11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/fs/buffer.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Copyright (C) 1991, 1992, 2002 Linus Torvalds 51da177e4SLinus Torvalds */ 61da177e4SLinus Torvalds 71da177e4SLinus Torvalds /* 81da177e4SLinus Torvalds * Start bdflush() with kernel_thread not syscall - Paul Gortmaker, 12/95 91da177e4SLinus Torvalds * 101da177e4SLinus Torvalds * Removed a lot of unnecessary code and simplified things now that 111da177e4SLinus Torvalds * the buffer cache isn't our primary cache - Andrew Tridgell 12/96 121da177e4SLinus Torvalds * 131da177e4SLinus Torvalds * Speed up hash, lru, and free list operations. Use gfp() for allocating 141da177e4SLinus Torvalds * hash table, use SLAB cache for buffer heads. SMP threading. -DaveM 151da177e4SLinus Torvalds * 161da177e4SLinus Torvalds * Added 32k buffer block sizes - these are required older ARM systems. - RMK 171da177e4SLinus Torvalds * 181da177e4SLinus Torvalds * async buffer flushing, 1999 Andrea Arcangeli <andrea@suse.de> 191da177e4SLinus Torvalds */ 201da177e4SLinus Torvalds 211da177e4SLinus Torvalds #include <linux/config.h> 221da177e4SLinus Torvalds #include <linux/kernel.h> 231da177e4SLinus Torvalds #include <linux/syscalls.h> 241da177e4SLinus Torvalds #include <linux/fs.h> 251da177e4SLinus Torvalds #include <linux/mm.h> 261da177e4SLinus Torvalds #include <linux/percpu.h> 271da177e4SLinus Torvalds #include <linux/slab.h> 281da177e4SLinus Torvalds #include <linux/smp_lock.h> 291da177e4SLinus Torvalds #include <linux/blkdev.h> 301da177e4SLinus Torvalds #include <linux/file.h> 311da177e4SLinus Torvalds #include <linux/quotaops.h> 321da177e4SLinus Torvalds #include <linux/highmem.h> 331da177e4SLinus Torvalds #include <linux/module.h> 341da177e4SLinus Torvalds #include <linux/writeback.h> 351da177e4SLinus Torvalds #include <linux/hash.h> 361da177e4SLinus Torvalds #include <linux/suspend.h> 371da177e4SLinus Torvalds #include <linux/buffer_head.h> 381da177e4SLinus Torvalds #include <linux/bio.h> 391da177e4SLinus Torvalds #include <linux/notifier.h> 401da177e4SLinus Torvalds #include <linux/cpu.h> 411da177e4SLinus Torvalds #include <linux/bitops.h> 421da177e4SLinus Torvalds #include <linux/mpage.h> 431da177e4SLinus Torvalds 441da177e4SLinus Torvalds static int fsync_buffers_list(spinlock_t *lock, struct list_head *list); 451da177e4SLinus Torvalds static void invalidate_bh_lrus(void); 461da177e4SLinus Torvalds 471da177e4SLinus Torvalds #define BH_ENTRY(list) list_entry((list), struct buffer_head, b_assoc_buffers) 481da177e4SLinus Torvalds 491da177e4SLinus Torvalds inline void 501da177e4SLinus Torvalds init_buffer(struct buffer_head *bh, bh_end_io_t *handler, void *private) 511da177e4SLinus Torvalds { 521da177e4SLinus Torvalds bh->b_end_io = handler; 531da177e4SLinus Torvalds bh->b_private = private; 541da177e4SLinus Torvalds } 551da177e4SLinus Torvalds 561da177e4SLinus Torvalds static int sync_buffer(void *word) 571da177e4SLinus Torvalds { 581da177e4SLinus Torvalds struct block_device *bd; 591da177e4SLinus Torvalds struct buffer_head *bh 601da177e4SLinus Torvalds = container_of(word, struct buffer_head, b_state); 611da177e4SLinus Torvalds 621da177e4SLinus Torvalds smp_mb(); 631da177e4SLinus Torvalds bd = bh->b_bdev; 641da177e4SLinus Torvalds if (bd) 651da177e4SLinus Torvalds blk_run_address_space(bd->bd_inode->i_mapping); 661da177e4SLinus Torvalds io_schedule(); 671da177e4SLinus Torvalds return 0; 681da177e4SLinus Torvalds } 691da177e4SLinus Torvalds 701da177e4SLinus Torvalds void fastcall __lock_buffer(struct buffer_head *bh) 711da177e4SLinus Torvalds { 721da177e4SLinus Torvalds wait_on_bit_lock(&bh->b_state, BH_Lock, sync_buffer, 731da177e4SLinus Torvalds TASK_UNINTERRUPTIBLE); 741da177e4SLinus Torvalds } 751da177e4SLinus Torvalds EXPORT_SYMBOL(__lock_buffer); 761da177e4SLinus Torvalds 771da177e4SLinus Torvalds void fastcall unlock_buffer(struct buffer_head *bh) 781da177e4SLinus Torvalds { 791da177e4SLinus Torvalds clear_buffer_locked(bh); 801da177e4SLinus Torvalds smp_mb__after_clear_bit(); 811da177e4SLinus Torvalds wake_up_bit(&bh->b_state, BH_Lock); 821da177e4SLinus Torvalds } 831da177e4SLinus Torvalds 841da177e4SLinus Torvalds /* 851da177e4SLinus Torvalds * Block until a buffer comes unlocked. This doesn't stop it 861da177e4SLinus Torvalds * from becoming locked again - you have to lock it yourself 871da177e4SLinus Torvalds * if you want to preserve its state. 881da177e4SLinus Torvalds */ 891da177e4SLinus Torvalds void __wait_on_buffer(struct buffer_head * bh) 901da177e4SLinus Torvalds { 911da177e4SLinus Torvalds wait_on_bit(&bh->b_state, BH_Lock, sync_buffer, TASK_UNINTERRUPTIBLE); 921da177e4SLinus Torvalds } 931da177e4SLinus Torvalds 941da177e4SLinus Torvalds static void 951da177e4SLinus Torvalds __clear_page_buffers(struct page *page) 961da177e4SLinus Torvalds { 971da177e4SLinus Torvalds ClearPagePrivate(page); 981da177e4SLinus Torvalds page->private = 0; 991da177e4SLinus Torvalds page_cache_release(page); 1001da177e4SLinus Torvalds } 1011da177e4SLinus Torvalds 1021da177e4SLinus Torvalds static void buffer_io_error(struct buffer_head *bh) 1031da177e4SLinus Torvalds { 1041da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 1051da177e4SLinus Torvalds 1061da177e4SLinus Torvalds printk(KERN_ERR "Buffer I/O error on device %s, logical block %Lu\n", 1071da177e4SLinus Torvalds bdevname(bh->b_bdev, b), 1081da177e4SLinus Torvalds (unsigned long long)bh->b_blocknr); 1091da177e4SLinus Torvalds } 1101da177e4SLinus Torvalds 1111da177e4SLinus Torvalds /* 1121da177e4SLinus Torvalds * Default synchronous end-of-IO handler.. Just mark it up-to-date and 1131da177e4SLinus Torvalds * unlock the buffer. This is what ll_rw_block uses too. 1141da177e4SLinus Torvalds */ 1151da177e4SLinus Torvalds void end_buffer_read_sync(struct buffer_head *bh, int uptodate) 1161da177e4SLinus Torvalds { 1171da177e4SLinus Torvalds if (uptodate) { 1181da177e4SLinus Torvalds set_buffer_uptodate(bh); 1191da177e4SLinus Torvalds } else { 1201da177e4SLinus Torvalds /* This happens, due to failed READA attempts. */ 1211da177e4SLinus Torvalds clear_buffer_uptodate(bh); 1221da177e4SLinus Torvalds } 1231da177e4SLinus Torvalds unlock_buffer(bh); 1241da177e4SLinus Torvalds put_bh(bh); 1251da177e4SLinus Torvalds } 1261da177e4SLinus Torvalds 1271da177e4SLinus Torvalds void end_buffer_write_sync(struct buffer_head *bh, int uptodate) 1281da177e4SLinus Torvalds { 1291da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 1301da177e4SLinus Torvalds 1311da177e4SLinus Torvalds if (uptodate) { 1321da177e4SLinus Torvalds set_buffer_uptodate(bh); 1331da177e4SLinus Torvalds } else { 1341da177e4SLinus Torvalds if (!buffer_eopnotsupp(bh) && printk_ratelimit()) { 1351da177e4SLinus Torvalds buffer_io_error(bh); 1361da177e4SLinus Torvalds printk(KERN_WARNING "lost page write due to " 1371da177e4SLinus Torvalds "I/O error on %s\n", 1381da177e4SLinus Torvalds bdevname(bh->b_bdev, b)); 1391da177e4SLinus Torvalds } 1401da177e4SLinus Torvalds set_buffer_write_io_error(bh); 1411da177e4SLinus Torvalds clear_buffer_uptodate(bh); 1421da177e4SLinus Torvalds } 1431da177e4SLinus Torvalds unlock_buffer(bh); 1441da177e4SLinus Torvalds put_bh(bh); 1451da177e4SLinus Torvalds } 1461da177e4SLinus Torvalds 1471da177e4SLinus Torvalds /* 1481da177e4SLinus Torvalds * Write out and wait upon all the dirty data associated with a block 1491da177e4SLinus Torvalds * device via its mapping. Does not take the superblock lock. 1501da177e4SLinus Torvalds */ 1511da177e4SLinus Torvalds int sync_blockdev(struct block_device *bdev) 1521da177e4SLinus Torvalds { 1531da177e4SLinus Torvalds int ret = 0; 1541da177e4SLinus Torvalds 1551da177e4SLinus Torvalds if (bdev) { 1561da177e4SLinus Torvalds int err; 1571da177e4SLinus Torvalds 1581da177e4SLinus Torvalds ret = filemap_fdatawrite(bdev->bd_inode->i_mapping); 1591da177e4SLinus Torvalds err = filemap_fdatawait(bdev->bd_inode->i_mapping); 1601da177e4SLinus Torvalds if (!ret) 1611da177e4SLinus Torvalds ret = err; 1621da177e4SLinus Torvalds } 1631da177e4SLinus Torvalds return ret; 1641da177e4SLinus Torvalds } 1651da177e4SLinus Torvalds EXPORT_SYMBOL(sync_blockdev); 1661da177e4SLinus Torvalds 1671da177e4SLinus Torvalds /* 1681da177e4SLinus Torvalds * Write out and wait upon all dirty data associated with this 1691da177e4SLinus Torvalds * superblock. Filesystem data as well as the underlying block 1701da177e4SLinus Torvalds * device. Takes the superblock lock. 1711da177e4SLinus Torvalds */ 1721da177e4SLinus Torvalds int fsync_super(struct super_block *sb) 1731da177e4SLinus Torvalds { 1741da177e4SLinus Torvalds sync_inodes_sb(sb, 0); 1751da177e4SLinus Torvalds DQUOT_SYNC(sb); 1761da177e4SLinus Torvalds lock_super(sb); 1771da177e4SLinus Torvalds if (sb->s_dirt && sb->s_op->write_super) 1781da177e4SLinus Torvalds sb->s_op->write_super(sb); 1791da177e4SLinus Torvalds unlock_super(sb); 1801da177e4SLinus Torvalds if (sb->s_op->sync_fs) 1811da177e4SLinus Torvalds sb->s_op->sync_fs(sb, 1); 1821da177e4SLinus Torvalds sync_blockdev(sb->s_bdev); 1831da177e4SLinus Torvalds sync_inodes_sb(sb, 1); 1841da177e4SLinus Torvalds 1851da177e4SLinus Torvalds return sync_blockdev(sb->s_bdev); 1861da177e4SLinus Torvalds } 1871da177e4SLinus Torvalds 1881da177e4SLinus Torvalds /* 1891da177e4SLinus Torvalds * Write out and wait upon all dirty data associated with this 1901da177e4SLinus Torvalds * device. Filesystem data as well as the underlying block 1911da177e4SLinus Torvalds * device. Takes the superblock lock. 1921da177e4SLinus Torvalds */ 1931da177e4SLinus Torvalds int fsync_bdev(struct block_device *bdev) 1941da177e4SLinus Torvalds { 1951da177e4SLinus Torvalds struct super_block *sb = get_super(bdev); 1961da177e4SLinus Torvalds if (sb) { 1971da177e4SLinus Torvalds int res = fsync_super(sb); 1981da177e4SLinus Torvalds drop_super(sb); 1991da177e4SLinus Torvalds return res; 2001da177e4SLinus Torvalds } 2011da177e4SLinus Torvalds return sync_blockdev(bdev); 2021da177e4SLinus Torvalds } 2031da177e4SLinus Torvalds 2041da177e4SLinus Torvalds /** 2051da177e4SLinus Torvalds * freeze_bdev -- lock a filesystem and force it into a consistent state 2061da177e4SLinus Torvalds * @bdev: blockdevice to lock 2071da177e4SLinus Torvalds * 2081da177e4SLinus Torvalds * This takes the block device bd_mount_sem to make sure no new mounts 2091da177e4SLinus Torvalds * happen on bdev until thaw_bdev() is called. 2101da177e4SLinus Torvalds * If a superblock is found on this device, we take the s_umount semaphore 2111da177e4SLinus Torvalds * on it to make sure nobody unmounts until the snapshot creation is done. 2121da177e4SLinus Torvalds */ 2131da177e4SLinus Torvalds struct super_block *freeze_bdev(struct block_device *bdev) 2141da177e4SLinus Torvalds { 2151da177e4SLinus Torvalds struct super_block *sb; 2161da177e4SLinus Torvalds 2171da177e4SLinus Torvalds down(&bdev->bd_mount_sem); 2181da177e4SLinus Torvalds sb = get_super(bdev); 2191da177e4SLinus Torvalds if (sb && !(sb->s_flags & MS_RDONLY)) { 2201da177e4SLinus Torvalds sb->s_frozen = SB_FREEZE_WRITE; 221d59dd462Sakpm@osdl.org smp_wmb(); 2221da177e4SLinus Torvalds 2231da177e4SLinus Torvalds sync_inodes_sb(sb, 0); 2241da177e4SLinus Torvalds DQUOT_SYNC(sb); 2251da177e4SLinus Torvalds 2261da177e4SLinus Torvalds lock_super(sb); 2271da177e4SLinus Torvalds if (sb->s_dirt && sb->s_op->write_super) 2281da177e4SLinus Torvalds sb->s_op->write_super(sb); 2291da177e4SLinus Torvalds unlock_super(sb); 2301da177e4SLinus Torvalds 2311da177e4SLinus Torvalds if (sb->s_op->sync_fs) 2321da177e4SLinus Torvalds sb->s_op->sync_fs(sb, 1); 2331da177e4SLinus Torvalds 2341da177e4SLinus Torvalds sync_blockdev(sb->s_bdev); 2351da177e4SLinus Torvalds sync_inodes_sb(sb, 1); 2361da177e4SLinus Torvalds 2371da177e4SLinus Torvalds sb->s_frozen = SB_FREEZE_TRANS; 238d59dd462Sakpm@osdl.org smp_wmb(); 2391da177e4SLinus Torvalds 2401da177e4SLinus Torvalds sync_blockdev(sb->s_bdev); 2411da177e4SLinus Torvalds 2421da177e4SLinus Torvalds if (sb->s_op->write_super_lockfs) 2431da177e4SLinus Torvalds sb->s_op->write_super_lockfs(sb); 2441da177e4SLinus Torvalds } 2451da177e4SLinus Torvalds 2461da177e4SLinus Torvalds sync_blockdev(bdev); 2471da177e4SLinus Torvalds return sb; /* thaw_bdev releases s->s_umount and bd_mount_sem */ 2481da177e4SLinus Torvalds } 2491da177e4SLinus Torvalds EXPORT_SYMBOL(freeze_bdev); 2501da177e4SLinus Torvalds 2511da177e4SLinus Torvalds /** 2521da177e4SLinus Torvalds * thaw_bdev -- unlock filesystem 2531da177e4SLinus Torvalds * @bdev: blockdevice to unlock 2541da177e4SLinus Torvalds * @sb: associated superblock 2551da177e4SLinus Torvalds * 2561da177e4SLinus Torvalds * Unlocks the filesystem and marks it writeable again after freeze_bdev(). 2571da177e4SLinus Torvalds */ 2581da177e4SLinus Torvalds void thaw_bdev(struct block_device *bdev, struct super_block *sb) 2591da177e4SLinus Torvalds { 2601da177e4SLinus Torvalds if (sb) { 2611da177e4SLinus Torvalds BUG_ON(sb->s_bdev != bdev); 2621da177e4SLinus Torvalds 2631da177e4SLinus Torvalds if (sb->s_op->unlockfs) 2641da177e4SLinus Torvalds sb->s_op->unlockfs(sb); 2651da177e4SLinus Torvalds sb->s_frozen = SB_UNFROZEN; 266d59dd462Sakpm@osdl.org smp_wmb(); 2671da177e4SLinus Torvalds wake_up(&sb->s_wait_unfrozen); 2681da177e4SLinus Torvalds drop_super(sb); 2691da177e4SLinus Torvalds } 2701da177e4SLinus Torvalds 2711da177e4SLinus Torvalds up(&bdev->bd_mount_sem); 2721da177e4SLinus Torvalds } 2731da177e4SLinus Torvalds EXPORT_SYMBOL(thaw_bdev); 2741da177e4SLinus Torvalds 2751da177e4SLinus Torvalds /* 2761da177e4SLinus Torvalds * sync everything. Start out by waking pdflush, because that writes back 2771da177e4SLinus Torvalds * all queues in parallel. 