_safe(bitmap->sysfs_can_clear); 1511935fe098SMike Snitzer } 1512935fe098SMike Snitzer 1513935fe098SMike Snitzer if (!success && !NEEDED(*bmc)) 1514935fe098SMike Snitzer *bmc |= NEEDED_MASK; 1515935fe098SMike Snitzer 1516935fe098SMike Snitzer if (COUNTER(*bmc) == COUNTER_MAX) 1517935fe098SMike Snitzer wake_up(&bitmap->overflow_wait); 1518935fe098SMike Snitzer 1519935fe098SMike Snitzer (*bmc)--; 1520935fe098SMike Snitzer if (*bmc <= 2) { 1521e64e4018SAndy Shevchenko md_bitmap_set_pending(&bitmap->counts, offset); 1522935fe098SMike Snitzer bitmap->allclean = 0; 1523935fe098SMike Snitzer } 1524935fe098SMike Snitzer spin_unlock_irqrestore(&bitmap->counts.lock, flags); 1525935fe098SMike Snitzer offset += blocks; 1526935fe098SMike Snitzer if (sectors > blocks) 1527935fe098SMike Snitzer sectors -= blocks; 1528935fe098SMike Snitzer else 1529935fe098SMike Snitzer sectors = 0; 1530935fe098SMike Snitzer } 1531935fe098SMike Snitzer } 1532e64e4018SAndy Shevchenko EXPORT_SYMBOL(md_bitmap_endwrite); 1533935fe098SMike Snitzer 1534935fe098SMike Snitzer static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, 1535935fe098SMike Snitzer int degraded) 1536935fe098SMike Snitzer { 1537935fe098SMike Snitzer bitmap_counter_t *bmc; 1538935fe098SMike Snitzer int rv; 1539935fe098SMike Snitzer if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */ 1540935fe098SMike Snitzer *blocks = 1024; 1541935fe098SMike Snitzer return 1; /* always resync if no bitmap */ 1542935fe098SMike Snitzer } 1543935fe098SMike Snitzer spin_lock_irq(&bitmap->counts.lock); 1544e64e4018SAndy Shevchenko bmc = md_bitmap_get_counter(&bitmap->counts, offset, blocks, 0); 1545935fe098SMike Snitzer rv = 0; 1546935fe098SMike Snitzer if (bmc) { 1547935fe098SMike Snitzer /* locked */ 1548935fe098SMike Snitzer if (RESYNC(*bmc)) 1549935fe098SMike Snitzer rv = 1; 1550935fe098SMike Snitzer else if (NEEDED(*bmc)) { 1551935fe098SMike Snitzer rv = 1; 1552935fe098SMike Snitzer if (!degraded) { /* don't set/clear bits if degraded */ 1553935fe098SMike Snitzer *bmc |= RESYNC_MASK; 1554935fe098SMike Snitzer *bmc &= ~NEEDED_MASK; 1555935fe098SMike Snitzer } 1556935fe098SMike Snitzer } 1557935fe098SMike Snitzer } 1558935fe098SMike Snitzer spin_unlock_irq(&bitmap->counts.lock); 1559935fe098SMike Snitzer return rv; 1560935fe098SMike Snitzer } 1561935fe098SMike Snitzer 1562e64e4018SAndy Shevchenko int md_bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, 1563935fe098SMike Snitzer int degraded) 1564935fe098SMike Snitzer { 1565935fe098SMike Snitzer /* bitmap_start_sync must always report on multiples of whole 1566935fe098SMike Snitzer * pages, otherwise resync (which is very PAGE_SIZE based) will 1567935fe098SMike Snitzer * get confused. 1568935fe098SMike Snitzer * So call __bitmap_start_sync repeatedly (if needed) until 1569935fe098SMike Snitzer * At least PAGE_SIZE>>9 blocks are covered. 1570935fe098SMike Snitzer * Return the 'or' of the result. 1571935fe098SMike Snitzer */ 1572935fe098SMike Snitzer int rv = 0; 1573935fe098SMike Snitzer sector_t blocks1; 1574935fe098SMike Snitzer 1575935fe098SMike Snitzer *blocks = 0; 1576935fe098SMike Snitzer while (*blocks < (PAGE_SIZE>>9)) { 1577935fe098SMike Snitzer rv |= __bitmap_start_sync(bitmap, offset, 1578935fe098SMike Snitzer &blocks1, degraded); 1579935fe098SMike Snitzer offset += blocks1; 1580935fe098SMike Snitzer *blocks += blocks1; 1581935fe098SMike Snitzer } 1582935fe098SMike Snitzer return rv; 1583935fe098SMike Snitzer } 1584e64e4018SAndy Shevchenko EXPORT_SYMBOL(md_bitmap_start_sync); 1585935fe098SMike Snitzer 1586e64e4018SAndy Shevchenko void md_bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted) 1587935fe098SMike Snitzer { 1588935fe098SMike Snitzer bitmap_counter_t *bmc; 1589935fe098SMike Snitzer unsigned long flags; 1590935fe098SMike Snitzer 1591935fe098SMike Snitzer if (bitmap == NULL) { 1592935fe098SMike Snitzer *blocks = 1024; 1593935fe098SMike Snitzer return; 1594935fe098SMike Snitzer } 1595935fe098SMike Snitzer spin_lock_irqsave(&bitmap->counts.lock, flags); 1596e64e4018SAndy Shevchenko bmc = md_bitmap_get_counter(&bitmap->counts, offset, blocks, 0); 1597935fe098SMike Snitzer if (bmc == NULL) 1598935fe098SMike Snitzer goto unlock; 1599935fe098SMike Snitzer /* locked */ 1600935fe098SMike Snitzer if (RESYNC(*bmc)) { 1601935fe098SMike Snitzer *bmc &= ~RESYNC_MASK; 1602935fe098SMike Snitzer 1603935fe098SMike Snitzer if (!NEEDED(*bmc) && aborted) 1604935fe098SMike Snitzer *bmc |= NEEDED_MASK; 1605935fe098SMike Snitzer else { 1606935fe098SMike Snitzer if (*bmc <= 2) { 1607e64e4018SAndy Shevchenko md_bitmap_set_pending(&bitmap->counts, offset); 1608935fe098SMike Snitzer bitmap->allclean = 0; 1609935fe098SMike Snitzer } 1610935fe098SMike Snitzer } 1611935fe098SMike Snitzer } 1612935fe098SMike Snitzer unlock: 1613935fe098SMike Snitzer spin_unlock_irqrestore(&bitmap->counts.lock, flags); 1614935fe098SMike Snitzer } 1615e64e4018SAndy Shevchenko EXPORT_SYMBOL(md_bitmap_end_sync); 1616935fe098SMike Snitzer 1617e64e4018SAndy Shevchenko void md_bitmap_close_sync(struct bitmap *bitmap) 1618935fe098SMike Snitzer { 1619935fe098SMike Snitzer /* Sync has finished, and any bitmap chunks that weren't synced 1620935fe098SMike Snitzer * properly have been aborted. It remains to us to clear the 1621935fe098SMike Snitzer * RESYNC bit wherever it is still on 1622935fe098SMike Snitzer */ 1623935fe098SMike Snitzer sector_t sector = 0; 1624935fe098SMike Snitzer sector_t blocks; 1625935fe098SMike Snitzer if (!bitmap) 1626935fe098SMike Snitzer return; 1627935fe098SMike Snitzer while (sector < bitmap->mddev->resync_max_sectors) { 1628e64e4018SAndy Shevchenko md_bitmap_end_sync(bitmap, sector, &blocks, 0); 1629935fe098SMike Snitzer sector += blocks; 1630935fe098SMike Snitzer } 1631935fe098SMike Snitzer } 1632e64e4018SAndy Shevchenko EXPORT_SYMBOL(md_bitmap_close_sync); 1633935fe098SMike Snitzer 1634e64e4018SAndy Shevchenko void md_bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector, bool force) 1635935fe098SMike Snitzer { 1636935fe098SMike Snitzer sector_t s = 0; 1637935fe098SMike Snitzer sector_t blocks; 1638935fe098SMike Snitzer 1639935fe098SMike Snitzer if (!bitmap) 1640935fe098SMike Snitzer return; 1641935fe098SMike Snitzer if (sector == 0) { 1642935fe098SMike Snitzer bitmap->last_end_sync = jiffies; 1643935fe098SMike Snitzer return; 1644935fe098SMike Snitzer } 1645935fe098SMike Snitzer if (!force && time_before(jiffies, (bitmap->last_end_sync 1646935fe098SMike Snitzer + bitmap->mddev->bitmap_info.daemon_sleep))) 1647935fe098SMike Snitzer return; 1648935fe098SMike Snitzer wait_event(bitmap->mddev->recovery_wait, 1649935fe098SMike Snitzer atomic_read(&bitmap->mddev->recovery_active) == 0); 1650935fe098SMike Snitzer 1651935fe098SMike Snitzer bitmap->mddev->curr_resync_completed = sector; 1652935fe098SMike Snitzer set_bit(MD_SB_CHANGE_CLEAN, &bitmap->mddev->sb_flags); 1653935fe098SMike Snitzer sector &= ~((1ULL << bitmap->counts.chunkshift) - 1); 1654935fe098SMike Snitzer s = 0; 1655935fe098SMike Snitzer while (s < sector && s < bitmap->mddev->resync_max_sectors) { 1656e64e4018SAndy Shevchenko md_bitmap_end_sync(bitmap, s, &blocks, 0); 1657935fe098SMike Snitzer s += blocks; 1658935fe098SMike Snitzer } 1659935fe098SMike Snitzer bitmap->last_end_sync = jiffies; 1660e1a86dbbSJunxiao Bi sysfs_notify_dirent_safe(bitmap->mddev->sysfs_completed); 