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 57935fe098SMike Snitzer if (page >= bitmap->pages) { 58935fe098SMike Snitzer /* This can happen if bitmap_start_sync goes beyond 59935fe098SMike Snitzer * End-of-device while looking for a whole page. 60935fe098SMike Snitzer * It is harmless. 61935fe098SMike Snitzer */ 62935fe098SMike Snitzer return -EINVAL; 63935fe098SMike Snitzer } 64935fe098SMike Snitzer 65935fe098SMike Snitzer if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */ 66935fe098SMike Snitzer return 0; 67935fe098SMike Snitzer 68935fe098SMike Snitzer if (bitmap->bp[page].map) /* page is already allocated, just return */ 69935fe098SMike Snitzer return 0; 70935fe098SMike Snitzer 71935fe098SMike Snitzer if (!create) 72935fe098SMike Snitzer return -ENOENT; 73935fe098SMike Snitzer 74935fe098SMike Snitzer /* this page has not been allocated yet */ 75935fe098SMike Snitzer 76935fe098SMike Snitzer spin_unlock_irq(&bitmap->lock); 77935fe098SMike Snitzer /* It is possible that this is being called inside a 78935fe098SMike Snitzer * prepare_to_wait/finish_wait loop from raid5c:make_request(). 79935fe098SMike Snitzer * In general it is not permitted to sleep in that context as it 80935fe098SMike Snitzer * can cause the loop to spin freely. 81935fe098SMike Snitzer * That doesn't apply here as we can only reach this point 82935fe098SMike Snitzer * once with any loop. 83935fe098SMike Snitzer * When this function completes, either bp[page].map or 84935fe098SMike Snitzer * bp[page].hijacked. In either case, this function will 85935fe098SMike Snitzer * abort before getting to this point again. So there is 86935fe098SMike Snitzer * no risk of a free-spin, and so it is safe to assert 87935fe098SMike Snitzer * that sleeping here is allowed. 88935fe098SMike Snitzer */ 89935fe098SMike Snitzer sched_annotate_sleep(); 90935fe098SMike Snitzer mappage = kzalloc(PAGE_SIZE, GFP_NOIO); 91935fe098SMike Snitzer spin_lock_irq(&bitmap->lock); 92935fe098SMike Snitzer 93935fe098SMike Snitzer if (mappage == NULL) { 94935fe098SMike Snitzer pr_debug("md/bitmap: map page allocation failed, hijacking\n"); 95935fe098SMike Snitzer /* We don't support hijack for cluster raid */ 96935fe098SMike Snitzer if (no_hijack) 97935fe098SMike Snitzer return -ENOMEM; 98935fe098SMike Snitzer /* failed - set the hijacked flag so that we can use the 99935fe098SMike Snitzer * pointer as a counter */ 100935fe098SMike Snitzer if (!bitmap->bp[page].map) 101935fe098SMike Snitzer bitmap->bp[page].hijacked = 1; 102935fe098SMike Snitzer } else if (bitmap->bp[page].map || 103935fe098SMike Snitzer bitmap->bp[page].hijacked) { 104935fe098SMike Snitzer /* somebody beat us to getting the page */ 105935fe098SMike Snitzer kfree(mappage); 106935fe098SMike Snitzer } else { 107935fe098SMike Snitzer 108935fe098SMike Snitzer /* no page was in place and we have one, so install it */ 109935fe098SMike Snitzer 110935fe098SMike Snitzer bitmap->bp[page].map = mappage; 111935fe098SMike Snitzer bitmap->missing_pages--; 112935fe098SMike Snitzer } 113935fe098SMike Snitzer return 0; 114935fe098SMike Snitzer } 115935fe098SMike Snitzer 116935fe098SMike Snitzer /* if page is completely empty, put it back on the free list, or dealloc it */ 117935fe098SMike Snitzer /* if page was hijacked, unmark the flag so it might get alloced next time */ 118935fe098SMike Snitzer /* Note: lock should be held when calling this */ 119e64e4018SAndy Shevchenko static void md_bitmap_checkfree(struct bitmap_counts *bitmap, unsigned long page) 120935fe098SMike Snitzer { 121935fe098SMike Snitzer char *ptr; 122935fe098SMike Snitzer 123935fe098SMike Snitzer if (bitmap->bp[page].count) /* page is still busy */ 124935fe098SMike Snitzer return; 125935fe098SMike Snitzer 126935fe098SMike Snitzer /* page is no longer in use, it can be released */ 127935fe098SMike Snitzer 128935fe098SMike Snitzer if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */ 129935fe098SMike Snitzer bitmap->bp[page].hijacked = 0; 130935fe098SMike Snitzer bitmap->bp[page].map = NULL; 131935fe098SMike Snitzer } else { 132935fe098SMike Snitzer /* normal case, free the page */ 133935fe098SMike Snitzer ptr = bitmap->bp[page].map; 134935fe098SMike Snitzer bitmap->bp[page].map = NULL; 135935fe098SMike Snitzer bitmap->missing_pages++; 136935fe098SMike Snitzer kfree(ptr); 137935fe098SMike Snitzer } 138935fe098SMike Snitzer } 139935fe098SMike Snitzer 140935fe098SMike Snitzer /* 141935fe098SMike Snitzer * bitmap file handling - read and write the bitmap file and its superblock 142935fe098SMike Snitzer */ 143935fe098SMike Snitzer 144935fe098SMike Snitzer /* 145935fe098SMike Snitzer * basic page I/O operations 146935fe098SMike Snitzer */ 147935fe098SMike Snitzer 148935fe098SMike Snitzer /* IO operations when bitmap is stored near all superblocks */ 149935fe098SMike Snitzer static int read_sb_page(struct mddev *mddev, loff_t offset, 150935fe098SMike Snitzer struct page *page, 151935fe098SMike Snitzer unsigned long index, int size) 152935fe098SMike Snitzer { 153935fe098SMike Snitzer /* choose a good rdev and read the page from there */ 154935fe098SMike Snitzer 155935fe098SMike Snitzer struct md_rdev *rdev; 156935fe098SMike Snitzer sector_t target; 157935fe098SMike Snitzer 158935fe098SMike Snitzer rdev_for_each(rdev, mddev) { 159935fe098SMike Snitzer if (! test_bit(In_sync, &rdev->flags) 160935fe098SMike Snitzer || test_bit(Faulty, &rdev->flags) 161935fe098SMike Snitzer || test_bit(Bitmap_sync, &rdev->flags)) 162935fe098SMike Snitzer continue; 163935fe098SMike Snitzer 164935fe098SMike Snitzer target = offset + index * (PAGE_SIZE/512); 165935fe098SMike Snitzer 166935fe098SMike Snitzer if (sync_page_io(rdev, target, 167935fe098SMike Snitzer roundup(size, bdev_logical_block_size(rdev->bdev)), 168935fe098SMike Snitzer page, REQ_OP_READ, 0, true)) { 169935fe098SMike Snitzer page->index = index; 170935fe098SMike Snitzer return 0; 171935fe098SMike Snitzer } 172935fe098SMike Snitzer } 173935fe098SMike Snitzer return -EIO; 174935fe098SMike Snitzer } 175935fe098SMike Snitzer 176935fe098SMike Snitzer static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev) 177935fe098SMike Snitzer { 178935fe098SMike Snitzer /* Iterate the disks of an mddev, using rcu to protect access to the 179935fe098SMike Snitzer * linked list, and raising the refcount of devices we return to ensure 180935fe098SMike Snitzer * they don't disappear while in use. 181935fe098SMike Snitzer * As devices are only added or removed when raid_disk is < 0 and 182935fe098SMike Snitzer * nr_pending is 0 and In_sync is clear, the entries we return will 183935fe098SMike Snitzer * still be in the same position on the list when we re-enter 184935fe098SMike Snitzer * list_for_each_entry_continue_rcu. 185935fe098SMike Snitzer * 186935fe098SMike Snitzer * Note that if entered with 'rdev == NULL' to start at the 187935fe098SMike Snitzer * beginning, we temporarily assign 'rdev' to an address which 188935fe098SMike Snitzer * isn't really an rdev, but which can be used by 189935fe098SMike Snitzer * list_for_each_entry_continue_rcu() to find the first entry. 190935fe098SMike Snitzer */ 191935fe098SMike Snitzer rcu_read_lock(); 192935fe098SMike Snitzer if (rdev == NULL) 193935fe098SMike Snitzer /* start at the beginning */ 194935fe098SMike Snitzer rdev = list_entry(&mddev->disks, struct md_rdev, same_set); 195935fe098SMike Snitzer else { 196935fe098SMike Snitzer /* release the previous rdev and start from there. */ 197935fe098SMike Snitzer rdev_dec_pending(rdev, mddev); 198935fe098SMike Snitzer } 199935fe098SMike Snitzer list_for_each_entry_continue_rcu(rdev, &mddev->disks, same_set) { 200935fe098SMike Snitzer if (rdev->raid_disk >= 0 && 201935fe098SMike Snitzer !test_bit(Faulty, &rdev->flags)) { 202935fe098SMike Snitzer /* this is a usable devices */ 203935fe098SMike Snitzer atomic_inc(&rdev->nr_pending); 204935fe098SMike Snitzer rcu_read_unlock(); 205935fe098SMike Snitzer return rdev; 206935fe098SMike Snitzer } 207935fe098SMike Snitzer } 208935fe098SMike Snitzer rcu_read_unlock(); 209935fe098SMike Snitzer return NULL; 210935fe098SMike Snitzer } 211935fe098SMike Snitzer 212935fe098SMike Snitzer static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait) 213935fe098SMike Snitzer { 214935fe098SMike Snitzer struct md_rdev *rdev; 215935fe098SMike Snitzer struct block_device *bdev; 216935fe098SMike Snitzer struct mddev *mddev = bitmap->mddev; 217935fe098SMike Snitzer struct bitmap_storage *store = &bitmap->storage; 218935fe098SMike Snitzer 219935fe098SMike Snitzer restart: 220935fe098SMike Snitzer rdev = NULL; 221935fe098SMike Snitzer while ((rdev = next_active_rdev(rdev, mddev)) != NULL) { 222935fe098SMike Snitzer int size = PAGE_SIZE; 223935fe098SMike Snitzer loff_t offset = mddev->bitmap_info.offset; 224935fe098SMike Snitzer 225935fe098SMike Snitzer bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev; 226935fe098SMike Snitzer 227935fe098SMike Snitzer if (page->index == store->file_pages-1) { 228935fe098SMike Snitzer int last_page_size = store->bytes & (PAGE_SIZE-1); 229935fe098SMike Snitzer if (last_page_size == 0) 230935fe098SMike Snitzer last_page_size = PAGE_SIZE; 231935fe098SMike Snitzer size = roundup(last_page_size, 232935fe098SMike Snitzer bdev_logical_block_size(bdev)); 233935fe098SMike Snitzer } 234935fe098SMike Snitzer /* Just make sure we aren't corrupting data or 235935fe098SMike Snitzer * metadata 236935fe098SMike Snitzer */ 237935fe098SMike Snitzer if (mddev->external) { 238935fe098SMike Snitzer /* Bitmap could be anywhere. */ 239935fe098SMike Snitzer if (rdev->sb_start + offset + (page->index 240935fe098SMike Snitzer * (PAGE_SIZE/512)) 241935fe098SMike Snitzer > rdev->data_offset 242935fe098SMike Snitzer && 243935fe098SMike Snitzer rdev->sb_start + offset 244935fe098SMike Snitzer < (rdev->data_offset + mddev->dev_sectors 245935fe098SMike Snitzer + (PAGE_SIZE/512))) 246935fe098SMike Snitzer goto bad_alignment; 247935fe098SMike Snitzer } else if (offset < 0) { 248935fe098SMike Snitzer /* DATA BITMAP METADATA */ 249935fe098SMike Snitzer if (offset 250935fe098SMike Snitzer + (long)(page->index * (PAGE_SIZE/512)) 251935fe098SMike Snitzer + size/512 > 0) 252935fe098SMike Snitzer /* bitmap runs in to metadata */ 253935fe098SMike Snitzer goto bad_alignment; 254935fe098SMike Snitzer if (rdev->data_offset + mddev->dev_sectors 255935fe098SMike Snitzer > rdev->sb_start + offset) 256935fe098SMike Snitzer /* data runs in to bitmap */ 257935fe098SMike Snitzer goto bad_alignment; 258935fe098SMike Snitzer } else if (rdev->sb_start < rdev->data_offset) { 259935fe098SMike Snitzer /* METADATA BITMAP DATA */ 260935fe098SMike Snitzer if (rdev->sb_start 261935fe098SMike Snitzer + offset 262935fe098SMike Snitzer + page->index*(PAGE_SIZE/512) + size/512 263935fe098SMike Snitzer > rdev->data_offset) 264935fe098SMike Snitzer /* bitmap runs in to data */ 265935fe098SMike Snitzer goto bad_alignment; 266935fe098SMike Snitzer } else { 267935fe098SMike Snitzer /* DATA METADATA BITMAP - no problems */ 268935fe098SMike Snitzer } 269935fe098SMike Snitzer md_super_write(mddev, rdev, 270935fe098SMike Snitzer rdev->sb_start + offset 271935fe098SMike Snitzer + page->index * (PAGE_SIZE/512), 272935fe098SMike Snitzer size, 273935fe098SMike Snitzer page); 274935fe098SMike Snitzer } 275935fe098SMike Snitzer 276935fe098SMike Snitzer if (wait && md_super_wait(mddev) < 0) 277935fe098SMike Snitzer goto restart; 278935fe098SMike Snitzer return 0; 279935fe098SMike Snitzer 280935fe098SMike Snitzer bad_alignment: 281935fe098SMike Snitzer return -EINVAL; 282935fe098SMike Snitzer } 283935fe098SMike Snitzer 284e64e4018SAndy Shevchenko static void md_bitmap_file_kick(struct bitmap *bitmap); 285935fe098SMike Snitzer /* 286935fe098SMike Snitzer * write out a page to a file 287935fe098SMike Snitzer */ 288935fe098SMike Snitzer static void write_page(struct bitmap *bitmap, struct page *page, int wait) 289935fe098SMike Snitzer { 290935fe098SMike Snitzer struct buffer_head *bh; 291935fe098SMike Snitzer 292935fe098SMike Snitzer if (bitmap->storage.file == NULL) { 293935fe098SMike Snitzer switch (write_sb_page(bitmap, page, wait)) { 294935fe098SMike Snitzer case -EINVAL: 295935fe098SMike Snitzer set_bit(BITMAP_WRITE_ERROR, &bitmap->flags); 296935fe098SMike Snitzer } 297935fe098SMike Snitzer } else { 298935fe098SMike Snitzer 299935fe098SMike Snitzer bh = page_buffers(page); 300935fe098SMike Snitzer 301935fe098SMike Snitzer while (bh && bh->b_blocknr) { 302935fe098SMike Snitzer atomic_inc(&bitmap->pending_writes); 303935fe098SMike Snitzer set_buffer_locked(bh); 304935fe098SMike Snitzer set_buffer_mapped(bh); 305935fe098SMike Snitzer submit_bh(REQ_OP_WRITE, REQ_SYNC, bh); 306935fe098SMike Snitzer bh = bh->b_this_page; 307935fe098SMike Snitzer } 308935fe098SMike Snitzer 309935fe098SMike Snitzer if (wait) 310935fe098SMike Snitzer wait_event(bitmap->write_wait, 311935fe098SMike Snitzer atomic_read(&bitmap->pending_writes)==0); 312935fe098SMike Snitzer } 313935fe098SMike Snitzer if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) 314e64e4018SAndy Shevchenko md_bitmap_file_kick(bitmap); 315935fe098SMike Snitzer } 316935fe098SMike Snitzer 317935fe098SMike Snitzer static void end_bitmap_write(struct buffer_head *bh, int uptodate) 318935fe098SMike Snitzer { 319935fe098SMike Snitzer struct bitmap *bitmap = bh->b_private; 320935fe098SMike Snitzer 321935fe098SMike Snitzer if (!uptodate) 322935fe098SMike Snitzer set_bit(BITMAP_WRITE_ERROR, &bitmap->flags); 323935fe098SMike Snitzer if (atomic_dec_and_test(&bitmap->pending_writes)) 324935fe098SMike Snitzer wake_up(&bitmap->write_wait); 325935fe098SMike Snitzer } 326935fe098SMike Snitzer 327935fe098SMike Snitzer /* copied from buffer.c */ 328935fe098SMike Snitzer static void 329935fe098SMike Snitzer __clear_page_buffers(struct page *page) 330935fe098SMike Snitzer { 331935fe098SMike Snitzer ClearPagePrivate(page); 332935fe098SMike Snitzer set_page_private(page, 0); 333935fe098SMike Snitzer put_page(page); 334935fe098SMike Snitzer } 335935fe098SMike Snitzer static void free_buffers(struct page *page) 336935fe098SMike Snitzer { 337935fe098SMike Snitzer struct buffer_head *bh; 338935fe098SMike Snitzer 339935fe098SMike Snitzer if (!PagePrivate(page)) 340935fe098SMike Snitzer return; 341935fe098SMike Snitzer 342935fe098SMike Snitzer bh = page_buffers(page); 343935fe098SMike Snitzer while (bh) { 344935fe098SMike Snitzer struct buffer_head *next = bh->b_this_page; 345935fe098SMike Snitzer free_buffer_head(bh); 346935fe098SMike Snitzer bh = next; 347935fe098SMike Snitzer } 348935fe098SMike Snitzer __clear_page_buffers(page); 349935fe098SMike Snitzer put_page(page); 350935fe098SMike Snitzer } 351935fe098SMike Snitzer 352935fe098SMike Snitzer /* read a page from a file. 353935fe098SMike Snitzer * We both read the page, and attach buffers to the page to record the 354935fe098SMike Snitzer * address of each block (using bmap). These addresses will be used 355935fe098SMike Snitzer * to write the block later, completely bypassing the filesystem. 356935fe098SMike Snitzer * This usage is similar to how swap files are handled, and allows us 357935fe098SMike Snitzer * to write to a file with no concerns of memory allocation failing. 