2781da177e4SLinus Torvalds */ 2791da177e4SLinus Torvalds static void do_sync(unsigned long wait) 2801da177e4SLinus Torvalds { 2811da177e4SLinus Torvalds wakeup_bdflush(0); 2821da177e4SLinus Torvalds sync_inodes(0); /* All mappings, inodes and their blockdevs */ 2831da177e4SLinus Torvalds DQUOT_SYNC(NULL); 2841da177e4SLinus Torvalds sync_supers(); /* Write the superblocks */ 2851da177e4SLinus Torvalds sync_filesystems(0); /* Start syncing the filesystems */ 2861da177e4SLinus Torvalds sync_filesystems(wait); /* Waitingly sync the filesystems */ 2871da177e4SLinus Torvalds sync_inodes(wait); /* Mappings, inodes and blockdevs, again. */ 2881da177e4SLinus Torvalds if (!wait) 2891da177e4SLinus Torvalds printk("Emergency Sync complete\n"); 2901da177e4SLinus Torvalds if (unlikely(laptop_mode)) 2911da177e4SLinus Torvalds laptop_sync_completion(); 2921da177e4SLinus Torvalds } 2931da177e4SLinus Torvalds 2941da177e4SLinus Torvalds asmlinkage long sys_sync(void) 2951da177e4SLinus Torvalds { 2961da177e4SLinus Torvalds do_sync(1); 2971da177e4SLinus Torvalds return 0; 2981da177e4SLinus Torvalds } 2991da177e4SLinus Torvalds 3001da177e4SLinus Torvalds void emergency_sync(void) 3011da177e4SLinus Torvalds { 3021da177e4SLinus Torvalds pdflush_operation(do_sync, 0); 3031da177e4SLinus Torvalds } 3041da177e4SLinus Torvalds 3051da177e4SLinus Torvalds /* 3061da177e4SLinus Torvalds * Generic function to fsync a file. 3071da177e4SLinus Torvalds * 3081da177e4SLinus Torvalds * filp may be NULL if called via the msync of a vma. 3091da177e4SLinus Torvalds */ 3101da177e4SLinus Torvalds 3111da177e4SLinus Torvalds int file_fsync(struct file *filp, struct dentry *dentry, int datasync) 3121da177e4SLinus Torvalds { 3131da177e4SLinus Torvalds struct inode * inode = dentry->d_inode; 3141da177e4SLinus Torvalds struct super_block * sb; 3151da177e4SLinus Torvalds int ret, err; 3161da177e4SLinus Torvalds 3171da177e4SLinus Torvalds /* sync the inode to buffers */ 3181da177e4SLinus Torvalds ret = write_inode_now(inode, 0); 3191da177e4SLinus Torvalds 3201da177e4SLinus Torvalds /* sync the superblock to buffers */ 3211da177e4SLinus Torvalds sb = inode->i_sb; 3221da177e4SLinus Torvalds lock_super(sb); 3231da177e4SLinus Torvalds if (sb->s_op->write_super) 3241da177e4SLinus Torvalds sb->s_op->write_super(sb); 3251da177e4SLinus Torvalds unlock_super(sb); 3261da177e4SLinus Torvalds 3271da177e4SLinus Torvalds /* .. finally sync the buffers to disk */ 3281da177e4SLinus Torvalds err = sync_blockdev(sb->s_bdev); 3291da177e4SLinus Torvalds if (!ret) 3301da177e4SLinus Torvalds ret = err; 3311da177e4SLinus Torvalds return ret; 3321da177e4SLinus Torvalds } 3331da177e4SLinus Torvalds 3341da177e4SLinus Torvalds asmlinkage long sys_fsync(unsigned int fd) 3351da177e4SLinus Torvalds { 3361da177e4SLinus Torvalds struct file * file; 3371da177e4SLinus Torvalds struct address_space *mapping; 3381da177e4SLinus Torvalds int ret, err; 3391da177e4SLinus Torvalds 3401da177e4SLinus Torvalds ret = -EBADF; 3411da177e4SLinus Torvalds file = fget(fd); 3421da177e4SLinus Torvalds if (!file) 3431da177e4SLinus Torvalds goto out; 3441da177e4SLinus Torvalds 3451da177e4SLinus Torvalds mapping = file->f_mapping; 3461da177e4SLinus Torvalds 3471da177e4SLinus Torvalds ret = -EINVAL; 3481da177e4SLinus Torvalds if (!file->f_op || !file->f_op->fsync) { 3491da177e4SLinus Torvalds /* Why? We can still call filemap_fdatawrite */ 3501da177e4SLinus Torvalds goto out_putf; 3511da177e4SLinus Torvalds } 3521da177e4SLinus Torvalds 3531da177e4SLinus Torvalds current->flags |= PF_SYNCWRITE; 3541da177e4SLinus Torvalds ret = filemap_fdatawrite(mapping); 3551da177e4SLinus Torvalds 3561da177e4SLinus Torvalds /* 3571da177e4SLinus Torvalds * We need to protect against concurrent writers, 3581da177e4SLinus Torvalds * which could cause livelocks in fsync_buffers_list 3591da177e4SLinus Torvalds */ 3601da177e4SLinus Torvalds down(&mapping->host->i_sem); 3611da177e4SLinus Torvalds err = file->f_op->fsync(file, file->f_dentry, 0); 3621da177e4SLinus Torvalds if (!ret) 3631da177e4SLinus Torvalds ret = err; 3641da177e4SLinus Torvalds up(&mapping->host->i_sem); 3651da177e4SLinus Torvalds err = filemap_fdatawait(mapping); 3661da177e4SLinus Torvalds if (!ret) 3671da177e4SLinus Torvalds ret = err; 3681da177e4SLinus Torvalds current->flags &= ~PF_SYNCWRITE; 3691da177e4SLinus Torvalds 3701da177e4SLinus Torvalds out_putf: 3711da177e4SLinus Torvalds fput(file); 3721da177e4SLinus Torvalds out: 3731da177e4SLinus Torvalds return ret; 3741da177e4SLinus Torvalds } 3751da177e4SLinus Torvalds 3761da177e4SLinus Torvalds asmlinkage long sys_fdatasync(unsigned int fd) 3771da177e4SLinus Torvalds { 3781da177e4SLinus Torvalds struct file * file; 3791da177e4SLinus Torvalds struct address_space *mapping; 3801da177e4SLinus Torvalds int ret, err; 3811da177e4SLinus Torvalds 3821da177e4SLinus Torvalds ret = -EBADF; 3831da177e4SLinus Torvalds file = fget(fd); 3841da177e4SLinus Torvalds if (!file) 3851da177e4SLinus Torvalds goto out; 3861da177e4SLinus Torvalds 3871da177e4SLinus Torvalds ret = -EINVAL; 3881da177e4SLinus Torvalds if (!file->f_op || !file->f_op->fsync) 3891da177e4SLinus Torvalds goto out_putf; 3901da177e4SLinus Torvalds 3911da177e4SLinus Torvalds mapping = file->f_mapping; 3921da177e4SLinus Torvalds 3931da177e4SLinus Torvalds current->flags |= PF_SYNCWRITE; 3941da177e4SLinus Torvalds ret = filemap_fdatawrite(mapping); 3951da177e4SLinus Torvalds down(&mapping->host->i_sem); 3961da177e4SLinus Torvalds err = file->f_op->fsync(file, file->f_dentry, 1); 3971da177e4SLinus Torvalds if (!ret) 3981da177e4SLinus Torvalds ret = err; 3991da177e4SLinus Torvalds up(&mapping->host->i_sem); 4001da177e4SLinus Torvalds err = filemap_fdatawait(mapping); 4011da177e4SLinus Torvalds if (!ret) 4021da177e4SLinus Torvalds ret = err; 4031da177e4SLinus Torvalds current->flags &= ~PF_SYNCWRITE; 4041da177e4SLinus Torvalds 4051da177e4SLinus Torvalds out_putf: 4061da177e4SLinus Torvalds fput(file); 4071da177e4SLinus Torvalds out: 4081da177e4SLinus Torvalds return ret; 4091da177e4SLinus Torvalds } 4101da177e4SLinus Torvalds 4111da177e4SLinus Torvalds /* 4121da177e4SLinus Torvalds * Various filesystems appear to want __find_get_block to be non-blocking. 4131da177e4SLinus Torvalds * But it's the page lock which protects the buffers. To get around this, 4141da177e4SLinus Torvalds * we get exclusion from try_to_free_buffers with the blockdev mapping's 4151da177e4SLinus Torvalds * private_lock. 4161da177e4SLinus Torvalds * 4171da177e4SLinus Torvalds * Hack idea: for the blockdev mapping, i_bufferlist_lock contention 4181da177e4SLinus Torvalds * may be quite high. This code could TryLock the page, and if that 4191da177e4SLinus Torvalds * succeeds, there is no need to take private_lock. (But if 4201da177e4SLinus Torvalds * private_lock is contended then so is mapping->tree_lock). 4211da177e4SLinus Torvalds */ 4221da177e4SLinus Torvalds static struct buffer_head * 4231da177e4SLinus Torvalds __find_get_block_slow(struct block_device *bdev, sector_t block, int unused) 4241da177e4SLinus Torvalds { 4251da177e4SLinus Torvalds struct inode *bd_inode = bdev->bd_inode; 4261da177e4SLinus Torvalds struct address_space *bd_mapping = bd_inode->i_mapping; 4271da177e4SLinus Torvalds struct buffer_head *ret = NULL; 4281da177e4SLinus Torvalds pgoff_t index; 4291da177e4SLinus Torvalds struct buffer_head *bh; 4301da177e4SLinus Torvalds struct buffer_head *head; 4311da177e4SLinus Torvalds struct page *page; 4321da177e4SLinus Torvalds int all_mapped = 1; 4331da177e4SLinus Torvalds 4341da177e4SLinus Torvalds index = block >> (PAGE_CACHE_SHIFT - bd_inode->i_blkbits); 4351da177e4SLinus Torvalds page = find_get_page(bd_mapping, index); 4361da177e4SLinus Torvalds if (!page) 4371da177e4SLinus Torvalds goto out; 4381da177e4SLinus Torvalds 4391da177e4SLinus Torvalds spin_lock(&bd_mapping->private_lock); 4401da177e4SLinus Torvalds if (!page_has_buffers(page)) 4411da177e4SLinus Torvalds goto out_unlock; 4421da177e4SLinus Torvalds head = page_buffers(page); 4431da177e4SLinus Torvalds bh = head; 4441da177e4SLinus Torvalds do { 4451da177e4SLinus Torvalds if (bh->b_blocknr == block) { 4461da177e4SLinus Torvalds ret = bh; 4471da177e4SLinus Torvalds get_bh(bh); 4481da177e4SLinus Torvalds goto out_unlock; 4491da177e4SLinus Torvalds } 4501da177e4SLinus Torvalds if (!buffer_mapped(bh)) 4511da177e4SLinus Torvalds all_mapped = 0; 4521da177e4SLinus Torvalds bh = bh->b_this_page; 4531da177e4SLinus Torvalds } while (bh != head); 4541da177e4SLinus Torvalds 4551da177e4SLinus Torvalds /* we might be here because some of the buffers on this page are 4561da177e4SLinus Torvalds * not mapped. This is due to various races between 4571da177e4SLinus Torvalds * file io on the block device and getblk. It gets dealt with 4581da177e4SLinus Torvalds * elsewhere, don't buffer_error if we had some unmapped buffers 4591da177e4SLinus Torvalds */ 4601da177e4SLinus Torvalds if (all_mapped) { 4611da177e4SLinus Torvalds printk("__find_get_block_slow() failed. " 4621da177e4SLinus Torvalds "block=%llu, b_blocknr=%llu\n", 4631da177e4SLinus Torvalds (unsigned long long)block, (unsigned long long)bh->b_blocknr); 4641da177e4SLinus Torvalds printk("b_state=0x%08lx, b_size=%u\n", bh->b_state, bh->b_size); 4651da177e4SLinus Torvalds printk("device blocksize: %d\n", 1 << bd_inode->i_blkbits); 4661da177e4SLinus Torvalds } 4671da177e4SLinus Torvalds out_unlock: 4681da177e4SLinus Torvalds spin_unlock(&bd_mapping->private_lock); 4691da177e4SLinus Torvalds page_cache_release(page); 4701da177e4SLinus Torvalds out: 4711da177e4SLinus Torvalds return ret; 4721da177e4SLinus Torvalds } 4731da177e4SLinus Torvalds 4741da177e4SLinus Torvalds /* If invalidate_buffers() will trash dirty buffers, it means some kind 4751da177e4SLinus Torvalds of fs corruption is going on. Trashing dirty data always imply losing 4761da177e4SLinus Torvalds information that was supposed to be just stored on the physical layer 4771da177e4SLinus Torvalds by the user. 4781da177e4SLinus Torvalds 4791da177e4SLinus Torvalds Thus invalidate_buffers in general usage is not allwowed to trash 4801da177e4SLinus Torvalds dirty buffers. For example ioctl(FLSBLKBUF) expects dirty data to 4811da177e4SLinus Torvalds be preserved. These buffers are simply skipped. 4821da177e4SLinus Torvalds 4831da177e4SLinus Torvalds We also skip buffers which are still in use. For example this can 4841da177e4SLinus Torvalds happen if a userspace program is reading the block device. 4851da177e4SLinus Torvalds 4861da177e4SLinus Torvalds NOTE: In the case where the user removed a removable-media-disk even if 4871da177e4SLinus Torvalds there's still dirty data not synced on disk (due a bug in the device driver 4881da177e4SLinus Torvalds or due an error of the user), by not destroying the dirty buffers we could 4891da177e4SLinus Torvalds generate corruption also on the next media inserted, thus a parameter is 4901da177e4SLinus Torvalds necessary to handle this case in the most safe way possible (trying 4911da177e4SLinus Torvalds to not corrupt also the new disk inserted with the data belonging to 4921da177e4SLinus Torvalds the old now corrupted disk). Also for the ramdisk the natural thing 4931da177e4SLinus Torvalds to do in order to release the ramdisk memory is to destroy dirty buffers. 4941da177e4SLinus Torvalds 4951da177e4SLinus Torvalds These are two special cases. Normal usage imply the device driver 4961da177e4SLinus Torvalds to issue a sync on the device (without waiting I/O completion) and 4971da177e4SLinus Torvalds then an invalidate_buffers call that doesn't trash dirty buffers. 4981da177e4SLinus Torvalds 4991da177e4SLinus Torvalds For handling cache coherency with the blkdev pagecache the 'update' case 5001da177e4SLinus Torvalds is been introduced. It is needed to re-read from disk any pinned 5011da177e4SLinus Torvalds buffer. NOTE: re-reading from disk is destructive so we can do it only 5021da177e4SLinus Torvalds when we assume nobody is changing the buffercache under our I/O and when 5031da177e4SLinus Torvalds we think the disk contains more recent information than the buffercache. 5041da177e4SLinus Torvalds The update == 1 pass marks the buffers we need to update, the update == 2 5051da177e4SLinus Torvalds pass does the actual I/O. */ 5061da177e4SLinus Torvalds void invalidate_bdev(struct block_device *bdev, int destroy_dirty_buffers) 5071da177e4SLinus Torvalds { 5081da177e4SLinus Torvalds invalidate_bh_lrus(); 5091da177e4SLinus Torvalds /* 5101da177e4SLinus Torvalds * FIXME: what about destroy_dirty_buffers? 5111da177e4SLinus Torvalds * We really want to use invalidate_inode_pages2() for 5121da177e4SLinus Torvalds * that, but not until that's cleaned up. 5131da177e4SLinus Torvalds */ 5141da177e4SLinus Torvalds invalidate_inode_pages(bdev->bd_inode->i_mapping); 5151da177e4SLinus Torvalds } 5161da177e4SLinus Torvalds 5171da177e4SLinus Torvalds /* 5181da177e4SLinus Torvalds * Kick pdflush then try to free up some ZONE_NORMAL memory. 