1661935fe098SMike Snitzer } 1662e64e4018SAndy Shevchenko EXPORT_SYMBOL(md_bitmap_cond_end_sync); 1663935fe098SMike Snitzer 1664e64e4018SAndy Shevchenko void md_bitmap_sync_with_cluster(struct mddev *mddev, 1665935fe098SMike Snitzer sector_t old_lo, sector_t old_hi, 1666935fe098SMike Snitzer sector_t new_lo, sector_t new_hi) 1667935fe098SMike Snitzer { 1668935fe098SMike Snitzer struct bitmap *bitmap = mddev->bitmap; 1669935fe098SMike Snitzer sector_t sector, blocks = 0; 1670935fe098SMike Snitzer 1671935fe098SMike Snitzer for (sector = old_lo; sector < new_lo; ) { 1672e64e4018SAndy Shevchenko md_bitmap_end_sync(bitmap, sector, &blocks, 0); 1673935fe098SMike Snitzer sector += blocks; 1674935fe098SMike Snitzer } 1675935fe098SMike Snitzer WARN((blocks > new_lo) && old_lo, "alignment is not correct for lo\n"); 1676935fe098SMike Snitzer 1677935fe098SMike Snitzer for (sector = old_hi; sector < new_hi; ) { 1678e64e4018SAndy Shevchenko md_bitmap_start_sync(bitmap, sector, &blocks, 0); 1679935fe098SMike Snitzer sector += blocks; 1680935fe098SMike Snitzer } 1681935fe098SMike Snitzer WARN((blocks > new_hi) && old_hi, "alignment is not correct for hi\n"); 1682935fe098SMike Snitzer } 1683e64e4018SAndy Shevchenko EXPORT_SYMBOL(md_bitmap_sync_with_cluster); 1684935fe098SMike Snitzer 1685e64e4018SAndy Shevchenko static void md_bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed) 1686935fe098SMike Snitzer { 1687935fe098SMike Snitzer /* For each chunk covered by any of these sectors, set the 1688935fe098SMike Snitzer * counter to 2 and possibly set resync_needed. They should all 1689935fe098SMike Snitzer * be 0 at this point 1690935fe098SMike Snitzer */ 1691935fe098SMike Snitzer 1692935fe098SMike Snitzer sector_t secs; 1693935fe098SMike Snitzer bitmap_counter_t *bmc; 1694935fe098SMike Snitzer spin_lock_irq(&bitmap->counts.lock); 1695e64e4018SAndy Shevchenko bmc = md_bitmap_get_counter(&bitmap->counts, offset, &secs, 1); 1696935fe098SMike Snitzer if (!bmc) { 1697935fe098SMike Snitzer spin_unlock_irq(&bitmap->counts.lock); 1698935fe098SMike Snitzer return; 1699935fe098SMike Snitzer } 1700935fe098SMike Snitzer if (!*bmc) { 1701935fe098SMike Snitzer *bmc = 2; 1702e64e4018SAndy Shevchenko md_bitmap_count_page(&bitmap->counts, offset, 1); 1703e64e4018SAndy Shevchenko md_bitmap_set_pending(&bitmap->counts, offset); 1704935fe098SMike Snitzer bitmap->allclean = 0; 1705935fe098SMike Snitzer } 1706935fe098SMike Snitzer if (needed) 1707935fe098SMike Snitzer *bmc |= NEEDED_MASK; 1708935fe098SMike Snitzer spin_unlock_irq(&bitmap->counts.lock); 1709935fe098SMike Snitzer } 1710935fe098SMike Snitzer 1711935fe098SMike Snitzer /* dirty the memory and file bits for bitmap chunks "s" to "e" */ 1712e64e4018SAndy Shevchenko void md_bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e) 1713935fe098SMike Snitzer { 1714935fe098SMike Snitzer unsigned long chunk; 1715935fe098SMike Snitzer 1716935fe098SMike Snitzer for (chunk = s; chunk <= e; chunk++) { 1717935fe098SMike Snitzer sector_t sec = (sector_t)chunk << bitmap->counts.chunkshift; 1718e64e4018SAndy Shevchenko md_bitmap_set_memory_bits(bitmap, sec, 1); 1719e64e4018SAndy Shevchenko md_bitmap_file_set_bit(bitmap, sec); 1720935fe098SMike Snitzer if (sec < bitmap->mddev->recovery_cp) 1721935fe098SMike Snitzer /* We are asserting that the array is dirty, 1722935fe098SMike Snitzer * so move the recovery_cp address back so 1723935fe098SMike Snitzer * that it is obvious that it is dirty 1724935fe098SMike Snitzer */ 1725935fe098SMike Snitzer bitmap->mddev->recovery_cp = sec; 1726935fe098SMike Snitzer } 1727935fe098SMike Snitzer } 1728935fe098SMike Snitzer 1729935fe098SMike Snitzer /* 1730935fe098SMike Snitzer * flush out any pending updates 1731935fe098SMike Snitzer */ 1732e64e4018SAndy Shevchenko void md_bitmap_flush(struct mddev *mddev) 1733935fe098SMike Snitzer { 1734935fe098SMike Snitzer struct bitmap *bitmap = mddev->bitmap; 1735935fe098SMike Snitzer long sleep; 1736935fe098SMike Snitzer 1737935fe098SMike Snitzer if (!bitmap) /* there was no bitmap */ 1738935fe098SMike Snitzer return; 1739935fe098SMike Snitzer 1740935fe098SMike Snitzer /* run the daemon_work three time to ensure everything is flushed 1741935fe098SMike Snitzer * that can be 1742935fe098SMike Snitzer */ 1743935fe098SMike Snitzer sleep = mddev->bitmap_info.daemon_sleep * 2; 1744935fe098SMike Snitzer bitmap->daemon_lastrun -= sleep; 1745e64e4018SAndy Shevchenko md_bitmap_daemon_work(mddev); 1746935fe098SMike Snitzer bitmap->daemon_lastrun -= sleep; 1747e64e4018SAndy Shevchenko md_bitmap_daemon_work(mddev); 1748935fe098SMike Snitzer bitmap->daemon_lastrun -= sleep; 1749e64e4018SAndy Shevchenko md_bitmap_daemon_work(mddev); 1750404a8ef5SSudhakar Panneerselvam if (mddev->bitmap_info.external) 1751404a8ef5SSudhakar Panneerselvam md_super_wait(mddev); 1752e64e4018SAndy Shevchenko md_bitmap_update_sb(bitmap); 1753935fe098SMike Snitzer } 1754935fe098SMike Snitzer 1755935fe098SMike Snitzer /* 1756935fe098SMike Snitzer * free memory that was allocated 1757935fe098SMike Snitzer */ 1758e64e4018SAndy Shevchenko void md_bitmap_free(struct bitmap *bitmap) 1759935fe098SMike Snitzer { 1760935fe098SMike Snitzer unsigned long k, pages; 1761935fe098SMike Snitzer struct bitmap_page *bp; 1762935fe098SMike Snitzer 1763935fe098SMike Snitzer if (!bitmap) /* there was no bitmap */ 1764935fe098SMike Snitzer return; 1765935fe098SMike Snitzer 1766935fe098SMike Snitzer if (bitmap->sysfs_can_clear) 1767935fe098SMike Snitzer sysfs_put(bitmap->sysfs_can_clear); 1768935fe098SMike Snitzer 1769935fe098SMike Snitzer if (mddev_is_clustered(bitmap->mddev) && bitmap->mddev->cluster_info && 1770935fe098SMike Snitzer bitmap->cluster_slot == md_cluster_ops->slot_number(bitmap->mddev)) 1771935fe098SMike Snitzer md_cluster_stop(bitmap->mddev); 1772935fe098SMike Snitzer 1773935fe098SMike Snitzer /* Shouldn't be needed - but just in case.... */ 1774935fe098SMike Snitzer wait_event(bitmap->write_wait, 1775935fe098SMike Snitzer atomic_read(&bitmap->pending_writes) == 0); 1776935fe098SMike Snitzer 1777935fe098SMike Snitzer /* release the bitmap file */ 1778e64e4018SAndy Shevchenko md_bitmap_file_unmap(&bitmap->storage); 1779935fe098SMike Snitzer 1780935fe098SMike Snitzer bp = bitmap->counts.bp; 1781935fe098SMike Snitzer pages = bitmap->counts.pages; 1782935fe098SMike Snitzer 1783935fe098SMike Snitzer /* free all allocated memory */ 1784935fe098SMike Snitzer 1785935fe098SMike Snitzer if (bp) /* deallocate the page memory */ 1786935fe098SMike Snitzer for (k = 0; k < pages; k++) 1787935fe098SMike Snitzer if (bp[k].map && !bp[k].hijacked) 1788935fe098SMike Snitzer kfree(bp[k].map); 1789935fe098SMike Snitzer kfree(bp); 1790935fe098SMike Snitzer kfree(bitmap); 1791935fe098SMike Snitzer } 1792e64e4018SAndy Shevchenko EXPORT_SYMBOL(md_bitmap_free); 1793935fe098SMike Snitzer 1794e64e4018SAndy Shevchenko void md_bitmap_wait_behind_writes(struct mddev *mddev) 1795935fe098SMike Snitzer { 1796935fe098SMike Snitzer struct bitmap *bitmap = mddev->bitmap; 1797935fe098SMike Snitzer 1798935fe098SMike Snitzer /* wait for behind writes to complete */ 1799935fe098SMike Snitzer if (bitmap && atomic_read(&bitmap->behind_writes) > 0) { 1800935fe098SMike Snitzer pr_debug("md:%s: behind writes in progress - waiting to stop.