358935fe098SMike Snitzer */ 359935fe098SMike Snitzer static int read_page(struct file *file, unsigned long index, 360935fe098SMike Snitzer struct bitmap *bitmap, 361935fe098SMike Snitzer unsigned long count, 362935fe098SMike Snitzer struct page *page) 363935fe098SMike Snitzer { 364935fe098SMike Snitzer int ret = 0; 365935fe098SMike Snitzer struct inode *inode = file_inode(file); 366935fe098SMike Snitzer struct buffer_head *bh; 367935fe098SMike Snitzer sector_t block; 368935fe098SMike Snitzer 369935fe098SMike Snitzer pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE, 370935fe098SMike Snitzer (unsigned long long)index << PAGE_SHIFT); 371935fe098SMike Snitzer 37247f521baSLinus Torvalds bh = alloc_page_buffers(page, 1<<inode->i_blkbits, false); 373935fe098SMike Snitzer if (!bh) { 374935fe098SMike Snitzer ret = -ENOMEM; 375935fe098SMike Snitzer goto out; 376935fe098SMike Snitzer } 377935fe098SMike Snitzer attach_page_buffers(page, bh); 378935fe098SMike Snitzer block = index << (PAGE_SHIFT - inode->i_blkbits); 379935fe098SMike Snitzer while (bh) { 380935fe098SMike Snitzer if (count == 0) 381935fe098SMike Snitzer bh->b_blocknr = 0; 382935fe098SMike Snitzer else { 383935fe098SMike Snitzer bh->b_blocknr = bmap(inode, block); 384935fe098SMike Snitzer if (bh->b_blocknr == 0) { 385935fe098SMike Snitzer /* Cannot use this file! */ 386935fe098SMike Snitzer ret = -EINVAL; 387935fe098SMike Snitzer goto out; 388935fe098SMike Snitzer } 389935fe098SMike Snitzer bh->b_bdev = inode->i_sb->s_bdev; 390935fe098SMike Snitzer if (count < (1<<inode->i_blkbits)) 391935fe098SMike Snitzer count = 0; 392935fe098SMike Snitzer else 393935fe098SMike Snitzer count -= (1<<inode->i_blkbits); 394935fe098SMike Snitzer 395935fe098SMike Snitzer bh->b_end_io = end_bitmap_write; 396935fe098SMike Snitzer bh->b_private = bitmap; 397935fe098SMike Snitzer atomic_inc(&bitmap->pending_writes); 398935fe098SMike Snitzer set_buffer_locked(bh); 399935fe098SMike Snitzer set_buffer_mapped(bh); 400935fe098SMike Snitzer submit_bh(REQ_OP_READ, 0, bh); 401935fe098SMike Snitzer } 402935fe098SMike Snitzer block++; 403935fe098SMike Snitzer bh = bh->b_this_page; 404935fe098SMike Snitzer } 405935fe098SMike Snitzer page->index = index; 406935fe098SMike Snitzer 407935fe098SMike Snitzer wait_event(bitmap->write_wait, 408935fe098SMike Snitzer atomic_read(&bitmap->pending_writes)==0); 409935fe098SMike Snitzer if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) 410935fe098SMike Snitzer ret = -EIO; 411935fe098SMike Snitzer out: 412935fe098SMike Snitzer if (ret) 413935fe098SMike Snitzer pr_err("md: bitmap read error: (%dB @ %llu): %d\n", 414935fe098SMike Snitzer (int)PAGE_SIZE, 415935fe098SMike Snitzer (unsigned long long)index << PAGE_SHIFT, 416935fe098SMike Snitzer ret); 417935fe098SMike Snitzer return ret; 418935fe098SMike Snitzer } 419935fe098SMike Snitzer 420935fe098SMike Snitzer /* 421935fe098SMike Snitzer * bitmap file superblock operations 422935fe098SMike Snitzer */ 423935fe098SMike Snitzer 424935fe098SMike Snitzer /* 425e64e4018SAndy Shevchenko * md_bitmap_wait_writes() should be called before writing any bitmap 426935fe098SMike Snitzer * blocks, to ensure previous writes, particularly from 427e64e4018SAndy Shevchenko * md_bitmap_daemon_work(), have completed. 428935fe098SMike Snitzer */ 429e64e4018SAndy Shevchenko static void md_bitmap_wait_writes(struct bitmap *bitmap) 430935fe098SMike Snitzer { 431935fe098SMike Snitzer if (bitmap->storage.file) 432935fe098SMike Snitzer wait_event(bitmap->write_wait, 433935fe098SMike Snitzer atomic_read(&bitmap->pending_writes)==0); 434935fe098SMike Snitzer else 435935fe098SMike Snitzer /* Note that we ignore the return value. The writes 436935fe098SMike Snitzer * might have failed, but that would just mean that 437935fe098SMike Snitzer * some bits which should be cleared haven't been, 438935fe098SMike Snitzer * which is safe. The relevant bitmap blocks will 439935fe098SMike Snitzer * probably get written again, but there is no great 440935fe098SMike Snitzer * loss if they aren't. 441935fe098SMike Snitzer */ 442935fe098SMike Snitzer md_super_wait(bitmap->mddev); 443935fe098SMike Snitzer } 444935fe098SMike Snitzer 445935fe098SMike Snitzer 446935fe098SMike Snitzer /* update the event counter and sync the superblock to disk */ 447e64e4018SAndy Shevchenko void md_bitmap_update_sb(struct bitmap *bitmap) 448935fe098SMike Snitzer { 449935fe098SMike Snitzer bitmap_super_t *sb; 450935fe098SMike Snitzer 451935fe098SMike Snitzer if (!bitmap || !bitmap->mddev) /* no bitmap for this array */ 452935fe098SMike Snitzer return; 453935fe098SMike Snitzer if (bitmap->mddev->bitmap_info.external) 454935fe098SMike Snitzer return; 455935fe098SMike Snitzer if (!bitmap->storage.sb_page) /* no superblock */ 456935fe098SMike Snitzer return; 457935fe098SMike Snitzer sb = kmap_atomic(bitmap->storage.sb_page); 458935fe098SMike Snitzer sb->events = cpu_to_le64(bitmap->mddev->events); 459935fe098SMike Snitzer if (bitmap->mddev->events < bitmap->events_cleared) 460935fe098SMike Snitzer /* rocking back to read-only */ 461935fe098SMike Snitzer bitmap->events_cleared = bitmap->mddev->events; 462935fe098SMike Snitzer sb->events_cleared = cpu_to_le64(bitmap->events_cleared); 46397f0eb9fSHou Tao /* 46497f0eb9fSHou Tao * clear BITMAP_WRITE_ERROR bit to protect against the case that 46597f0eb9fSHou Tao * a bitmap write error occurred but the later writes succeeded. 46697f0eb9fSHou Tao */ 46797f0eb9fSHou Tao sb->state = cpu_to_le32(bitmap->flags & ~BIT(BITMAP_WRITE_ERROR)); 468935fe098SMike Snitzer /* Just in case these have been changed via sysfs: */ 469935fe098SMike Snitzer sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ); 470935fe098SMike Snitzer sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind); 471935fe098SMike Snitzer /* This might have been changed by a reshape */ 472935fe098SMike Snitzer sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); 473935fe098SMike Snitzer sb->chunksize = cpu_to_le32(bitmap->mddev->bitmap_info.chunksize); 474935fe098SMike Snitzer sb->nodes = cpu_to_le32(bitmap->mddev->bitmap_info.nodes); 475935fe098SMike Snitzer sb->sectors_reserved = cpu_to_le32(bitmap->mddev-> 476935fe098SMike Snitzer bitmap_info.space); 477935fe098SMike Snitzer kunmap_atomic(sb); 478935fe098SMike Snitzer write_page(bitmap, bitmap->storage.sb_page, 1); 479935fe098SMike Snitzer } 480e64e4018SAndy Shevchenko EXPORT_SYMBOL(md_bitmap_update_sb); 481935fe098SMike Snitzer 482935fe098SMike Snitzer /* print out the bitmap file superblock */ 483e64e4018SAndy Shevchenko void md_bitmap_print_sb(struct bitmap *bitmap) 484935fe098SMike Snitzer { 485935fe098SMike Snitzer bitmap_super_t *sb; 486935fe098SMike Snitzer 487935fe098SMike Snitzer if (!bitmap || !bitmap->storage.sb_page) 488935fe098SMike Snitzer return; 489935fe098SMike Snitzer sb = kmap_atomic(bitmap->storage.sb_page); 490935fe098SMike Snitzer pr_debug("%s: bitmap file superblock:\n", bmname(bitmap)); 491935fe098SMike Snitzer pr_debug(" magic: %08x\n", le32_to_cpu(sb->magic)); 492935fe098SMike Snitzer pr_debug(" version: %d\n", le32_to_cpu(sb->version)); 493935fe098SMike Snitzer pr_debug(" uuid: %08x.%08x.%08x.%08x\n", 494c35403f8SChristoph Hellwig le32_to_cpu(*(__le32 *)(sb->uuid+0)), 495c35403f8SChristoph Hellwig le32_to_cpu(*(__le32 *)(sb->uuid+4)), 496c35403f8SChristoph Hellwig le32_to_cpu(*(__le32 *)(sb->uuid+8)), 497c35403f8SChristoph Hellwig le32_to_cpu(*(__le32 *)(sb->uuid+12))); 498935fe098SMike Snitzer pr_debug(" events: %llu\n", 499935fe098SMike Snitzer (unsigned long long) le64_to_cpu(sb->events)); 500935fe098SMike Snitzer pr_debug("events cleared: %llu\n", 501935fe098SMike Snitzer (unsigned long long) le64_to_cpu(sb->events_cleared)); 502935fe098SMike Snitzer pr_debug(" state: %08x\n", le32_to_cpu(sb->state)); 503935fe098SMike Snitzer pr_debug(" chunksize: %d B\n", le32_to_cpu(sb->chunksize)); 504935fe098SMike Snitzer pr_debug(" daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep)); 505935fe098SMike Snitzer pr_debug(" sync size: %llu KB\n", 506935fe098SMike Snitzer (unsigned long long)le64_to_cpu(sb->sync_size)/2); 507935fe098SMike Snitzer pr_debug("max write behind: %d\n", le32_to_cpu(sb->write_behind)); 508935fe098SMike Snitzer kunmap_atomic(sb); 509935fe098SMike Snitzer } 510935fe098SMike Snitzer 511935fe098SMike Snitzer /* 512935fe098SMike Snitzer * bitmap_new_disk_sb 513935fe098SMike Snitzer * @bitmap 514935fe098SMike Snitzer * 515935fe098SMike Snitzer * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb 516935fe098SMike Snitzer * reads and verifies the on-disk bitmap superblock and populates bitmap_info. 517935fe098SMike Snitzer * This function verifies 'bitmap_info' and populates the on-disk bitmap 518935fe098SMike Snitzer * structure, which is to be written to disk. 519935fe098SMike Snitzer * 520935fe098SMike Snitzer * Returns: 0 on success, -Exxx on error 521935fe098SMike Snitzer */ 522e64e4018SAndy Shevchenko static int md_bitmap_new_disk_sb(struct bitmap *bitmap) 523935fe098SMike Snitzer { 524935fe098SMike Snitzer bitmap_super_t *sb; 525935fe098SMike Snitzer unsigned long chunksize, daemon_sleep, write_behind; 526935fe098SMike Snitzer 527935fe098SMike Snitzer bitmap->storage.sb_page = alloc_page(GFP_KERNEL | __GFP_ZERO); 528935fe098SMike Snitzer if (bitmap->storage.sb_page == NULL) 529935fe098SMike Snitzer return -ENOMEM; 530935fe098SMike Snitzer bitmap->storage.sb_page->index = 0; 531935fe098SMike Snitzer 532935fe098SMike Snitzer sb = kmap_atomic(bitmap->storage.sb_page); 533935fe098SMike Snitzer 534935fe098SMike Snitzer sb->magic = cpu_to_le32(BITMAP_MAGIC); 535935fe098SMike Snitzer sb->version = cpu_to_le32(BITMAP_MAJOR_HI); 536935fe098SMike Snitzer 537935fe098SMike Snitzer chunksize = bitmap->mddev->bitmap_info.chunksize; 538935fe098SMike Snitzer BUG_ON(!chunksize); 539935fe098SMike Snitzer if (!is_power_of_2(chunksize)) { 540935fe098SMike Snitzer kunmap_atomic(sb); 541935fe098SMike Snitzer pr_warn("bitmap chunksize not a power of 2\n"); 542935fe098SMike Snitzer return -EINVAL; 543935fe098SMike Snitzer } 544935fe098SMike Snitzer sb->chunksize = cpu_to_le32(chunksize); 545935fe098SMike Snitzer 546935fe098SMike Snitzer daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep; 547935fe098SMike Snitzer if (!daemon_sleep || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) { 548935fe098SMike Snitzer pr_debug("Choosing daemon_sleep default (5 sec)\n"); 549935fe098SMike Snitzer daemon_sleep = 5 * HZ; 550935fe098SMike Snitzer } 551935fe098SMike Snitzer sb->daemon_sleep = cpu_to_le32(daemon_sleep); 552935fe098SMike Snitzer bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep; 553935fe098SMike Snitzer 554935fe098SMike Snitzer /* 555935fe098SMike Snitzer * FIXME: write_behind for RAID1. If not specified, what 556935fe098SMike Snitzer * is a good choice? We choose COUNTER_MAX / 2 arbitrarily. 557935fe098SMike Snitzer */ 558935fe098SMike Snitzer write_behind = bitmap->mddev->bitmap_info.max_write_behind; 559935fe098SMike Snitzer if (write_behind > COUNTER_MAX) 560935fe098SMike Snitzer write_behind = COUNTER_MAX / 2; 561935fe098SMike Snitzer sb->write_behind = cpu_to_le32(write_behind); 562935fe098SMike Snitzer bitmap->mddev->bitmap_info.max_write_behind = write_behind; 563935fe098SMike Snitzer 564935fe098SMike Snitzer /* keep the array size field of the bitmap superblock up to date */ 565935fe098SMike Snitzer sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); 566935fe098SMike Snitzer 567935fe098SMike Snitzer memcpy(sb->uuid, bitmap->mddev->uuid, 16); 568935fe098SMike Snitzer 569935fe098SMike Snitzer set_bit(BITMAP_STALE, &bitmap->flags); 570935fe098SMike Snitzer sb->state = cpu_to_le32(bitmap->flags); 571935fe098SMike Snitzer bitmap->events_cleared = bitmap->mddev->events; 572935fe098SMike Snitzer sb->events_cleared = cpu_to_le64(bitmap->mddev->events); 573935fe098SMike Snitzer bitmap->mddev->bitmap_info.nodes = 0; 574935fe098SMike Snitzer 575935fe098SMike Snitzer kunmap_atomic(sb); 576935fe098SMike Snitzer 577935fe098SMike Snitzer return 0; 578935fe098SMike Snitzer } 579935fe098SMike Snitzer 580935fe098SMike Snitzer /* read the superblock from the bitmap file and initialize some bitmap fields */ 581e64e4018SAndy Shevchenko static int md_bitmap_read_sb(struct bitmap *bitmap) 582935fe098SMike Snitzer { 583935fe098SMike Snitzer char *reason = NULL; 584935fe098SMike Snitzer bitmap_super_t *sb; 585935fe098SMike Snitzer unsigned long chunksize, daemon_sleep, write_behind; 586935fe098SMike Snitzer unsigned long long events; 587935fe098SMike Snitzer int nodes = 0; 588935fe098SMike Snitzer unsigned long sectors_reserved = 0; 589935fe098SMike Snitzer int err = -EINVAL; 590935fe098SMike Snitzer struct page *sb_page; 591935fe098SMike Snitzer loff_t offset = bitmap->mddev->bitmap_info.offset; 592935fe098SMike Snitzer 593935fe098SMike Snitzer if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) { 594935fe098SMike Snitzer chunksize = 128 * 1024 * 1024; 595935fe098SMike Snitzer daemon_sleep = 5 * HZ; 596935fe098SMike Snitzer write_behind = 0; 597935fe098SMike Snitzer set_bit(BITMAP_STALE, &bitmap->flags); 598935fe098SMike Snitzer err = 0; 599935fe098SMike Snitzer goto out_no_sb; 600935fe098SMike Snitzer } 601935fe098SMike Snitzer /* page 0 is the superblock, read it... */ 602935fe098SMike Snitzer sb_page = alloc_page(GFP_KERNEL); 603935fe098SMike Snitzer if (!sb_page) 604935fe098SMike Snitzer return -ENOMEM; 605935fe098SMike Snitzer bitmap->storage.sb_page = sb_page; 606935fe098SMike Snitzer 607935fe098SMike Snitzer re_read: 608935fe098SMike Snitzer /* If cluster_slot is set, the cluster is setup */ 609935fe098SMike Snitzer if (bitmap->cluster_slot >= 0) { 610935fe098SMike Snitzer sector_t bm_blocks = bitmap->mddev->resync_max_sectors; 611935fe098SMike Snitzer 612935fe098SMike Snitzer sector_div(bm_blocks, 613935fe098SMike Snitzer bitmap->mddev->bitmap_info.chunksize >> 9); 614935fe098SMike Snitzer /* bits to bytes */ 615935fe098SMike Snitzer bm_blocks = ((bm_blocks+7) >> 3) + sizeof(bitmap_super_t); 616935fe098SMike Snitzer /* to 4k blocks */ 617935fe098SMike Snitzer bm_blocks = DIV_ROUND_UP_SECTOR_T(bm_blocks, 4096); 618935fe098SMike Snitzer offset = bitmap->mddev->bitmap_info.offset + (bitmap->cluster_slot * (bm_blocks << 3)); 619935fe098SMike Snitzer pr_debug("%s:%d bm slot: %d offset: %llu\n", __func__, __LINE__, 620935fe098SMike Snitzer bitmap->cluster_slot, offset); 621935fe098SMike Snitzer } 622935fe098SMike Snitzer 623935fe098SMike Snitzer if (bitmap->storage.file) { 624935fe098SMike Snitzer loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host); 625935fe098SMike Snitzer int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize; 626935fe098SMike Snitzer 627935fe098SMike Snitzer err = read_page(bitmap->storage.file, 0, 628935fe098SMike Snitzer bitmap, bytes, sb_page); 629935fe098SMike Snitzer } else { 630935fe098SMike Snitzer err = read_sb_page(bitmap->mddev, 631935fe098SMike Snitzer offset, 632935fe098SMike Snitzer sb_page, 633935fe098SMike Snitzer 0, sizeof(bitmap_super_t)); 634935fe098SMike Snitzer } 635935fe098SMike Snitzer if (err) 636935fe098SMike Snitzer return err; 637935fe098SMike Snitzer 638935fe098SMike Snitzer err = -EINVAL; 639935fe098SMike Snitzer sb = kmap_atomic(sb_page); 640935fe098SMike Snitzer 641935fe098SMike Snitzer chunksize = le32_to_cpu(sb->chunksize); 642935fe098SMike Snitzer daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ; 643935fe098SMike Snitzer write_behind = le32_to_cpu(sb->write_behind); 644935fe098SMike Snitzer sectors_reserved = le32_to_cpu(sb->sectors_reserved); 645935fe098SMike Snitzer /* Setup nodes/clustername only if bitmap version is 646935fe098SMike Snitzer * cluster-compatible 647935fe098SMike Snitzer */ 648935fe098SMike Snitzer if (sb->version == cpu_to_le32(BITMAP_MAJOR_CLUSTERED)) { 649935fe098SMike Snitzer nodes = le32_to_cpu(sb->nodes); 650935fe098SMike Snitzer strlcpy(bitmap->mddev->bitmap_info.cluster_name, 651935fe098SMike Snitzer sb->cluster_name, 64); 652935fe098SMike Snitzer } 653935fe098SMike Snitzer 654935fe098SMike Snitzer /* verify that the bitmap-specific fields are valid */ 655935fe098SMike Snitzer if (sb->magic != cpu_to_le32(BITMAP_MAGIC)) 656935fe098SMike Snitzer reason = "bad magic"; 657935fe098SMike Snitzer else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO || 658935fe098SMike Snitzer le32_to_cpu(sb->version) > BITMAP_MAJOR_CLUSTERED) 659935fe098SMike Snitzer reason = "unrecognized superblock version"; 660935fe098SMike Snitzer else if (chunksize < 512) 661935fe098SMike Snitzer reason = "bitmap chunksize too small"; 662935fe098SMike Snitzer else if (!is_power_of_2(chunksize)) 663935fe098SMike Snitzer reason = "bitmap chunksize not a power of 2"; 664935fe098SMike Snitzer else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT) 665935fe098SMike Snitzer reason = "daemon sleep period out of range"; 666935fe098SMike Snitzer else if (write_behind > COUNTER_MAX) 667935fe098SMike Snitzer reason = "write-behind limit out of range (0 - 16383)"; 668935fe098SMike Snitzer if (reason) { 669935fe098SMike Snitzer pr_warn("%s: invalid bitmap file superblock: %s\n", 670935fe098SMike Snitzer bmname(bitmap), reason); 671935fe098SMike Snitzer goto out; 672935fe098SMike Snitzer } 673935fe098SMike Snitzer 674935fe098SMike Snitzer /* keep the array size field of the bitmap superblock up to date */ 675935fe098SMike Snitzer sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); 676935fe098SMike Snitzer 677935fe098SMike Snitzer if (bitmap->mddev->persistent) { 678935fe098SMike Snitzer /* 679935fe098SMike Snitzer * We