5191da177e4SLinus Torvalds */ 5201da177e4SLinus Torvalds static void free_more_memory(void) 5211da177e4SLinus Torvalds { 5221da177e4SLinus Torvalds struct zone **zones; 5231da177e4SLinus Torvalds pg_data_t *pgdat; 5241da177e4SLinus Torvalds 5251da177e4SLinus Torvalds wakeup_bdflush(1024); 5261da177e4SLinus Torvalds yield(); 5271da177e4SLinus Torvalds 5281da177e4SLinus Torvalds for_each_pgdat(pgdat) { 5291da177e4SLinus Torvalds zones = pgdat->node_zonelists[GFP_NOFS&GFP_ZONEMASK].zones; 5301da177e4SLinus Torvalds if (*zones) 5311da177e4SLinus Torvalds try_to_free_pages(zones, GFP_NOFS, 0); 5321da177e4SLinus Torvalds } 5331da177e4SLinus Torvalds } 5341da177e4SLinus Torvalds 5351da177e4SLinus Torvalds /* 5361da177e4SLinus Torvalds * I/O completion handler for block_read_full_page() - pages 5371da177e4SLinus Torvalds * which come unlocked at the end of I/O. 5381da177e4SLinus Torvalds */ 5391da177e4SLinus Torvalds static void end_buffer_async_read(struct buffer_head *bh, int uptodate) 5401da177e4SLinus Torvalds { 5411da177e4SLinus Torvalds static DEFINE_SPINLOCK(page_uptodate_lock); 5421da177e4SLinus Torvalds unsigned long flags; 5431da177e4SLinus Torvalds struct buffer_head *tmp; 5441da177e4SLinus Torvalds struct page *page; 5451da177e4SLinus Torvalds int page_uptodate = 1; 5461da177e4SLinus Torvalds 5471da177e4SLinus Torvalds BUG_ON(!buffer_async_read(bh)); 5481da177e4SLinus Torvalds 5491da177e4SLinus Torvalds page = bh->b_page; 5501da177e4SLinus Torvalds if (uptodate) { 5511da177e4SLinus Torvalds set_buffer_uptodate(bh); 5521da177e4SLinus Torvalds } else { 5531da177e4SLinus Torvalds clear_buffer_uptodate(bh); 5541da177e4SLinus Torvalds if (printk_ratelimit()) 5551da177e4SLinus Torvalds buffer_io_error(bh); 5561da177e4SLinus Torvalds SetPageError(page); 5571da177e4SLinus Torvalds } 5581da177e4SLinus Torvalds 5591da177e4SLinus Torvalds /* 5601da177e4SLinus Torvalds * Be _very_ careful from here on. Bad things can happen if 5611da177e4SLinus Torvalds * two buffer heads end IO at almost the same time and both 5621da177e4SLinus Torvalds * decide that the page is now completely done. 5631da177e4SLinus Torvalds */ 5641da177e4SLinus Torvalds spin_lock_irqsave(&page_uptodate_lock, flags); 5651da177e4SLinus Torvalds clear_buffer_async_read(bh); 5661da177e4SLinus Torvalds unlock_buffer(bh); 5671da177e4SLinus Torvalds tmp = bh; 5681da177e4SLinus Torvalds do { 5691da177e4SLinus Torvalds if (!buffer_uptodate(tmp)) 5701da177e4SLinus Torvalds page_uptodate = 0; 5711da177e4SLinus Torvalds if (buffer_async_read(tmp)) { 5721da177e4SLinus Torvalds BUG_ON(!buffer_locked(tmp)); 5731da177e4SLinus Torvalds goto still_busy; 5741da177e4SLinus Torvalds } 5751da177e4SLinus Torvalds tmp = tmp->b_this_page; 5761da177e4SLinus Torvalds } while (tmp != bh); 5771da177e4SLinus Torvalds spin_unlock_irqrestore(&page_uptodate_lock, flags); 5781da177e4SLinus Torvalds 5791da177e4SLinus Torvalds /* 5801da177e4SLinus Torvalds * If none of the buffers had errors and they are all 5811da177e4SLinus Torvalds * uptodate then we can set the page uptodate. 5821da177e4SLinus Torvalds */ 5831da177e4SLinus Torvalds if (page_uptodate && !PageError(page)) 5841da177e4SLinus Torvalds SetPageUptodate(page); 5851da177e4SLinus Torvalds unlock_page(page); 5861da177e4SLinus Torvalds return; 5871da177e4SLinus Torvalds 5881da177e4SLinus Torvalds still_busy: 5891da177e4SLinus Torvalds spin_unlock_irqrestore(&page_uptodate_lock, flags); 5901da177e4SLinus Torvalds return; 5911da177e4SLinus Torvalds } 5921da177e4SLinus Torvalds 5931da177e4SLinus Torvalds /* 5941da177e4SLinus Torvalds * Completion handler for block_write_full_page() - pages which are unlocked 5951da177e4SLinus Torvalds * during I/O, and which have PageWriteback cleared upon I/O completion. 5961da177e4SLinus Torvalds */ 5971da177e4SLinus Torvalds void end_buffer_async_write(struct buffer_head *bh, int uptodate) 5981da177e4SLinus Torvalds { 5991da177e4SLinus Torvalds char b[BDEVNAME_SIZE]; 6001da177e4SLinus Torvalds static DEFINE_SPINLOCK(page_uptodate_lock); 6011da177e4SLinus Torvalds unsigned long flags; 6021da177e4SLinus Torvalds struct buffer_head *tmp; 6031da177e4SLinus Torvalds struct page *page; 6041da177e4SLinus Torvalds 6051da177e4SLinus Torvalds BUG_ON(!buffer_async_write(bh)); 6061da177e4SLinus Torvalds 6071da177e4SLinus Torvalds page = bh->b_page; 6081da177e4SLinus Torvalds if (uptodate) { 6091da177e4SLinus Torvalds set_buffer_uptodate(bh); 6101da177e4SLinus Torvalds } else { 6111da177e4SLinus Torvalds if (printk_ratelimit()) { 6121da177e4SLinus Torvalds buffer_io_error(bh); 6131da177e4SLinus Torvalds printk(KERN_WARNING "lost page write due to " 6141da177e4SLinus Torvalds "I/O error on %s\n", 6151da177e4SLinus Torvalds bdevname(bh->b_bdev, b)); 6161da177e4SLinus Torvalds } 6171da177e4SLinus Torvalds set_bit(AS_EIO, &page->mapping->flags); 6181da177e4SLinus Torvalds clear_buffer_uptodate(bh); 6191da177e4SLinus Torvalds SetPageError(page); 6201da177e4SLinus Torvalds } 6211da177e4SLinus Torvalds 6221da177e4SLinus Torvalds spin_lock_irqsave(&page_uptodate_lock, flags); 6231da177e4SLinus Torvalds clear_buffer_async_write(bh); 6241da177e4SLinus Torvalds unlock_buffer(bh); 6251da177e4SLinus Torvalds tmp = bh->b_this_page; 6261da177e4SLinus Torvalds while (tmp != bh) { 6271da177e4SLinus Torvalds if (buffer_async_write(tmp)) { 6281da177e4SLinus Torvalds BUG_ON(!buffer_locked(tmp)); 6291da177e4SLinus Torvalds goto still_busy; 6301da177e4SLinus Torvalds } 6311da177e4SLinus Torvalds tmp = tmp->b_this_page; 6321da177e4SLinus Torvalds } 6331da177e4SLinus Torvalds spin_unlock_irqrestore(&page_uptodate_lock, flags); 6341da177e4SLinus Torvalds end_page_writeback(page); 6351da177e4SLinus Torvalds return; 6361da177e4SLinus Torvalds 6371da177e4SLinus Torvalds still_busy: 6381da177e4SLinus Torvalds spin_unlock_irqrestore(&page_uptodate_lock, flags); 6391da177e4SLinus Torvalds return; 6401da177e4SLinus Torvalds } 6411da177e4SLinus Torvalds 6421da177e4SLinus Torvalds /* 6431da177e4SLinus Torvalds * If a page's buffers are under async readin (end_buffer_async_read 6441da177e4SLinus Torvalds * completion) then there is a possibility that another thread of 6451da177e4SLinus Torvalds * control could lock one of the buffers after it has completed 6461da177e4SLinus Torvalds * but while some of the other buffers have not completed. This 6471da177e4SLinus Torvalds * locked buffer would confuse end_buffer_async_read() into not unlocking 6481da177e4SLinus Torvalds * the page. So the absence of BH_Async_Read tells end_buffer_async_read() 6491da177e4SLinus Torvalds * that this buffer is not under async I/O. 6501da177e4SLinus Torvalds * 6511da177e4SLinus Torvalds * The page comes unlocked when it has no locked buffer_async buffers 6521da177e4SLinus Torvalds * left. 6531da177e4SLinus Torvalds * 6541da177e4SLinus Torvalds * PageLocked prevents anyone starting new async I/O reads any of 6551da177e4SLinus Torvalds * the buffers. 6561da177e4SLinus Torvalds * 6571da177e4SLinus Torvalds * PageWriteback is used to prevent simultaneous writeout of the same 6581da177e4SLinus Torvalds * page. 6591da177e4SLinus Torvalds * 6601da177e4SLinus Torvalds * PageLocked prevents anyone from starting writeback of a page which is 6611da177e4SLinus Torvalds * under read I/O (PageWriteback is only ever set against a locked page). 6621da177e4SLinus Torvalds */ 6631da177e4SLinus Torvalds static void mark_buffer_async_read(struct buffer_head *bh) 6641da177e4SLinus Torvalds { 6651da177e4SLinus Torvalds bh->b_end_io = end_buffer_async_read; 6661da177e4SLinus Torvalds set_buffer_async_read(bh); 6671da177e4SLinus Torvalds } 6681da177e4SLinus Torvalds 6691da177e4SLinus Torvalds void mark_buffer_async_write(struct buffer_head *bh) 6701da177e4SLinus Torvalds { 6711da177e4SLinus Torvalds bh->b_end_io = end_buffer_async_write; 6721da177e4SLinus Torvalds set_buffer_async_write(bh); 6731da177e4SLinus Torvalds } 6741da177e4SLinus Torvalds EXPORT_SYMBOL(mark_buffer_async_write); 6751da177e4SLinus Torvalds 6761da177e4SLinus Torvalds 6771da177e4SLinus Torvalds /* 6781da177e4SLinus Torvalds * fs/buffer.c contains helper functions for buffer-backed address space's 6791da177e4SLinus Torvalds * fsync functions. A common requirement for buffer-based filesystems is 6801da177e4SLinus Torvalds * that certain data from the backing blockdev needs to be written out for 6811da177e4SLinus Torvalds * a successful fsync(). For example, ext2 indirect blocks need to be 6821da177e4SLinus Torvalds * written back and waited upon before fsync() returns. 6831da177e4SLinus Torvalds * 6841da177e4SLinus Torvalds * The functions mark_buffer_inode_dirty(), fsync_inode_buffers(), 6851da177e4SLinus Torvalds * inode_has_buffers() and invalidate_inode_buffers() are provided for the 6861da177e4SLinus Torvalds * management of a list of dependent buffers at ->i_mapping->private_list. 6871da177e4SLinus Torvalds * 6881da177e4SLinus Torvalds * Locking is a little subtle: try_to_free_buffers() will remove buffers 6891da177e4SLinus Torvalds * from their controlling inode's queue when they are being freed. But 6901da177e4SLinus Torvalds * try_to_free_buffers() will be operating against the *blockdev* mapping 6911da177e4SLinus Torvalds * at the time, not against the S_ISREG file which depends on those buffers. 6921da177e4SLinus Torvalds * So the locking for private_list is via the private_lock in the address_space 6931da177e4SLinus Torvalds * which backs the buffers. Which is different from the address_space 6941da177e4SLinus Torvalds * against which the buffers are listed. So for a particular address_space, 6951da177e4SLinus Torvalds * mapping->private_lock does *not* protect mapping->private_list! In fact, 6961da177e4SLinus Torvalds * mapping->private_list will always be protected by the backing blockdev's 6971da177e4SLinus Torvalds * ->private_lock. 6981da177e4SLinus Torvalds * 6991da177e4SLinus Torvalds * Which introduces a requirement: all buffers on an address_space's 7001da177e4SLinus Torvalds * ->private_list must be from the same address_space: the blockdev's. 7011da177e4SLinus Torvalds * 7021da177e4SLinus Torvalds * address_spaces which do not place buffers at ->private_list via these 7031da177e4SLinus Torvalds * utility functions are free to use private_lock and private_list for 7041da177e4SLinus Torvalds * whatever they want. The only requirement is that list_empty(private_list) 7051da177e4SLinus Torvalds * be true at clear_inode() time. 7061da177e4SLinus Torvalds * 7071da177e4SLinus Torvalds * FIXME: clear_inode should not call invalidate_inode_buffers(). The 7081da177e4SLinus Torvalds * filesystems should do that. invalidate_inode_buffers() should just go 7091da177e4SLinus Torvalds * BUG_ON(!list_empty). 7101da177e4SLinus Torvalds * 7111da177e4SLinus Torvalds * FIXME: mark_buffer_dirty_inode() is a data-plane operation. It should 7121da177e4SLinus Torvalds * take an address_space, not an inode. And it should be called 7131da177e4SLinus Torvalds * mark_buffer_dirty_fsync() to clearly define why those buffers are being 7141da177e4SLinus Torvalds * queued up. 7151da177e4SLinus Torvalds * 7161da177e4SLinus Torvalds * FIXME: mark_buffer_dirty_inode() doesn't need to add the buffer to the 7171da177e4SLinus Torvalds * list if it is already on a list. Because if the buffer is on a list, 7181da177e4SLinus Torvalds * it *must* already be on the right one. If not, the filesystem is being 7191da177e4SLinus Torvalds * silly. This will save a ton of locking. But first we have to ensure 7201da177e4SLinus Torvalds * that buffers are taken *off* the old inode's list when they are freed 7211da177e4SLinus Torvalds * (presumably in truncate). That requires careful auditing of all 7221da177e4SLinus Torvalds * filesystems (do it inside bforget()). It could also be done by bringing 7231da177e4SLinus Torvalds * b_inode back. 7241da177e4SLinus Torvalds */ 7251da177e4SLinus Torvalds 7261da177e4SLinus Torvalds /* 7271da177e4SLinus Torvalds * The buffer's backing address_space's private_lock must be held 7281da177e4SLinus Torvalds */ 7291da177e4SLinus Torvalds static inline void __remove_assoc_queue(struct buffer_head *bh) 7301da177e4SLinus Torvalds { 7311da177e4SLinus Torvalds list_del_init(&bh->b_assoc_buffers); 7321da177e4SLinus Torvalds } 7331da177e4SLinus Torvalds 7341da177e4SLinus Torvalds int inode_has_buffers(struct inode *inode) 7351da177e4SLinus Torvalds { 7361da177e4SLinus Torvalds return !list_empty(&inode->i_data.private_list); 7371da177e4SLinus Torvalds } 7381da177e4SLinus Torvalds 7391da177e4SLinus Torvalds /* 7401da177e4SLinus Torvalds * osync is designed to support O_SYNC io. It waits synchronously for 7411da177e4SLinus Torvalds * all already-submitted IO to complete, but does not queue any new 7421da177e4SLinus Torvalds * writes to the disk. 7431da177e4SLinus Torvalds * 7441da177e4SLinus Torvalds * To do O_SYNC writes, just queue the buffer writes with ll_rw_block as 7451da177e4SLinus Torvalds * you dirty the buffers, and then use osync_inode_buffers to wait for 7461da177e4SLinus Torvalds * completion. Any other dirty buffers which are not yet queued for 7471da177e4SLinus Torvalds * write will not be flushed to disk by the osync. 7481da177e4SLinus Torvalds */ 7491da177e4SLinus Torvalds static int osync_buffers_list(spinlock_t *lock, struct list_head *list) 7501da177e4SLinus Torvalds { 7511da177e4SLinus Torvalds struct buffer_head *bh; 7521da177e4SLinus Torvalds struct list_head *p; 7531da177e4SLinus Torvalds int err = 0; 7541da177e4SLinus Torvalds 7551da177e4SLinus Torvalds spin_lock(lock); 7561da177e4SLinus Torvalds repeat: 7571da177e4SLinus Torvalds list_for_each_prev(p, list) { 7581da177e4SLinus Torvalds bh = BH_ENTRY(p); 7591da177e4SLinus Torvalds if (buffer_locked(bh)) { 7601da177e4SLinus Torvalds get_bh(bh); 7611da177e4SLinus Torvalds spin_unlock(lock); 7621da177e4SLinus Torvalds wait_on_buffer(bh); 7631da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 7641da177e4SLinus Torvalds err = -EIO; 7651da177e4SLinus Torvalds brelse(bh); 7661da177e4SLinus Torvalds spin_lock(lock); 7671da177e4SLinus Torvalds goto repeat; 7681da177e4SLinus Torvalds } 7691da177e4SLinus Torvalds } 7701da177e4SLinus Torvalds spin_unlock(lock); 7711da177e4SLinus Torvalds return err; 7721da177e4SLinus Torvalds } 7731da177e4SLinus Torvalds 7741da177e4SLinus Torvalds /** 7751da177e4SLinus Torvalds * sync_mapping_buffers - write out and wait upon a mapping's "associated" 7761da177e4SLinus Torvalds * buffers 777*67be2dd1SMartin Waitz * @mapping: the mapping which wants those buffers written 7781da177e4SLinus Torvalds * 7791da177e4SLinus Torvalds * Starts I/O against the buffers at mapping->private_list, and waits upon 7801da177e4SLinus Torvalds * that I/O. 7811da177e4SLinus Torvalds * 782*67be2dd1SMartin Waitz * Basically, this is a convenience function for fsync(). 