\n", 1801935fe098SMike Snitzer mdname(mddev)); 1802935fe098SMike Snitzer /* need to kick something here to make sure I/O goes? */ 1803935fe098SMike Snitzer wait_event(bitmap->behind_wait, 1804935fe098SMike Snitzer atomic_read(&bitmap->behind_writes) == 0); 1805935fe098SMike Snitzer } 1806935fe098SMike Snitzer } 1807935fe098SMike Snitzer 1808e64e4018SAndy Shevchenko void md_bitmap_destroy(struct mddev *mddev) 1809935fe098SMike Snitzer { 1810935fe098SMike Snitzer struct bitmap *bitmap = mddev->bitmap; 1811935fe098SMike Snitzer 1812935fe098SMike Snitzer if (!bitmap) /* there was no bitmap */ 1813935fe098SMike Snitzer return; 1814935fe098SMike Snitzer 1815e64e4018SAndy Shevchenko md_bitmap_wait_behind_writes(mddev); 181669b00b5bSGuoqing Jiang if (!mddev->serialize_policy) 181769b00b5bSGuoqing Jiang mddev_destroy_serial_pool(mddev, NULL, true); 1818935fe098SMike Snitzer 1819935fe098SMike Snitzer mutex_lock(&mddev->bitmap_info.mutex); 1820935fe098SMike Snitzer spin_lock(&mddev->lock); 1821935fe098SMike Snitzer mddev->bitmap = NULL; /* disconnect from the md device */ 1822935fe098SMike Snitzer spin_unlock(&mddev->lock); 1823935fe098SMike Snitzer mutex_unlock(&mddev->bitmap_info.mutex); 1824935fe098SMike Snitzer if (mddev->thread) 1825935fe098SMike Snitzer mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; 1826935fe098SMike Snitzer 1827e64e4018SAndy Shevchenko md_bitmap_free(bitmap); 1828935fe098SMike Snitzer } 1829935fe098SMike Snitzer 1830935fe098SMike Snitzer /* 1831935fe098SMike Snitzer * initialize the bitmap structure 1832935fe098SMike Snitzer * if this returns an error, bitmap_destroy must be called to do clean up 1833935fe098SMike Snitzer * once mddev->bitmap is set 1834935fe098SMike Snitzer */ 1835e64e4018SAndy Shevchenko struct bitmap *md_bitmap_create(struct mddev *mddev, int slot) 1836935fe098SMike Snitzer { 1837935fe098SMike Snitzer struct bitmap *bitmap; 1838935fe098SMike Snitzer sector_t blocks = mddev->resync_max_sectors; 1839935fe098SMike Snitzer struct file *file = mddev->bitmap_info.file; 1840935fe098SMike Snitzer int err; 1841935fe098SMike Snitzer struct kernfs_node *bm = NULL; 1842935fe098SMike Snitzer 1843935fe098SMike Snitzer BUILD_BUG_ON(sizeof(bitmap_super_t) != 256); 1844935fe098SMike Snitzer 1845935fe098SMike Snitzer BUG_ON(file && mddev->bitmap_info.offset); 1846935fe098SMike Snitzer 1847230b55faSNeilBrown if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) { 1848230b55faSNeilBrown pr_notice("md/raid:%s: array with journal cannot have bitmap\n", 1849230b55faSNeilBrown mdname(mddev)); 1850230b55faSNeilBrown return ERR_PTR(-EBUSY); 1851230b55faSNeilBrown } 1852230b55faSNeilBrown 1853935fe098SMike Snitzer bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL); 1854935fe098SMike Snitzer if (!bitmap) 1855935fe098SMike Snitzer return ERR_PTR(-ENOMEM); 1856935fe098SMike Snitzer 1857935fe098SMike Snitzer spin_lock_init(&bitmap->counts.lock); 1858935fe098SMike Snitzer atomic_set(&bitmap->pending_writes, 0); 1859935fe098SMike Snitzer init_waitqueue_head(&bitmap->write_wait); 1860935fe098SMike Snitzer init_waitqueue_head(&bitmap->overflow_wait); 1861935fe098SMike Snitzer init_waitqueue_head(&bitmap->behind_wait); 1862935fe098SMike Snitzer 1863935fe098SMike Snitzer bitmap->mddev = mddev; 1864935fe098SMike Snitzer bitmap->cluster_slot = slot; 1865935fe098SMike Snitzer 1866935fe098SMike Snitzer if (mddev->kobj.sd) 1867935fe098SMike Snitzer bm = sysfs_get_dirent(mddev->kobj.sd, "bitmap"); 1868935fe098SMike Snitzer if (bm) { 1869935fe098SMike Snitzer bitmap->sysfs_can_clear = sysfs_get_dirent(bm, "can_clear"); 1870935fe098SMike Snitzer sysfs_put(bm); 1871935fe098SMike Snitzer } else 1872935fe098SMike Snitzer bitmap->sysfs_can_clear = NULL; 1873935fe098SMike Snitzer 1874935fe098SMike Snitzer bitmap->storage.file = file; 1875935fe098SMike Snitzer if (file) { 1876935fe098SMike Snitzer get_file(file); 1877935fe098SMike Snitzer /* As future accesses to this file will use bmap, 1878935fe098SMike Snitzer * and bypass the page cache, we must sync the file 1879935fe098SMike Snitzer * first. 1880935fe098SMike Snitzer */ 1881935fe098SMike Snitzer vfs_fsync(file, 1); 1882935fe098SMike Snitzer } 1883935fe098SMike Snitzer /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */ 1884935fe098SMike Snitzer if (!mddev->bitmap_info.external) { 1885935fe098SMike Snitzer /* 1886935fe098SMike Snitzer * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is 1887935fe098SMike Snitzer * instructing us to create a new on-disk bitmap instance. 1888935fe098SMike Snitzer */ 1889935fe098SMike Snitzer if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags)) 1890e64e4018SAndy Shevchenko err = md_bitmap_new_disk_sb(bitmap); 1891935fe098SMike Snitzer else 1892e64e4018SAndy Shevchenko err = md_bitmap_read_sb(bitmap); 1893935fe098SMike Snitzer } else { 1894935fe098SMike Snitzer err = 0; 1895935fe098SMike Snitzer if (mddev->bitmap_info.chunksize == 0 || 1896935fe098SMike Snitzer mddev->bitmap_info.daemon_sleep == 0) 1897935fe098SMike Snitzer /* chunksize and time_base need to be 1898935fe098SMike Snitzer * set first. */ 1899935fe098SMike Snitzer err = -EINVAL; 1900935fe098SMike Snitzer } 1901935fe098SMike Snitzer if (err) 1902935fe098SMike Snitzer goto error; 1903935fe098SMike Snitzer 1904935fe098SMike Snitzer bitmap->daemon_lastrun = jiffies; 1905e64e4018SAndy Shevchenko err = md_bitmap_resize(bitmap, blocks, mddev->bitmap_info.chunksize, 1); 1906935fe098SMike Snitzer if (err) 1907935fe098SMike Snitzer goto error; 1908935fe098SMike Snitzer 1909935fe098SMike Snitzer pr_debug("created bitmap (%lu pages) for device %s\n", 1910935fe098SMike Snitzer bitmap->counts.pages, bmname(bitmap)); 1911935fe098SMike Snitzer 1912935fe098SMike Snitzer err = test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0; 1913935fe098SMike Snitzer if (err) 1914935fe098SMike Snitzer goto error; 1915935fe098SMike Snitzer 1916935fe098SMike Snitzer return bitmap; 1917935fe098SMike Snitzer error: 1918e64e4018SAndy Shevchenko md_bitmap_free(bitmap); 1919935fe098SMike Snitzer return ERR_PTR(err); 1920935fe098SMike Snitzer } 1921935fe098SMike Snitzer 1922e64e4018SAndy Shevchenko int md_bitmap_load(struct mddev *mddev) 1923935fe098SMike Snitzer { 1924935fe098SMike Snitzer int err = 0; 1925935fe098SMike Snitzer sector_t start = 0; 1926935fe098SMike Snitzer sector_t sector = 0; 1927935fe098SMike Snitzer struct bitmap *bitmap = mddev->bitmap; 1928617b194aSGuoqing Jiang struct md_rdev *rdev; 1929935fe098SMike Snitzer 1930935fe098SMike Snitzer if (!bitmap) 1931935fe098SMike Snitzer goto out; 1932935fe098SMike Snitzer 1933617b194aSGuoqing Jiang rdev_for_each(rdev, mddev) 1934404659cfSGuoqing Jiang mddev_create_serial_pool(mddev, rdev, true); 1935617b194aSGuoqing Jiang 1936935fe098SMike Snitzer if (mddev_is_clustered(mddev)) 1937935fe098SMike Snitzer md_cluster_ops->load_bitmaps(mddev, mddev->bitmap_info.nodes); 1938935fe098SMike Snitzer 1939935fe098SMike Snitzer /* Clear out old bitmap info first: Either there is none, or we 1940935fe098SMike Snitzer * are resuming after someone else has possibly changed things, 1941935fe098SMike Snitzer * so we should forget old cached info. 1942935fe098SMike Snitzer * All chunks should be clean, but some might need_sync. 1943935fe098SMike Snitzer */ 1944935fe098SMike Snitzer while (sector < mddev->resync_max_sectors) { 1945935fe098SMike Snitzer sector_t blocks; 1946e64e4018SAndy