have a persistent array superblock, so compare the 680935fe098SMike Snitzer * bitmap's UUID and event counter to the mddev's 681935fe098SMike Snitzer */ 682935fe098SMike Snitzer if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) { 683935fe098SMike Snitzer pr_warn("%s: bitmap superblock UUID mismatch\n", 684935fe098SMike Snitzer bmname(bitmap)); 685935fe098SMike Snitzer goto out; 686935fe098SMike Snitzer } 687935fe098SMike Snitzer events = le64_to_cpu(sb->events); 688935fe098SMike Snitzer if (!nodes && (events < bitmap->mddev->events)) { 689935fe098SMike Snitzer pr_warn("%s: bitmap file is out of date (%llu < %llu) -- forcing full recovery\n", 690935fe098SMike Snitzer bmname(bitmap), events, 691935fe098SMike Snitzer (unsigned long long) bitmap->mddev->events); 692935fe098SMike Snitzer set_bit(BITMAP_STALE, &bitmap->flags); 693935fe098SMike Snitzer } 694935fe098SMike Snitzer } 695935fe098SMike Snitzer 696935fe098SMike Snitzer /* assign fields using values from superblock */ 697935fe098SMike Snitzer bitmap->flags |= le32_to_cpu(sb->state); 698935fe098SMike Snitzer if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN) 699935fe098SMike Snitzer set_bit(BITMAP_HOSTENDIAN, &bitmap->flags); 700935fe098SMike Snitzer bitmap->events_cleared = le64_to_cpu(sb->events_cleared); 701935fe098SMike Snitzer strlcpy(bitmap->mddev->bitmap_info.cluster_name, sb->cluster_name, 64); 702935fe098SMike Snitzer err = 0; 703935fe098SMike Snitzer 704935fe098SMike Snitzer out: 705935fe098SMike Snitzer kunmap_atomic(sb); 706935fe098SMike Snitzer /* Assigning chunksize is required for "re_read" */ 707935fe098SMike Snitzer bitmap->mddev->bitmap_info.chunksize = chunksize; 708935fe098SMike Snitzer if (err == 0 && nodes && (bitmap->cluster_slot < 0)) { 709935fe098SMike Snitzer err = md_setup_cluster(bitmap->mddev, nodes); 710935fe098SMike Snitzer if (err) { 711935fe098SMike Snitzer pr_warn("%s: Could not setup cluster service (%d)\n", 712935fe098SMike Snitzer bmname(bitmap), err); 713935fe098SMike Snitzer goto out_no_sb; 714935fe098SMike Snitzer } 715935fe098SMike Snitzer bitmap->cluster_slot = md_cluster_ops->slot_number(bitmap->mddev); 716935fe098SMike Snitzer goto re_read; 717935fe098SMike Snitzer } 718935fe098SMike Snitzer 719935fe098SMike Snitzer 720935fe098SMike Snitzer out_no_sb: 721935fe098SMike Snitzer if (test_bit(BITMAP_STALE, &bitmap->flags)) 722935fe098SMike Snitzer bitmap->events_cleared = bitmap->mddev->events; 723935fe098SMike Snitzer bitmap->mddev->bitmap_info.chunksize = chunksize; 724935fe098SMike Snitzer bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep; 725935fe098SMike Snitzer bitmap->mddev->bitmap_info.max_write_behind = write_behind; 726935fe098SMike Snitzer bitmap->mddev->bitmap_info.nodes = nodes; 727935fe098SMike Snitzer if (bitmap->mddev->bitmap_info.space == 0 || 728935fe098SMike Snitzer bitmap->mddev->bitmap_info.space > sectors_reserved) 729935fe098SMike Snitzer bitmap->mddev->bitmap_info.space = sectors_reserved; 730935fe098SMike Snitzer if (err) { 731e64e4018SAndy Shevchenko md_bitmap_print_sb(bitmap); 732935fe098SMike Snitzer if (bitmap->cluster_slot < 0) 733935fe098SMike Snitzer md_cluster_stop(bitmap->mddev); 734935fe098SMike Snitzer } 735935fe098SMike Snitzer return err; 736935fe098SMike Snitzer } 737935fe098SMike Snitzer 738935fe098SMike Snitzer /* 739935fe098SMike Snitzer * general bitmap file operations 740935fe098SMike Snitzer */ 741935fe098SMike Snitzer 742935fe098SMike Snitzer /* 743935fe098SMike Snitzer * on-disk bitmap: 744935fe098SMike Snitzer * 745935fe098SMike Snitzer * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap 746935fe098SMike Snitzer * file a page at a time. There's a superblock at the start of the file. 747935fe098SMike Snitzer */ 748935fe098SMike Snitzer /* calculate the index of the page that contains this bit */ 749935fe098SMike Snitzer static inline unsigned long file_page_index(struct bitmap_storage *store, 750935fe098SMike Snitzer unsigned long chunk) 751935fe098SMike Snitzer { 752935fe098SMike Snitzer if (store->sb_page) 753935fe098SMike Snitzer chunk += sizeof(bitmap_super_t) << 3; 754935fe098SMike Snitzer return chunk >> PAGE_BIT_SHIFT; 755935fe098SMike Snitzer } 756935fe098SMike Snitzer 757935fe098SMike Snitzer /* calculate the (bit) offset of this bit within a page */ 758935fe098SMike Snitzer static inline unsigned long file_page_offset(struct bitmap_storage *store, 759935fe098SMike Snitzer unsigned long chunk) 760935fe098SMike Snitzer { 761935fe098SMike Snitzer if (store->sb_page) 762935fe098SMike Snitzer chunk += sizeof(bitmap_super_t) << 3; 763935fe098SMike Snitzer return chunk & (PAGE_BITS - 1); 764935fe098SMike Snitzer } 765935fe098SMike Snitzer 766935fe098SMike Snitzer /* 767935fe098SMike Snitzer * return a pointer to the page in the filemap that contains the given bit 768935fe098SMike Snitzer * 769935fe098SMike Snitzer */ 770935fe098SMike Snitzer static inline struct page *filemap_get_page(struct bitmap_storage *store, 771935fe098SMike Snitzer unsigned long chunk) 772935fe098SMike Snitzer { 773935fe098SMike Snitzer if (file_page_index(store, chunk) >= store->file_pages) 774935fe098SMike Snitzer return NULL; 775935fe098SMike Snitzer return store->filemap[file_page_index(store, chunk)]; 776935fe098SMike Snitzer } 777935fe098SMike Snitzer 778e64e4018SAndy Shevchenko static int md_bitmap_storage_alloc(struct bitmap_storage *store, 779935fe098SMike Snitzer unsigned long chunks, int with_super, 780935fe098SMike Snitzer int slot_number) 781935fe098SMike Snitzer { 782935fe098SMike Snitzer int pnum, offset = 0; 783935fe098SMike Snitzer unsigned long num_pages; 784935fe098SMike Snitzer unsigned long bytes; 785935fe098SMike Snitzer 786935fe098SMike Snitzer bytes = DIV_ROUND_UP(chunks, 8); 787935fe098SMike Snitzer if (with_super) 788935fe098SMike Snitzer bytes += sizeof(bitmap_super_t); 789935fe098SMike Snitzer 790935fe098SMike Snitzer num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE); 791935fe098SMike Snitzer offset = slot_number * num_pages; 792935fe098SMike Snitzer 7936da2ec56SKees Cook store->filemap = kmalloc_array(num_pages, sizeof(struct page *), 7946da2ec56SKees Cook GFP_KERNEL); 795935fe098SMike Snitzer if (!store->filemap) 796935fe098SMike Snitzer return -ENOMEM; 797935fe098SMike Snitzer 798935fe098SMike Snitzer if (with_super && !store->sb_page) { 799935fe098SMike Snitzer store->sb_page = alloc_page(GFP_KERNEL|__GFP_ZERO); 800935fe098SMike Snitzer if (store->sb_page == NULL) 801935fe098SMike Snitzer return -ENOMEM; 802935fe098SMike Snitzer } 803935fe098SMike Snitzer 804935fe098SMike Snitzer pnum = 0; 805935fe098SMike Snitzer if (store->sb_page) { 806935fe098SMike Snitzer store->filemap[0] = store->sb_page; 807935fe098SMike Snitzer pnum = 1; 808935fe098SMike Snitzer store->sb_page->index = offset; 809935fe098SMike Snitzer } 810935fe098SMike Snitzer 811935fe098SMike Snitzer for ( ; pnum < num_pages; pnum++) { 812935fe098SMike Snitzer store->filemap[pnum] = alloc_page(GFP_KERNEL|__GFP_ZERO); 813935fe098SMike Snitzer if (!store->filemap[pnum]) { 814935fe098SMike Snitzer store->file_pages = pnum; 815935fe098SMike Snitzer return -ENOMEM; 816935fe098SMike Snitzer } 817935fe098SMike Snitzer store->filemap[pnum]->index = pnum + offset; 818935fe098SMike Snitzer } 819935fe098SMike Snitzer store->file_pages = pnum; 820935fe098SMike Snitzer 821935fe098SMike Snitzer /* We need 4 bits per page, rounded up to a multiple 822935fe098SMike Snitzer * of sizeof(unsigned long) */ 823935fe098SMike Snitzer store->filemap_attr = kzalloc( 824935fe098SMike Snitzer roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)), 825935fe098SMike Snitzer GFP_KERNEL); 826935fe098SMike Snitzer if (!store->filemap_attr) 827935fe098SMike Snitzer return -ENOMEM; 828935fe098SMike Snitzer 829935fe098SMike Snitzer store->bytes = bytes; 830935fe098SMike Snitzer 831935fe098SMike Snitzer return 0; 832935fe098SMike Snitzer } 833935fe098SMike Snitzer 834e64e4018SAndy Shevchenko static void md_bitmap_file_unmap(struct bitmap_storage *store) 835935fe098SMike Snitzer { 836935fe098SMike Snitzer struct page **map, *sb_page; 837935fe098SMike Snitzer int pages; 838935fe098SMike Snitzer struct file *file; 839935fe098SMike Snitzer 840935fe098SMike Snitzer file = store->file; 841935fe098SMike Snitzer map = store->filemap; 842935fe098SMike Snitzer pages = store->file_pages; 843935fe098SMike Snitzer sb_page = store->sb_page; 844935fe098SMike Snitzer 845935fe098SMike Snitzer while (pages--) 846935fe098SMike Snitzer if (map[pages] != sb_page) /* 0 is sb_page, release it below */ 847935fe098SMike Snitzer free_buffers(map[pages]); 848935fe098SMike Snitzer kfree(map); 849935fe098SMike Snitzer kfree(store->filemap_attr); 850935fe098SMike Snitzer 851935fe098SMike Snitzer if (sb_page) 852935fe098SMike Snitzer free_buffers(sb_page); 853935fe098SMike Snitzer 854935fe098SMike Snitzer if (file) { 855935fe098SMike Snitzer struct inode *inode = file_inode(file); 856935fe098SMike Snitzer invalidate_mapping_pages(inode->i_mapping, 0, -1); 857935fe098SMike Snitzer fput(file); 858935fe098SMike Snitzer } 859935fe098SMike Snitzer } 860935fe098SMike Snitzer 861935fe098SMike Snitzer /* 862935fe098SMike Snitzer * bitmap_file_kick - if an error occurs while manipulating the bitmap file 863935fe098SMike Snitzer * then it is no longer reliable, so we stop using it and we mark the file 864935fe098SMike Snitzer * as failed in the superblock 865935fe098SMike Snitzer */ 866e64e4018SAndy Shevchenko static void md_bitmap_file_kick(struct bitmap *bitmap) 867935fe098SMike Snitzer { 868935fe098SMike Snitzer char *path, *ptr = NULL; 869935fe098SMike Snitzer 870935fe098SMike Snitzer if (!test_and_set_bit(BITMAP_STALE, &bitmap->flags)) { 871e64e4018SAndy Shevchenko md_bitmap_update_sb(bitmap); 872935fe098SMike Snitzer 873935fe098SMike Snitzer if (bitmap->storage.file) { 874935fe098SMike Snitzer path = kmalloc(PAGE_SIZE, GFP_KERNEL); 875935fe098SMike Snitzer if (path) 876935fe098SMike Snitzer ptr = file_path(bitmap->storage.file, 877935fe098SMike Snitzer path, PAGE_SIZE); 878935fe098SMike Snitzer 879935fe098SMike Snitzer pr_warn("%s: kicking failed bitmap file %s from array!\n", 880935fe098SMike Snitzer bmname(bitmap), IS_ERR(ptr) ? "" : ptr); 881935fe098SMike Snitzer 882935fe098SMike Snitzer kfree(path); 883935fe098SMike Snitzer } else 884935fe098SMike Snitzer pr_warn("%s: disabling internal bitmap due to errors\n", 885935fe098SMike Snitzer bmname(bitmap)); 886935fe098SMike Snitzer } 887935fe098SMike Snitzer } 888935fe098SMike Snitzer 889935fe098SMike Snitzer enum bitmap_page_attr { 890935fe098SMike Snitzer BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */ 891935fe098SMike Snitzer BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned. 892935fe098SMike Snitzer * i.e. counter is 1 or 2. */ 893935fe098SMike Snitzer BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */ 894935fe098SMike Snitzer }; 895935fe098SMike Snitzer 896935fe098SMike Snitzer static inline void set_page_attr(struct bitmap *bitmap, int pnum, 897935fe098SMike Snitzer enum bitmap_page_attr attr) 898935fe098SMike Snitzer { 899935fe098SMike Snitzer set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); 900935fe098SMike Snitzer } 901935fe098SMike Snitzer 902935fe098SMike Snitzer static inline void clear_page_attr(struct bitmap *bitmap, int pnum, 903935fe098SMike Snitzer enum bitmap_page_attr attr) 904935fe098SMike Snitzer { 905935fe098SMike Snitzer clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); 906935fe098SMike Snitzer } 907935fe098SMike Snitzer 908935fe098SMike Snitzer static inline int test_page_attr(struct bitmap *bitmap, int pnum, 909935fe098SMike Snitzer enum bitmap_page_attr attr) 910935fe098SMike Snitzer { 911935fe098SMike Snitzer return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); 912935fe098SMike Snitzer } 913935fe098SMike Snitzer 914935fe098SMike Snitzer static inline int test_and_clear_page_attr(struct bitmap *bitmap, int pnum, 915935fe098SMike Snitzer enum bitmap_page_attr attr) 916935fe098SMike Snitzer { 917935fe098SMike Snitzer return test_and_clear_bit((pnum<<2) + attr, 918935fe098SMike Snitzer bitmap->storage.filemap_attr); 919935fe098SMike Snitzer } 920935fe098SMike Snitzer /* 921935fe098SMike Snitzer * bitmap_file_set_bit -- called before performing a write to the md device 922935fe098SMike Snitzer * to set (and eventually sync) a particular bit in the bitmap file 923935fe098SMike Snitzer * 924935fe098SMike Snitzer * we set the bit immediately, then we record the page number so that 925935fe098SMike Snitzer * when an unplug occurs, we can flush the dirty pages out to disk 926935fe098SMike Snitzer */ 927e64e4018SAndy Shevchenko static void md_bitmap_file_set_bit(struct bitmap *bitmap, sector_t block) 928935fe098SMike Snitzer { 929935fe098SMike Snitzer unsigned long bit; 930935fe098SMike Snitzer struct page *page; 931935fe098SMike Snitzer void *kaddr; 932935fe098SMike Snitzer unsigned long chunk = block >> bitmap->counts.chunkshift; 933935fe098SMike Snitzer struct bitmap_storage *store = &bitmap->storage; 934935fe098SMike Snitzer unsigned long node_offset = 0; 935935fe098SMike Snitzer 936935fe098SMike Snitzer if (mddev_is_clustered(bitmap->mddev)) 937935fe098SMike Snitzer node_offset = bitmap->cluster_slot * store->file_pages; 938935fe098SMike Snitzer 939935fe098SMike Snitzer page = filemap_get_page(&bitmap->storage, chunk); 940935fe098SMike Snitzer if (!page) 941935fe098SMike Snitzer return; 942935fe098SMike Snitzer bit = file_page_offset(&bitmap->storage, chunk); 943935fe098SMike Snitzer 944935fe098SMike Snitzer /* set the bit */ 945935fe098SMike Snitzer kaddr = kmap_atomic(page); 946935fe098SMike Snitzer if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) 947935fe098SMike Snitzer set_bit(bit, kaddr); 948935fe098SMike Snitzer else 949935fe098SMike Snitzer set_bit_le(bit, kaddr); 950935fe098SMike Snitzer kunmap_atomic(kaddr); 951935fe098SMike Snitzer pr_debug("set file bit %lu page %lu\n", bit, page->index); 952935fe098SMike Snitzer /* record page number so it gets flushed to disk when unplug occurs */ 953935fe098SMike Snitzer set_page_attr(bitmap, page->index - node_offset, BITMAP_PAGE_DIRTY); 954935fe098SMike Snitzer } 955935fe098SMike Snitzer 956e64e4018SAndy Shevchenko static void md_bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block) 957935fe098SMike Snitzer { 958935fe098SMike Snitzer unsigned long bit; 959935fe098SMike Snitzer struct page *page; 960935fe098SMike Snitzer void *paddr; 961935fe098SMike Snitzer unsigned long chunk = block >> bitmap->counts.chunkshift; 962935fe098SMike Snitzer struct bitmap_storage *store = &bitmap->storage; 963935fe098SMike Snitzer unsigned long node_offset = 0; 964935fe098SMike Snitzer 965935fe098SMike Snitzer if (mddev_is_clustered(bitmap->mddev)) 966935fe098SMike Snitzer node_offset = bitmap->cluster_slot * store->file_pages; 967935fe098SMike Snitzer 968935fe098SMike Snitzer page = filemap_get_page(&bitmap->storage, chunk); 969935fe098SMike Snitzer if (!page) 970935fe098SMike Snitzer return; 971935fe098SMike Snitzer bit = file_page_offset(&bitmap->storage, chunk); 972935fe098SMike Snitzer paddr = kmap_atomic(page); 973935fe098SMike Snitzer if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) 974935fe098SMike Snitzer clear_bit(bit, paddr); 975935fe098SMike Snitzer else 976935fe098SMike Snitzer clear_bit_le(bit, paddr); 977935fe098SMike Snitzer kunmap_atomic(paddr); 978935fe098SMike Snitzer if (!test_page_attr(bitmap, page->index - node_offset, BITMAP_PAGE_NEEDWRITE)) { 979935fe098SMike Snitzer set_page_attr(bitmap, page->index - node_offset, BITMAP_PAGE_PENDING); 980935fe098SMike Snitzer bitmap->allclean = 0; 981935fe098SMike Snitzer } 982935fe098SMike Snitzer } 983935fe098SMike Snitzer 984e64e4018SAndy Shevchenko static int md_bitmap_file_test_bit(struct bitmap *bitmap, sector_t block) 985935fe098SMike Snitzer { 986935fe098SMike Snitzer unsigned long bit; 987935fe098SMike Snitzer struct page *page; 988935fe098SMike Snitzer void *paddr; 989935fe098SMike Snitzer unsigned long chunk = block >> bitmap->counts.chunkshift; 990935fe098SMike Snitzer int set = 0; 991935fe098SMike Snitzer 992935fe098SMike Snitzer page = filemap_get_page(&bitmap->storage, chunk); 993935fe098SMike Snitzer if (!page) 994935fe098SMike Snitzer return -EINVAL; 995935fe098SMike Snitzer bit = file_page_offset(&bitmap->storage, chunk); 996935fe098SMike Snitzer