783*67be2dd1SMartin Waitz * @mapping is a file or directory which needs those buffers to be written for 784*67be2dd1SMartin Waitz * a successful fsync(). 7851da177e4SLinus Torvalds */ 7861da177e4SLinus Torvalds int sync_mapping_buffers(struct address_space *mapping) 7871da177e4SLinus Torvalds { 7881da177e4SLinus Torvalds struct address_space *buffer_mapping = mapping->assoc_mapping; 7891da177e4SLinus Torvalds 7901da177e4SLinus Torvalds if (buffer_mapping == NULL || list_empty(&mapping->private_list)) 7911da177e4SLinus Torvalds return 0; 7921da177e4SLinus Torvalds 7931da177e4SLinus Torvalds return fsync_buffers_list(&buffer_mapping->private_lock, 7941da177e4SLinus Torvalds &mapping->private_list); 7951da177e4SLinus Torvalds } 7961da177e4SLinus Torvalds EXPORT_SYMBOL(sync_mapping_buffers); 7971da177e4SLinus Torvalds 7981da177e4SLinus Torvalds /* 7991da177e4SLinus Torvalds * Called when we've recently written block `bblock', and it is known that 8001da177e4SLinus Torvalds * `bblock' was for a buffer_boundary() buffer. This means that the block at 8011da177e4SLinus Torvalds * `bblock + 1' is probably a dirty indirect block. Hunt it down and, if it's 8021da177e4SLinus Torvalds * dirty, schedule it for IO. So that indirects merge nicely with their data. 8031da177e4SLinus Torvalds */ 8041da177e4SLinus Torvalds void write_boundary_block(struct block_device *bdev, 8051da177e4SLinus Torvalds sector_t bblock, unsigned blocksize) 8061da177e4SLinus Torvalds { 8071da177e4SLinus Torvalds struct buffer_head *bh = __find_get_block(bdev, bblock + 1, blocksize); 8081da177e4SLinus Torvalds if (bh) { 8091da177e4SLinus Torvalds if (buffer_dirty(bh)) 8101da177e4SLinus Torvalds ll_rw_block(WRITE, 1, &bh); 8111da177e4SLinus Torvalds put_bh(bh); 8121da177e4SLinus Torvalds } 8131da177e4SLinus Torvalds } 8141da177e4SLinus Torvalds 8151da177e4SLinus Torvalds void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode) 8161da177e4SLinus Torvalds { 8171da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 8181da177e4SLinus Torvalds struct address_space *buffer_mapping = bh->b_page->mapping; 8191da177e4SLinus Torvalds 8201da177e4SLinus Torvalds mark_buffer_dirty(bh); 8211da177e4SLinus Torvalds if (!mapping->assoc_mapping) { 8221da177e4SLinus Torvalds mapping->assoc_mapping = buffer_mapping; 8231da177e4SLinus Torvalds } else { 8241da177e4SLinus Torvalds if (mapping->assoc_mapping != buffer_mapping) 8251da177e4SLinus Torvalds BUG(); 8261da177e4SLinus Torvalds } 8271da177e4SLinus Torvalds if (list_empty(&bh->b_assoc_buffers)) { 8281da177e4SLinus Torvalds spin_lock(&buffer_mapping->private_lock); 8291da177e4SLinus Torvalds list_move_tail(&bh->b_assoc_buffers, 8301da177e4SLinus Torvalds &mapping->private_list); 8311da177e4SLinus Torvalds spin_unlock(&buffer_mapping->private_lock); 8321da177e4SLinus Torvalds } 8331da177e4SLinus Torvalds } 8341da177e4SLinus Torvalds EXPORT_SYMBOL(mark_buffer_dirty_inode); 8351da177e4SLinus Torvalds 8361da177e4SLinus Torvalds /* 8371da177e4SLinus Torvalds * Add a page to the dirty page list. 8381da177e4SLinus Torvalds * 8391da177e4SLinus Torvalds * It is a sad fact of life that this function is called from several places 8401da177e4SLinus Torvalds * deeply under spinlocking. It may not sleep. 8411da177e4SLinus Torvalds * 8421da177e4SLinus Torvalds * If the page has buffers, the uptodate buffers are set dirty, to preserve 8431da177e4SLinus Torvalds * dirty-state coherency between the page and the buffers. It the page does 8441da177e4SLinus Torvalds * not have buffers then when they are later attached they will all be set 8451da177e4SLinus Torvalds * dirty. 8461da177e4SLinus Torvalds * 8471da177e4SLinus Torvalds * The buffers are dirtied before the page is dirtied. There's a small race 8481da177e4SLinus Torvalds * window in which a writepage caller may see the page cleanness but not the 8491da177e4SLinus Torvalds * buffer dirtiness. That's fine. If this code were to set the page dirty 8501da177e4SLinus Torvalds * before the buffers, a concurrent writepage caller could clear the page dirty 8511da177e4SLinus Torvalds * bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean 8521da177e4SLinus Torvalds * page on the dirty page list. 8531da177e4SLinus Torvalds * 8541da177e4SLinus Torvalds * We use private_lock to lock against try_to_free_buffers while using the 8551da177e4SLinus Torvalds * page's buffer list. Also use this to protect against clean buffers being 8561da177e4SLinus Torvalds * added to the page after it was set dirty. 8571da177e4SLinus Torvalds * 8581da177e4SLinus Torvalds * FIXME: may need to call ->reservepage here as well. That's rather up to the 8591da177e4SLinus Torvalds * address_space though. 8601da177e4SLinus Torvalds */ 8611da177e4SLinus Torvalds int __set_page_dirty_buffers(struct page *page) 8621da177e4SLinus Torvalds { 8631da177e4SLinus Torvalds struct address_space * const mapping = page->mapping; 8641da177e4SLinus Torvalds 8651da177e4SLinus Torvalds spin_lock(&mapping->private_lock); 8661da177e4SLinus Torvalds if (page_has_buffers(page)) { 8671da177e4SLinus Torvalds struct buffer_head *head = page_buffers(page); 8681da177e4SLinus Torvalds struct buffer_head *bh = head; 8691da177e4SLinus Torvalds 8701da177e4SLinus Torvalds do { 8711da177e4SLinus Torvalds set_buffer_dirty(bh); 8721da177e4SLinus Torvalds bh = bh->b_this_page; 8731da177e4SLinus Torvalds } while (bh != head); 8741da177e4SLinus Torvalds } 8751da177e4SLinus Torvalds spin_unlock(&mapping->private_lock); 8761da177e4SLinus Torvalds 8771da177e4SLinus Torvalds if (!TestSetPageDirty(page)) { 8781da177e4SLinus Torvalds write_lock_irq(&mapping->tree_lock); 8791da177e4SLinus Torvalds if (page->mapping) { /* Race with truncate? */ 8801da177e4SLinus Torvalds if (mapping_cap_account_dirty(mapping)) 8811da177e4SLinus Torvalds inc_page_state(nr_dirty); 8821da177e4SLinus Torvalds radix_tree_tag_set(&mapping->page_tree, 8831da177e4SLinus Torvalds page_index(page), 8841da177e4SLinus Torvalds PAGECACHE_TAG_DIRTY); 8851da177e4SLinus Torvalds } 8861da177e4SLinus Torvalds write_unlock_irq(&mapping->tree_lock); 8871da177e4SLinus Torvalds __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); 8881da177e4SLinus Torvalds } 8891da177e4SLinus Torvalds 8901da177e4SLinus Torvalds return 0; 8911da177e4SLinus Torvalds } 8921da177e4SLinus Torvalds EXPORT_SYMBOL(__set_page_dirty_buffers); 8931da177e4SLinus Torvalds 8941da177e4SLinus Torvalds /* 8951da177e4SLinus Torvalds * Write out and wait upon a list of buffers. 8961da177e4SLinus Torvalds * 8971da177e4SLinus Torvalds * We have conflicting pressures: we want to make sure that all 8981da177e4SLinus Torvalds * initially dirty buffers get waited on, but that any subsequently 8991da177e4SLinus Torvalds * dirtied buffers don't. After all, we don't want fsync to last 9001da177e4SLinus Torvalds * forever if somebody is actively writing to the file. 9011da177e4SLinus Torvalds * 9021da177e4SLinus Torvalds * Do this in two main stages: first we copy dirty buffers to a 9031da177e4SLinus Torvalds * temporary inode list, queueing the writes as we go. Then we clean 9041da177e4SLinus Torvalds * up, waiting for those writes to complete. 9051da177e4SLinus Torvalds * 9061da177e4SLinus Torvalds * During this second stage, any subsequent updates to the file may end 9071da177e4SLinus Torvalds * up refiling the buffer on the original inode's dirty list again, so 9081da177e4SLinus Torvalds * there is a chance we will end up with a buffer queued for write but 9091da177e4SLinus Torvalds * not yet completed on that list. So, as a final cleanup we go through 9101da177e4SLinus Torvalds * the osync code to catch these locked, dirty buffers without requeuing 9111da177e4SLinus Torvalds * any newly dirty buffers for write. 9121da177e4SLinus Torvalds */ 9131da177e4SLinus Torvalds static int fsync_buffers_list(spinlock_t *lock, struct list_head *list) 9141da177e4SLinus Torvalds { 9151da177e4SLinus Torvalds struct buffer_head *bh; 9161da177e4SLinus Torvalds struct list_head tmp; 9171da177e4SLinus Torvalds int err = 0, err2; 9181da177e4SLinus Torvalds 9191da177e4SLinus Torvalds INIT_LIST_HEAD(&tmp); 9201da177e4SLinus Torvalds 9211da177e4SLinus Torvalds spin_lock(lock); 9221da177e4SLinus Torvalds while (!list_empty(list)) { 9231da177e4SLinus Torvalds bh = BH_ENTRY(list->next); 9241da177e4SLinus Torvalds list_del_init(&bh->b_assoc_buffers); 9251da177e4SLinus Torvalds if (buffer_dirty(bh) || buffer_locked(bh)) { 9261da177e4SLinus Torvalds list_add(&bh->b_assoc_buffers, &tmp); 9271da177e4SLinus Torvalds if (buffer_dirty(bh)) { 9281da177e4SLinus Torvalds get_bh(bh); 9291da177e4SLinus Torvalds spin_unlock(lock); 9301da177e4SLinus Torvalds /* 9311da177e4SLinus Torvalds * Ensure any pending I/O completes so that 9321da177e4SLinus Torvalds * ll_rw_block() actually writes the current 9331da177e4SLinus Torvalds * contents - it is a noop if I/O is still in 9341da177e4SLinus Torvalds * flight on potentially older contents. 9351da177e4SLinus Torvalds */ 9361da177e4SLinus Torvalds wait_on_buffer(bh); 9371da177e4SLinus Torvalds ll_rw_block(WRITE, 1, &bh); 9381da177e4SLinus Torvalds brelse(bh); 9391da177e4SLinus Torvalds spin_lock(lock); 9401da177e4SLinus Torvalds } 9411da177e4SLinus Torvalds } 9421da177e4SLinus Torvalds } 9431da177e4SLinus Torvalds 9441da177e4SLinus Torvalds while (!list_empty(&tmp)) { 9451da177e4SLinus Torvalds bh = BH_ENTRY(tmp.prev); 9461da177e4SLinus Torvalds __remove_assoc_queue(bh); 9471da177e4SLinus Torvalds get_bh(bh); 9481da177e4SLinus Torvalds spin_unlock(lock); 9491da177e4SLinus Torvalds wait_on_buffer(bh); 9501da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 9511da177e4SLinus Torvalds err = -EIO; 9521da177e4SLinus Torvalds brelse(bh); 9531da177e4SLinus Torvalds spin_lock(lock); 9541da177e4SLinus Torvalds } 9551da177e4SLinus Torvalds 9561da177e4SLinus Torvalds spin_unlock(lock); 9571da177e4SLinus Torvalds err2 = osync_buffers_list(lock, list); 9581da177e4SLinus Torvalds if (err) 9591da177e4SLinus Torvalds return err; 9601da177e4SLinus Torvalds else 9611da177e4SLinus Torvalds return err2; 9621da177e4SLinus Torvalds } 9631da177e4SLinus Torvalds 9641da177e4SLinus Torvalds /* 9651da177e4SLinus Torvalds * Invalidate any and all dirty buffers on a given inode. We are 9661da177e4SLinus Torvalds * probably unmounting the fs, but that doesn't mean we have already 9671da177e4SLinus Torvalds * done a sync(). Just drop the buffers from the inode list. 9681da177e4SLinus Torvalds * 9691da177e4SLinus Torvalds * NOTE: we take the inode's blockdev's mapping's private_lock. Which 9701da177e4SLinus Torvalds * assumes that all the buffers are against the blockdev. Not true 9711da177e4SLinus Torvalds * for reiserfs. 9721da177e4SLinus Torvalds */ 9731da177e4SLinus Torvalds void invalidate_inode_buffers(struct inode *inode) 9741da177e4SLinus Torvalds { 9751da177e4SLinus Torvalds if (inode_has_buffers(inode)) { 9761da177e4SLinus Torvalds struct address_space *mapping = &inode->i_data; 9771da177e4SLinus Torvalds struct list_head *list = &mapping->private_list; 9781da177e4SLinus Torvalds struct address_space *buffer_mapping = mapping->assoc_mapping; 9791da177e4SLinus Torvalds 9801da177e4SLinus Torvalds spin_lock(&buffer_mapping->private_lock); 9811da177e4SLinus Torvalds while (!list_empty(list)) 9821da177e4SLinus Torvalds __remove_assoc_queue(BH_ENTRY(list->next)); 9831da177e4SLinus Torvalds spin_unlock(&buffer_mapping->private_lock); 9841da177e4SLinus Torvalds } 9851da177e4SLinus Torvalds } 9861da177e4SLinus Torvalds 9871da177e4SLinus Torvalds /* 9881da177e4SLinus Torvalds * Remove any clean buffers from the inode's buffer list. This is called 9891da177e4SLinus Torvalds * when we're trying to free the inode itself. Those buffers can pin it. 9901da177e4SLinus Torvalds * 9911da177e4SLinus Torvalds * Returns true if all buffers were removed. 9921da177e4SLinus Torvalds */ 9931da177e4SLinus Torvalds int remove_inode_buffers(struct inode *inode) 9941da177e4SLinus Torvalds { 9951da177e4SLinus Torvalds int ret = 1; 9961da177e4SLinus Torvalds 9971da177e4SLinus Torvalds if (inode_has_buffers(inode)) { 9981da177e4SLinus Torvalds struct address_space *mapping = &inode->i_data; 9991da177e4SLinus Torvalds struct list_head *list = &mapping->private_list; 10001da177e4SLinus Torvalds struct address_space *buffer_mapping = mapping->assoc_mapping; 10011da177e4SLinus Torvalds 10021da177e4SLinus Torvalds spin_lock(&buffer_mapping->private_lock); 10031da177e4SLinus Torvalds while (!list_empty(list)) { 10041da177e4SLinus Torvalds struct buffer_head *bh = BH_ENTRY(list->next); 10051da177e4SLinus Torvalds if (buffer_dirty(bh)) { 10061da177e4SLinus Torvalds ret = 0; 10071da177e4SLinus Torvalds break; 10081da177e4SLinus Torvalds } 10091da177e4SLinus Torvalds __remove_assoc_queue(bh); 10101da177e4SLinus Torvalds } 10111da177e4SLinus Torvalds spin_unlock(&buffer_mapping->private_lock); 10121da177e4SLinus Torvalds } 10131da177e4SLinus Torvalds return ret; 10141da177e4SLinus Torvalds } 10151da177e4SLinus Torvalds 10161da177e4SLinus Torvalds /* 10171da177e4SLinus Torvalds * Create the appropriate buffers when given a page for data area and 10181da177e4SLinus Torvalds * the size of each buffer.. Use the bh->b_this_page linked list to 10191da177e4SLinus Torvalds * follow the buffers created. Return NULL if unable to create more 10201da177e4SLinus Torvalds * buffers. 10211da177e4SLinus Torvalds * 10221da177e4SLinus Torvalds * The retry flag is used to differentiate async IO (paging, swapping) 10231da177e4SLinus Torvalds * which may not fail from ordinary buffer allocations. 