Shevchenko md_bitmap_start_sync(bitmap, sector, &blocks, 0); 1947935fe098SMike Snitzer sector += blocks; 1948935fe098SMike Snitzer } 1949e64e4018SAndy Shevchenko md_bitmap_close_sync(bitmap); 1950935fe098SMike Snitzer 1951935fe098SMike Snitzer if (mddev->degraded == 0 1952935fe098SMike Snitzer || bitmap->events_cleared == mddev->events) 1953935fe098SMike Snitzer /* no need to keep dirty bits to optimise a 1954935fe098SMike Snitzer * re-add of a missing device */ 1955935fe098SMike Snitzer start = mddev->recovery_cp; 1956935fe098SMike Snitzer 1957935fe098SMike Snitzer mutex_lock(&mddev->bitmap_info.mutex); 1958e64e4018SAndy Shevchenko err = md_bitmap_init_from_disk(bitmap, start); 1959935fe098SMike Snitzer mutex_unlock(&mddev->bitmap_info.mutex); 1960935fe098SMike Snitzer 1961935fe098SMike Snitzer if (err) 1962935fe098SMike Snitzer goto out; 1963935fe098SMike Snitzer clear_bit(BITMAP_STALE, &bitmap->flags); 1964935fe098SMike Snitzer 1965935fe098SMike Snitzer /* Kick recovery in case any bits were set */ 1966935fe098SMike Snitzer set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery); 1967935fe098SMike Snitzer 1968935fe098SMike Snitzer mddev->thread->timeout = mddev->bitmap_info.daemon_sleep; 1969935fe098SMike Snitzer md_wakeup_thread(mddev->thread); 1970935fe098SMike Snitzer 1971e64e4018SAndy Shevchenko md_bitmap_update_sb(bitmap); 1972935fe098SMike Snitzer 1973935fe098SMike Snitzer if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) 1974935fe098SMike Snitzer err = -EIO; 1975935fe098SMike Snitzer out: 1976935fe098SMike Snitzer return err; 1977935fe098SMike Snitzer } 1978e64e4018SAndy Shevchenko EXPORT_SYMBOL_GPL(md_bitmap_load); 1979935fe098SMike Snitzer 19801383b347SZhao Heming /* caller need to free returned bitmap with md_bitmap_free() */ 1981935fe098SMike Snitzer struct bitmap *get_bitmap_from_slot(struct mddev *mddev, int slot) 1982935fe098SMike Snitzer { 1983935fe098SMike Snitzer int rv = 0; 1984935fe098SMike Snitzer struct bitmap *bitmap; 1985935fe098SMike Snitzer 1986e64e4018SAndy Shevchenko bitmap = md_bitmap_create(mddev, slot); 1987935fe098SMike Snitzer if (IS_ERR(bitmap)) { 1988935fe098SMike Snitzer rv = PTR_ERR(bitmap); 1989935fe098SMike Snitzer return ERR_PTR(rv); 1990935fe098SMike Snitzer } 1991935fe098SMike Snitzer 1992e64e4018SAndy Shevchenko rv = md_bitmap_init_from_disk(bitmap, 0); 1993935fe098SMike Snitzer if (rv) { 1994e64e4018SAndy Shevchenko md_bitmap_free(bitmap); 1995935fe098SMike Snitzer return ERR_PTR(rv); 1996935fe098SMike Snitzer } 1997935fe098SMike Snitzer 1998935fe098SMike Snitzer return bitmap; 1999935fe098SMike Snitzer } 2000935fe098SMike Snitzer EXPORT_SYMBOL(get_bitmap_from_slot); 2001935fe098SMike Snitzer 2002935fe098SMike Snitzer /* Loads the bitmap associated with slot and copies the resync information 2003935fe098SMike Snitzer * to our bitmap 2004935fe098SMike Snitzer */ 2005e64e4018SAndy Shevchenko int md_bitmap_copy_from_slot(struct mddev *mddev, int slot, 2006935fe098SMike Snitzer sector_t *low, sector_t *high, bool clear_bits) 2007935fe098SMike Snitzer { 2008935fe098SMike Snitzer int rv = 0, i, j; 2009935fe098SMike Snitzer sector_t block, lo = 0, hi = 0; 2010935fe098SMike Snitzer struct bitmap_counts *counts; 2011935fe098SMike Snitzer struct bitmap *bitmap; 2012935fe098SMike Snitzer 2013935fe098SMike Snitzer bitmap = get_bitmap_from_slot(mddev, slot); 2014935fe098SMike Snitzer if (IS_ERR(bitmap)) { 2015935fe098SMike Snitzer pr_err("%s can't get bitmap from slot %d\n", __func__, slot); 2016935fe098SMike Snitzer return -1; 2017935fe098SMike Snitzer } 2018935fe098SMike Snitzer 2019935fe098SMike Snitzer counts = &bitmap->counts; 2020935fe098SMike Snitzer for (j = 0; j < counts->chunks; j++) { 2021935fe098SMike Snitzer block = (sector_t)j << counts->chunkshift; 2022e64e4018SAndy Shevchenko if (md_bitmap_file_test_bit(bitmap, block)) { 2023935fe098SMike Snitzer if (!lo) 2024935fe098SMike Snitzer lo = block; 2025935fe098SMike Snitzer hi = block; 2026e64e4018SAndy Shevchenko md_bitmap_file_clear_bit(bitmap, block); 2027e64e4018SAndy Shevchenko md_bitmap_set_memory_bits(mddev->bitmap, block, 1); 2028e64e4018SAndy Shevchenko md_bitmap_file_set_bit(mddev->bitmap, block); 2029935fe098SMike Snitzer } 2030935fe098SMike Snitzer } 2031935fe098SMike Snitzer 2032935fe098SMike Snitzer if (clear_bits) { 2033e64e4018SAndy Shevchenko md_bitmap_update_sb(bitmap); 2034935fe098SMike Snitzer /* BITMAP_PAGE_PENDING is set, but bitmap_unplug needs 2035935fe098SMike Snitzer * BITMAP_PAGE_DIRTY or _NEEDWRITE to write ... */ 2036935fe098SMike Snitzer for (i = 0; i < bitmap->storage.file_pages; i++) 2037935fe098SMike Snitzer if (test_page_attr(bitmap, i, BITMAP_PAGE_PENDING)) 2038935fe098SMike Snitzer set_page_attr(bitmap, i, BITMAP_PAGE_NEEDWRITE); 2039e64e4018SAndy Shevchenko md_bitmap_unplug(bitmap); 2040935fe098SMike Snitzer } 2041e64e4018SAndy Shevchenko md_bitmap_unplug(mddev->bitmap); 2042935fe098SMike Snitzer *low = lo; 2043935fe098SMike Snitzer *high = hi; 20441383b347SZhao Heming md_bitmap_free(bitmap); 2045935fe098SMike Snitzer 2046935fe098SMike Snitzer return rv; 2047935fe098SMike Snitzer } 2048e64e4018SAndy Shevchenko EXPORT_SYMBOL_GPL(md_bitmap_copy_from_slot); 2049935fe098SMike Snitzer 2050935fe098SMike Snitzer 2051e64e4018SAndy Shevchenko void md_bitmap_status(struct seq_file *seq, struct bitmap *bitmap) 2052935fe098SMike Snitzer { 2053935fe098SMike Snitzer unsigned long chunk_kb; 2054935fe098SMike Snitzer struct bitmap_counts *counts; 2055935fe098SMike Snitzer 2056935fe098SMike Snitzer if (!bitmap) 2057935fe098SMike Snitzer return; 2058935fe098SMike Snitzer 2059935fe098SMike Snitzer counts = &bitmap->counts; 2060935fe098SMike Snitzer 2061935fe098SMike Snitzer chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10; 2062935fe098SMike Snitzer seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], " 2063935fe098SMike Snitzer "%lu%s chunk", 2064935fe098SMike Snitzer counts->pages - counts->missing_pages, 2065935fe098SMike Snitzer counts->pages, 2066935fe098SMike Snitzer (counts->pages - counts->missing_pages) 2067935fe098SMike Snitzer << (PAGE_SHIFT - 10), 2068935fe098SMike Snitzer chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize, 2069935fe098SMike Snitzer chunk_kb ? "KB" : "B"); 2070935fe098SMike Snitzer if (bitmap->storage.file) { 2071935fe098SMike Snitzer seq_printf(seq, ", file: "); 2072935fe098SMike Snitzer seq_file_path(seq, bitmap->storage.file, " \t\n"); 2073935fe098SMike Snitzer } 2074935fe098SMike Snitzer 2075935fe098SMike Snitzer seq_printf(seq, "\n"); 2076935fe098SMike Snitzer } 2077935fe098SMike Snitzer 2078e64e4018SAndy Shevchenko int md_bitmap_resize(struct bitmap *bitmap, sector_t blocks, 2079935fe098SMike Snitzer int chunksize, int init) 2080935fe098SMike Snitzer { 2081935fe098SMike Snitzer /* If chunk_size is 0, choose an appropriate chunk size. 2082935fe098SMike Snitzer * Then possibly allocate new storage space. 2083935fe098SMike Snitzer * Then quiesce, copy bits, replace bitmap, and re-start 2084935fe098SMike Snitzer * 2085935fe098SMike Snitzer * This function is called both to set up the initial bitmap 2086935fe098SMike Snitzer * and to resize the bitmap while the array is active. 2087935fe098SMike Snitzer * If this happens as a result of the array being resized, 2088935fe098SMike Snitzer * chunksize will be zero, and we need to choose a suitable 2089935fe098SMike Snitzer * chunksize, otherwise we use what we are given. 