paddr = kmap_atomic(page); 997935fe098SMike Snitzer if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) 998935fe098SMike Snitzer set = test_bit(bit, paddr); 999935fe098SMike Snitzer else 1000935fe098SMike Snitzer set = test_bit_le(bit, paddr); 1001935fe098SMike Snitzer kunmap_atomic(paddr); 1002935fe098SMike Snitzer return set; 1003935fe098SMike Snitzer } 1004935fe098SMike Snitzer 1005935fe098SMike Snitzer 1006935fe098SMike Snitzer /* this gets called when the md device is ready to unplug its underlying 1007935fe098SMike Snitzer * (slave) device queues -- before we let any writes go down, we need to 1008935fe098SMike Snitzer * sync the dirty pages of the bitmap file to disk */ 1009e64e4018SAndy Shevchenko void md_bitmap_unplug(struct bitmap *bitmap) 1010935fe098SMike Snitzer { 1011935fe098SMike Snitzer unsigned long i; 1012935fe098SMike Snitzer int dirty, need_write; 1013935fe098SMike Snitzer int writing = 0; 1014935fe098SMike Snitzer 1015935fe098SMike Snitzer if (!bitmap || !bitmap->storage.filemap || 1016935fe098SMike Snitzer test_bit(BITMAP_STALE, &bitmap->flags)) 1017935fe098SMike Snitzer return; 1018935fe098SMike Snitzer 1019935fe098SMike Snitzer /* look at each page to see if there are any set bits that need to be 1020935fe098SMike Snitzer * flushed out to disk */ 1021935fe098SMike Snitzer for (i = 0; i < bitmap->storage.file_pages; i++) { 1022935fe098SMike Snitzer if (!bitmap->storage.filemap) 1023935fe098SMike Snitzer return; 1024935fe098SMike Snitzer dirty = test_and_clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY); 1025935fe098SMike Snitzer need_write = test_and_clear_page_attr(bitmap, i, 1026935fe098SMike Snitzer BITMAP_PAGE_NEEDWRITE); 1027935fe098SMike Snitzer if (dirty || need_write) { 1028935fe098SMike Snitzer if (!writing) { 1029e64e4018SAndy Shevchenko md_bitmap_wait_writes(bitmap); 1030935fe098SMike Snitzer if (bitmap->mddev->queue) 1031935fe098SMike Snitzer blk_add_trace_msg(bitmap->mddev->queue, 1032935fe098SMike Snitzer "md bitmap_unplug"); 1033935fe098SMike Snitzer } 1034935fe098SMike Snitzer clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING); 1035935fe098SMike Snitzer write_page(bitmap, bitmap->storage.filemap[i], 0); 1036935fe098SMike Snitzer writing = 1; 1037935fe098SMike Snitzer } 1038935fe098SMike Snitzer } 1039935fe098SMike Snitzer if (writing) 1040e64e4018SAndy Shevchenko md_bitmap_wait_writes(bitmap); 1041935fe098SMike Snitzer 1042935fe098SMike Snitzer if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) 1043e64e4018SAndy Shevchenko md_bitmap_file_kick(bitmap); 1044935fe098SMike Snitzer } 1045e64e4018SAndy Shevchenko EXPORT_SYMBOL(md_bitmap_unplug); 1046935fe098SMike Snitzer 1047e64e4018SAndy Shevchenko static void md_bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed); 1048935fe098SMike Snitzer /* * bitmap_init_from_disk -- called at bitmap_create time to initialize 1049935fe098SMike Snitzer * the in-memory bitmap from the on-disk bitmap -- also, sets up the 1050935fe098SMike Snitzer * memory mapping of the bitmap file 1051935fe098SMike Snitzer * Special cases: 1052935fe098SMike Snitzer * if there's no bitmap file, or if the bitmap file had been 1053935fe098SMike Snitzer * previously kicked from the array, we mark all the bits as 1054935fe098SMike Snitzer * 1's in order to cause a full resync. 1055935fe098SMike Snitzer * 1056935fe098SMike Snitzer * We ignore all bits for sectors that end earlier than 'start'. 1057935fe098SMike Snitzer * This is used when reading an out-of-date bitmap... 1058935fe098SMike Snitzer */ 1059e64e4018SAndy Shevchenko static int md_bitmap_init_from_disk(struct bitmap *bitmap, sector_t start) 1060935fe098SMike Snitzer { 1061935fe098SMike Snitzer unsigned long i, chunks, index, oldindex, bit, node_offset = 0; 1062935fe098SMike Snitzer struct page *page = NULL; 1063935fe098SMike Snitzer unsigned long bit_cnt = 0; 1064935fe098SMike Snitzer struct file *file; 1065935fe098SMike Snitzer unsigned long offset; 1066935fe098SMike Snitzer int outofdate; 1067935fe098SMike Snitzer int ret = -ENOSPC; 1068935fe098SMike Snitzer void *paddr; 1069935fe098SMike Snitzer struct bitmap_storage *store = &bitmap->storage; 1070935fe098SMike Snitzer 1071935fe098SMike Snitzer chunks = bitmap->counts.chunks; 1072935fe098SMike Snitzer file = store->file; 1073935fe098SMike Snitzer 1074935fe098SMike Snitzer if (!file && !bitmap->mddev->bitmap_info.offset) { 1075935fe098SMike Snitzer /* No permanent bitmap - fill with '1s'. */ 1076935fe098SMike Snitzer store->filemap = NULL; 1077935fe098SMike Snitzer store->file_pages = 0; 1078935fe098SMike Snitzer for (i = 0; i < chunks ; i++) { 1079935fe098SMike Snitzer /* if the disk bit is set, set the memory bit */ 1080935fe098SMike Snitzer int needed = ((sector_t)(i+1) << (bitmap->counts.chunkshift) 1081935fe098SMike Snitzer >= start); 1082e64e4018SAndy Shevchenko md_bitmap_set_memory_bits(bitmap, 1083935fe098SMike Snitzer (sector_t)i << bitmap->counts.chunkshift, 1084935fe098SMike Snitzer needed); 1085935fe098SMike Snitzer } 1086935fe098SMike Snitzer return 0; 1087935fe098SMike Snitzer } 1088935fe098SMike Snitzer 1089935fe098SMike Snitzer outofdate = test_bit(BITMAP_STALE, &bitmap->flags); 1090935fe098SMike Snitzer if (outofdate) 1091935fe098SMike Snitzer pr_warn("%s: bitmap file is out of date, doing full recovery\n", bmname(bitmap)); 1092935fe098SMike Snitzer 1093935fe098SMike Snitzer if (file && i_size_read(file->f_mapping->host) < store->bytes) { 1094935fe098SMike Snitzer pr_warn("%s: bitmap file too short %lu < %lu\n", 1095935fe098SMike Snitzer bmname(bitmap), 1096935fe098SMike Snitzer (unsigned long) i_size_read(file->f_mapping->host), 1097935fe098SMike Snitzer store->bytes); 1098935fe098SMike Snitzer goto err; 1099935fe098SMike Snitzer } 1100935fe098SMike Snitzer 1101935fe098SMike Snitzer oldindex = ~0L; 1102935fe098SMike Snitzer offset = 0; 1103935fe098SMike Snitzer if (!bitmap->mddev->bitmap_info.external) 1104935fe098SMike Snitzer offset = sizeof(bitmap_super_t); 1105935fe098SMike Snitzer 1106935fe098SMike Snitzer if (mddev_is_clustered(bitmap->mddev)) 1107935fe098SMike Snitzer node_offset = bitmap->cluster_slot * (DIV_ROUND_UP(store->bytes, PAGE_SIZE)); 1108935fe098SMike Snitzer 1109935fe098SMike Snitzer for (i = 0; i < chunks; i++) { 1110935fe098SMike Snitzer int b; 1111935fe098SMike Snitzer index = file_page_index(&bitmap->storage, i); 1112935fe098SMike Snitzer bit = file_page_offset(&bitmap->storage, i); 1113935fe098SMike Snitzer if (index != oldindex) { /* this is a new page, read it in */ 1114935fe098SMike Snitzer int count; 1115935fe098SMike Snitzer /* unmap the old page, we're done with it */ 1116935fe098SMike Snitzer if (index == store->file_pages-1) 1117935fe098SMike Snitzer count = store->bytes - index * PAGE_SIZE; 1118935fe098SMike Snitzer else 1119935fe098SMike Snitzer count = PAGE_SIZE; 1120935fe098SMike Snitzer page = store->filemap[index]; 1121935fe098SMike Snitzer if (file) 1122935fe098SMike Snitzer ret = read_page(file, index, bitmap, 1123935fe098SMike Snitzer count, page); 1124935fe098SMike Snitzer else 1125935fe098SMike Snitzer ret = read_sb_page( 1126935fe098SMike Snitzer bitmap->mddev, 1127935fe098SMike Snitzer bitmap->mddev->bitmap_info.offset, 1128935fe098SMike Snitzer page, 1129935fe098SMike Snitzer index + node_offset, count); 1130935fe098SMike Snitzer 1131935fe098SMike Snitzer if (ret) 1132935fe098SMike Snitzer goto err; 1133935fe098SMike Snitzer 1134935fe098SMike Snitzer oldindex = index; 1135935fe098SMike Snitzer 1136935fe098SMike Snitzer if (outofdate) { 1137935fe098SMike Snitzer /* 1138935fe098SMike Snitzer * if bitmap is out of date, dirty the 1139935fe098SMike Snitzer * whole page and write it out 1140935fe098SMike Snitzer */ 1141935fe098SMike Snitzer paddr = kmap_atomic(page); 1142935fe098SMike Snitzer memset(paddr + offset, 0xff, 1143935fe098SMike Snitzer PAGE_SIZE - offset); 1144935fe098SMike Snitzer kunmap_atomic(paddr); 1145935fe098SMike Snitzer write_page(bitmap, page, 1); 1146935fe098SMike Snitzer 1147935fe098SMike Snitzer ret = -EIO; 1148935fe098SMike Snitzer if (test_bit(BITMAP_WRITE_ERROR, 1149935fe098SMike Snitzer &bitmap->flags)) 1150935fe098SMike Snitzer goto err; 1151935fe098SMike Snitzer } 1152935fe098SMike Snitzer } 1153935fe098SMike Snitzer paddr = kmap_atomic(page); 1154935fe098SMike Snitzer if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) 1155935fe098SMike Snitzer b = test_bit(bit, paddr); 1156935fe098SMike Snitzer else 1157935fe098SMike Snitzer b = test_bit_le(bit, paddr); 1158935fe098SMike Snitzer kunmap_atomic(paddr); 1159935fe098SMike Snitzer if (b) { 1160935fe098SMike Snitzer /* if the disk bit is set, set the memory bit */ 1161935fe098SMike Snitzer int needed = ((sector_t)(i+1) << bitmap->counts.chunkshift 1162935fe098SMike Snitzer >= start); 1163e64e4018SAndy Shevchenko md_bitmap_set_memory_bits(bitmap, 1164935fe098SMike Snitzer (sector_t)i << bitmap->counts.chunkshift, 1165935fe098SMike Snitzer needed); 1166935fe098SMike Snitzer bit_cnt++; 1167935fe098SMike Snitzer } 1168935fe098SMike Snitzer offset = 0; 1169935fe098SMike Snitzer } 1170935fe098SMike Snitzer 1171935fe098SMike Snitzer pr_debug("%s: bitmap initialized from disk: read %lu pages, set %lu of %lu bits\n", 1172935fe098SMike Snitzer bmname(bitmap), store->file_pages, 1173935fe098SMike Snitzer bit_cnt, chunks); 1174935fe098SMike Snitzer 1175935fe098SMike Snitzer return 0; 1176935fe098SMike Snitzer 1177935fe098SMike Snitzer err: 1178935fe098SMike Snitzer pr_warn("%s: bitmap initialisation failed: %d\n", 1179935fe098SMike Snitzer bmname(bitmap), ret); 1180935fe098SMike Snitzer return ret; 1181935fe098SMike Snitzer } 1182935fe098SMike Snitzer 1183e64e4018SAndy Shevchenko void md_bitmap_write_all(struct bitmap *bitmap) 1184935fe098SMike Snitzer { 1185935fe098SMike Snitzer /* We don't actually write all bitmap blocks here, 1186935fe098SMike Snitzer * just flag them as needing to be written 1187935fe098SMike Snitzer */ 1188935fe098SMike Snitzer int i; 1189935fe098SMike Snitzer 1190935fe098SMike Snitzer if (!bitmap || !bitmap->storage.filemap) 1191935fe098SMike Snitzer return; 1192935fe098SMike Snitzer if (bitmap->storage.file) 1193935fe098SMike Snitzer /* Only one copy, so nothing needed */ 1194935fe098SMike Snitzer return; 1195935fe098SMike Snitzer 1196935fe098SMike Snitzer for (i = 0; i < bitmap->storage.file_pages; i++) 1197935fe098SMike Snitzer set_page_attr(bitmap, i, 1198935fe098SMike Snitzer BITMAP_PAGE_NEEDWRITE); 1199935fe098SMike Snitzer bitmap->allclean = 0; 1200935fe098SMike Snitzer } 1201935fe098SMike Snitzer 1202e64e4018SAndy Shevchenko static void md_bitmap_count_page(struct bitmap_counts *bitmap, 1203935fe098SMike Snitzer sector_t offset, int inc) 1204935fe098SMike Snitzer { 1205935fe098SMike Snitzer sector_t chunk = offset >> bitmap->chunkshift; 1206935fe098SMike Snitzer unsigned long page = chunk >> PAGE_COUNTER_SHIFT; 1207935fe098SMike Snitzer bitmap->bp[page].count += inc; 1208e64e4018SAndy Shevchenko md_bitmap_checkfree(bitmap, page); 1209935fe098SMike Snitzer } 1210935fe098SMike Snitzer 1211e64e4018SAndy Shevchenko static void md_bitmap_set_pending(struct bitmap_counts *bitmap, sector_t offset) 1212935fe098SMike Snitzer { 1213935fe098SMike Snitzer sector_t chunk = offset >> bitmap->chunkshift; 1214935fe098SMike Snitzer unsigned long page = chunk >> PAGE_COUNTER_SHIFT; 1215935fe098SMike Snitzer struct bitmap_page *bp = &bitmap->bp[page]; 1216935fe098SMike Snitzer 1217935fe098SMike Snitzer if (!bp->pending) 1218935fe098SMike Snitzer bp->pending = 1; 1219935fe098SMike Snitzer } 1220935fe098SMike Snitzer 1221e64e4018SAndy Shevchenko static bitmap_counter_t *md_bitmap_get_counter(struct bitmap_counts *bitmap, 1222935fe098SMike Snitzer sector_t offset, sector_t *blocks, 1223935fe098SMike Snitzer int create); 1224935fe098SMike Snitzer 1225935fe098SMike Snitzer /* 1226935fe098SMike Snitzer * bitmap daemon -- periodically wakes up to clean bits and flush pages 1227935fe098SMike Snitzer * out to disk 1228935fe098SMike Snitzer */ 1229935fe098SMike Snitzer 1230e64e4018SAndy Shevchenko void md_bitmap_daemon_work(struct mddev *mddev) 1231935fe098SMike Snitzer { 1232935fe098SMike Snitzer struct bitmap *bitmap; 1233935fe098SMike Snitzer unsigned long j; 1234935fe098SMike Snitzer unsigned long nextpage; 1235935fe098SMike Snitzer sector_t blocks; 1236935fe098SMike Snitzer struct bitmap_counts *counts; 1237935fe098SMike Snitzer 1238935fe098SMike Snitzer /* Use a mutex to guard daemon_work against 1239935fe098SMike Snitzer * bitmap_destroy. 1240935fe098SMike Snitzer */ 1241935fe098SMike Snitzer mutex_lock(&mddev->bitmap_info.mutex); 1242935fe098SMike Snitzer bitmap = mddev->bitmap; 1243935fe098SMike Snitzer if (bitmap == NULL) { 1244935fe098SMike Snitzer mutex_unlock(&mddev->bitmap_info.mutex); 1245935fe098SMike Snitzer return; 1246935fe098SMike Snitzer } 1247935fe098SMike Snitzer if (time_before(jiffies, bitmap->daemon_lastrun 1248935fe098SMike Snitzer + mddev->bitmap_info.daemon_sleep)) 1249935fe098SMike Snitzer goto done; 1250935fe098SMike Snitzer 1251935fe098SMike Snitzer bitmap->daemon_lastrun = jiffies; 1252935fe098SMike Snitzer if (bitmap->allclean) { 1253935fe098SMike Snitzer mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; 1254935fe098SMike Snitzer goto done; 1255935fe098SMike Snitzer } 1256935fe098SMike Snitzer bitmap->allclean = 1; 1257935fe098SMike Snitzer 1258935fe098SMike Snitzer if (bitmap->mddev->queue) 1259935fe098SMike Snitzer blk_add_trace_msg(bitmap->mddev->queue, 1260935fe098SMike Snitzer "md bitmap_daemon_work"); 1261935fe098SMike Snitzer 1262935fe098SMike Snitzer /* Any file-page which is PENDING now needs to be written. 1263935fe098SMike Snitzer * So set NEEDWRITE now, then after we make any last-minute changes 1264935fe098SMike Snitzer * we will write it. 1265935fe098SMike Snitzer */ 1266935fe098SMike Snitzer for (j = 0; j < bitmap->storage.file_pages; j++) 1267935fe098SMike Snitzer if (test_and_clear_page_attr(bitmap, j, 1268935fe098SMike Snitzer BITMAP_PAGE_PENDING)) 1269935fe098SMike Snitzer set_page_attr(bitmap, j, 1270935fe098SMike Snitzer BITMAP_PAGE_NEEDWRITE); 1271935fe098SMike Snitzer 1272935fe098SMike Snitzer if (bitmap->need_sync && 1273935fe098SMike Snitzer mddev->bitmap_info.external == 0) { 1274935fe098SMike Snitzer /* Arrange for superblock update as well as 1275935fe098SMike Snitzer * other changes */ 1276935fe098SMike Snitzer bitmap_super_t *sb; 1277935fe098SMike Snitzer bitmap->need_sync = 0; 1278935fe098SMike Snitzer if (bitmap->storage.filemap) { 1279935fe098SMike Snitzer sb = kmap_atomic(bitmap->storage.sb_page); 1280935fe098SMike Snitzer sb->events_cleared = 1281935fe098SMike Snitzer cpu_to_le64(bitmap->events_cleared); 1282935fe098SMike Snitzer kunmap_atomic(sb); 1283935fe098SMike Snitzer set_page_attr(bitmap, 0, 1284935fe098SMike Snitzer BITMAP_PAGE_NEEDWRITE); 1285935fe098SMike Snitzer } 1286935fe098SMike Snitzer } 1287935fe098SMike Snitzer /* Now look at the bitmap counters and if any are '2' or '1', 1288935fe098SMike Snitzer * decrement and handle accordingly. 1289935fe098SMike Snitzer */ 1290935fe098SMike Snitzer counts = &bitmap->counts; 1291935fe098SMike Snitzer spin_lock_irq(&counts->lock); 1292935fe098SMike Snitzer nextpage = 0; 1293935fe098SMike Snitzer for (j = 0; j < counts->chunks; j++) { 1294935fe098SMike Snitzer bitmap_counter_t *bmc; 1295935fe098SMike Snitzer sector_t block = (sector_t)j << counts->chunkshift; 1296935fe098SMike Snitzer 1297935fe098SMike Snitzer if (j == nextpage) { 1298935fe098SMike Snitzer nextpage += PAGE_COUNTER_RATIO; 1299935fe098SMike Snitzer if (!counts->bp[j >> PAGE_COUNTER_SHIFT].pending) { 1300935fe098SMike Snitzer j |= PAGE_COUNTER_MASK; 1301935fe098SMike Snitzer continue; 1302935fe098SMike Snitzer } 1303935fe098SMike Snitzer counts->bp[j >> PAGE_COUNTER_SHIFT].pending = 0; 1304935fe098SMike Snitzer } 1305935fe098SMike Snitzer 1306e64e4018SAndy Shevchenko bmc = md_bitmap_get_counter(counts, block, &blocks, 0); 1307935fe098SMike Snitzer if (!bmc) { 1308935fe098SMike Snitzer j |= PAGE_COUNTER_MASK; 1309935fe098SMike Snitzer continue; 1310935fe098SMike Snitzer } 1311935fe098SMike Snitzer if (*bmc == 1 && !bitmap->need_sync) { 1312935fe098SMike Snitzer /* We can clear the bit */ 1313935fe098SMike Snitzer *bmc = 0; 1314e64e4018SAndy Shevchenko md_bitmap_count_page(counts, block, -1); 1315e64e4018SAndy Shevchenko md_bitmap_file_clear_bit(bitmap, block); 1316935fe098SMike Snitzer } else if (*bmc && *bmc <= 2) { 1317935fe098SMike Snitzer *bmc = 1; 1318e64e4018SAndy Shevchenko md_bitmap_set_pending(counts, block); 1319935fe098SMike Snitzer bitmap->allclean = 0; 1320935fe098SMike Snitzer } 1321935fe098SMike Snitzer } 1322935fe098SMike Snitzer spin_unlock_irq(&counts->lock); 1323935fe098SMike Snitzer 1324e64e4018SAndy Shevchenko md_bitmap_wait_writes(bitmap); 1325935fe098SMike Snitzer /* Now start writeout on any page in NEEDWRITE that isn't DIRTY. 