10241da177e4SLinus Torvalds */ 10251da177e4SLinus Torvalds struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size, 10261da177e4SLinus Torvalds int retry) 10271da177e4SLinus Torvalds { 10281da177e4SLinus Torvalds struct buffer_head *bh, *head; 10291da177e4SLinus Torvalds long offset; 10301da177e4SLinus Torvalds 10311da177e4SLinus Torvalds try_again: 10321da177e4SLinus Torvalds head = NULL; 10331da177e4SLinus Torvalds offset = PAGE_SIZE; 10341da177e4SLinus Torvalds while ((offset -= size) >= 0) { 10351da177e4SLinus Torvalds bh = alloc_buffer_head(GFP_NOFS); 10361da177e4SLinus Torvalds if (!bh) 10371da177e4SLinus Torvalds goto no_grow; 10381da177e4SLinus Torvalds 10391da177e4SLinus Torvalds bh->b_bdev = NULL; 10401da177e4SLinus Torvalds bh->b_this_page = head; 10411da177e4SLinus Torvalds bh->b_blocknr = -1; 10421da177e4SLinus Torvalds head = bh; 10431da177e4SLinus Torvalds 10441da177e4SLinus Torvalds bh->b_state = 0; 10451da177e4SLinus Torvalds atomic_set(&bh->b_count, 0); 10461da177e4SLinus Torvalds bh->b_size = size; 10471da177e4SLinus Torvalds 10481da177e4SLinus Torvalds /* Link the buffer to its page */ 10491da177e4SLinus Torvalds set_bh_page(bh, page, offset); 10501da177e4SLinus Torvalds 10511da177e4SLinus Torvalds bh->b_end_io = NULL; 10521da177e4SLinus Torvalds } 10531da177e4SLinus Torvalds return head; 10541da177e4SLinus Torvalds /* 10551da177e4SLinus Torvalds * In case anything failed, we just free everything we got. 10561da177e4SLinus Torvalds */ 10571da177e4SLinus Torvalds no_grow: 10581da177e4SLinus Torvalds if (head) { 10591da177e4SLinus Torvalds do { 10601da177e4SLinus Torvalds bh = head; 10611da177e4SLinus Torvalds head = head->b_this_page; 10621da177e4SLinus Torvalds free_buffer_head(bh); 10631da177e4SLinus Torvalds } while (head); 10641da177e4SLinus Torvalds } 10651da177e4SLinus Torvalds 10661da177e4SLinus Torvalds /* 10671da177e4SLinus Torvalds * Return failure for non-async IO requests. Async IO requests 10681da177e4SLinus Torvalds * are not allowed to fail, so we have to wait until buffer heads 10691da177e4SLinus Torvalds * become available. But we don't want tasks sleeping with 10701da177e4SLinus Torvalds * partially complete buffers, so all were released above. 10711da177e4SLinus Torvalds */ 10721da177e4SLinus Torvalds if (!retry) 10731da177e4SLinus Torvalds return NULL; 10741da177e4SLinus Torvalds 10751da177e4SLinus Torvalds /* We're _really_ low on memory. Now we just 10761da177e4SLinus Torvalds * wait for old buffer heads to become free due to 10771da177e4SLinus Torvalds * finishing IO. Since this is an async request and 10781da177e4SLinus Torvalds * the reserve list is empty, we're sure there are 10791da177e4SLinus Torvalds * async buffer heads in use. 10801da177e4SLinus Torvalds */ 10811da177e4SLinus Torvalds free_more_memory(); 10821da177e4SLinus Torvalds goto try_again; 10831da177e4SLinus Torvalds } 10841da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(alloc_page_buffers); 10851da177e4SLinus Torvalds 10861da177e4SLinus Torvalds static inline void 10871da177e4SLinus Torvalds link_dev_buffers(struct page *page, struct buffer_head *head) 10881da177e4SLinus Torvalds { 10891da177e4SLinus Torvalds struct buffer_head *bh, *tail; 10901da177e4SLinus Torvalds 10911da177e4SLinus Torvalds bh = head; 10921da177e4SLinus Torvalds do { 10931da177e4SLinus Torvalds tail = bh; 10941da177e4SLinus Torvalds bh = bh->b_this_page; 10951da177e4SLinus Torvalds } while (bh); 10961da177e4SLinus Torvalds tail->b_this_page = head; 10971da177e4SLinus Torvalds attach_page_buffers(page, head); 10981da177e4SLinus Torvalds } 10991da177e4SLinus Torvalds 11001da177e4SLinus Torvalds /* 11011da177e4SLinus Torvalds * Initialise the state of a blockdev page's buffers. 11021da177e4SLinus Torvalds */ 11031da177e4SLinus Torvalds static void 11041da177e4SLinus Torvalds init_page_buffers(struct page *page, struct block_device *bdev, 11051da177e4SLinus Torvalds sector_t block, int size) 11061da177e4SLinus Torvalds { 11071da177e4SLinus Torvalds struct buffer_head *head = page_buffers(page); 11081da177e4SLinus Torvalds struct buffer_head *bh = head; 11091da177e4SLinus Torvalds int uptodate = PageUptodate(page); 11101da177e4SLinus Torvalds 11111da177e4SLinus Torvalds do { 11121da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 11131da177e4SLinus Torvalds init_buffer(bh, NULL, NULL); 11141da177e4SLinus Torvalds bh->b_bdev = bdev; 11151da177e4SLinus Torvalds bh->b_blocknr = block; 11161da177e4SLinus Torvalds if (uptodate) 11171da177e4SLinus Torvalds set_buffer_uptodate(bh); 11181da177e4SLinus Torvalds set_buffer_mapped(bh); 11191da177e4SLinus Torvalds } 11201da177e4SLinus Torvalds block++; 11211da177e4SLinus Torvalds bh = bh->b_this_page; 11221da177e4SLinus Torvalds } while (bh != head); 11231da177e4SLinus Torvalds } 11241da177e4SLinus Torvalds 11251da177e4SLinus Torvalds /* 11261da177e4SLinus Torvalds * Create the page-cache page that contains the requested block. 11271da177e4SLinus Torvalds * 11281da177e4SLinus Torvalds * This is user purely for blockdev mappings. 11291da177e4SLinus Torvalds */ 11301da177e4SLinus Torvalds static struct page * 11311da177e4SLinus Torvalds grow_dev_page(struct block_device *bdev, sector_t block, 11321da177e4SLinus Torvalds pgoff_t index, int size) 11331da177e4SLinus Torvalds { 11341da177e4SLinus Torvalds struct inode *inode = bdev->bd_inode; 11351da177e4SLinus Torvalds struct page *page; 11361da177e4SLinus Torvalds struct buffer_head *bh; 11371da177e4SLinus Torvalds 11381da177e4SLinus Torvalds page = find_or_create_page(inode->i_mapping, index, GFP_NOFS); 11391da177e4SLinus Torvalds if (!page) 11401da177e4SLinus Torvalds return NULL; 11411da177e4SLinus Torvalds 11421da177e4SLinus Torvalds if (!PageLocked(page)) 11431da177e4SLinus Torvalds BUG(); 11441da177e4SLinus Torvalds 11451da177e4SLinus Torvalds if (page_has_buffers(page)) { 11461da177e4SLinus Torvalds bh = page_buffers(page); 11471da177e4SLinus Torvalds if (bh->b_size == size) { 11481da177e4SLinus Torvalds init_page_buffers(page, bdev, block, size); 11491da177e4SLinus Torvalds return page; 11501da177e4SLinus Torvalds } 11511da177e4SLinus Torvalds if (!try_to_free_buffers(page)) 11521da177e4SLinus Torvalds goto failed; 11531da177e4SLinus Torvalds } 11541da177e4SLinus Torvalds 11551da177e4SLinus Torvalds /* 11561da177e4SLinus Torvalds * Allocate some buffers for this page 11571da177e4SLinus Torvalds */ 11581da177e4SLinus Torvalds bh = alloc_page_buffers(page, size, 0); 11591da177e4SLinus Torvalds if (!bh) 11601da177e4SLinus Torvalds goto failed; 11611da177e4SLinus Torvalds 11621da177e4SLinus Torvalds /* 11631da177e4SLinus Torvalds * Link the page to the buffers and initialise them. Take the 11641da177e4SLinus Torvalds * lock to be atomic wrt __find_get_block(), which does not 11651da177e4SLinus Torvalds * run under the page lock. 11661da177e4SLinus Torvalds */ 11671da177e4SLinus Torvalds spin_lock(&inode->i_mapping->private_lock); 11681da177e4SLinus Torvalds link_dev_buffers(page, bh); 11691da177e4SLinus Torvalds init_page_buffers(page, bdev, block, size); 11701da177e4SLinus Torvalds spin_unlock(&inode->i_mapping->private_lock); 11711da177e4SLinus Torvalds return page; 11721da177e4SLinus Torvalds 11731da177e4SLinus Torvalds failed: 11741da177e4SLinus Torvalds BUG(); 11751da177e4SLinus Torvalds unlock_page(page); 11761da177e4SLinus Torvalds page_cache_release(page); 11771da177e4SLinus Torvalds return NULL; 11781da177e4SLinus Torvalds } 11791da177e4SLinus Torvalds 11801da177e4SLinus Torvalds /* 11811da177e4SLinus Torvalds * Create buffers for the specified block device block's page. If 11821da177e4SLinus Torvalds * that page was dirty, the buffers are set dirty also. 11831da177e4SLinus Torvalds * 11841da177e4SLinus Torvalds * Except that's a bug. Attaching dirty buffers to a dirty 11851da177e4SLinus Torvalds * blockdev's page can result in filesystem corruption, because 11861da177e4SLinus Torvalds * some of those buffers may be aliases of filesystem data. 11871da177e4SLinus Torvalds * grow_dev_page() will go BUG() if this happens. 11881da177e4SLinus Torvalds */ 11891da177e4SLinus Torvalds static inline int 11901da177e4SLinus Torvalds grow_buffers(struct block_device *bdev, sector_t block, int size) 11911da177e4SLinus Torvalds { 11921da177e4SLinus Torvalds struct page *page; 11931da177e4SLinus Torvalds pgoff_t index; 11941da177e4SLinus Torvalds int sizebits; 11951da177e4SLinus Torvalds 11961da177e4SLinus Torvalds sizebits = -1; 11971da177e4SLinus Torvalds do { 11981da177e4SLinus Torvalds sizebits++; 11991da177e4SLinus Torvalds } while ((size << sizebits) < PAGE_SIZE); 12001da177e4SLinus Torvalds 12011da177e4SLinus Torvalds index = block >> sizebits; 12021da177e4SLinus Torvalds block = index << sizebits; 12031da177e4SLinus Torvalds 12041da177e4SLinus Torvalds /* Create a page with the proper size buffers.. */ 12051da177e4SLinus Torvalds page = grow_dev_page(bdev, block, index, size); 12061da177e4SLinus Torvalds if (!page) 12071da177e4SLinus Torvalds return 0; 12081da177e4SLinus Torvalds unlock_page(page); 12091da177e4SLinus Torvalds page_cache_release(page); 12101da177e4SLinus Torvalds return 1; 12111da177e4SLinus Torvalds } 12121da177e4SLinus Torvalds 12131da177e4SLinus Torvalds struct buffer_head * 12141da177e4SLinus Torvalds __getblk_slow(struct block_device *bdev, sector_t block, int size) 12151da177e4SLinus Torvalds { 12161da177e4SLinus Torvalds /* Size must be multiple of hard sectorsize */ 12171da177e4SLinus Torvalds if (unlikely(size & (bdev_hardsect_size(bdev)-1) || 12181da177e4SLinus Torvalds (size < 512 || size > PAGE_SIZE))) { 12191da177e4SLinus Torvalds printk(KERN_ERR "getblk(): invalid block size %d requested\n", 12201da177e4SLinus Torvalds size); 12211da177e4SLinus Torvalds printk(KERN_ERR "hardsect size: %d\n", 12221da177e4SLinus Torvalds bdev_hardsect_size(bdev)); 12231da177e4SLinus Torvalds 12241da177e4SLinus Torvalds dump_stack(); 12251da177e4SLinus Torvalds return NULL; 12261da177e4SLinus Torvalds } 12271da177e4SLinus Torvalds 12281da177e4SLinus Torvalds for (;;) { 12291da177e4SLinus Torvalds struct buffer_head * bh; 12301da177e4SLinus Torvalds 12311da177e4SLinus Torvalds bh = __find_get_block(bdev, block, size); 12321da177e4SLinus Torvalds if (bh) 12331da177e4SLinus Torvalds return bh; 12341da177e4SLinus Torvalds 12351da177e4SLinus Torvalds if (!grow_buffers(bdev, block, size)) 12361da177e4SLinus Torvalds free_more_memory(); 12371da177e4SLinus Torvalds } 12381da177e4SLinus Torvalds } 12391da177e4SLinus Torvalds 12401da177e4SLinus Torvalds /* 12411da177e4SLinus Torvalds * The relationship between dirty buffers and dirty pages: 12421da177e4SLinus Torvalds * 12431da177e4SLinus Torvalds * Whenever a page has any dirty buffers, the page's dirty bit is set, and 12441da177e4SLinus Torvalds * the page is tagged dirty in its radix tree. 12451da177e4SLinus Torvalds * 12461da177e4SLinus Torvalds * At all times, the dirtiness of the buffers represents the dirtiness of 12471da177e4SLinus Torvalds * subsections of the page. If the page has buffers, the page dirty bit is 12481da177e4SLinus Torvalds * merely a hint about the true dirty state. 12491da177e4SLinus Torvalds * 12501da177e4SLinus Torvalds * When a page is set dirty in its entirety, all its buffers are marked dirty 12511da177e4SLinus Torvalds * (if the page has buffers). 12521da177e4SLinus Torvalds * 12531da177e4SLinus Torvalds * When a buffer is marked dirty, its page is dirtied, but the page's other 12541da177e4SLinus Torvalds * buffers are not. 12551da177e4SLinus Torvalds * 12561da177e4SLinus Torvalds * Also. When blockdev buffers are explicitly read with bread(), they 12571da177e4SLinus Torvalds * individually become uptodate. But their backing page remains not 12581da177e4SLinus Torvalds * uptodate - even if all of its buffers are uptodate. A subsequent 12591da177e4SLinus Torvalds * block_read_full_page() against that page will discover all the uptodate 12601da177e4SLinus Torvalds * buffers, will set the page uptodate and will perform no I/O. 12611da177e4SLinus Torvalds */ 12621da177e4SLinus Torvalds 12631da177e4SLinus Torvalds /** 12641da177e4SLinus Torvalds * mark_buffer_dirty - mark a buffer_head as needing writeout 1265*67be2dd1SMartin Waitz * @bh: the buffer_head to mark dirty 12661da177e4SLinus Torvalds * 12671da177e4SLinus Torvalds * mark_buffer_dirty() will set the dirty bit against the buffer, then set its 12681da177e4SLinus Torvalds * backing page dirty, then tag the page as dirty in its address_space's radix 12691da177e4SLinus Torvalds * tree and then attach the address_space's inode to its superblock's dirty 12701da177e4SLinus Torvalds * inode list. 12711da177e4SLinus Torvalds * 12721da177e4SLinus Torvalds * mark_buffer_dirty() is atomic. It takes bh->b_page->mapping->private_lock, 12731da177e4SLinus Torvalds * mapping->tree_lock and the global inode_lock. 12741da177e4SLinus Torvalds */ 12751da177e4SLinus Torvalds void fastcall mark_buffer_dirty(struct buffer_head *bh) 12761da177e4SLinus Torvalds { 12771da177e4SLinus Torvalds if (!buffer_dirty(bh) && !test_set_buffer_dirty(bh)) 12781da177e4SLinus Torvalds __set_page_dirty_nobuffers(bh->b_page); 12791da177e4SLinus Torvalds } 12801da177e4SLinus Torvalds 12811da177e4SLinus Torvalds /* 12821da177e4SLinus Torvalds * Decrement a buffer_head's reference count. If all buffers against a page 12831da177e4SLinus Torvalds * have zero reference count, are clean and unlocked, and if the page is clean 12841da177e4SLinus Torvalds * and unlocked then try_to_free_buffers() may strip the buffers from the page 12851da177e4SLinus Torvalds * in preparation for freeing it (sometimes, rarely, buffers are removed from 12861da177e4SLinus Torvalds * a page but it ends up not being freed, and buffers may later be reattached). 