2090935fe098SMike Snitzer */ 2091935fe098SMike Snitzer struct bitmap_storage store; 2092935fe098SMike Snitzer struct bitmap_counts old_counts; 2093935fe098SMike Snitzer unsigned long chunks; 2094935fe098SMike Snitzer sector_t block; 2095935fe098SMike Snitzer sector_t old_blocks, new_blocks; 2096935fe098SMike Snitzer int chunkshift; 2097935fe098SMike Snitzer int ret = 0; 2098935fe098SMike Snitzer long pages; 2099935fe098SMike Snitzer struct bitmap_page *new_bp; 2100935fe098SMike Snitzer 2101935fe098SMike Snitzer if (bitmap->storage.file && !init) { 2102935fe098SMike Snitzer pr_info("md: cannot resize file-based bitmap\n"); 2103935fe098SMike Snitzer return -EINVAL; 2104935fe098SMike Snitzer } 2105935fe098SMike Snitzer 2106935fe098SMike Snitzer if (chunksize == 0) { 2107935fe098SMike Snitzer /* If there is enough space, leave the chunk size unchanged, 2108935fe098SMike Snitzer * else increase by factor of two until there is enough space. 2109935fe098SMike Snitzer */ 2110935fe098SMike Snitzer long bytes; 2111935fe098SMike Snitzer long space = bitmap->mddev->bitmap_info.space; 2112935fe098SMike Snitzer 2113935fe098SMike Snitzer if (space == 0) { 2114935fe098SMike Snitzer /* We don't know how much space there is, so limit 2115935fe098SMike Snitzer * to current size - in sectors. 2116935fe098SMike Snitzer */ 2117935fe098SMike Snitzer bytes = DIV_ROUND_UP(bitmap->counts.chunks, 8); 2118935fe098SMike Snitzer if (!bitmap->mddev->bitmap_info.external) 2119935fe098SMike Snitzer bytes += sizeof(bitmap_super_t); 2120935fe098SMike Snitzer space = DIV_ROUND_UP(bytes, 512); 2121935fe098SMike Snitzer bitmap->mddev->bitmap_info.space = space; 2122935fe098SMike Snitzer } 2123935fe098SMike Snitzer chunkshift = bitmap->counts.chunkshift; 2124935fe098SMike Snitzer chunkshift--; 2125935fe098SMike Snitzer do { 2126935fe098SMike Snitzer /* 'chunkshift' is shift from block size to chunk size */ 2127935fe098SMike Snitzer chunkshift++; 2128935fe098SMike Snitzer chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift); 2129935fe098SMike Snitzer bytes = DIV_ROUND_UP(chunks, 8); 2130935fe098SMike Snitzer if (!bitmap->mddev->bitmap_info.external) 2131935fe098SMike Snitzer bytes += sizeof(bitmap_super_t); 213245552111SFlorian-Ewald Mueller } while (bytes > (space << 9) && (chunkshift + BITMAP_BLOCK_SHIFT) < 213345552111SFlorian-Ewald Mueller (BITS_PER_BYTE * sizeof(((bitmap_super_t *)0)->chunksize) - 1)); 2134935fe098SMike Snitzer } else 2135935fe098SMike Snitzer chunkshift = ffz(~chunksize) - BITMAP_BLOCK_SHIFT; 2136935fe098SMike Snitzer 2137935fe098SMike Snitzer chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift); 2138935fe098SMike Snitzer memset(&store, 0, sizeof(store)); 2139935fe098SMike Snitzer if (bitmap->mddev->bitmap_info.offset || bitmap->mddev->bitmap_info.file) 2140e64e4018SAndy Shevchenko ret = md_bitmap_storage_alloc(&store, chunks, 2141935fe098SMike Snitzer !bitmap->mddev->bitmap_info.external, 2142935fe098SMike Snitzer mddev_is_clustered(bitmap->mddev) 2143935fe098SMike Snitzer ? bitmap->cluster_slot : 0); 2144935fe098SMike Snitzer if (ret) { 2145e64e4018SAndy Shevchenko md_bitmap_file_unmap(&store); 2146935fe098SMike Snitzer goto err; 2147935fe098SMike Snitzer } 2148935fe098SMike Snitzer 2149935fe098SMike Snitzer pages = DIV_ROUND_UP(chunks, PAGE_COUNTER_RATIO); 2150935fe098SMike Snitzer 21516396bb22SKees Cook new_bp = kcalloc(pages, sizeof(*new_bp), GFP_KERNEL); 2152935fe098SMike Snitzer ret = -ENOMEM; 2153935fe098SMike Snitzer if (!new_bp) { 2154e64e4018SAndy Shevchenko md_bitmap_file_unmap(&store); 2155935fe098SMike Snitzer goto err; 2156935fe098SMike Snitzer } 2157935fe098SMike Snitzer 2158935fe098SMike Snitzer if (!init) 2159935fe098SMike Snitzer bitmap->mddev->pers->quiesce(bitmap->mddev, 1); 2160935fe098SMike Snitzer 2161935fe098SMike Snitzer store.file = bitmap->storage.file; 2162935fe098SMike Snitzer bitmap->storage.file = NULL; 2163935fe098SMike Snitzer 2164935fe098SMike Snitzer if (store.sb_page && bitmap->storage.sb_page) 2165935fe098SMike Snitzer memcpy(page_address(store.sb_page), 2166935fe098SMike Snitzer page_address(bitmap->storage.sb_page), 2167935fe098SMike Snitzer sizeof(bitmap_super_t)); 2168fadcbd29SGuoqing Jiang spin_lock_irq(&bitmap->counts.lock); 2169e64e4018SAndy Shevchenko md_bitmap_file_unmap(&bitmap->storage); 2170935fe098SMike Snitzer bitmap->storage = store; 2171935fe098SMike Snitzer 2172935fe098SMike Snitzer old_counts = bitmap->counts; 2173935fe098SMike Snitzer bitmap->counts.bp = new_bp; 2174935fe098SMike Snitzer bitmap->counts.pages = pages; 2175935fe098SMike Snitzer bitmap->counts.missing_pages = pages; 2176935fe098SMike Snitzer bitmap->counts.chunkshift = chunkshift; 2177935fe098SMike Snitzer bitmap->counts.chunks = chunks; 217845552111SFlorian-Ewald Mueller bitmap->mddev->bitmap_info.chunksize = 1UL << (chunkshift + 2179935fe098SMike Snitzer BITMAP_BLOCK_SHIFT); 2180935fe098SMike Snitzer 2181935fe098SMike Snitzer blocks = min(old_counts.chunks << old_counts.chunkshift, 2182935fe098SMike Snitzer chunks << chunkshift); 2183935fe098SMike Snitzer 2184935fe098SMike Snitzer /* For cluster raid, need to pre-allocate bitmap */ 2185935fe098SMike Snitzer if (mddev_is_clustered(bitmap->mddev)) { 2186935fe098SMike Snitzer unsigned long page; 2187935fe098SMike Snitzer for (page = 0; page < pages; page++) { 2188e64e4018SAndy Shevchenko ret = md_bitmap_checkpage(&bitmap->counts, page, 1, 1); 2189935fe098SMike Snitzer if (ret) { 2190935fe098SMike Snitzer unsigned long k; 2191935fe098SMike Snitzer 2192935fe098SMike Snitzer /* deallocate the page memory */ 2193935fe098SMike Snitzer for (k = 0; k < page; k++) { 2194935fe098SMike Snitzer kfree(new_bp[k].map); 2195935fe098SMike Snitzer } 21960868b99cSZdenek Kabelac kfree(new_bp); 2197935fe098SMike Snitzer 2198935fe098SMike Snitzer /* restore some fields from old_counts */ 2199935fe098SMike Snitzer bitmap->counts.bp = old_counts.bp; 2200935fe098SMike Snitzer bitmap->counts.pages = old_counts.pages; 2201935fe098SMike Snitzer bitmap->counts.missing_pages = old_counts.pages; 2202935fe098SMike Snitzer bitmap->counts.chunkshift = old_counts.chunkshift; 2203935fe098SMike Snitzer bitmap->counts.chunks = old_counts.chunks; 220445552111SFlorian-Ewald Mueller bitmap->mddev->bitmap_info.chunksize = 220545552111SFlorian-Ewald Mueller 1UL << (old_counts.chunkshift + BITMAP_BLOCK_SHIFT); 2206935fe098SMike Snitzer blocks = old_counts.chunks << old_counts.chunkshift; 2207935fe098SMike Snitzer pr_warn("Could not pre-allocate in-memory bitmap for cluster raid\n"); 2208935fe098SMike Snitzer break; 2209935fe098SMike Snitzer } else 2210935fe098SMike Snitzer bitmap->counts.bp[page].count += 1; 2211935fe098SMike Snitzer } 2212935fe098SMike Snitzer } 2213935fe098SMike Snitzer 2214935fe098SMike Snitzer for (block = 0; block < blocks; ) { 2215935fe098SMike Snitzer bitmap_counter_t *bmc_old, *bmc_new; 2216935fe098SMike Snitzer int set; 2217935fe098SMike Snitzer 2218e64e4018SAndy Shevchenko bmc_old = md_bitmap_get_counter(&old_counts, block, &old_blocks, 0); 2219935fe098SMike Snitzer set = bmc_old && NEEDED(*bmc_old); 2220935fe098SMike Snitzer 2221935fe098SMike Snitzer if (set) { 2222e64e4018SAndy Shevchenko bmc_new = md_bitmap_get_counter(&bitmap->counts, block, &new_blocks, 