1326935fe098SMike Snitzer * DIRTY pages need to be written by bitmap_unplug so it can wait 1327935fe098SMike Snitzer * for them. 1328935fe098SMike Snitzer * If we find any DIRTY page we stop there and let bitmap_unplug 1329935fe098SMike Snitzer * handle all the rest. This is important in the case where 1330935fe098SMike Snitzer * the first blocking holds the superblock and it has been updated. 1331935fe098SMike Snitzer * We mustn't write any other blocks before the superblock. 1332935fe098SMike Snitzer */ 1333935fe098SMike Snitzer for (j = 0; 1334935fe098SMike Snitzer j < bitmap->storage.file_pages 1335935fe098SMike Snitzer && !test_bit(BITMAP_STALE, &bitmap->flags); 1336935fe098SMike Snitzer j++) { 1337935fe098SMike Snitzer if (test_page_attr(bitmap, j, 1338935fe098SMike Snitzer BITMAP_PAGE_DIRTY)) 1339935fe098SMike Snitzer /* bitmap_unplug will handle the rest */ 1340935fe098SMike Snitzer break; 1341935fe098SMike Snitzer if (test_and_clear_page_attr(bitmap, j, 1342935fe098SMike Snitzer BITMAP_PAGE_NEEDWRITE)) { 1343935fe098SMike Snitzer write_page(bitmap, bitmap->storage.filemap[j], 0); 1344935fe098SMike Snitzer } 1345935fe098SMike Snitzer } 1346935fe098SMike Snitzer 1347935fe098SMike Snitzer done: 1348935fe098SMike Snitzer if (bitmap->allclean == 0) 1349935fe098SMike Snitzer mddev->thread->timeout = 1350935fe098SMike Snitzer mddev->bitmap_info.daemon_sleep; 1351935fe098SMike Snitzer mutex_unlock(&mddev->bitmap_info.mutex); 1352935fe098SMike Snitzer } 1353935fe098SMike Snitzer 1354e64e4018SAndy Shevchenko static bitmap_counter_t *md_bitmap_get_counter(struct bitmap_counts *bitmap, 1355935fe098SMike Snitzer sector_t offset, sector_t *blocks, 1356935fe098SMike Snitzer int create) 1357935fe098SMike Snitzer __releases(bitmap->lock) 1358935fe098SMike Snitzer __acquires(bitmap->lock) 1359935fe098SMike Snitzer { 1360935fe098SMike Snitzer /* If 'create', we might release the lock and reclaim it. 1361935fe098SMike Snitzer * The lock must have been taken with interrupts enabled. 1362935fe098SMike Snitzer * If !create, we don't release the lock. 1363935fe098SMike Snitzer */ 1364935fe098SMike Snitzer sector_t chunk = offset >> bitmap->chunkshift; 1365935fe098SMike Snitzer unsigned long page = chunk >> PAGE_COUNTER_SHIFT; 1366935fe098SMike Snitzer unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT; 1367935fe098SMike Snitzer sector_t csize; 1368935fe098SMike Snitzer int err; 1369935fe098SMike Snitzer 1370e64e4018SAndy Shevchenko err = md_bitmap_checkpage(bitmap, page, create, 0); 1371935fe098SMike Snitzer 1372935fe098SMike Snitzer if (bitmap->bp[page].hijacked || 1373935fe098SMike Snitzer bitmap->bp[page].map == NULL) 1374935fe098SMike Snitzer csize = ((sector_t)1) << (bitmap->chunkshift + 1375935fe098SMike Snitzer PAGE_COUNTER_SHIFT - 1); 1376935fe098SMike Snitzer else 1377935fe098SMike Snitzer csize = ((sector_t)1) << bitmap->chunkshift; 1378935fe098SMike Snitzer *blocks = csize - (offset & (csize - 1)); 1379935fe098SMike Snitzer 1380935fe098SMike Snitzer if (err < 0) 1381935fe098SMike Snitzer return NULL; 1382935fe098SMike Snitzer 1383935fe098SMike Snitzer /* now locked ... */ 1384935fe098SMike Snitzer 1385935fe098SMike Snitzer if (bitmap->bp[page].hijacked) { /* hijacked pointer */ 1386935fe098SMike Snitzer /* should we use the first or second counter field 1387935fe098SMike Snitzer * of the hijacked pointer? */ 1388935fe098SMike Snitzer int hi = (pageoff > PAGE_COUNTER_MASK); 1389935fe098SMike Snitzer return &((bitmap_counter_t *) 1390935fe098SMike Snitzer &bitmap->bp[page].map)[hi]; 1391935fe098SMike Snitzer } else /* page is allocated */ 1392935fe098SMike Snitzer return (bitmap_counter_t *) 1393935fe098SMike Snitzer &(bitmap->bp[page].map[pageoff]); 1394935fe098SMike Snitzer } 1395935fe098SMike Snitzer 1396e64e4018SAndy Shevchenko int md_bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind) 1397935fe098SMike Snitzer { 1398935fe098SMike Snitzer if (!bitmap) 1399935fe098SMike Snitzer return 0; 1400935fe098SMike Snitzer 1401935fe098SMike Snitzer if (behind) { 1402935fe098SMike Snitzer int bw; 1403935fe098SMike Snitzer atomic_inc(&bitmap->behind_writes); 1404935fe098SMike Snitzer bw = atomic_read(&bitmap->behind_writes); 1405935fe098SMike Snitzer if (bw > bitmap->behind_writes_used) 1406935fe098SMike Snitzer bitmap->behind_writes_used = bw; 1407935fe098SMike Snitzer 1408935fe098SMike Snitzer pr_debug("inc write-behind count %d/%lu\n", 1409935fe098SMike Snitzer bw, bitmap->mddev->bitmap_info.max_write_behind); 1410935fe098SMike Snitzer } 1411935fe098SMike Snitzer 1412935fe098SMike Snitzer while (sectors) { 1413935fe098SMike Snitzer sector_t blocks; 1414935fe098SMike Snitzer bitmap_counter_t *bmc; 1415935fe098SMike Snitzer 1416935fe098SMike Snitzer spin_lock_irq(&bitmap->counts.lock); 1417e64e4018SAndy Shevchenko bmc = md_bitmap_get_counter(&bitmap->counts, offset, &blocks, 1); 1418935fe098SMike Snitzer if (!bmc) { 1419935fe098SMike Snitzer spin_unlock_irq(&bitmap->counts.lock); 1420935fe098SMike Snitzer return 0; 1421935fe098SMike Snitzer } 1422935fe098SMike Snitzer 1423935fe098SMike Snitzer if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) { 1424935fe098SMike Snitzer DEFINE_WAIT(__wait); 1425935fe098SMike Snitzer /* note that it is safe to do the prepare_to_wait 1426935fe098SMike Snitzer * after the test as long as we do it before dropping 1427935fe098SMike Snitzer * the spinlock. 1428935fe098SMike Snitzer */ 1429935fe098SMike Snitzer prepare_to_wait(&bitmap->overflow_wait, &__wait, 1430935fe098SMike Snitzer TASK_UNINTERRUPTIBLE); 1431935fe098SMike Snitzer spin_unlock_irq(&bitmap->counts.lock); 1432935fe098SMike Snitzer schedule(); 1433935fe098SMike Snitzer finish_wait(&bitmap->overflow_wait, &__wait); 1434935fe098SMike Snitzer continue; 1435935fe098SMike Snitzer } 1436935fe098SMike Snitzer 1437935fe098SMike Snitzer switch (*bmc) { 1438935fe098SMike Snitzer case 0: 1439e64e4018SAndy Shevchenko md_bitmap_file_set_bit(bitmap, offset); 1440e64e4018SAndy Shevchenko md_bitmap_count_page(&bitmap->counts, offset, 1); 1441935fe098SMike Snitzer /* fall through */ 1442935fe098SMike Snitzer case 1: 1443935fe098SMike Snitzer *bmc = 2; 1444935fe098SMike Snitzer } 1445935fe098SMike Snitzer 1446935fe098SMike Snitzer (*bmc)++; 1447935fe098SMike Snitzer 1448935fe098SMike Snitzer spin_unlock_irq(&bitmap->counts.lock); 1449935fe098SMike Snitzer 1450935fe098SMike Snitzer offset += blocks; 1451935fe098SMike Snitzer if (sectors > blocks) 1452935fe098SMike Snitzer sectors -= blocks; 1453935fe098SMike Snitzer else 1454935fe098SMike Snitzer sectors = 0; 1455935fe098SMike Snitzer } 1456935fe098SMike Snitzer return 0; 1457935fe098SMike Snitzer } 1458e64e4018SAndy Shevchenko EXPORT_SYMBOL(md_bitmap_startwrite); 1459935fe098SMike Snitzer 1460e64e4018SAndy Shevchenko void md_bitmap_endwrite(struct bitmap *bitmap, sector_t offset, 1461e64e4018SAndy Shevchenko unsigned long sectors, int success, int behind) 1462935fe098SMike Snitzer { 1463935fe098SMike Snitzer if (!bitmap) 1464935fe098SMike Snitzer return; 1465935fe098SMike Snitzer if (behind) { 1466935fe098SMike Snitzer if (atomic_dec_and_test(&bitmap->behind_writes)) 1467935fe098SMike Snitzer wake_up(&bitmap->behind_wait); 1468935fe098SMike Snitzer pr_debug("dec write-behind count %d/%lu\n", 1469935fe098SMike Snitzer atomic_read(&bitmap->behind_writes), 1470935fe098SMike Snitzer bitmap->mddev->bitmap_info.max_write_behind); 1471935fe098SMike Snitzer } 1472935fe098SMike Snitzer 1473935fe098SMike Snitzer while (sectors) { 1474935fe098SMike Snitzer sector_t blocks; 1475935fe098SMike Snitzer unsigned long flags; 1476935fe098SMike Snitzer bitmap_counter_t *bmc; 1477935fe098SMike Snitzer 1478935fe098SMike Snitzer spin_lock_irqsave(&bitmap->counts.lock, flags); 1479e64e4018SAndy Shevchenko bmc = md_bitmap_get_counter(&bitmap->counts, offset, &blocks, 0); 1480935fe098SMike Snitzer if (!bmc) { 1481935fe098SMike Snitzer spin_unlock_irqrestore(&bitmap->counts.lock, flags); 1482935fe098SMike Snitzer return; 1483935fe098SMike Snitzer } 1484935fe098SMike Snitzer 1485935fe098SMike Snitzer if (success && !bitmap->mddev->degraded && 1486935fe098SMike Snitzer bitmap->events_cleared < bitmap->mddev->events) { 1487935fe098SMike Snitzer bitmap->events_cleared = bitmap->mddev->events; 1488935fe098SMike Snitzer bitmap->need_sync = 1; 1489935fe098SMike Snitzer sysfs_notify_dirent_safe(bitmap->sysfs_can_clear); 1490935fe098SMike Snitzer } 1491935fe098SMike Snitzer 1492935fe098SMike Snitzer if (!success && !NEEDED(*bmc)) 1493935fe098SMike Snitzer *bmc |= NEEDED_MASK; 1494935fe098SMike Snitzer 1495935fe098SMike Snitzer if (COUNTER(*bmc) == COUNTER_MAX) 1496935fe098SMike Snitzer wake_up(&bitmap->overflow_wait); 1497935fe098SMike Snitzer 1498935fe098SMike Snitzer (*bmc)--; 1499935fe098SMike Snitzer if (*bmc <= 2) { 1500e64e4018SAndy Shevchenko md_bitmap_set_pending(&bitmap->counts, offset); 1501935fe098SMike Snitzer bitmap->allclean = 0; 1502935fe098SMike Snitzer } 1503935fe098SMike Snitzer spin_unlock_irqrestore(&bitmap->counts.lock, flags); 1504935fe098SMike Snitzer offset += blocks; 1505935fe098SMike Snitzer if (sectors > blocks) 1506935fe098SMike Snitzer sectors -= blocks; 1507935fe098SMike Snitzer else 1508935fe098SMike Snitzer sectors = 0; 1509935fe098SMike Snitzer } 1510935fe098SMike Snitzer } 1511e64e4018SAndy Shevchenko EXPORT_SYMBOL(md_bitmap_endwrite); 1512935fe098SMike Snitzer 1513935fe098SMike Snitzer static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, 1514935fe098SMike Snitzer int degraded) 1515935fe098SMike Snitzer { 1516935fe098SMike Snitzer bitmap_counter_t *bmc; 1517935fe098SMike Snitzer int rv; 1518935fe098SMike Snitzer if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */ 1519935fe098SMike Snitzer *blocks = 1024; 1520935fe098SMike Snitzer return 1; /* always resync if no bitmap */ 1521935fe098SMike Snitzer } 1522935fe098SMike Snitzer spin_lock_irq(&bitmap->counts.lock); 1523e64e4018SAndy Shevchenko bmc = md_bitmap_get_counter(&bitmap->counts, offset, blocks, 0); 1524935fe098SMike Snitzer rv = 0; 1525935fe098SMike Snitzer if (bmc) { 1526935fe098SMike Snitzer /* locked */ 1527935fe098SMike Snitzer if (RESYNC(*bmc)) 1528935fe098SMike Snitzer rv = 1; 1529935fe098SMike Snitzer else if (NEEDED(*bmc)) { 1530935fe098SMike Snitzer rv = 1; 1531935fe098SMike Snitzer if (!degraded) { /* don't set/clear bits if degraded */ 1532935fe098SMike Snitzer *bmc |= RESYNC_MASK; 1533935fe098SMike Snitzer *bmc &= ~NEEDED_MASK; 1534935fe098SMike Snitzer } 1535935fe098SMike Snitzer } 1536935fe098SMike Snitzer } 1537935fe098SMike Snitzer spin_unlock_irq(&bitmap->counts.lock); 1538935fe098SMike Snitzer return rv; 1539935fe098SMike Snitzer } 1540935fe098SMike Snitzer 1541e64e4018SAndy Shevchenko int md_bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, 1542935fe098SMike Snitzer int degraded) 1543935fe098SMike Snitzer { 1544935fe098SMike Snitzer /* bitmap_start_sync must always report on multiples of whole 1545935fe098SMike Snitzer * pages, otherwise resync (which is very PAGE_SIZE based) will 1546935fe098SMike Snitzer * get confused. 1547935fe098SMike Snitzer * So call __bitmap_start_sync repeatedly (if needed) until 1548935fe098SMike Snitzer * At least PAGE_SIZE>>9 blocks are covered. 1549935fe098SMike Snitzer * Return the 'or' of the result. 1550935fe098SMike Snitzer */ 1551935fe098SMike Snitzer int rv = 0; 1552935fe098SMike Snitzer sector_t blocks1; 1553935fe098SMike Snitzer 1554935fe098SMike Snitzer *blocks = 0; 1555935fe098SMike Snitzer while (*blocks < (PAGE_SIZE>>9)) { 1556935fe098SMike Snitzer rv |= __bitmap_start_sync(bitmap, offset, 1557935fe098SMike Snitzer &blocks1, degraded); 1558935fe098SMike Snitzer offset += blocks1; 1559935fe098SMike Snitzer *blocks += blocks1; 1560935fe098SMike Snitzer } 1561935fe098SMike Snitzer return rv; 1562935fe098SMike Snitzer } 1563e64e4018SAndy Shevchenko EXPORT_SYMBOL(md_bitmap_start_sync); 1564935fe098SMike Snitzer 1565e64e4018SAndy Shevchenko void md_bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted) 1566935fe098SMike Snitzer { 1567935fe098SMike Snitzer bitmap_counter_t *bmc; 1568935fe098SMike Snitzer unsigned long flags; 1569935fe098SMike Snitzer 1570935fe098SMike Snitzer if (bitmap == NULL) { 1571935fe098SMike Snitzer *blocks = 1024; 1572935fe098SMike Snitzer return; 1573935fe098SMike Snitzer } 1574935fe098SMike Snitzer spin_lock_irqsave(&bitmap->counts.lock, flags); 1575e64e4018SAndy Shevchenko bmc = md_bitmap_get_counter(&bitmap->counts, offset, blocks, 0); 1576935fe098SMike Snitzer if (bmc == NULL) 1577935fe098SMike Snitzer goto unlock; 1578935fe098SMike Snitzer /* locked */ 1579935fe098SMike Snitzer if (RESYNC(*bmc)) { 1580935fe098SMike Snitzer *bmc &= ~RESYNC_MASK; 1581935fe098SMike Snitzer 1582935fe098SMike Snitzer if (!NEEDED(*bmc) && aborted) 1583935fe098SMike Snitzer *bmc |= NEEDED_MASK; 1584935fe098SMike Snitzer else { 1585935fe098SMike Snitzer if (*bmc <= 2) { 1586e64e4018SAndy Shevchenko md_bitmap_set_pending(&bitmap->counts, offset); 1587935fe098SMike Snitzer bitmap->allclean = 0; 1588935fe098SMike Snitzer } 1589935fe098SMike Snitzer } 1590935fe098SMike Snitzer } 1591935fe098SMike Snitzer unlock: 1592935fe098SMike Snitzer spin_unlock_irqrestore(&bitmap->counts.lock, flags); 1593935fe098SMike Snitzer } 1594e64e4018SAndy Shevchenko EXPORT_SYMBOL(md_bitmap_end_sync); 1595935fe098SMike Snitzer 1596e64e4018SAndy Shevchenko void md_bitmap_close_sync(struct bitmap *bitmap) 1597935fe098SMike Snitzer { 1598935fe098SMike Snitzer /* Sync has finished, and any bitmap chunks that weren't synced 1599935fe098SMike Snitzer * properly have been aborted. It remains to us to clear the 1600935fe098SMike Snitzer * RESYNC bit wherever it is still on 1601935fe098SMike Snitzer */ 1602935fe098SMike Snitzer sector_t sector = 0; 1603935fe098SMike Snitzer sector_t blocks; 1604935fe098SMike Snitzer if (!bitmap) 1605935fe098SMike Snitzer return; 1606935fe098SMike Snitzer while (sector < bitmap->mddev->resync_max_sectors) { 1607e64e4018SAndy Shevchenko md_bitmap_end_sync(bitmap, sector, &blocks, 0); 1608935fe098SMike Snitzer sector += blocks; 1609935fe098SMike Snitzer } 1610935fe098SMike Snitzer } 1611e64e4018SAndy Shevchenko EXPORT_SYMBOL(md_bitmap_close_sync); 1612935fe098SMike Snitzer 1613e64e4018SAndy Shevchenko void md_bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector, bool force) 1614935fe098SMike Snitzer { 1615935fe098SMike Snitzer sector_t s = 0; 1616935fe098SMike Snitzer sector_t blocks; 1617935fe098SMike Snitzer 1618935fe098SMike Snitzer if (!bitmap) 1619935fe098SMike Snitzer return; 1620935fe098SMike Snitzer if (sector == 0) { 1621935fe098SMike Snitzer bitmap->last_end_sync = jiffies; 1622935fe098SMike Snitzer return; 1623935fe098SMike Snitzer } 1624935fe098SMike Snitzer if (!force && time_before(jiffies, (bitmap->last_end_sync 1625935fe098SMike Snitzer + bitmap->mddev->bitmap_info.daemon_sleep))) 1626935fe098SMike Snitzer return; 1627935fe098SMike Snitzer wait_event(bitmap->mddev->recovery_wait, 1628935fe098SMike Snitzer atomic_read(&bitmap->mddev->recovery_active) == 0); 1629935fe098SMike Snitzer 1630935fe098SMike Snitzer bitmap->mddev->curr_resync_completed = sector; 1631935fe098SMike Snitzer set_bit(MD_SB_CHANGE_CLEAN, &bitmap->mddev->sb_flags); 1632935fe098SMike Snitzer sector &= ~((1ULL << bitmap->counts.chunkshift) - 1); 1633935fe098SMike Snitzer s = 0; 1634935fe098SMike Snitzer while (s < sector && s < bitmap->mddev->resync_max_sectors) { 1635e64e4018SAndy Shevchenko md_bitmap_end_sync(bitmap, s, &blocks, 0); 1636935fe098SMike Snitzer s += blocks; 1637935fe098SMike Snitzer } 1638935fe098SMike Snitzer bitmap->last_end_sync = jiffies; 1639935fe098SMike Snitzer sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed"); 1640935fe098SMike Snitzer } 1641e64e4018SAndy Shevchenko EXPORT_SYMBOL(md_bitmap_cond_end_sync); 1642935fe098SMike Snitzer 1643e64e4018SAndy Shevchenko void md_bitmap_sync_with_cluster(struct mddev *mddev, 1644935fe098SMike Snitzer sector_t old_lo, sector_t old_hi, 1645935fe098SMike Snitzer sector_t new_lo, sector_t new_hi) 1646935fe098SMike Snitzer { 1647935fe098SMike Snitzer struct bitmap *bitmap = mddev->bitmap; 1648935fe098SMike Snitzer sector_t sector, blocks = 0; 1649935fe098SMike Snitzer 1650935fe098SMike Snitzer for (sector = old_lo; sector < new_lo; ) { 1651e64e4018SAndy Shevchenko md_bitmap_end_sync(bitmap, sector, &blocks, 0); 1652935fe098SMike Snitzer sector += blocks; 1653935fe098SMike Snitzer } 1654935fe098SMike Snitzer WARN((blocks > new_lo) && old_lo, "alignment is not correct for lo\n"); 1655935fe098SMike Snitzer 1656935fe098SMike Snitzer for (sector = old_hi; sector < new_hi; ) { 1657e64e4018SAndy Shevchenko md_bitmap_start_sync(bitmap, sector, &blocks, 0); 1658935fe098SMike Snitzer sector += blocks; 1659935fe098SMike Snitzer } 1660935fe098SMike Snitzer WARN((blocks > new_hi) && old_hi, "alignment is not correct for hi\n"); 1661935fe098SMike Snitzer } 1662e64e4018SAndy Shevchenko EXPORT_SYMBOL(md_bitmap_sync_with_cluster); 1663935fe098SMike Snitzer 1664e64e4018SAndy Shevchenko static void md_bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed) 1665935fe098SMike Snitzer { 1666935fe098SMike Snitzer /* For each chunk covered by any of these sectors, set the 1667935fe098SMike Snitzer * counter to 2 and possibly set resync_needed. They should all 1668935fe098SMike Snitzer * be 0 at this point 1669935fe098SMike Snitzer */ 1670935fe098SMike Snitzer 1671935fe098SMike Snitzer sector_t secs; 1672935fe098SMike Snitzer bitmap_counter_t *bmc; 1673935fe098SMike Snitzer spin_lock_irq(&bitmap->counts.lock); 1674e64e4018SAndy Shevchenko bmc = md_bitmap_get_counter(&bitmap->counts, offset, &secs, 1); 1675935fe098SMike Snitzer if (!bmc) { 1676935fe098SMike Snitzer spin_unlock_irq(&bitmap->counts.lock); 1677935fe098SMike Snitzer return; 1678935fe098SMike Snitzer } 1679935fe098SMike Snitzer if (!