12871da177e4SLinus Torvalds */ 12881da177e4SLinus Torvalds void __brelse(struct buffer_head * buf) 12891da177e4SLinus Torvalds { 12901da177e4SLinus Torvalds if (atomic_read(&buf->b_count)) { 12911da177e4SLinus Torvalds put_bh(buf); 12921da177e4SLinus Torvalds return; 12931da177e4SLinus Torvalds } 12941da177e4SLinus Torvalds printk(KERN_ERR "VFS: brelse: Trying to free free buffer\n"); 12951da177e4SLinus Torvalds WARN_ON(1); 12961da177e4SLinus Torvalds } 12971da177e4SLinus Torvalds 12981da177e4SLinus Torvalds /* 12991da177e4SLinus Torvalds * bforget() is like brelse(), except it discards any 13001da177e4SLinus Torvalds * potentially dirty data. 13011da177e4SLinus Torvalds */ 13021da177e4SLinus Torvalds void __bforget(struct buffer_head *bh) 13031da177e4SLinus Torvalds { 13041da177e4SLinus Torvalds clear_buffer_dirty(bh); 13051da177e4SLinus Torvalds if (!list_empty(&bh->b_assoc_buffers)) { 13061da177e4SLinus Torvalds struct address_space *buffer_mapping = bh->b_page->mapping; 13071da177e4SLinus Torvalds 13081da177e4SLinus Torvalds spin_lock(&buffer_mapping->private_lock); 13091da177e4SLinus Torvalds list_del_init(&bh->b_assoc_buffers); 13101da177e4SLinus Torvalds spin_unlock(&buffer_mapping->private_lock); 13111da177e4SLinus Torvalds } 13121da177e4SLinus Torvalds __brelse(bh); 13131da177e4SLinus Torvalds } 13141da177e4SLinus Torvalds 13151da177e4SLinus Torvalds static struct buffer_head *__bread_slow(struct buffer_head *bh) 13161da177e4SLinus Torvalds { 13171da177e4SLinus Torvalds lock_buffer(bh); 13181da177e4SLinus Torvalds if (buffer_uptodate(bh)) { 13191da177e4SLinus Torvalds unlock_buffer(bh); 13201da177e4SLinus Torvalds return bh; 13211da177e4SLinus Torvalds } else { 13221da177e4SLinus Torvalds get_bh(bh); 13231da177e4SLinus Torvalds bh->b_end_io = end_buffer_read_sync; 13241da177e4SLinus Torvalds submit_bh(READ, bh); 13251da177e4SLinus Torvalds wait_on_buffer(bh); 13261da177e4SLinus Torvalds if (buffer_uptodate(bh)) 13271da177e4SLinus Torvalds return bh; 13281da177e4SLinus Torvalds } 13291da177e4SLinus Torvalds brelse(bh); 13301da177e4SLinus Torvalds return NULL; 13311da177e4SLinus Torvalds } 13321da177e4SLinus Torvalds 13331da177e4SLinus Torvalds /* 13341da177e4SLinus Torvalds * Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block(). 13351da177e4SLinus Torvalds * The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their 13361da177e4SLinus Torvalds * refcount elevated by one when they're in an LRU. A buffer can only appear 13371da177e4SLinus Torvalds * once in a particular CPU's LRU. A single buffer can be present in multiple 13381da177e4SLinus Torvalds * CPU's LRUs at the same time. 13391da177e4SLinus Torvalds * 13401da177e4SLinus Torvalds * This is a transparent caching front-end to sb_bread(), sb_getblk() and 13411da177e4SLinus Torvalds * sb_find_get_block(). 13421da177e4SLinus Torvalds * 13431da177e4SLinus Torvalds * The LRUs themselves only need locking against invalidate_bh_lrus. We use 13441da177e4SLinus Torvalds * a local interrupt disable for that. 13451da177e4SLinus Torvalds */ 13461da177e4SLinus Torvalds 13471da177e4SLinus Torvalds #define BH_LRU_SIZE 8 13481da177e4SLinus Torvalds 13491da177e4SLinus Torvalds struct bh_lru { 13501da177e4SLinus Torvalds struct buffer_head *bhs[BH_LRU_SIZE]; 13511da177e4SLinus Torvalds }; 13521da177e4SLinus Torvalds 13531da177e4SLinus Torvalds static DEFINE_PER_CPU(struct bh_lru, bh_lrus) = {{ NULL }}; 13541da177e4SLinus Torvalds 13551da177e4SLinus Torvalds #ifdef CONFIG_SMP 13561da177e4SLinus Torvalds #define bh_lru_lock() local_irq_disable() 13571da177e4SLinus Torvalds #define bh_lru_unlock() local_irq_enable() 13581da177e4SLinus Torvalds #else 13591da177e4SLinus Torvalds #define bh_lru_lock() preempt_disable() 13601da177e4SLinus Torvalds #define bh_lru_unlock() preempt_enable() 13611da177e4SLinus Torvalds #endif 13621da177e4SLinus Torvalds 13631da177e4SLinus Torvalds static inline void check_irqs_on(void) 13641da177e4SLinus Torvalds { 13651da177e4SLinus Torvalds #ifdef irqs_disabled 13661da177e4SLinus Torvalds BUG_ON(irqs_disabled()); 13671da177e4SLinus Torvalds #endif 13681da177e4SLinus Torvalds } 13691da177e4SLinus Torvalds 13701da177e4SLinus Torvalds /* 13711da177e4SLinus Torvalds * The LRU management algorithm is dopey-but-simple. Sorry. 13721da177e4SLinus Torvalds */ 13731da177e4SLinus Torvalds static void bh_lru_install(struct buffer_head *bh) 13741da177e4SLinus Torvalds { 13751da177e4SLinus Torvalds struct buffer_head *evictee = NULL; 13761da177e4SLinus Torvalds struct bh_lru *lru; 13771da177e4SLinus Torvalds 13781da177e4SLinus Torvalds check_irqs_on(); 13791da177e4SLinus Torvalds bh_lru_lock(); 13801da177e4SLinus Torvalds lru = &__get_cpu_var(bh_lrus); 13811da177e4SLinus Torvalds if (lru->bhs[0] != bh) { 13821da177e4SLinus Torvalds struct buffer_head *bhs[BH_LRU_SIZE]; 13831da177e4SLinus Torvalds int in; 13841da177e4SLinus Torvalds int out = 0; 13851da177e4SLinus Torvalds 13861da177e4SLinus Torvalds get_bh(bh); 13871da177e4SLinus Torvalds bhs[out++] = bh; 13881da177e4SLinus Torvalds for (in = 0; in < BH_LRU_SIZE; in++) { 13891da177e4SLinus Torvalds struct buffer_head *bh2 = lru->bhs[in]; 13901da177e4SLinus Torvalds 13911da177e4SLinus Torvalds if (bh2 == bh) { 13921da177e4SLinus Torvalds __brelse(bh2); 13931da177e4SLinus Torvalds } else { 13941da177e4SLinus Torvalds if (out >= BH_LRU_SIZE) { 13951da177e4SLinus Torvalds BUG_ON(evictee != NULL); 13961da177e4SLinus Torvalds evictee = bh2; 13971da177e4SLinus Torvalds } else { 13981da177e4SLinus Torvalds bhs[out++] = bh2; 13991da177e4SLinus Torvalds } 14001da177e4SLinus Torvalds } 14011da177e4SLinus Torvalds } 14021da177e4SLinus Torvalds while (out < BH_LRU_SIZE) 14031da177e4SLinus Torvalds bhs[out++] = NULL; 14041da177e4SLinus Torvalds memcpy(lru->bhs, bhs, sizeof(bhs)); 14051da177e4SLinus Torvalds } 14061da177e4SLinus Torvalds bh_lru_unlock(); 14071da177e4SLinus Torvalds 14081da177e4SLinus Torvalds if (evictee) 14091da177e4SLinus Torvalds __brelse(evictee); 14101da177e4SLinus Torvalds } 14111da177e4SLinus Torvalds 14121da177e4SLinus Torvalds /* 14131da177e4SLinus Torvalds * Look up the bh in this cpu's LRU. If it's there, move it to the head. 14141da177e4SLinus Torvalds */ 14151da177e4SLinus Torvalds static inline struct buffer_head * 14161da177e4SLinus Torvalds lookup_bh_lru(struct block_device *bdev, sector_t block, int size) 14171da177e4SLinus Torvalds { 14181da177e4SLinus Torvalds struct buffer_head *ret = NULL; 14191da177e4SLinus Torvalds struct bh_lru *lru; 14201da177e4SLinus Torvalds int i; 14211da177e4SLinus Torvalds 14221da177e4SLinus Torvalds check_irqs_on(); 14231da177e4SLinus Torvalds bh_lru_lock(); 14241da177e4SLinus Torvalds lru = &__get_cpu_var(bh_lrus); 14251da177e4SLinus Torvalds for (i = 0; i < BH_LRU_SIZE; i++) { 14261da177e4SLinus Torvalds struct buffer_head *bh = lru->bhs[i]; 14271da177e4SLinus Torvalds 14281da177e4SLinus Torvalds if (bh && bh->b_bdev == bdev && 14291da177e4SLinus Torvalds bh->b_blocknr == block && bh->b_size == size) { 14301da177e4SLinus Torvalds if (i) { 14311da177e4SLinus Torvalds while (i) { 14321da177e4SLinus Torvalds lru->bhs[i] = lru->bhs[i - 1]; 14331da177e4SLinus Torvalds i--; 14341da177e4SLinus Torvalds } 14351da177e4SLinus Torvalds lru->bhs[0] = bh; 14361da177e4SLinus Torvalds } 14371da177e4SLinus Torvalds get_bh(bh); 14381da177e4SLinus Torvalds ret = bh; 14391da177e4SLinus Torvalds break; 14401da177e4SLinus Torvalds } 14411da177e4SLinus Torvalds } 14421da177e4SLinus Torvalds bh_lru_unlock(); 14431da177e4SLinus Torvalds return ret; 14441da177e4SLinus Torvalds } 14451da177e4SLinus Torvalds 14461da177e4SLinus Torvalds /* 14471da177e4SLinus Torvalds * Perform a pagecache lookup for the matching buffer. If it's there, refresh 14481da177e4SLinus Torvalds * it in the LRU and mark it as accessed. If it is not present then return 14491da177e4SLinus Torvalds * NULL 14501da177e4SLinus Torvalds */ 14511da177e4SLinus Torvalds struct buffer_head * 14521da177e4SLinus Torvalds __find_get_block(struct block_device *bdev, sector_t block, int size) 14531da177e4SLinus Torvalds { 14541da177e4SLinus Torvalds struct buffer_head *bh = lookup_bh_lru(bdev, block, size); 14551da177e4SLinus Torvalds 14561da177e4SLinus Torvalds if (bh == NULL) { 14571da177e4SLinus Torvalds bh = __find_get_block_slow(bdev, block, size); 14581da177e4SLinus Torvalds if (bh) 14591da177e4SLinus Torvalds bh_lru_install(bh); 14601da177e4SLinus Torvalds } 14611da177e4SLinus Torvalds if (bh) 14621da177e4SLinus Torvalds touch_buffer(bh); 14631da177e4SLinus Torvalds return bh; 14641da177e4SLinus Torvalds } 14651da177e4SLinus Torvalds EXPORT_SYMBOL(__find_get_block); 14661da177e4SLinus Torvalds 14671da177e4SLinus Torvalds /* 14681da177e4SLinus Torvalds * __getblk will locate (and, if necessary, create) the buffer_head 14691da177e4SLinus Torvalds * which corresponds to the passed block_device, block and size. The 14701da177e4SLinus Torvalds * returned buffer has its reference count incremented. 14711da177e4SLinus Torvalds * 14721da177e4SLinus Torvalds * __getblk() cannot fail - it just keeps trying. If you pass it an 14731da177e4SLinus Torvalds * illegal block number, __getblk() will happily return a buffer_head 14741da177e4SLinus Torvalds * which represents the non-existent block. Very weird. 14751da177e4SLinus Torvalds * 14761da177e4SLinus Torvalds * __getblk() will lock up the machine if grow_dev_page's try_to_free_buffers() 14771da177e4SLinus Torvalds * attempt is failing. FIXME, perhaps? 14781da177e4SLinus Torvalds */ 14791da177e4SLinus Torvalds struct buffer_head * 14801da177e4SLinus Torvalds __getblk(struct block_device *bdev, sector_t block, int size) 14811da177e4SLinus Torvalds { 14821da177e4SLinus Torvalds struct buffer_head *bh = __find_get_block(bdev, block, size); 14831da177e4SLinus Torvalds 14841da177e4SLinus Torvalds might_sleep(); 14851da177e4SLinus Torvalds if (bh == NULL) 14861da177e4SLinus Torvalds bh = __getblk_slow(bdev, block, size); 14871da177e4SLinus Torvalds return bh; 14881da177e4SLinus Torvalds } 14891da177e4SLinus Torvalds EXPORT_SYMBOL(__getblk); 14901da177e4SLinus Torvalds 14911da177e4SLinus Torvalds /* 14921da177e4SLinus Torvalds * Do async read-ahead on a buffer.. 14931da177e4SLinus Torvalds */ 14941da177e4SLinus Torvalds void __breadahead(struct block_device *bdev, sector_t block, int size) 14951da177e4SLinus Torvalds { 14961da177e4SLinus Torvalds struct buffer_head *bh = __getblk(bdev, block, size); 14971da177e4SLinus Torvalds ll_rw_block(READA, 1, &bh); 14981da177e4SLinus Torvalds brelse(bh); 14991da177e4SLinus Torvalds } 15001da177e4SLinus Torvalds EXPORT_SYMBOL(__breadahead); 15011da177e4SLinus Torvalds 15021da177e4SLinus Torvalds /** 15031da177e4SLinus Torvalds * __bread() - reads a specified block and returns the bh 1504*67be2dd1SMartin Waitz * @bdev: the block_device to read from 15051da177e4SLinus Torvalds * @block: number of block 15061da177e4SLinus Torvalds * @size: size (in bytes) to read 15071da177e4SLinus Torvalds * 15081da177e4SLinus Torvalds * Reads a specified block, and returns buffer head that contains it. 15091da177e4SLinus Torvalds * It returns NULL if the block was unreadable. 15101da177e4SLinus Torvalds */ 15111da177e4SLinus Torvalds struct buffer_head * 15121da177e4SLinus Torvalds __bread(struct block_device *bdev, sector_t block, int size) 15131da177e4SLinus Torvalds { 15141da177e4SLinus Torvalds struct buffer_head *bh = __getblk(bdev, block, size); 15151da177e4SLinus Torvalds 15161da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 15171da177e4SLinus Torvalds bh = __bread_slow(bh); 15181da177e4SLinus Torvalds return bh; 15191da177e4SLinus Torvalds } 15201da177e4SLinus Torvalds EXPORT_SYMBOL(__bread); 15211da177e4SLinus Torvalds 15221da177e4SLinus Torvalds /* 15231da177e4SLinus Torvalds * invalidate_bh_lrus() is called rarely - but not only at unmount. 15241da177e4SLinus Torvalds * This doesn't race because it runs in each cpu either in irq 15251da177e4SLinus Torvalds * or with preempt disabled. 15261da177e4SLinus Torvalds */ 15271da177e4SLinus Torvalds static void invalidate_bh_lru(void *arg) 15281da177e4SLinus Torvalds { 15291da177e4SLinus Torvalds struct bh_lru *b = &get_cpu_var(bh_lrus); 15301da177e4SLinus Torvalds int i; 15311da177e4SLinus Torvalds 15321da177e4SLinus Torvalds for (i = 0; i < BH_LRU_SIZE; i++) { 15331da177e4SLinus Torvalds brelse(b->bhs[i]); 15341da177e4SLinus Torvalds b->bhs[i] = NULL; 15351da177e4SLinus Torvalds } 15361da177e4SLinus Torvalds put_cpu_var(bh_lrus); 15371da177e4SLinus Torvalds } 15381da177e4SLinus Torvalds 15391da177e4SLinus Torvalds static void invalidate_bh_lrus(void) 15401da177e4SLinus Torvalds { 15411da177e4SLinus Torvalds on_each_cpu(invalidate_bh_lru, NULL, 1, 1); 15421da177e4SLinus Torvalds } 15431da177e4SLinus Torvalds 15441da177e4SLinus Torvalds void set_bh_page(struct buffer_head *bh, 15451da177e4SLinus Torvalds struct page *page, unsigned long offset) 15461da177e4SLinus Torvalds { 15471da177e4SLinus Torvalds bh->b_page = page; 15481da177e4SLinus Torvalds if (offset >= PAGE_SIZE) 15491da177e4SLinus Torvalds BUG(); 15501da177e4SLinus Torvalds if (PageHighMem(page)) 15511da177e4SLinus Torvalds /* 15521da177e4SLinus Torvalds * This catches illegal uses and preserves the offset: 15531da177e4SLinus Torvalds */ 15541da177e4SLinus Torvalds bh->b_data = (char *)(0 + offset); 15551da177e4SLinus Torvalds else 15561da177e4SLinus Torvalds bh->b_data = page_address(page) + offset; 15571da177e4SLinus Torvalds } 15581da177e4SLinus Torvalds EXPORT_SYMBOL(set_bh_page); 15591da177e4SLinus Torvalds 15601da177e4SLinus Torvalds /* 15611da177e4SLinus Torvalds * Called when truncating a buffer on a page completely. 