1); 22233bd548e5SLi Zhong if (bmc_new) { 2224935fe098SMike Snitzer if (*bmc_new == 0) { 2225935fe098SMike Snitzer /* need to set on-disk bits too. */ 2226935fe098SMike Snitzer sector_t end = block + new_blocks; 2227935fe098SMike Snitzer sector_t start = block >> chunkshift; 22283bd548e5SLi Zhong 2229935fe098SMike Snitzer start <<= chunkshift; 2230935fe098SMike Snitzer while (start < end) { 2231e64e4018SAndy Shevchenko md_bitmap_file_set_bit(bitmap, block); 2232935fe098SMike Snitzer start += 1 << chunkshift; 2233935fe098SMike Snitzer } 2234935fe098SMike Snitzer *bmc_new = 2; 2235e64e4018SAndy Shevchenko md_bitmap_count_page(&bitmap->counts, block, 1); 2236e64e4018SAndy Shevchenko md_bitmap_set_pending(&bitmap->counts, block); 2237935fe098SMike Snitzer } 2238935fe098SMike Snitzer *bmc_new |= NEEDED_MASK; 22393bd548e5SLi Zhong } 2240935fe098SMike Snitzer if (new_blocks < old_blocks) 2241935fe098SMike Snitzer old_blocks = new_blocks; 2242935fe098SMike Snitzer } 2243935fe098SMike Snitzer block += old_blocks; 2244935fe098SMike Snitzer } 2245935fe098SMike Snitzer 22460868b99cSZdenek Kabelac if (bitmap->counts.bp != old_counts.bp) { 22470868b99cSZdenek Kabelac unsigned long k; 22480868b99cSZdenek Kabelac for (k = 0; k < old_counts.pages; k++) 22490868b99cSZdenek Kabelac if (!old_counts.bp[k].hijacked) 22500868b99cSZdenek Kabelac kfree(old_counts.bp[k].map); 22510868b99cSZdenek Kabelac kfree(old_counts.bp); 22520868b99cSZdenek Kabelac } 22530868b99cSZdenek Kabelac 2254935fe098SMike Snitzer if (!init) { 2255935fe098SMike Snitzer int i; 2256935fe098SMike Snitzer while (block < (chunks << chunkshift)) { 2257935fe098SMike Snitzer bitmap_counter_t *bmc; 2258e64e4018SAndy Shevchenko bmc = md_bitmap_get_counter(&bitmap->counts, block, &new_blocks, 1); 2259935fe098SMike Snitzer if (bmc) { 2260935fe098SMike Snitzer /* new space. It needs to be resynced, so 2261935fe098SMike Snitzer * we set NEEDED_MASK. 2262935fe098SMike Snitzer */ 2263935fe098SMike Snitzer if (*bmc == 0) { 2264935fe098SMike Snitzer *bmc = NEEDED_MASK | 2; 2265e64e4018SAndy Shevchenko md_bitmap_count_page(&bitmap->counts, block, 1); 2266e64e4018SAndy Shevchenko md_bitmap_set_pending(&bitmap->counts, block); 2267935fe098SMike Snitzer } 2268935fe098SMike Snitzer } 2269935fe098SMike Snitzer block += new_blocks; 2270935fe098SMike Snitzer } 2271935fe098SMike Snitzer for (i = 0; i < bitmap->storage.file_pages; i++) 2272935fe098SMike Snitzer set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY); 2273935fe098SMike Snitzer } 2274935fe098SMike Snitzer spin_unlock_irq(&bitmap->counts.lock); 2275935fe098SMike Snitzer 2276935fe098SMike Snitzer if (!init) { 2277e64e4018SAndy Shevchenko md_bitmap_unplug(bitmap); 2278935fe098SMike Snitzer bitmap->mddev->pers->quiesce(bitmap->mddev, 0); 2279935fe098SMike Snitzer } 2280935fe098SMike Snitzer ret = 0; 2281935fe098SMike Snitzer err: 2282935fe098SMike Snitzer return ret; 2283935fe098SMike Snitzer } 2284e64e4018SAndy Shevchenko EXPORT_SYMBOL_GPL(md_bitmap_resize); 2285935fe098SMike Snitzer 2286935fe098SMike Snitzer static ssize_t 2287935fe098SMike Snitzer location_show(struct mddev *mddev, char *page) 2288935fe098SMike Snitzer { 2289935fe098SMike Snitzer ssize_t len; 2290935fe098SMike Snitzer if (mddev->bitmap_info.file) 2291935fe098SMike Snitzer len = sprintf(page, "file"); 2292935fe098SMike Snitzer else if (mddev->bitmap_info.offset) 2293935fe098SMike Snitzer len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset); 2294935fe098SMike Snitzer else 2295935fe098SMike Snitzer len = sprintf(page, "none"); 2296935fe098SMike Snitzer len += sprintf(page+len, "\n"); 2297935fe098SMike Snitzer return len; 2298935fe098SMike Snitzer } 2299935fe098SMike Snitzer 2300935fe098SMike Snitzer static ssize_t 2301935fe098SMike Snitzer location_store(struct mddev *mddev, const char *buf, size_t len) 2302935fe098SMike Snitzer { 2303935fe098SMike Snitzer int rv; 2304935fe098SMike Snitzer 2305935fe098SMike Snitzer rv = mddev_lock(mddev); 2306935fe098SMike Snitzer if (rv) 2307935fe098SMike Snitzer return rv; 2308935fe098SMike Snitzer if (mddev->pers) { 2309935fe098SMike Snitzer if (!mddev->pers->quiesce) { 2310935fe098SMike Snitzer rv = -EBUSY; 2311935fe098SMike Snitzer goto out; 2312935fe098SMike Snitzer } 2313935fe098SMike Snitzer if (mddev->recovery || mddev->sync_thread) { 2314935fe098SMike Snitzer rv = -EBUSY; 2315935fe098SMike Snitzer goto out; 2316935fe098SMike Snitzer } 2317935fe098SMike Snitzer } 2318935fe098SMike Snitzer 2319935fe098SMike Snitzer if (mddev->bitmap || mddev->bitmap_info.file || 2320935fe098SMike Snitzer mddev->bitmap_info.offset) { 2321935fe098SMike Snitzer /* bitmap already configured. Only option is to clear it */ 2322935fe098SMike Snitzer if (strncmp(buf, "none", 4) != 0) { 2323935fe098SMike Snitzer rv = -EBUSY; 2324935fe098SMike Snitzer goto out; 2325935fe098SMike Snitzer } 2326935fe098SMike Snitzer if (mddev->pers) { 2327f8f83d8fSJack Wang mddev_suspend(mddev); 2328e64e4018SAndy Shevchenko md_bitmap_destroy(mddev); 2329f8f83d8fSJack Wang mddev_resume(mddev); 2330935fe098SMike Snitzer } 2331935fe098SMike Snitzer mddev->bitmap_info.offset = 0; 2332935fe098SMike Snitzer if (mddev->bitmap_info.file) { 2333935fe098SMike Snitzer struct file *f = mddev->bitmap_info.file; 2334935fe098SMike Snitzer mddev->bitmap_info.file = NULL; 2335935fe098SMike Snitzer fput(f); 2336935fe098SMike Snitzer } 2337935fe098SMike Snitzer } else { 2338935fe098SMike Snitzer /* No bitmap, OK to set a location */ 2339935fe098SMike Snitzer long long offset; 2340935fe098SMike Snitzer if (strncmp(buf, "none", 4) == 0) 2341935fe098SMike Snitzer /* nothing to be done */; 2342935fe098SMike Snitzer else if (strncmp(buf, "file:", 5) == 0) { 2343935fe098SMike Snitzer /* Not supported yet */ 2344935fe098SMike Snitzer rv = -EINVAL; 2345935fe098SMike Snitzer goto out; 2346935fe098SMike Snitzer } else { 2347935fe098SMike Snitzer if (buf[0] == '+') 2348935fe098SMike Snitzer rv = kstrtoll(buf+1, 10, &offset); 2349935fe098SMike Snitzer else 2350935fe098SMike Snitzer rv = kstrtoll(buf, 10, &offset); 2351935fe098SMike Snitzer if (rv) 2352935fe098SMike Snitzer goto out; 2353935fe098SMike Snitzer if (offset == 0) { 2354935fe098SMike Snitzer rv = -EINVAL; 2355935fe098SMike Snitzer goto out; 2356935fe098SMike Snitzer } 2357935fe098SMike Snitzer if (mddev->bitmap_info.external == 0 && 2358935fe098SMike Snitzer mddev->major_version == 0 && 2359935fe098SMike Snitzer offset != mddev->bitmap_info.default_offset) { 2360935fe098SMike Snitzer rv = -EINVAL; 2361935fe098SMike Snitzer goto out; 2362935fe098SMike Snitzer } 2363935fe098SMike Snitzer mddev->bitmap_info.offset = offset; 2364935fe098SMike Snitzer if (mddev->pers) { 2365935fe098SMike Snitzer struct bitmap *bitmap; 2366e64e4018SAndy Shevchenko bitmap = md_bitmap_create(mddev, -1); 2367f8f83d8fSJack Wang mddev_suspend(mddev); 2368935fe098SMike Snitzer if (IS_ERR(bitmap)) 2369935fe098SMike Snitzer rv = PTR_ERR(bitmap); 2370935fe098SMike Snitzer else { 2371935fe098SMike Snitzer mddev->bitmap = bitmap; 2372e64e4018SAndy Shevchenko rv = md_bitmap_load(mddev); 2373935fe098SMike Snitzer if (rv) 2374935fe098SMike Snitzer mddev->bitmap_info.offset = 0; 2375935fe098SMike Snitzer } 2376935fe098SMike Snitzer if (rv) { 2377e64e4018SAndy Shevchenko md_bitmap_destroy(mddev); 2378f8f83d8fSJack Wang mddev_resume(mddev); 2379935fe098SMike Snitzer goto out; 2380935fe098SMike Snitzer } 2381f8f83d8fSJack Wang mddev_resume(mddev); 2382935fe098SMike Snitzer } 2383935fe098SMike Snitzer } 2384935fe098SMike Snitzer } 2385935fe098SMike Snitzer if (!mddev->external) { 2386935fe098SMike Snitzer /* Ensure new bitmap info is stored in 2387935fe098SMike Snitzer * metadata promptly. 