*bmc) { 1680935fe098SMike Snitzer *bmc = 2; 1681e64e4018SAndy Shevchenko md_bitmap_count_page(&bitmap->counts, offset, 1); 1682e64e4018SAndy Shevchenko md_bitmap_set_pending(&bitmap->counts, offset); 1683935fe098SMike Snitzer bitmap->allclean = 0; 1684935fe098SMike Snitzer } 1685935fe098SMike Snitzer if (needed) 1686935fe098SMike Snitzer *bmc |= NEEDED_MASK; 1687935fe098SMike Snitzer spin_unlock_irq(&bitmap->counts.lock); 1688935fe098SMike Snitzer } 1689935fe098SMike Snitzer 1690935fe098SMike Snitzer /* dirty the memory and file bits for bitmap chunks "s" to "e" */ 1691e64e4018SAndy Shevchenko void md_bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e) 1692935fe098SMike Snitzer { 1693935fe098SMike Snitzer unsigned long chunk; 1694935fe098SMike Snitzer 1695935fe098SMike Snitzer for (chunk = s; chunk <= e; chunk++) { 1696935fe098SMike Snitzer sector_t sec = (sector_t)chunk << bitmap->counts.chunkshift; 1697e64e4018SAndy Shevchenko md_bitmap_set_memory_bits(bitmap, sec, 1); 1698e64e4018SAndy Shevchenko md_bitmap_file_set_bit(bitmap, sec); 1699935fe098SMike Snitzer if (sec < bitmap->mddev->recovery_cp) 1700935fe098SMike Snitzer /* We are asserting that the array is dirty, 1701935fe098SMike Snitzer * so move the recovery_cp address back so 1702935fe098SMike Snitzer * that it is obvious that it is dirty 1703935fe098SMike Snitzer */ 1704935fe098SMike Snitzer bitmap->mddev->recovery_cp = sec; 1705935fe098SMike Snitzer } 1706935fe098SMike Snitzer } 1707935fe098SMike Snitzer 1708935fe098SMike Snitzer /* 1709935fe098SMike Snitzer * flush out any pending updates 1710935fe098SMike Snitzer */ 1711e64e4018SAndy Shevchenko void md_bitmap_flush(struct mddev *mddev) 1712935fe098SMike Snitzer { 1713935fe098SMike Snitzer struct bitmap *bitmap = mddev->bitmap; 1714935fe098SMike Snitzer long sleep; 1715935fe098SMike Snitzer 1716935fe098SMike Snitzer if (!bitmap) /* there was no bitmap */ 1717935fe098SMike Snitzer return; 1718935fe098SMike Snitzer 1719935fe098SMike Snitzer /* run the daemon_work three time to ensure everything is flushed 1720935fe098SMike Snitzer * that can be 1721935fe098SMike Snitzer */ 1722935fe098SMike Snitzer sleep = mddev->bitmap_info.daemon_sleep * 2; 1723935fe098SMike Snitzer bitmap->daemon_lastrun -= sleep; 1724e64e4018SAndy Shevchenko md_bitmap_daemon_work(mddev); 1725935fe098SMike Snitzer bitmap->daemon_lastrun -= sleep; 1726e64e4018SAndy Shevchenko md_bitmap_daemon_work(mddev); 1727935fe098SMike Snitzer bitmap->daemon_lastrun -= sleep; 1728e64e4018SAndy Shevchenko md_bitmap_daemon_work(mddev); 1729e64e4018SAndy Shevchenko md_bitmap_update_sb(bitmap); 1730935fe098SMike Snitzer } 1731935fe098SMike Snitzer 1732935fe098SMike Snitzer /* 1733935fe098SMike Snitzer * free memory that was allocated 1734935fe098SMike Snitzer */ 1735e64e4018SAndy Shevchenko void md_bitmap_free(struct bitmap *bitmap) 1736935fe098SMike Snitzer { 1737935fe098SMike Snitzer unsigned long k, pages; 1738935fe098SMike Snitzer struct bitmap_page *bp; 1739935fe098SMike Snitzer 1740935fe098SMike Snitzer if (!bitmap) /* there was no bitmap */ 1741935fe098SMike Snitzer return; 1742935fe098SMike Snitzer 1743935fe098SMike Snitzer if (bitmap->sysfs_can_clear) 1744935fe098SMike Snitzer sysfs_put(bitmap->sysfs_can_clear); 1745935fe098SMike Snitzer 1746935fe098SMike Snitzer if (mddev_is_clustered(bitmap->mddev) && bitmap->mddev->cluster_info && 1747935fe098SMike Snitzer bitmap->cluster_slot == md_cluster_ops->slot_number(bitmap->mddev)) 1748935fe098SMike Snitzer md_cluster_stop(bitmap->mddev); 1749935fe098SMike Snitzer 1750935fe098SMike Snitzer /* Shouldn't be needed - but just in case.... */ 1751935fe098SMike Snitzer wait_event(bitmap->write_wait, 1752935fe098SMike Snitzer atomic_read(&bitmap->pending_writes) == 0); 1753935fe098SMike Snitzer 1754935fe098SMike Snitzer /* release the bitmap file */ 1755e64e4018SAndy Shevchenko md_bitmap_file_unmap(&bitmap->storage); 1756935fe098SMike Snitzer 1757935fe098SMike Snitzer bp = bitmap->counts.bp; 1758935fe098SMike Snitzer pages = bitmap->counts.pages; 1759935fe098SMike Snitzer 1760935fe098SMike Snitzer /* free all allocated memory */ 1761935fe098SMike Snitzer 1762935fe098SMike Snitzer if (bp) /* deallocate the page memory */ 1763935fe098SMike Snitzer for (k = 0; k < pages; k++) 1764935fe098SMike Snitzer if (bp[k].map && !bp[k].hijacked) 1765935fe098SMike Snitzer kfree(bp[k].map); 1766935fe098SMike Snitzer kfree(bp); 1767935fe098SMike Snitzer kfree(bitmap); 1768935fe098SMike Snitzer } 1769e64e4018SAndy Shevchenko EXPORT_SYMBOL(md_bitmap_free); 1770935fe098SMike Snitzer 1771e64e4018SAndy Shevchenko void md_bitmap_wait_behind_writes(struct mddev *mddev) 1772935fe098SMike Snitzer { 1773935fe098SMike Snitzer struct bitmap *bitmap = mddev->bitmap; 1774935fe098SMike Snitzer 1775935fe098SMike Snitzer /* wait for behind writes to complete */ 1776935fe098SMike Snitzer if (bitmap && atomic_read(&bitmap->behind_writes) > 0) { 1777935fe098SMike Snitzer pr_debug("md:%s: behind writes in progress - waiting to stop.\n", 1778935fe098SMike Snitzer mdname(mddev)); 1779935fe098SMike Snitzer /* need to kick something here to make sure I/O goes? */ 1780935fe098SMike Snitzer wait_event(bitmap->behind_wait, 1781935fe098SMike Snitzer atomic_read(&bitmap->behind_writes) == 0); 1782935fe098SMike Snitzer } 1783935fe098SMike Snitzer } 1784935fe098SMike Snitzer 1785e64e4018SAndy Shevchenko void md_bitmap_destroy(struct mddev *mddev) 1786935fe098SMike Snitzer { 1787935fe098SMike Snitzer struct bitmap *bitmap = mddev->bitmap; 1788935fe098SMike Snitzer 1789935fe098SMike Snitzer if (!bitmap) /* there was no bitmap */ 1790935fe098SMike Snitzer return; 1791935fe098SMike Snitzer 1792e64e4018SAndy Shevchenko md_bitmap_wait_behind_writes(mddev); 1793935fe098SMike Snitzer 1794935fe098SMike Snitzer mutex_lock(&mddev->bitmap_info.mutex); 1795935fe098SMike Snitzer spin_lock(&mddev->lock); 1796935fe098SMike Snitzer mddev->bitmap = NULL; /* disconnect from the md device */ 1797935fe098SMike Snitzer spin_unlock(&mddev->lock); 1798935fe098SMike Snitzer mutex_unlock(&mddev->bitmap_info.mutex); 1799935fe098SMike Snitzer if (mddev->thread) 1800935fe098SMike Snitzer mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; 1801935fe098SMike Snitzer 1802e64e4018SAndy Shevchenko md_bitmap_free(bitmap); 1803935fe098SMike Snitzer } 1804935fe098SMike Snitzer 1805935fe098SMike Snitzer /* 1806935fe098SMike Snitzer * initialize the bitmap structure 1807935fe098SMike Snitzer * if this returns an error, bitmap_destroy must be called to do clean up 1808935fe098SMike Snitzer * once mddev->bitmap is set 1809935fe098SMike Snitzer */ 1810e64e4018SAndy Shevchenko struct bitmap *md_bitmap_create(struct mddev *mddev, int slot) 1811935fe098SMike Snitzer { 1812935fe098SMike Snitzer struct bitmap *bitmap; 1813935fe098SMike Snitzer sector_t blocks = mddev->resync_max_sectors; 1814935fe098SMike Snitzer struct file *file = mddev->bitmap_info.file; 1815935fe098SMike Snitzer int err; 1816935fe098SMike Snitzer struct kernfs_node *bm = NULL; 1817935fe098SMike Snitzer 1818935fe098SMike Snitzer BUILD_BUG_ON(sizeof(bitmap_super_t) != 256); 1819935fe098SMike Snitzer 1820935fe098SMike Snitzer BUG_ON(file && mddev->bitmap_info.offset); 1821935fe098SMike Snitzer 1822230b55faSNeilBrown if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) { 1823230b55faSNeilBrown pr_notice("md/raid:%s: array with journal cannot have bitmap\n", 1824230b55faSNeilBrown mdname(mddev)); 1825230b55faSNeilBrown return ERR_PTR(-EBUSY); 1826230b55faSNeilBrown } 1827230b55faSNeilBrown 1828935fe098SMike Snitzer bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL); 1829935fe098SMike Snitzer if (!bitmap) 1830935fe098SMike Snitzer return ERR_PTR(-ENOMEM); 1831935fe098SMike Snitzer 1832935fe098SMike Snitzer spin_lock_init(&bitmap->counts.lock); 1833935fe098SMike Snitzer atomic_set(&bitmap->pending_writes, 0); 1834935fe098SMike Snitzer init_waitqueue_head(&bitmap->write_wait); 1835935fe098SMike Snitzer init_waitqueue_head(&bitmap->overflow_wait); 1836935fe098SMike Snitzer init_waitqueue_head(&bitmap->behind_wait); 1837935fe098SMike Snitzer 1838935fe098SMike Snitzer bitmap->mddev = mddev; 1839935fe098SMike Snitzer bitmap->cluster_slot = slot; 1840935fe098SMike Snitzer 1841935fe098SMike Snitzer if (mddev->kobj.sd) 1842935fe098SMike Snitzer bm = sysfs_get_dirent(mddev->kobj.sd, "bitmap"); 1843935fe098SMike Snitzer if (bm) { 1844935fe098SMike Snitzer bitmap->sysfs_can_clear = sysfs_get_dirent(bm, "can_clear"); 1845935fe098SMike Snitzer sysfs_put(bm); 1846935fe098SMike Snitzer } else 1847935fe098SMike Snitzer bitmap->sysfs_can_clear = NULL; 1848935fe098SMike Snitzer 1849935fe098SMike Snitzer bitmap->storage.file = file; 1850935fe098SMike Snitzer if (file) { 1851935fe098SMike Snitzer get_file(file); 1852935fe098SMike Snitzer /* As future accesses to this file will use bmap, 1853935fe098SMike Snitzer * and bypass the page cache, we must sync the file 1854935fe098SMike Snitzer * first. 1855935fe098SMike Snitzer */ 1856935fe098SMike Snitzer vfs_fsync(file, 1); 1857935fe098SMike Snitzer } 1858935fe098SMike Snitzer /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */ 1859935fe098SMike Snitzer if (!mddev->bitmap_info.external) { 1860935fe098SMike Snitzer /* 1861935fe098SMike Snitzer * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is 1862935fe098SMike Snitzer * instructing us to create a new on-disk bitmap instance. 1863935fe098SMike Snitzer */ 1864935fe098SMike Snitzer if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags)) 1865e64e4018SAndy Shevchenko err = md_bitmap_new_disk_sb(bitmap); 1866935fe098SMike Snitzer else 1867e64e4018SAndy Shevchenko err = md_bitmap_read_sb(bitmap); 1868935fe098SMike Snitzer } else { 1869935fe098SMike Snitzer err = 0; 1870935fe098SMike Snitzer if (mddev->bitmap_info.chunksize == 0 || 1871935fe098SMike Snitzer mddev->bitmap_info.daemon_sleep == 0) 1872935fe098SMike Snitzer /* chunksize and time_base need to be 1873935fe098SMike Snitzer * set first. */ 1874935fe098SMike Snitzer err = -EINVAL; 1875935fe098SMike Snitzer } 1876935fe098SMike Snitzer if (err) 1877935fe098SMike Snitzer goto error; 1878935fe098SMike Snitzer 1879935fe098SMike Snitzer bitmap->daemon_lastrun = jiffies; 1880e64e4018SAndy Shevchenko err = md_bitmap_resize(bitmap, blocks, mddev->bitmap_info.chunksize, 1); 1881935fe098SMike Snitzer if (err) 1882935fe098SMike Snitzer goto error; 1883935fe098SMike Snitzer 1884935fe098SMike Snitzer pr_debug("created bitmap (%lu pages) for device %s\n", 1885935fe098SMike Snitzer bitmap->counts.pages, bmname(bitmap)); 1886935fe098SMike Snitzer 1887935fe098SMike Snitzer err = test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0; 1888935fe098SMike Snitzer if (err) 1889935fe098SMike Snitzer goto error; 1890935fe098SMike Snitzer 1891935fe098SMike Snitzer return bitmap; 1892935fe098SMike Snitzer error: 1893e64e4018SAndy Shevchenko md_bitmap_free(bitmap); 1894935fe098SMike Snitzer return ERR_PTR(err); 1895935fe098SMike Snitzer } 1896935fe098SMike Snitzer 1897e64e4018SAndy Shevchenko int md_bitmap_load(struct mddev *mddev) 1898935fe098SMike Snitzer { 1899935fe098SMike Snitzer int err = 0; 1900935fe098SMike Snitzer sector_t start = 0; 1901935fe098SMike Snitzer sector_t sector = 0; 1902935fe098SMike Snitzer struct bitmap *bitmap = mddev->bitmap; 1903935fe098SMike Snitzer 1904935fe098SMike Snitzer if (!bitmap) 1905935fe098SMike Snitzer goto out; 1906935fe098SMike Snitzer 1907935fe098SMike Snitzer if (mddev_is_clustered(mddev)) 1908935fe098SMike Snitzer md_cluster_ops->load_bitmaps(mddev, mddev->bitmap_info.nodes); 1909935fe098SMike Snitzer 1910935fe098SMike Snitzer /* Clear out old bitmap info first: Either there is none, or we 1911935fe098SMike Snitzer * are resuming after someone else has possibly changed things, 1912935fe098SMike Snitzer * so we should forget old cached info. 1913935fe098SMike Snitzer * All chunks should be clean, but some might need_sync. 1914935fe098SMike Snitzer */ 1915935fe098SMike Snitzer while (sector < mddev->resync_max_sectors) { 1916935fe098SMike Snitzer sector_t blocks; 1917e64e4018SAndy Shevchenko md_bitmap_start_sync(bitmap, sector, &blocks, 0); 1918935fe098SMike Snitzer sector += blocks; 1919935fe098SMike Snitzer } 1920e64e4018SAndy Shevchenko md_bitmap_close_sync(bitmap); 1921935fe098SMike Snitzer 1922935fe098SMike Snitzer if (mddev->degraded == 0 1923935fe098SMike Snitzer || bitmap->events_cleared == mddev->events) 1924935fe098SMike Snitzer /* no need to keep dirty bits to optimise a 1925935fe098SMike Snitzer * re-add of a missing device */ 1926935fe098SMike Snitzer start = mddev->recovery_cp; 1927935fe098SMike Snitzer 1928935fe098SMike Snitzer mutex_lock(&mddev->bitmap_info.mutex); 1929e64e4018SAndy Shevchenko err = md_bitmap_init_from_disk(bitmap, start); 1930935fe098SMike Snitzer mutex_unlock(&mddev->bitmap_info.mutex); 1931935fe098SMike Snitzer 1932935fe098SMike Snitzer if (err) 1933935fe098SMike Snitzer goto out; 1934935fe098SMike Snitzer clear_bit(BITMAP_STALE, &bitmap->flags); 1935935fe098SMike Snitzer 1936935fe098SMike Snitzer /* Kick recovery in case any bits were set */ 1937935fe098SMike Snitzer set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery); 1938935fe098SMike Snitzer 1939935fe098SMike Snitzer mddev->thread->timeout = mddev->bitmap_info.daemon_sleep; 1940935fe098SMike Snitzer md_wakeup_thread(mddev->thread); 1941935fe098SMike Snitzer 1942e64e4018SAndy Shevchenko md_bitmap_update_sb(bitmap); 1943935fe098SMike Snitzer 1944935fe098SMike Snitzer if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) 1945935fe098SMike Snitzer err = -EIO; 1946935fe098SMike Snitzer out: 1947935fe098SMike Snitzer return err; 1948935fe098SMike Snitzer } 1949e64e4018SAndy Shevchenko EXPORT_SYMBOL_GPL(md_bitmap_load); 1950935fe098SMike Snitzer 1951935fe098SMike Snitzer struct bitmap *get_bitmap_from_slot(struct mddev *mddev, int slot) 1952935fe098SMike Snitzer { 1953935fe098SMike Snitzer int rv = 0; 1954935fe098SMike Snitzer struct bitmap *bitmap; 1955935fe098SMike Snitzer 1956e64e4018SAndy Shevchenko bitmap = md_bitmap_create(mddev, slot); 1957935fe098SMike Snitzer if (IS_ERR(bitmap)) { 1958935fe098SMike Snitzer rv = PTR_ERR(bitmap); 1959935fe098SMike Snitzer return ERR_PTR(rv); 1960935fe098SMike Snitzer } 1961935fe098SMike Snitzer 1962e64e4018SAndy Shevchenko rv = md_bitmap_init_from_disk(bitmap, 0); 1963935fe098SMike Snitzer if (rv) { 1964e64e4018SAndy Shevchenko md_bitmap_free(bitmap); 1965935fe098SMike