15621da177e4SLinus Torvalds */ 15631da177e4SLinus Torvalds static inline void discard_buffer(struct buffer_head * bh) 15641da177e4SLinus Torvalds { 15651da177e4SLinus Torvalds lock_buffer(bh); 15661da177e4SLinus Torvalds clear_buffer_dirty(bh); 15671da177e4SLinus Torvalds bh->b_bdev = NULL; 15681da177e4SLinus Torvalds clear_buffer_mapped(bh); 15691da177e4SLinus Torvalds clear_buffer_req(bh); 15701da177e4SLinus Torvalds clear_buffer_new(bh); 15711da177e4SLinus Torvalds clear_buffer_delay(bh); 15721da177e4SLinus Torvalds unlock_buffer(bh); 15731da177e4SLinus Torvalds } 15741da177e4SLinus Torvalds 15751da177e4SLinus Torvalds /** 15761da177e4SLinus Torvalds * try_to_release_page() - release old fs-specific metadata on a page 15771da177e4SLinus Torvalds * 15781da177e4SLinus Torvalds * @page: the page which the kernel is trying to free 15791da177e4SLinus Torvalds * @gfp_mask: memory allocation flags (and I/O mode) 15801da177e4SLinus Torvalds * 15811da177e4SLinus Torvalds * The address_space is to try to release any data against the page 15821da177e4SLinus Torvalds * (presumably at page->private). If the release was successful, return `1'. 15831da177e4SLinus Torvalds * Otherwise return zero. 15841da177e4SLinus Torvalds * 15851da177e4SLinus Torvalds * The @gfp_mask argument specifies whether I/O may be performed to release 15861da177e4SLinus Torvalds * this page (__GFP_IO), and whether the call may block (__GFP_WAIT). 15871da177e4SLinus Torvalds * 15881da177e4SLinus Torvalds * NOTE: @gfp_mask may go away, and this function may become non-blocking. 15891da177e4SLinus Torvalds */ 15901da177e4SLinus Torvalds int try_to_release_page(struct page *page, int gfp_mask) 15911da177e4SLinus Torvalds { 15921da177e4SLinus Torvalds struct address_space * const mapping = page->mapping; 15931da177e4SLinus Torvalds 15941da177e4SLinus Torvalds BUG_ON(!PageLocked(page)); 15951da177e4SLinus Torvalds if (PageWriteback(page)) 15961da177e4SLinus Torvalds return 0; 15971da177e4SLinus Torvalds 15981da177e4SLinus Torvalds if (mapping && mapping->a_ops->releasepage) 15991da177e4SLinus Torvalds return mapping->a_ops->releasepage(page, gfp_mask); 16001da177e4SLinus Torvalds return try_to_free_buffers(page); 16011da177e4SLinus Torvalds } 16021da177e4SLinus Torvalds EXPORT_SYMBOL(try_to_release_page); 16031da177e4SLinus Torvalds 16041da177e4SLinus Torvalds /** 16051da177e4SLinus Torvalds * block_invalidatepage - invalidate part of all of a buffer-backed page 16061da177e4SLinus Torvalds * 16071da177e4SLinus Torvalds * @page: the page which is affected 16081da177e4SLinus Torvalds * @offset: the index of the truncation point 16091da177e4SLinus Torvalds * 16101da177e4SLinus Torvalds * block_invalidatepage() is called when all or part of the page has become 16111da177e4SLinus Torvalds * invalidatedby a truncate operation. 16121da177e4SLinus Torvalds * 16131da177e4SLinus Torvalds * block_invalidatepage() does not have to release all buffers, but it must 16141da177e4SLinus Torvalds * ensure that no dirty buffer is left outside @offset and that no I/O 16151da177e4SLinus Torvalds * is underway against any of the blocks which are outside the truncation 16161da177e4SLinus Torvalds * point. Because the caller is about to free (and possibly reuse) those 16171da177e4SLinus Torvalds * blocks on-disk. 16181da177e4SLinus Torvalds */ 16191da177e4SLinus Torvalds int block_invalidatepage(struct page *page, unsigned long offset) 16201da177e4SLinus Torvalds { 16211da177e4SLinus Torvalds struct buffer_head *head, *bh, *next; 16221da177e4SLinus Torvalds unsigned int curr_off = 0; 16231da177e4SLinus Torvalds int ret = 1; 16241da177e4SLinus Torvalds 16251da177e4SLinus Torvalds BUG_ON(!PageLocked(page)); 16261da177e4SLinus Torvalds if (!page_has_buffers(page)) 16271da177e4SLinus Torvalds goto out; 16281da177e4SLinus Torvalds 16291da177e4SLinus Torvalds head = page_buffers(page); 16301da177e4SLinus Torvalds bh = head; 16311da177e4SLinus Torvalds do { 16321da177e4SLinus Torvalds unsigned int next_off = curr_off + bh->b_size; 16331da177e4SLinus Torvalds next = bh->b_this_page; 16341da177e4SLinus Torvalds 16351da177e4SLinus Torvalds /* 16361da177e4SLinus Torvalds * is this block fully invalidated? 16371da177e4SLinus Torvalds */ 16381da177e4SLinus Torvalds if (offset <= curr_off) 16391da177e4SLinus Torvalds discard_buffer(bh); 16401da177e4SLinus Torvalds curr_off = next_off; 16411da177e4SLinus Torvalds bh = next; 16421da177e4SLinus Torvalds } while (bh != head); 16431da177e4SLinus Torvalds 16441da177e4SLinus Torvalds /* 16451da177e4SLinus Torvalds * We release buffers only if the entire page is being invalidated. 16461da177e4SLinus Torvalds * The get_block cached value has been unconditionally invalidated, 16471da177e4SLinus Torvalds * so real IO is not possible anymore. 16481da177e4SLinus Torvalds */ 16491da177e4SLinus Torvalds if (offset == 0) 16501da177e4SLinus Torvalds ret = try_to_release_page(page, 0); 16511da177e4SLinus Torvalds out: 16521da177e4SLinus Torvalds return ret; 16531da177e4SLinus Torvalds } 16541da177e4SLinus Torvalds EXPORT_SYMBOL(block_invalidatepage); 16551da177e4SLinus Torvalds 16561da177e4SLinus Torvalds /* 16571da177e4SLinus Torvalds * We attach and possibly dirty the buffers atomically wrt 16581da177e4SLinus Torvalds * __set_page_dirty_buffers() via private_lock. try_to_free_buffers 16591da177e4SLinus Torvalds * is already excluded via the page lock. 16601da177e4SLinus Torvalds */ 16611da177e4SLinus Torvalds void create_empty_buffers(struct page *page, 16621da177e4SLinus Torvalds unsigned long blocksize, unsigned long b_state) 16631da177e4SLinus Torvalds { 16641da177e4SLinus Torvalds struct buffer_head *bh, *head, *tail; 16651da177e4SLinus Torvalds 16661da177e4SLinus Torvalds head = alloc_page_buffers(page, blocksize, 1); 16671da177e4SLinus Torvalds bh = head; 16681da177e4SLinus Torvalds do { 16691da177e4SLinus Torvalds bh->b_state |= b_state; 16701da177e4SLinus Torvalds tail = bh; 16711da177e4SLinus Torvalds bh = bh->b_this_page; 16721da177e4SLinus Torvalds } while (bh); 16731da177e4SLinus Torvalds tail->b_this_page = head; 16741da177e4SLinus Torvalds 16751da177e4SLinus Torvalds spin_lock(&page->mapping->private_lock); 16761da177e4SLinus Torvalds if (PageUptodate(page) || PageDirty(page)) { 16771da177e4SLinus Torvalds bh = head; 16781da177e4SLinus Torvalds do { 16791da177e4SLinus Torvalds if (PageDirty(page)) 16801da177e4SLinus Torvalds set_buffer_dirty(bh); 16811da177e4SLinus Torvalds if (PageUptodate(page)) 16821da177e4SLinus Torvalds set_buffer_uptodate(bh); 16831da177e4SLinus Torvalds bh = bh->b_this_page; 16841da177e4SLinus Torvalds } while (bh != head); 16851da177e4SLinus Torvalds } 16861da177e4SLinus Torvalds attach_page_buffers(page, head); 16871da177e4SLinus Torvalds spin_unlock(&page->mapping->private_lock); 16881da177e4SLinus Torvalds } 16891da177e4SLinus Torvalds EXPORT_SYMBOL(create_empty_buffers); 16901da177e4SLinus Torvalds 16911da177e4SLinus Torvalds /* 16921da177e4SLinus Torvalds * We are taking a block for data and we don't want any output from any 16931da177e4SLinus Torvalds * buffer-cache aliases starting from return from that function and 16941da177e4SLinus Torvalds * until the moment when something will explicitly mark the buffer 16951da177e4SLinus Torvalds * dirty (hopefully that will not happen until we will free that block ;-) 16961da177e4SLinus Torvalds * We don't even need to mark it not-uptodate - nobody can expect 16971da177e4SLinus Torvalds * anything from a newly allocated buffer anyway. We used to used 16981da177e4SLinus Torvalds * unmap_buffer() for such invalidation, but that was wrong. We definitely 16991da177e4SLinus Torvalds * don't want to mark the alias unmapped, for example - it would confuse 17001da177e4SLinus Torvalds * anyone who might pick it with bread() afterwards... 17011da177e4SLinus Torvalds * 17021da177e4SLinus Torvalds * Also.. Note that bforget() doesn't lock the buffer. So there can 17031da177e4SLinus Torvalds * be writeout I/O going on against recently-freed buffers. We don't 17041da177e4SLinus Torvalds * wait on that I/O in bforget() - it's more efficient to wait on the I/O 17051da177e4SLinus Torvalds * only if we really need to. That happens here. 17061da177e4SLinus Torvalds */ 17071da177e4SLinus Torvalds void unmap_underlying_metadata(struct block_device *bdev, sector_t block) 17081da177e4SLinus Torvalds { 17091da177e4SLinus Torvalds struct buffer_head *old_bh; 17101da177e4SLinus Torvalds 17111da177e4SLinus Torvalds might_sleep(); 17121da177e4SLinus Torvalds 17131da177e4SLinus Torvalds old_bh = __find_get_block_slow(bdev, block, 0); 17141da177e4SLinus Torvalds if (old_bh) { 17151da177e4SLinus Torvalds clear_buffer_dirty(old_bh); 17161da177e4SLinus Torvalds wait_on_buffer(old_bh); 17171da177e4SLinus Torvalds clear_buffer_req(old_bh); 17181da177e4SLinus Torvalds __brelse(old_bh); 17191da177e4SLinus Torvalds } 17201da177e4SLinus Torvalds } 17211da177e4SLinus Torvalds EXPORT_SYMBOL(unmap_underlying_metadata); 17221da177e4SLinus Torvalds 17231da177e4SLinus Torvalds /* 17241da177e4SLinus Torvalds * NOTE! All mapped/uptodate combinations are valid: 17251da177e4SLinus Torvalds * 17261da177e4SLinus Torvalds * Mapped Uptodate Meaning 17271da177e4SLinus Torvalds * 17281da177e4SLinus Torvalds * No No "unknown" - must do get_block() 17291da177e4SLinus Torvalds * No Yes "hole" - zero-filled 17301da177e4SLinus Torvalds * Yes No "allocated" - allocated on disk, not read in 17311da177e4SLinus Torvalds * Yes Yes "valid" - allocated and up-to-date in memory. 17321da177e4SLinus Torvalds * 17331da177e4SLinus Torvalds * "Dirty" is valid only with the last case (mapped+uptodate). 17341da177e4SLinus Torvalds */ 17351da177e4SLinus Torvalds 17361da177e4SLinus Torvalds /* 17371da177e4SLinus Torvalds * While block_write_full_page is writing back the dirty buffers under 17381da177e4SLinus Torvalds * the page lock, whoever dirtied the buffers may decide to clean them 17391da177e4SLinus Torvalds * again at any time. We handle that by only looking at the buffer 17401da177e4SLinus Torvalds * state inside lock_buffer(). 17411da177e4SLinus Torvalds * 17421da177e4SLinus Torvalds * If block_write_full_page() is called for regular writeback 17431da177e4SLinus Torvalds * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a 17441da177e4SLinus Torvalds * locked buffer. This only can happen if someone has written the buffer 17451da177e4SLinus Torvalds * directly, with submit_bh(). At the address_space level PageWriteback 17461da177e4SLinus Torvalds * prevents this contention from occurring. 17471da177e4SLinus Torvalds */ 17481da177e4SLinus Torvalds static int __block_write_full_page(struct inode *inode, struct page *page, 17491da177e4SLinus Torvalds get_block_t *get_block, struct writeback_control *wbc) 17501da177e4SLinus Torvalds { 17511da177e4SLinus Torvalds int err; 17521da177e4SLinus Torvalds sector_t block; 17531da177e4SLinus Torvalds sector_t last_block; 17541da177e4SLinus Torvalds struct buffer_head *bh, *head; 17551da177e4SLinus Torvalds int nr_underway = 0; 17561da177e4SLinus Torvalds 17571da177e4SLinus Torvalds BUG_ON(!PageLocked(page)); 17581da177e4SLinus Torvalds 17591da177e4SLinus Torvalds last_block = (i_size_read(inode) - 1) >> inode->i_blkbits; 17601da177e4SLinus Torvalds 17611da177e4SLinus Torvalds if (!page_has_buffers(page)) { 17621da177e4SLinus Torvalds create_empty_buffers(page, 1 << inode->i_blkbits, 17631da177e4SLinus Torvalds (1 << BH_Dirty)|(1 << BH_Uptodate)); 17641da177e4SLinus Torvalds } 17651da177e4SLinus Torvalds 17661da177e4SLinus Torvalds /* 17671da177e4SLinus Torvalds * Be very careful. We have no exclusion from __set_page_dirty_buffers 17681da177e4SLinus Torvalds * here, and the (potentially unmapped) buffers may become dirty at 17691da177e4SLinus Torvalds * any time. If a buffer becomes dirty here after we've inspected it 17701da177e4SLinus Torvalds * then we just miss that fact, and the page stays dirty. 17711da177e4SLinus Torvalds * 17721da177e4SLinus Torvalds * Buffers outside i_size may be dirtied by __set_page_dirty_buffers; 17731da177e4SLinus Torvalds * handle that here by just cleaning them. 17741da177e4SLinus Torvalds */ 17751da177e4SLinus Torvalds 17761da177e4SLinus Torvalds block = page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits); 17771da177e4SLinus Torvalds head = page_buffers(page); 17781da177e4SLinus Torvalds bh = head; 17791da177e4SLinus Torvalds 17801da177e4SLinus Torvalds /* 17811da177e4SLinus Torvalds * Get all the dirty buffers mapped to disk addresses and 17821da177e4SLinus Torvalds * handle any aliases from the underlying blockdev's mapping. 17831da177e4SLinus Torvalds */ 17841da177e4SLinus Torvalds do { 17851da177e4SLinus Torvalds if (block > last_block) { 17861da177e4SLinus Torvalds /* 17871da177e4SLinus Torvalds * mapped buffers outside i_size will occur, because 17881da177e4SLinus Torvalds * this page can be outside i_size when there is a 17891da177e4SLinus Torvalds * truncate in progress. 17901da177e4SLinus Torvalds */ 17911da177e4SLinus Torvalds /* 17921da177e4SLinus Torvalds * The buffer was zeroed by block_write_full_page() 17931da177e4SLinus Torvalds */ 17941da177e4SLinus Torvalds clear_buffer_dirty(bh); 17951da177e4SLinus Torvalds set_buffer_uptodate(bh); 17961da177e4SLinus Torvalds } else if (!buffer_mapped(bh) && buffer_dirty(bh)) { 17971da177e4SLinus Torvalds err = get_block(inode, block, bh, 1); 17981da177e4SLinus Torvalds if (err) 17991da177e4SLinus Torvalds goto recover; 18001da177e4SLinus Torvalds if (buffer_new(bh)) { 18011da177e4SLinus Torvalds /* blockdev mappings never come here */ 18021da177e4SLinus Torvalds clear_buffer_new(bh); 18031da177e4SLinus Torvalds unmap_underlying_metadata(bh->b_bdev, 18041da177e4SLinus Torvalds bh->b_blocknr); 18051da177e4SLinus Torvalds } 18061da177e4SLinus Torvalds } 18071da177e4SLinus Torvalds bh = bh->b_this_page; 18081da177e4SLinus Torvalds block++; 18091da177e4SLinus Torvalds } while (bh != head); 18101da177e4SLinus Torvalds 18111da177e4SLinus Torvalds do { 18121da177e4SLinus Torvalds get_bh(bh); 18131da177e4SLinus Torvalds if (!buffer_mapped(bh)) 18141da177e4SLinus Torvalds continue; 18151da177e4SLinus Torvalds /* 18161da177e4SLinus Torvalds * If it's a fully non-blocking write attempt and we cannot 18171da177e4SLinus Torvalds * lock the buffer then redirty the page. Note that this can 18181da177e4SLinus Torvalds * potentially cause a busy-wait loop from pdflush and kswapd 18191da177e4SLinus Torvalds * activity, but those code paths have their own higher-level 18201da177e4SLinus Torvalds * throttling. 