2388935fe098SMike Snitzer */ 2389935fe098SMike Snitzer set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); 2390935fe098SMike Snitzer md_wakeup_thread(mddev->thread); 2391935fe098SMike Snitzer } 2392935fe098SMike Snitzer rv = 0; 2393935fe098SMike Snitzer out: 2394935fe098SMike Snitzer mddev_unlock(mddev); 2395935fe098SMike Snitzer if (rv) 2396935fe098SMike Snitzer return rv; 2397935fe098SMike Snitzer return len; 2398935fe098SMike Snitzer } 2399935fe098SMike Snitzer 2400935fe098SMike Snitzer static struct md_sysfs_entry bitmap_location = 2401935fe098SMike Snitzer __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store); 2402935fe098SMike Snitzer 2403935fe098SMike Snitzer /* 'bitmap/space' is the space available at 'location' for the 2404935fe098SMike Snitzer * bitmap. This allows the kernel to know when it is safe to 2405935fe098SMike Snitzer * resize the bitmap to match a resized array. 2406935fe098SMike Snitzer */ 2407935fe098SMike Snitzer static ssize_t 2408935fe098SMike Snitzer space_show(struct mddev *mddev, char *page) 2409935fe098SMike Snitzer { 2410935fe098SMike Snitzer return sprintf(page, "%lu\n", mddev->bitmap_info.space); 2411935fe098SMike Snitzer } 2412935fe098SMike Snitzer 2413935fe098SMike Snitzer static ssize_t 2414935fe098SMike Snitzer space_store(struct mddev *mddev, const char *buf, size_t len) 2415935fe098SMike Snitzer { 2416935fe098SMike Snitzer unsigned long sectors; 2417935fe098SMike Snitzer int rv; 2418935fe098SMike Snitzer 2419935fe098SMike Snitzer rv = kstrtoul(buf, 10, §ors); 2420935fe098SMike Snitzer if (rv) 2421935fe098SMike Snitzer return rv; 2422935fe098SMike Snitzer 2423935fe098SMike Snitzer if (sectors == 0) 2424935fe098SMike Snitzer return -EINVAL; 2425935fe098SMike Snitzer 2426935fe098SMike Snitzer if (mddev->bitmap && 2427935fe098SMike Snitzer sectors < (mddev->bitmap->storage.bytes + 511) >> 9) 2428935fe098SMike Snitzer return -EFBIG; /* Bitmap is too big for this small space */ 2429935fe098SMike Snitzer 2430935fe098SMike Snitzer /* could make sure it isn't too big, but that isn't really 2431935fe098SMike Snitzer * needed - user-space should be careful. 2432935fe098SMike Snitzer */ 2433935fe098SMike Snitzer mddev->bitmap_info.space = sectors; 2434935fe098SMike Snitzer return len; 2435935fe098SMike Snitzer } 2436935fe098SMike Snitzer 2437935fe098SMike Snitzer static struct md_sysfs_entry bitmap_space = 2438935fe098SMike Snitzer __ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store); 2439935fe098SMike Snitzer 2440935fe098SMike Snitzer static ssize_t 2441935fe098SMike Snitzer timeout_show(struct mddev *mddev, char *page) 2442935fe098SMike Snitzer { 2443935fe098SMike Snitzer ssize_t len; 2444935fe098SMike Snitzer unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ; 2445935fe098SMike Snitzer unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ; 2446935fe098SMike Snitzer 2447935fe098SMike Snitzer len = sprintf(page, "%lu", secs); 2448935fe098SMike Snitzer if (jifs) 2449935fe098SMike Snitzer len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs)); 2450935fe098SMike Snitzer len += sprintf(page+len, "\n"); 2451935fe098SMike Snitzer return len; 2452935fe098SMike Snitzer } 2453935fe098SMike Snitzer 2454935fe098SMike Snitzer static ssize_t 2455935fe098SMike Snitzer timeout_store(struct mddev *mddev, const char *buf, size_t len) 2456935fe098SMike Snitzer { 2457935fe098SMike Snitzer /* timeout can be set at any time */ 2458935fe098SMike Snitzer unsigned long timeout; 2459935fe098SMike Snitzer int rv = strict_strtoul_scaled(buf, &timeout, 4); 2460935fe098SMike Snitzer if (rv) 2461935fe098SMike Snitzer return rv; 2462935fe098SMike Snitzer 2463935fe098SMike Snitzer /* just to make sure we don't overflow... */ 2464935fe098SMike Snitzer if (timeout >= LONG_MAX / HZ) 2465935fe098SMike Snitzer return -EINVAL; 2466935fe098SMike Snitzer 2467935fe098SMike Snitzer timeout = timeout * HZ / 10000; 2468935fe098SMike Snitzer 2469935fe098SMike Snitzer if (timeout >= MAX_SCHEDULE_TIMEOUT) 2470935fe098SMike Snitzer timeout = MAX_SCHEDULE_TIMEOUT-1; 2471935fe098SMike Snitzer if (timeout < 1) 2472935fe098SMike Snitzer timeout = 1; 2473935fe098SMike Snitzer mddev->bitmap_info.daemon_sleep = timeout; 2474935fe098SMike Snitzer if (mddev->thread) { 2475935fe098SMike Snitzer /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then 2476935fe098SMike Snitzer * the bitmap is all clean and we don't need to 2477935fe098SMike Snitzer * adjust the timeout right now 2478935fe098SMike Snitzer */ 2479*c333673aSYu Kuai if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) 2480935fe098SMike Snitzer mddev->thread->timeout = timeout; 2481*c333673aSYu Kuai } 2482*c333673aSYu Kuai 2483935fe098SMike Snitzer md_wakeup_thread(mddev->thread); 2484935fe098SMike Snitzer return len; 2485935fe098SMike Snitzer } 2486935fe098SMike Snitzer 2487935fe098SMike Snitzer static struct md_sysfs_entry bitmap_timeout = 2488935fe098SMike Snitzer __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store); 2489935fe098SMike Snitzer 2490935fe098SMike Snitzer static ssize_t 2491935fe098SMike Snitzer backlog_show(struct mddev *mddev, char *page) 2492935fe098SMike Snitzer { 2493935fe098SMike Snitzer return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind); 2494935fe098SMike Snitzer } 2495935fe098SMike Snitzer 2496935fe098SMike Snitzer static ssize_t 2497935fe098SMike Snitzer backlog_store(struct mddev *mddev, const char *buf, size_t len) 2498935fe098SMike Snitzer { 2499935fe098SMike Snitzer unsigned long backlog; 250010c92fcaSGuoqing Jiang unsigned long old_mwb = mddev->bitmap_info.max_write_behind; 25018c13ab11SGuoqing Jiang struct md_rdev *rdev; 25028c13ab11SGuoqing Jiang bool has_write_mostly = false; 2503935fe098SMike Snitzer int rv = kstrtoul(buf, 10, &backlog); 2504935fe098SMike Snitzer if (rv) 2505935fe098SMike Snitzer return rv; 2506935fe098SMike Snitzer if (backlog > COUNTER_MAX) 2507935fe098SMike Snitzer return -EINVAL; 25088c13ab11SGuoqing Jiang 25098c13ab11SGuoqing Jiang /* 25108c13ab11SGuoqing Jiang * Without write mostly device, it doesn't make sense to set 25118c13ab11SGuoqing Jiang * backlog for max_write_behind. 