Snitzer return ERR_PTR(rv); 1966935fe098SMike Snitzer } 1967935fe098SMike Snitzer 1968935fe098SMike Snitzer return bitmap; 1969935fe098SMike Snitzer } 1970935fe098SMike Snitzer EXPORT_SYMBOL(get_bitmap_from_slot); 1971935fe098SMike Snitzer 1972935fe098SMike Snitzer /* Loads the bitmap associated with slot and copies the resync information 1973935fe098SMike Snitzer * to our bitmap 1974935fe098SMike Snitzer */ 1975e64e4018SAndy Shevchenko int md_bitmap_copy_from_slot(struct mddev *mddev, int slot, 1976935fe098SMike Snitzer sector_t *low, sector_t *high, bool clear_bits) 1977935fe098SMike Snitzer { 1978935fe098SMike Snitzer int rv = 0, i, j; 1979935fe098SMike Snitzer sector_t block, lo = 0, hi = 0; 1980935fe098SMike Snitzer struct bitmap_counts *counts; 1981935fe098SMike Snitzer struct bitmap *bitmap; 1982935fe098SMike Snitzer 1983935fe098SMike Snitzer bitmap = get_bitmap_from_slot(mddev, slot); 1984935fe098SMike Snitzer if (IS_ERR(bitmap)) { 1985935fe098SMike Snitzer pr_err("%s can't get bitmap from slot %d\n", __func__, slot); 1986935fe098SMike Snitzer return -1; 1987935fe098SMike Snitzer } 1988935fe098SMike Snitzer 1989935fe098SMike Snitzer counts = &bitmap->counts; 1990935fe098SMike Snitzer for (j = 0; j < counts->chunks; j++) { 1991935fe098SMike Snitzer block = (sector_t)j << counts->chunkshift; 1992e64e4018SAndy Shevchenko if (md_bitmap_file_test_bit(bitmap, block)) { 1993935fe098SMike Snitzer if (!lo) 1994935fe098SMike Snitzer lo = block; 1995935fe098SMike Snitzer hi = block; 1996e64e4018SAndy Shevchenko md_bitmap_file_clear_bit(bitmap, block); 1997e64e4018SAndy Shevchenko md_bitmap_set_memory_bits(mddev->bitmap, block, 1); 1998e64e4018SAndy Shevchenko md_bitmap_file_set_bit(mddev->bitmap, block); 1999935fe098SMike Snitzer } 2000935fe098SMike Snitzer } 2001935fe098SMike Snitzer 2002935fe098SMike Snitzer if (clear_bits) { 2003e64e4018SAndy Shevchenko md_bitmap_update_sb(bitmap); 2004935fe098SMike Snitzer /* BITMAP_PAGE_PENDING is set, but bitmap_unplug needs 2005935fe098SMike Snitzer * BITMAP_PAGE_DIRTY or _NEEDWRITE to write ... */ 2006935fe098SMike Snitzer for (i = 0; i < bitmap->storage.file_pages; i++) 2007935fe098SMike Snitzer if (test_page_attr(bitmap, i, BITMAP_PAGE_PENDING)) 2008935fe098SMike Snitzer set_page_attr(bitmap, i, BITMAP_PAGE_NEEDWRITE); 2009e64e4018SAndy Shevchenko md_bitmap_unplug(bitmap); 2010935fe098SMike Snitzer } 2011e64e4018SAndy Shevchenko md_bitmap_unplug(mddev->bitmap); 2012935fe098SMike Snitzer *low = lo; 2013935fe098SMike Snitzer *high = hi; 2014935fe098SMike Snitzer 2015935fe098SMike Snitzer return rv; 2016935fe098SMike Snitzer } 2017e64e4018SAndy Shevchenko EXPORT_SYMBOL_GPL(md_bitmap_copy_from_slot); 2018935fe098SMike Snitzer 2019935fe098SMike Snitzer 2020e64e4018SAndy Shevchenko void md_bitmap_status(struct seq_file *seq, struct bitmap *bitmap) 2021935fe098SMike Snitzer { 2022935fe098SMike Snitzer unsigned long chunk_kb; 2023935fe098SMike Snitzer struct bitmap_counts *counts; 2024935fe098SMike Snitzer 2025935fe098SMike Snitzer if (!bitmap) 2026935fe098SMike Snitzer return; 2027935fe098SMike Snitzer 2028935fe098SMike Snitzer counts = &bitmap->counts; 2029935fe098SMike Snitzer 2030935fe098SMike Snitzer chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10; 2031935fe098SMike Snitzer seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], " 2032935fe098SMike Snitzer "%lu%s chunk", 2033935fe098SMike Snitzer counts->pages - counts->missing_pages, 2034935fe098SMike Snitzer counts->pages, 2035935fe098SMike Snitzer (counts->pages - counts->missing_pages) 2036935fe098SMike Snitzer << (PAGE_SHIFT - 10), 2037935fe098SMike Snitzer chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize, 2038935fe098SMike Snitzer chunk_kb ? "KB" : "B"); 2039935fe098SMike Snitzer if (bitmap->storage.file) { 2040935fe098SMike Snitzer seq_printf(seq, ", file: "); 2041935fe098SMike Snitzer seq_file_path(seq, bitmap->storage.file, " \t\n"); 2042935fe098SMike Snitzer } 2043935fe098SMike Snitzer 2044935fe098SMike Snitzer seq_printf(seq, "\n"); 2045935fe098SMike Snitzer } 2046935fe098SMike Snitzer 2047e64e4018SAndy Shevchenko int md_bitmap_resize(struct bitmap *bitmap, sector_t blocks, 2048935fe098SMike Snitzer int chunksize, int init) 2049935fe098SMike Snitzer { 2050935fe098SMike Snitzer /* If chunk_size is 0, choose an appropriate chunk size. 2051935fe098SMike Snitzer * Then possibly allocate new storage space. 2052935fe098SMike Snitzer * Then quiesce, copy bits, replace bitmap, and re-start 2053935fe098SMike Snitzer * 2054935fe098SMike Snitzer * This function is called both to set up the initial bitmap 2055935fe098SMike Snitzer * and to resize the bitmap while the array is active. 2056935fe098SMike Snitzer * If this happens as a result of the array being resized, 2057935fe098SMike Snitzer * chunksize will be zero, and we need to choose a suitable 2058935fe098SMike Snitzer * chunksize, otherwise we use what we are given. 2059935fe098SMike Snitzer */ 2060935fe098SMike Snitzer struct bitmap_storage store; 2061935fe098SMike Snitzer struct bitmap_counts old_counts; 2062935fe098SMike Snitzer unsigned long chunks; 2063935fe098SMike Snitzer sector_t block; 2064935fe098SMike Snitzer sector_t old_blocks, new_blocks; 2065935fe098SMike Snitzer int chunkshift; 2066935fe098SMike Snitzer int ret = 0; 2067935fe098SMike Snitzer long pages; 2068935fe098SMike Snitzer struct bitmap_page *new_bp; 2069935fe098SMike Snitzer 2070935fe098SMike Snitzer if (bitmap->storage.file && !init) { 2071935fe098SMike Snitzer pr_info("md: cannot resize file-based bitmap\n"); 2072935fe098SMike Snitzer return -EINVAL; 2073935fe098SMike Snitzer } 2074935fe098SMike Snitzer 2075935fe098SMike Snitzer if (chunksize == 0) { 2076935fe098SMike Snitzer /* If there is enough space, leave the chunk size unchanged, 2077935fe098SMike Snitzer * else increase by factor of two until there is enough space. 2078935fe098SMike Snitzer */ 2079935fe098SMike Snitzer long bytes; 2080935fe098SMike Snitzer long space = bitmap->mddev->bitmap_info.space; 2081935fe098SMike Snitzer 2082935fe098SMike Snitzer if (space == 0) { 2083935fe098SMike Snitzer /* We don't know how much space there is, so limit 2084935fe098SMike Snitzer * to current size - in sectors. 2085935fe098SMike Snitzer */ 2086935fe098SMike Snitzer bytes = DIV_ROUND_UP(bitmap->counts.chunks, 8); 2087935fe098SMike Snitzer if (!bitmap->mddev->bitmap_info.external) 2088935fe098SMike Snitzer bytes += sizeof(bitmap_super_t); 2089935fe098SMike Snitzer space = DIV_ROUND_UP(bytes, 512); 2090935fe098SMike Snitzer bitmap->mddev->bitmap_info.space = space; 2091935fe098SMike Snitzer } 2092935fe098SMike Snitzer chunkshift = bitmap->counts.chunkshift; 2093935fe098SMike Snitzer chunkshift--; 2094935fe098SMike Snitzer do { 2095935fe098SMike Snitzer /* 'chunkshift' is shift from block size to chunk size */ 2096935fe098SMike Snitzer chunkshift++; 2097935fe098SMike Snitzer chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift); 2098935fe098SMike Snitzer bytes = DIV_ROUND_UP(chunks, 8); 2099935fe098SMike Snitzer if (!bitmap->mddev->bitmap_info.external) 2100935fe098SMike Snitzer bytes += sizeof(bitmap_super_t); 2101935fe098SMike Snitzer } while (bytes > (space << 9)); 2102935fe098SMike Snitzer } else 2103935fe098SMike Snitzer chunkshift = ffz(~chunksize) - BITMAP_BLOCK_SHIFT; 2104935fe098SMike Snitzer 2105935fe098SMike Snitzer chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift); 2106935fe098SMike Snitzer memset(&store, 0, sizeof(store)); 2107935fe098SMike Snitzer if (bitmap->mddev->bitmap_info.offset || bitmap->mddev->bitmap_info.file) 2108e64e4018SAndy Shevchenko ret = md_bitmap_storage_alloc(&store, chunks, 2109935fe098SMike Snitzer !bitmap->mddev->bitmap_info.external, 2110935fe098SMike Snitzer mddev_is_clustered(bitmap->mddev) 2111935fe098SMike Snitzer ? bitmap->cluster_slot : 0); 2112935fe098SMike Snitzer if (ret) { 2113e64e4018SAndy Shevchenko md_bitmap_file_unmap(&store); 2114935fe098SMike Snitzer goto err; 2115935fe098SMike Snitzer } 2116935fe098SMike Snitzer 2117935fe098SMike Snitzer pages = DIV_ROUND_UP(chunks, PAGE_COUNTER_RATIO); 2118935fe098SMike Snitzer 21196396bb22SKees Cook new_bp = kcalloc(pages, sizeof(*new_bp), GFP_KERNEL); 2120935fe098SMike Snitzer ret = -ENOMEM; 2121935fe098SMike Snitzer if (!new_bp) { 2122e64e4018SAndy Shevchenko md_bitmap_file_unmap(&store); 2123935fe098SMike Snitzer goto err; 2124935fe098SMike Snitzer } 2125935fe098SMike Snitzer 2126935fe098SMike Snitzer if (!init) 2127935fe098SMike Snitzer bitmap->mddev->pers->quiesce(bitmap->mddev, 1); 2128935fe098SMike Snitzer 2129935fe098SMike Snitzer store.file = bitmap->storage.file; 2130935fe098SMike Snitzer bitmap->storage.file = NULL; 2131935fe098SMike Snitzer 2132935fe098SMike Snitzer if (store.sb_page && bitmap->storage.sb_page) 2133935fe098SMike Snitzer memcpy(page_address(store.sb_page), 2134935fe098SMike Snitzer page_address(bitmap->storage.sb_page), 2135935fe098SMike Snitzer sizeof(bitmap_super_t)); 2136e64e4018SAndy Shevchenko md_bitmap_file_unmap(&bitmap->storage); 2137935fe098SMike Snitzer bitmap->storage = store; 2138935fe098SMike Snitzer 2139935fe098SMike Snitzer old_counts = bitmap->counts; 2140935fe098SMike Snitzer bitmap->counts.bp = new_bp; 2141935fe098SMike Snitzer bitmap->counts.pages = pages; 2142935fe098SMike Snitzer bitmap->counts.missing_pages = pages; 2143935fe098SMike Snitzer bitmap->counts.chunkshift = chunkshift; 2144935fe098SMike Snitzer bitmap->counts.chunks = chunks; 2145935fe098SMike Snitzer bitmap->mddev->bitmap_info.chunksize = 1 << (chunkshift + 2146935fe098SMike Snitzer BITMAP_BLOCK_SHIFT); 2147935fe098SMike Snitzer 2148935fe098SMike Snitzer blocks = min(old_counts.chunks << old_counts.chunkshift, 2149935fe098SMike Snitzer chunks << chunkshift); 2150935fe098SMike Snitzer 2151935fe098SMike Snitzer spin_lock_irq(&bitmap->counts.lock); 2152935fe098SMike Snitzer /* For cluster raid, need to pre-allocate bitmap */ 2153935fe098SMike Snitzer if (mddev_is_clustered(bitmap->mddev)) { 2154935fe098SMike Snitzer unsigned long page; 2155935fe098SMike Snitzer for (page = 0; page < pages; page++) { 2156e64e4018SAndy Shevchenko ret = md_bitmap_checkpage(&bitmap->counts, page, 1, 1); 2157935fe098SMike Snitzer if (ret) { 2158935fe098SMike Snitzer unsigned long k; 2159935fe098SMike Snitzer 2160935fe098SMike Snitzer /* deallocate the page memory */ 2161935fe098SMike Snitzer for (k = 0; k < page; k++) { 2162935fe098SMike Snitzer kfree(new_bp[k].map); 2163935fe098SMike Snitzer } 21640868b99cSZdenek Kabelac kfree(new_bp); 2165935fe098SMike Snitzer 2166935fe098SMike Snitzer /* restore some fields from old_counts */ 2167935fe098SMike Snitzer bitmap->counts.bp = old_counts.bp; 2168935fe098SMike Snitzer bitmap->counts.pages = old_counts.pages; 2169935fe098SMike Snitzer bitmap->counts.missing_pages = old_counts.pages; 2170935fe098SMike Snitzer bitmap->counts.chunkshift = old_counts.chunkshift; 2171935fe098SMike Snitzer bitmap->counts.chunks = old_counts.chunks; 2172935fe098SMike Snitzer bitmap->mddev->bitmap_info.chunksize = 1 << (old_counts.chunkshift + 2173935fe098SMike Snitzer BITMAP_BLOCK_SHIFT); 2174935fe098SMike Snitzer blocks = old_counts.chunks << old_counts.chunkshift; 2175935fe098SMike Snitzer pr_warn("Could not pre-allocate in-memory bitmap for cluster raid\n"); 2176935fe098SMike Snitzer break; 2177935fe098SMike Snitzer } else 2178935fe098SMike Snitzer bitmap->counts.bp[page].count += 1; 2179935fe098SMike Snitzer } 2180935fe098SMike Snitzer } 2181935fe098SMike Snitzer 2182935fe098SMike Snitzer for (block = 0; block < blocks; ) { 2183935fe098SMike Snitzer bitmap_counter_t *bmc_old, *bmc_new; 2184935fe098SMike Snitzer int set; 2185935fe098SMike Snitzer 2186e64e4018SAndy Shevchenko bmc_old = md_bitmap_get_counter(&old_counts, block, &old_blocks, 0); 2187935fe098SMike Snitzer set = bmc_old && NEEDED(*bmc_old); 2188935fe098SMike Snitzer 2189935fe098SMike Snitzer if (set) { 2190e64e4018SAndy Shevchenko bmc_new = md_bitmap_get_counter(&bitmap->counts, block, &new_blocks, 1); 2191935fe098SMike Snitzer if (*bmc_new == 0) { 2192935fe098SMike Snitzer /* need to set on-disk bits too. */ 2193935fe098SMike Snitzer sector_t end = block + new_blocks; 2194935fe098SMike Snitzer sector_t start = block >> chunkshift; 2195935fe098SMike Snitzer start <<= chunkshift; 2196935fe098SMike Snitzer while (start < end) { 2197e64e4018SAndy Shevchenko md_bitmap_file_set_bit(bitmap, block); 2198935fe098SMike Snitzer start += 1 << chunkshift; 2199935fe098SMike Snitzer } 2200935fe098SMike Snitzer *bmc_new = 2; 2201e64e4018SAndy Shevchenko md_bitmap_count_page(&bitmap->counts, block, 1); 2202e64e4018SAndy Shevchenko md_bitmap_set_pending(&bitmap->counts, block); 2203935fe098SMike Snitzer } 2204935fe098SMike Snitzer *bmc_new |= NEEDED_MASK; 2205935fe098SMike Snitzer if (new_blocks < old_blocks) 2206935fe098SMike Snitzer old_blocks = new_blocks; 2207935fe098SMike Snitzer } 2208935fe098SMike Snitzer block += old_blocks; 2209935fe098SMike Snitzer } 2210935fe098SMike Snitzer 22110868b99cSZdenek Kabelac if (bitmap->counts.bp != old_counts.bp) { 22120868b99cSZdenek Kabelac unsigned long k; 22130868b99cSZdenek Kabelac for (k = 0; k < old_counts.pages; k++) 22140868b99cSZdenek Kabelac if (!old_counts.bp[k].hijacked) 22150868b99cSZdenek Kabelac kfree(old_counts.bp[k].map); 22160868b99cSZdenek Kabelac kfree(old_counts.bp); 22170868b99cSZdenek Kabelac } 22180868b99cSZdenek Kabelac 2219935fe098SMike Snitzer if (!init) { 2220935fe098SMike Snitzer int i; 2221935fe098SMike Snitzer while (block < (chunks << chunkshift)) { 2222935fe098SMike Snitzer bitmap_counter_t *bmc; 2223e64e4018SAndy Shevchenko bmc = md_bitmap_get_counter(&bitmap->counts, block, &new_blocks, 1); 2224935fe098SMike Snitzer if (bmc) { 2225935fe098SMike Snitzer /* new space. It needs to be resynced, so 2226935fe098SMike Snitzer * we set NEEDED_MASK. 2227935fe098SMike Snitzer */ 2228935fe098SMike Snitzer if (*bmc == 0) { 2229935fe098SMike Snitzer *bmc = NEEDED_MASK | 2; 2230e64e4018SAndy Shevchenko md_bitmap_count_page(&bitmap->counts, block, 1); 2231e64e4018SAndy Shevchenko md_bitmap_set_pending(&bitmap->counts, block); 2232935fe098SMike Snitzer } 2233935fe098SMike Snitzer } 2234935fe098SMike Snitzer block += new_blocks; 2235935fe098SMike Snitzer } 2236935fe098SMike Snitzer for (i = 0; i < bitmap->storage.file_pages; i++) 2237935fe098SMike Snitzer set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY); 2238935fe098SMike Snitzer } 2239935fe098SMike Snitzer spin_unlock_irq(&bitmap->counts.lock); 2240935fe098SMike Snitzer 2241935fe098SMike Snitzer if (!init) { 2242e64e4018SAndy Shevchenko md_bitmap_unplug(bitmap); 2243935fe098SMike Snitzer bitmap->mddev->pers->quiesce(bitmap->mddev, 0); 2244935fe098SMike Snitzer } 2245935fe098SMike Snitzer ret = 0; 2246935fe098SMike Snitzer err: 2247935fe098SMike Snitzer return ret; 2248935fe098SMike Snitzer } 2249e64e4018SAndy Shevchenko EXPORT_SYMBOL_GPL(md_bitmap_resize); 2250935fe098SMike Snitzer 2251935fe098SMike Snitzer static ssize_t 2252935fe098SMike Snitzer location_show(struct mddev *mddev, char *page) 2253935fe098SMike Snitzer { 2254935fe098SMike Snitzer ssize_t len; 2255935fe098SMike Snitzer if (mddev->bitmap_info.file) 2256935fe098SMike Snitzer len = sprintf(page, "file"); 2257935fe098SMike Snitzer else if (mddev->bitmap_info.offset) 2258935fe098SMike Snitzer len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset); 2259935fe098SMike Snitzer else 2260935fe098SMike Snitzer len = sprintf(page, "none"); 2261935fe098SMike Snitzer len += sprintf(page+len, "\n"); 2262935fe098SMike Snitzer return len; 2263935fe098SMike Snitzer } 2264935fe098SMike Snitzer 2265935fe098SMike Snitzer static ssize_t 2266935fe098SMike Snitzer location_store(struct mddev *mddev, const char *buf, size_t len) 2267935fe098SMike Snitzer { 2268935fe098SMike Snitzer int rv; 2269935fe098SMike Snitzer 2270935fe098SMike Snitzer rv = mddev_lock(mddev); 2271935fe098SMike Snitzer if (rv) 2272935fe098SMike Snitzer return rv; 2273935fe098SMike Snitzer if (mddev->pers) { 2274935fe098SMike Snitzer if (!mddev->pers->quiesce) { 2275935fe098SMike Snitzer rv = -EBUSY; 2276935fe098SMike Snitzer goto out; 2277935fe098SMike Snitzer } 2278935fe098SMike Snitzer if (mddev->recovery || mddev->sync_thread) { 2279935fe098SMike Snitzer rv = -EBUSY; 2280935fe098SMike Snitzer goto out; 2281935fe098SMike Snitzer } 2282935fe098SMike Snitzer } 2283935fe098SMike Snitzer 2284935fe098SMike Snitzer if (mddev->bitmap || mddev->bitmap_info.file || 2285935fe098SMike Snitzer mddev->bitmap_info.offset) { 2286935fe098SMike Snitzer /* bitmap already configured. Only option is to clear it */ 2287935fe098SMike Snitzer if (strncmp(buf, "none", 4) != 0) { 2288935fe098SMike Snitzer rv = -EBUSY; 2289935fe098SMike Snitzer goto out; 2290935fe098SMike Snitzer } 2291935fe098SMike Snitzer if (mddev->pers) { 2292f8f83d8fSJack Wang mddev_suspend(mddev); 2293e64e4018SAndy Shevchenko md_bitmap_destroy(mddev); 2294f8f83d8fSJack Wang mddev_resume(mddev); 2295935fe098SMike Snitzer } 2296935fe098SMike Snitzer mddev->bitmap_info.offset = 0; 2297935fe098SMike Snitzer if (mddev->bitmap_info.file) { 2298935fe098SMike Snitzer struct file *f = mddev->bitmap_info.file; 2299935fe098SMike Snitzer mddev->bitmap_info.file = NULL; 2300935fe098SMike Snitzer fput(f); 2301935fe098SMike Snitzer } 2302935fe098SMike Snitzer } else { 2303935fe098SMike Snitzer /* No bitmap, OK to set a location */ 2304935fe098SMike Snitzer long long offset; 2305935fe098SMike Snitzer if (strncmp(buf, "none", 4) == 0) 2306935fe098SMike Snitzer /* nothing to be done */; 2307935fe098SMike Snitzer else if (strncmp(buf, "file:", 5) == 0) { 2308935fe098SMike Snitzer /* Not supported yet */ 2309935fe098SMike Snitzer rv = -EINVAL; 2310935fe098SMike Snitzer goto out; 2311935fe098SMike Snitzer } else { 2312935fe098SMike Snitzer if (buf[0] == '+') 2313935fe098SMike Snitzer rv = kstrtoll(buf+1, 10, &offset); 2314935fe098SMike Snitzer else 2315935fe098SMike Snitzer rv = kstrtoll(buf, 10, &offset); 2316935fe098SMike Snitzer if (rv) 2317935fe098SMike Snitzer goto out; 2318935fe098SMike Snitzer if (offset == 0) { 2319935fe098SMike Snitzer rv = -EINVAL; 2320935fe098SMike Snitzer goto out; 2321935fe098SMike Snitzer } 2322935fe098SMike Snitzer if (mddev->bitmap_info.external == 0 && 2323935fe098SMike Snitzer mddev->major_version == 0 && 2324935fe098SMike Snitzer offset != mddev->bitmap_info.default_offset) { 2325935fe098SMike Snitzer rv = -EINVAL; 2326935fe098SMike Snitzer goto out; 2327935fe098SMike Snitzer } 2328935fe098SMike Snitzer mddev->bitmap_info.offset = offset; 2329935fe098SMike Snitzer if (mddev->pers) { 2330935fe098SMike Snitzer struct bitmap *bitmap; 2331e64e4018SAndy Shevchenko bitmap = md_bitmap_create(mddev, -1); 2332f8f83d8fSJack Wang mddev_suspend(mddev); 2333935fe098SMike Snitzer if (IS_ERR(bitmap)) 2334935fe098SMike Snitzer rv = PTR_ERR(bitmap); 2335935fe098SMike Snitzer else { 2336935fe098SMike Snitzer mddev->bitmap = bitmap; 2337e64e4018SAndy Shevchenko rv = md_bitmap_load(mddev); 2338935fe098SMike Snitzer if (rv) 2339935fe098SMike Snitzer mddev->bitmap_info.offset = 0; 2340935fe098SMike Snitzer } 2341935fe098SMike Snitzer if (rv) { 2342e64e4018SAndy Shevchenko md_bitmap_destroy(mddev); 2343f8f83d8fSJack Wang mddev_resume(mddev); 2344935fe098SMike Snitzer goto out; 2345935fe098SMike Snitzer } 2346f8f83d8fSJack Wang mddev_resume(mddev); 2347935fe098SMike Snitzer } 2348935fe098SMike Snitzer } 2349935fe098SMike Snitzer } 2350935fe098SMike Snitzer if (!mddev->external) { 2351935fe098SMike Snitzer /* Ensure new bitmap info is stored in 2352935fe098SMike Snitzer * metadata promptly. 2353935fe098SMike Snitzer */ 2354935fe098SMike Snitzer set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); 2355935fe098SMike Snitzer md_wakeup_thread(mddev->thread); 2356935fe098SMike Snitzer } 2357935fe098SMike Snitzer rv = 0; 2358935fe098SMike Snitzer out: 2359935fe098SMike Snitzer mddev_unlock(mddev); 2360935fe098SMike Snitzer if (rv) 2361935fe098SMike Snitzer return rv; 2362935fe098SMike Snitzer return len; 2363935fe098SMike Snitzer } 2364935fe098SMike Snitzer 2365935fe098SMike Snitzer static struct md_sysfs_entry bitmap_location = 2366935fe098SMike Snitzer __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store); 2367935fe098SMike Snitzer 2368935fe098SMike Snitzer /* 'bitmap/space' is the space available at 'location' for the 2369935fe098SMike Snitzer * bitmap. This allows the kernel to know when it is safe to 2370935fe098SMike Snitzer * resize the bitmap to match a resized array. 2371935fe098SMike Snitzer */ 2372935fe098SMike Snitzer static ssize_t 2373935fe098SMike Snitzer space_show(struct mddev *mddev, char *page) 2374935fe098SMike Snitzer { 2375935fe098SMike Snitzer return sprintf(page, "%lu\n", mddev->bitmap_info.space); 2376935fe098SMike Snitzer } 2377935fe098SMike Snitzer 2378935fe098SMike Snitzer static ssize_t 2379935fe098SMike Snitzer space_store(struct mddev *mddev, const char *buf, size_t len) 2380935fe098SMike Snitzer { 2381935fe098SMike Snitzer unsigned long sectors; 2382935fe098SMike Snitzer int rv; 2383935fe098SMike Snitzer 2384935fe098SMike Snitzer rv = kstrtoul(buf, 10, §ors); 2385935fe098SMike Snitzer if (rv) 2386935fe098SMike Snitzer return rv; 2387935fe098SMike Snitzer 2388935fe098SMike Snitzer if (sectors == 0) 2389935fe098SMike Snitzer return -EINVAL; 2390935fe098SMike Snitzer 2391935fe098SMike Snitzer if (mddev->bitmap && 2392935fe098SMike Snitzer sectors < (mddev->bitmap->storage.bytes + 511) >> 9) 2393935fe098SMike Snitzer return -EFBIG; /* Bitmap is too big for this small space */ 2394935fe098SMike Snitzer 2395935fe098SMike Snitzer /* could make sure it isn't too big, but that isn't really 2396935fe098SMike Snitzer * needed - user-space should be careful. 2397935fe098SMike Snitzer */ 2398935fe098SMike Snitzer mddev->bitmap_info.space = sectors; 2399935fe098SMike Snitzer return len; 2400935fe098SMike Snitzer } 2401935fe098SMike Snitzer 2402935fe098SMike Snitzer static struct md_sysfs_entry bitmap_space = 2403935fe098SMike Snitzer __ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store); 2404935fe098SMike Snitzer 2405935fe098SMike Snitzer static ssize_t 2406935fe098SMike Snitzer timeout_show(struct mddev *mddev, char *page) 2407935fe098SMike Snitzer { 2408935fe098SMike Snitzer ssize_t len; 2409935fe098SMike Snitzer unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ; 2410935fe098SMike Snitzer unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ; 2411935fe098SMike Snitzer 2412935fe098SMike Snitzer len = sprintf(page, "%lu", secs); 2413935fe098SMike Snitzer if (jifs) 2414935fe098SMike Snitzer len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs)); 2415935fe098SMike Snitzer len += sprintf(page+len, "\n"); 2416935fe098SMike Snitzer return len; 2417935fe098SMike Snitzer } 2418935fe098SMike Snitzer 2419935fe098SMike Snitzer static ssize_t 2420935fe098SMike Snitzer timeout_store(struct mddev *mddev, const char *buf, size_t len) 2421935fe098SMike Snitzer { 2422935fe098SMike Snitzer /* timeout can be set at any time */ 2423935fe098SMike Snitzer unsigned long timeout; 2424935fe098SMike Snitzer int rv = strict_strtoul_scaled(buf, &timeout, 4); 2425935fe098SMike Snitzer if (rv) 2426935fe098SMike Snitzer return rv; 2427935fe098SMike Snitzer 2428935fe098SMike Snitzer /* just to make sure we don't overflow... */ 2429935fe098SMike Snitzer if (timeout >= LONG_MAX / HZ) 2430935fe098SMike Snitzer return -EINVAL; 2431935fe098SMike Snitzer 2432935fe098SMike Snitzer timeout = timeout * HZ / 10000; 2433935fe098SMike Snitzer 2434935fe098SMike Snitzer if (timeout >= MAX_SCHEDULE_TIMEOUT) 2435935fe098SMike Snitzer timeout = MAX_SCHEDULE_TIMEOUT-1; 2436935fe098SMike Snitzer if (timeout < 1) 2437935fe098SMike Snitzer timeout = 1; 2438935fe098SMike Snitzer mddev->bitmap_info.daemon_sleep = timeout; 2439935fe098SMike Snitzer if (mddev->thread) { 2440935fe098SMike Snitzer /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then 2441935fe098SMike Snitzer * the bitmap is all clean and we don't need to 2442935fe098SMike Snitzer * adjust the timeout right now 2443935fe098SMike Snitzer */ 2444935fe098SMike Snitzer if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) { 2445935fe098SMike Snitzer mddev->thread->timeout = timeout; 2446935fe098SMike Snitzer md_wakeup_thread(mddev->thread); 2447935fe098SMike Snitzer } 2448935fe098SMike Snitzer } 2449935fe098SMike Snitzer return len; 2450935fe098SMike Snitzer } 2451935fe098SMike Snitzer 2452935fe098SMike Snitzer static struct md_sysfs_entry bitmap_timeout = 2453935fe098SMike Snitzer __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store); 2454935fe098SMike Snitzer 2455935fe098SMike Snitzer static ssize_t 2456935fe098SMike Snitzer backlog_show(struct mddev *mddev, char *page) 2457935fe098SMike Snitzer { 2458935fe098SMike Snitzer return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind); 2459935fe098SMike Snitzer } 2460935fe098SMike Snitzer 2461935fe098SMike Snitzer static ssize_t 2462935fe098SMike Snitzer backlog_store(struct mddev *mddev, const char *buf, size_t len) 2463935fe098SMike Snitzer { 2464935fe098SMike Snitzer unsigned long backlog; 246510c92fcaSGuoqing Jiang unsigned long old_mwb = mddev->bitmap_info.max_write_behind; 2466935fe098SMike Snitzer int rv = kstrtoul(buf, 10, &backlog); 2467935fe098SMike Snitzer if (rv) 2468935fe098SMike Snitzer return rv; 2469935fe098SMike Snitzer if (backlog > COUNTER_MAX) 2470935fe098SMike Snitzer return -EINVAL; 2471935fe098SMike Snitzer mddev->bitmap_info.max_write_behind = backlog; 247210c92fcaSGuoqing Jiang if (!backlog && mddev->wb_info_pool) { 247310c92fcaSGuoqing Jiang /* wb_info_pool is not needed if backlog is zero */ 247410c92fcaSGuoqing Jiang mempool_destroy(mddev->wb_info_pool); 247510c92fcaSGuoqing Jiang mddev->wb_info_pool = NULL; 247610c92fcaSGuoqing Jiang } else if (backlog && !mddev->wb_info_pool) { 247710c92fcaSGuoqing Jiang /* wb_info_pool is needed since backlog is not zero */ 247810c92fcaSGuoqing Jiang struct md_rdev *rdev; 247910c92fcaSGuoqing Jiang 248010c92fcaSGuoqing Jiang rdev_for_each(rdev, mddev) 248110c92fcaSGuoqing Jiang mddev_create_wb_pool(mddev, rdev, false); 248210c92fcaSGuoqing Jiang } 248310c92fcaSGuoqing Jiang if (old_mwb != backlog) 248410c92fcaSGuoqing Jiang md_bitmap_update_sb(mddev->bitmap); 2485935fe098SMike Snitzer return len; 2486935fe098SMike Snitzer } 2487935fe098SMike Snitzer 2488935fe098SMike Snitzer static struct md_sysfs_entry bitmap_backlog = 2489935fe098SMike Snitzer __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store); 2490935fe098SMike Snitzer 2491935fe098SMike Snitzer static ssize_t 2492935fe098SMike Snitzer chunksize_show(struct mddev *mddev, char *page) 2493935fe098SMike Snitzer { 2494935fe098SMike Snitzer return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize); 2495935fe098SMike Snitzer } 2496935fe098SMike Snitzer 2497935fe098SMike Snitzer static ssize_t 2498935fe098SMike Snitzer chunksize_store(struct mddev *mddev, const char *buf, size_t len) 2499935fe098SMike Snitzer { 2500935fe098SMike Snitzer /* Can only be changed when no bitmap is active */ 2501935fe098SMike Snitzer int rv; 2502935fe098SMike Snitzer unsigned long csize; 2503935fe098SMike Snitzer if (mddev->bitmap) 2504935fe098SMike Snitzer return -EBUSY; 2505935fe098SMike Snitzer rv = kstrtoul(buf, 10, &csize); 2506935fe098SMike Snitzer if (rv) 2507935fe098SMike Snitzer return rv; 2508935fe098SMike Snitzer if (csize < 512 || 2509935fe098SMike Snitzer !is_power_of_2(csize)) 2510935fe098SMike Snitzer return -EINVAL; 2511935fe098SMike Snitzer mddev->bitmap_info.chunksize = csize; 2512935fe098SMike Snitzer return len; 2513935fe098SMike Snitzer } 2514935fe098SMike Snitzer 2515935fe098SMike Snitzer static struct md_sysfs_entry bitmap_chunksize = 2516935fe098SMike Snitzer __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store); 2517935fe098SMike Snitzer 2518935fe098SMike Snitzer static ssize_t metadata_show(struct mddev *mddev, char *page) 2519935fe098SMike Snitzer { 2520935fe098SMike Snitzer if (mddev_is_clustered(mddev)) 2521935fe098SMike Snitzer return sprintf(page, "clustered\n"); 2522935fe098SMike Snitzer return sprintf(page, "%s\n", (mddev->bitmap_info.external 2523935fe098SMike Snitzer ? "external" : "internal")); 2524935fe098SMike Snitzer } 2525935fe098SMike Snitzer 2526935fe098SMike Snitzer static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len) 2527935fe098SMike Snitzer { 2528935fe098SMike Snitzer if (mddev->bitmap || 2529935fe098SMike Snitzer mddev->bitmap_info.file || 2530935fe098SMike Snitzer mddev->bitmap_info.offset) 2531935fe098SMike Snitzer return -EBUSY; 2532935fe098SMike Snitzer if (strncmp(buf, "external", 8) == 0) 2533935fe098SMike Snitzer mddev->bitmap_info.external = 1; 2534935fe098SMike Snitzer else if ((strncmp(buf, "internal", 8) == 0) || 2535935fe098SMike Snitzer (strncmp(buf, "clustered", 9) == 0)) 2536935fe098SMike Snitzer mddev->bitmap_info.external = 0; 2537935fe098SMike Snitzer else 2538935fe098SMike Snitzer return -EINVAL; 2539935fe098SMike Snitzer return len; 2540935fe098SMike Snitzer } 2541935fe098SMike Snitzer 2542935fe098SMike Snitzer static struct md_sysfs_entry bitmap_metadata = 2543935fe098SMike Snitzer __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store); 2544935fe098SMike Snitzer 2545935fe098SMike Snitzer static ssize_t can_clear_show(struct mddev *mddev, char *page) 2546935fe098SMike Snitzer { 2547935fe098SMike Snitzer int len; 2548935fe098SMike Snitzer spin_lock(&mddev->lock); 2549935fe098SMike Snitzer if (mddev->bitmap) 2550935fe098SMike Snitzer len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ? 2551935fe098SMike Snitzer "false" : "true")); 2552935fe098SMike Snitzer else 2553935fe098SMike Snitzer len = sprintf(page, "\n"); 2554935fe098SMike Snitzer spin_unlock(&mddev->lock); 2555935fe098SMike Snitzer return len; 2556935fe098SMike Snitzer } 2557935fe098SMike Snitzer 2558935fe098SMike Snitzer static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len) 2559935fe098SMike Snitzer { 2560935fe098SMike Snitzer if (mddev->bitmap == NULL) 2561935fe098SMike Snitzer return -ENOENT; 2562935fe098SMike Snitzer if (strncmp(buf, "false", 5) == 0) 2563935fe098SMike Snitzer mddev->bitmap->need_sync = 1; 2564935fe098SMike Snitzer else if (strncmp(buf, "true", 4) == 0) { 2565935fe098SMike Snitzer if (mddev->degraded) 2566935fe098SMike Snitzer return -EBUSY; 2567935fe098SMike Snitzer mddev->bitmap->need_sync = 0; 2568935fe098SMike Snitzer } else 2569935fe098SMike Snitzer return -EINVAL; 2570935fe098SMike Snitzer return len; 2571935fe098SMike Snitzer } 2572935fe098SMike Snitzer 2573935fe098SMike Snitzer static struct md_sysfs_entry bitmap_can_clear = 2574935fe098SMike Snitzer __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store); 2575935fe098SMike Snitzer 2576935fe098SMike Snitzer static ssize_t 2577935fe098SMike Snitzer behind_writes_used_show(struct mddev *mddev, char *page) 2578935fe098SMike Snitzer { 2579935fe098SMike Snitzer ssize_t ret; 2580935fe098SMike Snitzer spin_lock(&mddev->lock); 2581935fe098SMike Snitzer if (mddev->bitmap == NULL) 2582935fe098SMike Snitzer ret = sprintf(page, "0\n"); 2583935fe098SMike Snitzer else 2584935fe098SMike Snitzer ret = sprintf(page, "%lu\n", 2585935fe098SMike Snitzer mddev->bitmap->behind_writes_used); 2586935fe098SMike Snitzer spin_unlock(&mddev->lock); 2587935fe098SMike Snitzer return ret; 2588935fe098SMike Snitzer } 2589935fe098SMike Snitzer 2590935fe098SMike Snitzer static ssize_t 2591935fe098SMike Snitzer behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len) 2592935fe098SMike Snitzer { 2593935fe098SMike Snitzer if (mddev->bitmap) 2594935fe098SMike Snitzer mddev->bitmap->behind_writes_used = 0; 2595935fe098SMike Snitzer return len; 2596935fe098SMike Snitzer } 2597935fe098SMike Snitzer 2598935fe098SMike Snitzer static struct md_sysfs_entry max_backlog_used = 2599935fe098SMike Snitzer __ATTR(max_backlog_used, S_IRUGO | S_IWUSR, 2600935fe098SMike Snitzer behind_writes_used_show, behind_writes_used_reset); 2601935fe098SMike Snitzer 2602935fe098SMike Snitzer static struct attribute *md_bitmap_attrs[] = { 2603935fe098SMike Snitzer &bitmap_location.attr, 2604935fe098SMike Snitzer &bitmap_space.attr, 2605935fe098SMike Snitzer &bitmap_timeout.attr, 2606935fe098SMike Snitzer &bitmap_backlog.attr, 2607935fe098SMike Snitzer &bitmap_chunksize.attr, 2608935fe098SMike Snitzer &bitmap_metadata.attr, 2609935fe098SMike Snitzer &bitmap_can_clear.attr, 2610935fe098SMike Snitzer &max_backlog_used.attr, 2611935fe098SMike Snitzer NULL 2612935fe098SMike Snitzer }; 2613935fe098SMike Snitzer struct attribute_group md_bitmap_group = { 2614935fe098SMike Snitzer .name = "bitmap", 2615935fe098SMike Snitzer .attrs = md_bitmap_attrs, 2616935fe098SMike Snitzer }; 2617935fe098SMike Snitzer 2618