25128c13ab11SGuoqing Jiang */ 25138c13ab11SGuoqing Jiang rdev_for_each(rdev, mddev) { 25148c13ab11SGuoqing Jiang if (test_bit(WriteMostly, &rdev->flags)) { 25158c13ab11SGuoqing Jiang has_write_mostly = true; 25168c13ab11SGuoqing Jiang break; 25178c13ab11SGuoqing Jiang } 25188c13ab11SGuoqing Jiang } 25198c13ab11SGuoqing Jiang if (!has_write_mostly) { 25208c13ab11SGuoqing Jiang pr_warn_ratelimited("%s: can't set backlog, no write mostly device available\n", 25218c13ab11SGuoqing Jiang mdname(mddev)); 25228c13ab11SGuoqing Jiang return -EINVAL; 25238c13ab11SGuoqing Jiang } 25248c13ab11SGuoqing Jiang 2525935fe098SMike Snitzer mddev->bitmap_info.max_write_behind = backlog; 2526404659cfSGuoqing Jiang if (!backlog && mddev->serial_info_pool) { 2527404659cfSGuoqing Jiang /* serial_info_pool is not needed if backlog is zero */ 252869b00b5bSGuoqing Jiang if (!mddev->serialize_policy) 252969b00b5bSGuoqing Jiang mddev_destroy_serial_pool(mddev, NULL, false); 2530404659cfSGuoqing Jiang } else if (backlog && !mddev->serial_info_pool) { 2531404659cfSGuoqing Jiang /* serial_info_pool is needed since backlog is not zero */ 253210c92fcaSGuoqing Jiang struct md_rdev *rdev; 253310c92fcaSGuoqing Jiang 253410c92fcaSGuoqing Jiang rdev_for_each(rdev, mddev) 2535404659cfSGuoqing Jiang mddev_create_serial_pool(mddev, rdev, false); 253610c92fcaSGuoqing Jiang } 253710c92fcaSGuoqing Jiang if (old_mwb != backlog) 253810c92fcaSGuoqing Jiang md_bitmap_update_sb(mddev->bitmap); 2539935fe098SMike Snitzer return len; 2540935fe098SMike Snitzer } 2541935fe098SMike Snitzer 2542935fe098SMike Snitzer static struct md_sysfs_entry bitmap_backlog = 2543935fe098SMike Snitzer __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store); 2544935fe098SMike Snitzer 2545935fe098SMike Snitzer static ssize_t 2546935fe098SMike Snitzer chunksize_show(struct mddev *mddev, char *page) 2547935fe098SMike Snitzer { 2548935fe098SMike Snitzer return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize); 2549935fe098SMike Snitzer } 2550935fe098SMike Snitzer 2551935fe098SMike Snitzer static ssize_t 2552935fe098SMike Snitzer chunksize_store(struct mddev *mddev, const char *buf, size_t len) 2553935fe098SMike Snitzer { 2554935fe098SMike Snitzer /* Can only be changed when no bitmap is active */ 2555935fe098SMike Snitzer int rv; 2556935fe098SMike Snitzer unsigned long csize; 2557935fe098SMike Snitzer if (mddev->bitmap) 2558935fe098SMike Snitzer return -EBUSY; 2559935fe098SMike Snitzer rv = kstrtoul(buf, 10, &csize); 2560935fe098SMike Snitzer if (rv) 2561935fe098SMike Snitzer return rv; 2562935fe098SMike Snitzer if (csize < 512 || 2563935fe098SMike Snitzer !is_power_of_2(csize)) 2564935fe098SMike Snitzer return -EINVAL; 256545552111SFlorian-Ewald Mueller if (BITS_PER_LONG > 32 && csize >= (1ULL << (BITS_PER_BYTE * 256645552111SFlorian-Ewald Mueller sizeof(((bitmap_super_t *)0)->chunksize)))) 256745552111SFlorian-Ewald Mueller return -EOVERFLOW; 2568935fe098SMike Snitzer mddev->bitmap_info.chunksize = csize; 2569935fe098SMike Snitzer return len; 2570935fe098SMike Snitzer } 2571935fe098SMike Snitzer 2572935fe098SMike Snitzer static struct md_sysfs_entry bitmap_chunksize = 2573935fe098SMike Snitzer __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store); 2574935fe098SMike Snitzer 2575935fe098SMike Snitzer static ssize_t metadata_show(struct mddev *mddev, char *page) 2576935fe098SMike Snitzer { 2577935fe098SMike Snitzer if (mddev_is_clustered(mddev)) 2578935fe098SMike Snitzer return sprintf(page, "clustered\n"); 2579935fe098SMike Snitzer return sprintf(page, "%s\n", (mddev->bitmap_info.external 2580935fe098SMike Snitzer ? "external" : "internal")); 2581935fe098SMike Snitzer } 2582935fe098SMike Snitzer 2583935fe098SMike Snitzer static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len) 2584935fe098SMike Snitzer { 2585935fe098SMike Snitzer if (mddev->bitmap || 2586935fe098SMike Snitzer mddev->bitmap_info.file || 2587935fe098SMike Snitzer mddev->bitmap_info.offset) 2588935fe098SMike Snitzer return -EBUSY; 2589935fe098SMike Snitzer if (strncmp(buf, "external", 8) == 0) 2590935fe098SMike Snitzer mddev->bitmap_info.external = 1; 2591935fe098SMike Snitzer else if ((strncmp(buf, "internal", 8) == 0) || 2592935fe098SMike Snitzer (strncmp(buf, "clustered", 9) == 0)) 2593935fe098SMike Snitzer mddev->bitmap_info.external = 0; 2594935fe098SMike Snitzer else 2595935fe098SMike Snitzer return -EINVAL; 2596935fe098SMike Snitzer return len; 2597935fe098SMike Snitzer } 2598935fe098SMike Snitzer 2599935fe098SMike Snitzer static struct md_sysfs_entry bitmap_metadata = 2600935fe098SMike Snitzer __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store); 2601935fe098SMike Snitzer 2602935fe098SMike Snitzer static ssize_t can_clear_show(struct mddev *mddev, char *page) 2603935fe098SMike Snitzer { 2604935fe098SMike Snitzer int len; 2605935fe098SMike Snitzer spin_lock(&mddev->lock); 2606935fe098SMike Snitzer if (mddev->bitmap) 2607935fe098SMike Snitzer len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ? 2608935fe098SMike Snitzer "false" : "true")); 2609935fe098SMike Snitzer else 2610935fe098SMike Snitzer len = sprintf(page, "\n"); 2611935fe098SMike Snitzer spin_unlock(&mddev->lock); 2612935fe098SMike Snitzer return len; 2613935fe098SMike Snitzer } 2614935fe098SMike Snitzer 2615935fe098SMike Snitzer static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len) 2616935fe098SMike Snitzer { 2617935fe098SMike Snitzer if (mddev->bitmap == NULL) 2618935fe098SMike Snitzer return -ENOENT; 2619935fe098SMike Snitzer if (strncmp(buf, "false", 5) == 0) 2620935fe098SMike Snitzer mddev->bitmap->need_sync = 1; 2621935fe098SMike Snitzer else if (strncmp(buf, "true", 4) == 0) { 2622935fe098SMike Snitzer if (mddev->degraded) 2623935fe098SMike Snitzer return -EBUSY; 2624935fe098SMike Snitzer mddev->bitmap->need_sync = 0; 2625935fe098SMike Snitzer } else 2626935fe098SMike Snitzer return -EINVAL; 2627935fe098SMike Snitzer return len; 2628935fe098SMike Snitzer } 2629935fe098SMike Snitzer 2630935fe098SMike Snitzer static struct md_sysfs_entry bitmap_can_clear = 2631935fe098SMike Snitzer __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store); 2632935fe098SMike Snitzer 2633935fe098SMike Snitzer static ssize_t 2634935fe098SMike Snitzer behind_writes_used_show(struct mddev *mddev, char *page) 2635935fe098SMike Snitzer { 2636935fe098SMike Snitzer ssize_t ret; 2637935fe098SMike Snitzer spin_lock(&mddev->lock); 2638935fe098SMike Snitzer if (mddev->bitmap == NULL) 2639935fe098SMike Snitzer ret = sprintf(page, "0\n"); 2640935fe098SMike Snitzer else 2641935fe098SMike Snitzer ret = sprintf(page, "%lu\n", 2642935fe098SMike Snitzer mddev->bitmap->behind_writes_used); 2643935fe098SMike Snitzer spin_unlock(&mddev->lock); 2644935fe098SMike Snitzer return ret; 2645935fe098SMike Snitzer } 2646935fe098SMike Snitzer 2647935fe098SMike Snitzer static ssize_t 2648935fe098SMike Snitzer behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len) 2649935fe098SMike Snitzer { 2650935fe098SMike Snitzer if (mddev->bitmap) 2651935fe098SMike Snitzer mddev->bitmap->behind_writes_used = 0; 2652935fe098SMike Snitzer return len; 2653935fe098SMike Snitzer } 2654935fe098SMike Snitzer 2655935fe098SMike Snitzer static struct md_sysfs_entry max_backlog_used = 2656935fe098SMike Snitzer __ATTR(max_backlog_used, S_IRUGO | S_IWUSR, 2657935fe098SMike Snitzer behind_writes_used_show, behind_writes_used_reset); 2658935fe098SMike Snitzer 2659935fe098SMike Snitzer static struct attribute *md_bitmap_attrs[] = { 2660935fe098SMike Snitzer &bitmap_location.attr, 2661935fe098SMike Snitzer &bitmap_space.attr, 2662935fe098SMike Snitzer &bitmap_timeout.attr, 2663935fe098SMike Snitzer &bitmap_backlog.attr, 2664935fe098SMike Snitzer &bitmap_chunksize.attr, 2665935fe098SMike Snitzer &bitmap_metadata.attr, 2666935fe098SMike Snitzer &bitmap_can_clear.attr, 2667935fe098SMike Snitzer &max_backlog_used.attr, 2668935fe098SMike Snitzer NULL 2669935fe098SMike Snitzer }; 2670c32dc040SRikard Falkeborn const struct attribute_group md_bitmap_group = { 2671935fe098SMike Snitzer .name = "bitmap", 2672935fe098SMike Snitzer .attrs = md_bitmap_attrs, 2673935fe098SMike Snitzer }; 2674