18211da177e4SLinus Torvalds */ 18221da177e4SLinus Torvalds if (wbc->sync_mode != WB_SYNC_NONE || !wbc->nonblocking) { 18231da177e4SLinus Torvalds lock_buffer(bh); 18241da177e4SLinus Torvalds } else if (test_set_buffer_locked(bh)) { 18251da177e4SLinus Torvalds redirty_page_for_writepage(wbc, page); 18261da177e4SLinus Torvalds continue; 18271da177e4SLinus Torvalds } 18281da177e4SLinus Torvalds if (test_clear_buffer_dirty(bh)) { 18291da177e4SLinus Torvalds mark_buffer_async_write(bh); 18301da177e4SLinus Torvalds } else { 18311da177e4SLinus Torvalds unlock_buffer(bh); 18321da177e4SLinus Torvalds } 18331da177e4SLinus Torvalds } while ((bh = bh->b_this_page) != head); 18341da177e4SLinus Torvalds 18351da177e4SLinus Torvalds /* 18361da177e4SLinus Torvalds * The page and its buffers are protected by PageWriteback(), so we can 18371da177e4SLinus Torvalds * drop the bh refcounts early. 18381da177e4SLinus Torvalds */ 18391da177e4SLinus Torvalds BUG_ON(PageWriteback(page)); 18401da177e4SLinus Torvalds set_page_writeback(page); 18411da177e4SLinus Torvalds unlock_page(page); 18421da177e4SLinus Torvalds 18431da177e4SLinus Torvalds do { 18441da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 18451da177e4SLinus Torvalds if (buffer_async_write(bh)) { 18461da177e4SLinus Torvalds submit_bh(WRITE, bh); 18471da177e4SLinus Torvalds nr_underway++; 18481da177e4SLinus Torvalds } 18491da177e4SLinus Torvalds put_bh(bh); 18501da177e4SLinus Torvalds bh = next; 18511da177e4SLinus Torvalds } while (bh != head); 18521da177e4SLinus Torvalds 18531da177e4SLinus Torvalds err = 0; 18541da177e4SLinus Torvalds done: 18551da177e4SLinus Torvalds if (nr_underway == 0) { 18561da177e4SLinus Torvalds /* 18571da177e4SLinus Torvalds * The page was marked dirty, but the buffers were 18581da177e4SLinus Torvalds * clean. Someone wrote them back by hand with 18591da177e4SLinus Torvalds * ll_rw_block/submit_bh. A rare case. 18601da177e4SLinus Torvalds */ 18611da177e4SLinus Torvalds int uptodate = 1; 18621da177e4SLinus Torvalds do { 18631da177e4SLinus Torvalds if (!buffer_uptodate(bh)) { 18641da177e4SLinus Torvalds uptodate = 0; 18651da177e4SLinus Torvalds break; 18661da177e4SLinus Torvalds } 18671da177e4SLinus Torvalds bh = bh->b_this_page; 18681da177e4SLinus Torvalds } while (bh != head); 18691da177e4SLinus Torvalds if (uptodate) 18701da177e4SLinus Torvalds SetPageUptodate(page); 18711da177e4SLinus Torvalds end_page_writeback(page); 18721da177e4SLinus Torvalds /* 18731da177e4SLinus Torvalds * The page and buffer_heads can be released at any time from 18741da177e4SLinus Torvalds * here on. 18751da177e4SLinus Torvalds */ 18761da177e4SLinus Torvalds wbc->pages_skipped++; /* We didn't write this page */ 18771da177e4SLinus Torvalds } 18781da177e4SLinus Torvalds return err; 18791da177e4SLinus Torvalds 18801da177e4SLinus Torvalds recover: 18811da177e4SLinus Torvalds /* 18821da177e4SLinus Torvalds * ENOSPC, or some other error. We may already have added some 18831da177e4SLinus Torvalds * blocks to the file, so we need to write these out to avoid 18841da177e4SLinus Torvalds * exposing stale data. 18851da177e4SLinus Torvalds * The page is currently locked and not marked for writeback 18861da177e4SLinus Torvalds */ 18871da177e4SLinus Torvalds bh = head; 18881da177e4SLinus Torvalds /* Recovery: lock and submit the mapped buffers */ 18891da177e4SLinus Torvalds do { 18901da177e4SLinus Torvalds get_bh(bh); 18911da177e4SLinus Torvalds if (buffer_mapped(bh) && buffer_dirty(bh)) { 18921da177e4SLinus Torvalds lock_buffer(bh); 18931da177e4SLinus Torvalds mark_buffer_async_write(bh); 18941da177e4SLinus Torvalds } else { 18951da177e4SLinus Torvalds /* 18961da177e4SLinus Torvalds * The buffer may have been set dirty during 18971da177e4SLinus Torvalds * attachment to a dirty page. 18981da177e4SLinus Torvalds */ 18991da177e4SLinus Torvalds clear_buffer_dirty(bh); 19001da177e4SLinus Torvalds } 19011da177e4SLinus Torvalds } while ((bh = bh->b_this_page) != head); 19021da177e4SLinus Torvalds SetPageError(page); 19031da177e4SLinus Torvalds BUG_ON(PageWriteback(page)); 19041da177e4SLinus Torvalds set_page_writeback(page); 19051da177e4SLinus Torvalds unlock_page(page); 19061da177e4SLinus Torvalds do { 19071da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 19081da177e4SLinus Torvalds if (buffer_async_write(bh)) { 19091da177e4SLinus Torvalds clear_buffer_dirty(bh); 19101da177e4SLinus Torvalds submit_bh(WRITE, bh); 19111da177e4SLinus Torvalds nr_underway++; 19121da177e4SLinus Torvalds } 19131da177e4SLinus Torvalds put_bh(bh); 19141da177e4SLinus Torvalds bh = next; 19151da177e4SLinus Torvalds } while (bh != head); 19161da177e4SLinus Torvalds goto done; 19171da177e4SLinus Torvalds } 19181da177e4SLinus Torvalds 19191da177e4SLinus Torvalds static int __block_prepare_write(struct inode *inode, struct page *page, 19201da177e4SLinus Torvalds unsigned from, unsigned to, get_block_t *get_block) 19211da177e4SLinus Torvalds { 19221da177e4SLinus Torvalds unsigned block_start, block_end; 19231da177e4SLinus Torvalds sector_t block; 19241da177e4SLinus Torvalds int err = 0; 19251da177e4SLinus Torvalds unsigned blocksize, bbits; 19261da177e4SLinus Torvalds struct buffer_head *bh, *head, *wait[2], **wait_bh=wait; 19271da177e4SLinus Torvalds 19281da177e4SLinus Torvalds BUG_ON(!PageLocked(page)); 19291da177e4SLinus Torvalds BUG_ON(from > PAGE_CACHE_SIZE); 19301da177e4SLinus Torvalds BUG_ON(to > PAGE_CACHE_SIZE); 19311da177e4SLinus Torvalds BUG_ON(from > to); 19321da177e4SLinus Torvalds 19331da177e4SLinus Torvalds blocksize = 1 << inode->i_blkbits; 19341da177e4SLinus Torvalds if (!page_has_buffers(page)) 19351da177e4SLinus Torvalds create_empty_buffers(page, blocksize, 0); 19361da177e4SLinus Torvalds head = page_buffers(page); 19371da177e4SLinus Torvalds 19381da177e4SLinus Torvalds bbits = inode->i_blkbits; 19391da177e4SLinus Torvalds block = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits); 19401da177e4SLinus Torvalds 19411da177e4SLinus Torvalds for(bh = head, block_start = 0; bh != head || !block_start; 19421da177e4SLinus Torvalds block++, block_start=block_end, bh = bh->b_this_page) { 19431da177e4SLinus Torvalds block_end = block_start + blocksize; 19441da177e4SLinus Torvalds if (block_end <= from || block_start >= to) { 19451da177e4SLinus Torvalds if (PageUptodate(page)) { 19461da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 19471da177e4SLinus Torvalds set_buffer_uptodate(bh); 19481da177e4SLinus Torvalds } 19491da177e4SLinus Torvalds continue; 19501da177e4SLinus Torvalds } 19511da177e4SLinus Torvalds if (buffer_new(bh)) 19521da177e4SLinus Torvalds clear_buffer_new(bh); 19531da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 19541da177e4SLinus Torvalds err = get_block(inode, block, bh, 1); 19551da177e4SLinus Torvalds if (err) 19561da177e4SLinus Torvalds goto out; 19571da177e4SLinus Torvalds if (buffer_new(bh)) { 19581da177e4SLinus Torvalds clear_buffer_new(bh); 19591da177e4SLinus Torvalds unmap_underlying_metadata(bh->b_bdev, 19601da177e4SLinus Torvalds bh->b_blocknr); 19611da177e4SLinus Torvalds if (PageUptodate(page)) { 19621da177e4SLinus Torvalds set_buffer_uptodate(bh); 19631da177e4SLinus Torvalds continue; 19641da177e4SLinus Torvalds } 19651da177e4SLinus Torvalds if (block_end > to || block_start < from) { 19661da177e4SLinus Torvalds void *kaddr; 19671da177e4SLinus Torvalds 19681da177e4SLinus Torvalds kaddr = kmap_atomic(page, KM_USER0); 19691da177e4SLinus Torvalds if (block_end > to) 19701da177e4SLinus Torvalds memset(kaddr+to, 0, 19711da177e4SLinus Torvalds block_end-to); 19721da177e4SLinus Torvalds if (block_start < from) 19731da177e4SLinus Torvalds memset(kaddr+block_start, 19741da177e4SLinus Torvalds 0, from-block_start); 19751da177e4SLinus Torvalds flush_dcache_page(page); 19761da177e4SLinus Torvalds kunmap_atomic(kaddr, KM_USER0); 19771da177e4SLinus Torvalds } 19781da177e4SLinus Torvalds continue; 19791da177e4SLinus Torvalds } 19801da177e4SLinus Torvalds } 19811da177e4SLinus Torvalds if (PageUptodate(page)) { 19821da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 19831da177e4SLinus Torvalds set_buffer_uptodate(bh); 19841da177e4SLinus Torvalds continue; 19851da177e4SLinus Torvalds } 19861da177e4SLinus Torvalds if (!buffer_uptodate(bh) && !buffer_delay(bh) && 19871da177e4SLinus Torvalds (block_start < from || block_end > to)) { 19881da177e4SLinus Torvalds ll_rw_block(READ, 1, &bh); 19891da177e4SLinus Torvalds *wait_bh++=bh; 19901da177e4SLinus Torvalds } 19911da177e4SLinus Torvalds } 19921da177e4SLinus Torvalds /* 19931da177e4SLinus Torvalds * If we issued read requests - let them complete. 19941da177e4SLinus Torvalds */ 19951da177e4SLinus Torvalds while(wait_bh > wait) { 19961da177e4SLinus Torvalds wait_on_buffer(*--wait_bh); 19971da177e4SLinus Torvalds if (!buffer_uptodate(*wait_bh)) 19981da177e4SLinus Torvalds return -EIO; 19991da177e4SLinus Torvalds } 20001da177e4SLinus Torvalds return 0; 20011da177e4SLinus Torvalds out: 20021da177e4SLinus Torvalds /* 20031da177e4SLinus Torvalds * Zero out any newly allocated blocks to avoid exposing stale 20041da177e4SLinus Torvalds * data. If BH_New is set, we know that the block was newly 20051da177e4SLinus Torvalds * allocated in the above loop. 20061da177e4SLinus Torvalds */ 20071da177e4SLinus Torvalds bh = head; 20081da177e4SLinus Torvalds block_start = 0; 20091da177e4SLinus Torvalds do { 20101da177e4SLinus Torvalds block_end = block_start+blocksize; 20111da177e4SLinus Torvalds if (block_end <= from) 20121da177e4SLinus Torvalds goto next_bh; 20131da177e4SLinus Torvalds if (block_start >= to) 20141da177e4SLinus Torvalds break; 20151da177e4SLinus Torvalds if (buffer_new(bh)) { 20161da177e4SLinus Torvalds void *kaddr; 20171da177e4SLinus Torvalds 20181da177e4SLinus Torvalds clear_buffer_new(bh); 20191da177e4SLinus Torvalds kaddr = kmap_atomic(page, KM_USER0); 20201da177e4SLinus Torvalds memset(kaddr+block_start, 0, bh->b_size); 20211da177e4SLinus Torvalds kunmap_atomic(kaddr, KM_USER0); 20221da177e4SLinus Torvalds set_buffer_uptodate(bh); 20231da177e4SLinus Torvalds mark_buffer_dirty(bh); 20241da177e4SLinus Torvalds } 20251da177e4SLinus Torvalds next_bh: 20261da177e4SLinus Torvalds block_start = block_end; 20271da177e4SLinus Torvalds bh = bh->b_this_page; 20281da177e4SLinus Torvalds } while (bh != head); 20291da177e4SLinus Torvalds return err; 20301da177e4SLinus Torvalds } 20311da177e4SLinus Torvalds 20321da177e4SLinus Torvalds static int __block_commit_write(struct inode *inode, struct page *page, 20331da177e4SLinus Torvalds unsigned from, unsigned to) 20341da177e4SLinus Torvalds { 20351da177e4SLinus Torvalds unsigned block_start, block_end; 20361da177e4SLinus Torvalds int partial = 0; 20371da177e4SLinus Torvalds unsigned blocksize; 20381da177e4SLinus Torvalds struct buffer_head *bh, *head; 20391da177e4SLinus Torvalds 20401da177e4SLinus Torvalds blocksize = 1 << inode->i_blkbits; 20411da177e4SLinus Torvalds 20421da177e4SLinus Torvalds for(bh = head = page_buffers(page), block_start = 0; 20431da177e4SLinus Torvalds bh != head || !block_start; 20441da177e4SLinus Torvalds block_start=block_end, bh = bh->b_this_page) { 20451da177e4SLinus Torvalds block_end = block_start + blocksize; 20461da177e4SLinus Torvalds if (block_end <= from || block_start >= to) { 20471da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 20481da177e4SLinus Torvalds partial = 1; 20491da177e4SLinus Torvalds } else { 20501da177e4SLinus Torvalds set_buffer_uptodate(bh); 20511da177e4SLinus Torvalds mark_buffer_dirty(bh); 20521da177e4SLinus Torvalds } 20531da177e4SLinus Torvalds } 20541da177e4SLinus Torvalds 20551da177e4SLinus Torvalds /* 20561da177e4SLinus Torvalds * If this is a partial write which happened to make all buffers 20571da177e4SLinus Torvalds * uptodate then we can optimize away a bogus readpage() for 20581da177e4SLinus Torvalds * the next read(). Here we 'discover' whether the page went 20591da177e4SLinus Torvalds * uptodate as a result of this (potentially partial) write. 20601da177e4SLinus Torvalds */ 20611da177e4SLinus Torvalds if (!partial) 20621da177e4SLinus Torvalds SetPageUptodate(page); 20631da177e4SLinus Torvalds return 0; 20641da177e4SLinus Torvalds } 20651da177e4SLinus Torvalds 20661da177e4SLinus Torvalds /* 20671da177e4SLinus Torvalds * Generic "read page" function for block devices that have the normal 20681da177e4SLinus Torvalds * get_block functionality. This is most of the block device filesystems. 20691da177e4SLinus Torvalds * Reads the page asynchronously --- the unlock_buffer() and 20701da177e4SLinus Torvalds * set/clear_buffer_uptodate() functions propagate buffer state into the 20711da177e4SLinus Torvalds * page struct once IO has completed. 20721da177e4SLinus Torvalds */ 20731da177e4SLinus Torvalds int block_read_full_page(struct page *page, get_block_t *get_block) 20741da177e4SLinus Torvalds { 20751da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 20761da177e4SLinus Torvalds sector_t iblock, lblock; 20771da177e4SLinus Torvalds struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE]; 20781da177e4SLinus Torvalds unsigned int blocksize; 20791da177e4SLinus Torvalds int nr, i; 20801da177e4SLinus Torvalds int fully_mapped = 1; 20811da177e4SLinus Torvalds 2082cd7619d6SMatt Mackall BUG_ON(!PageLocked(page)); 20831da177e4SLinus Torvalds blocksize = 1 << inode->i_blkbits; 20841da177e4SLinus Torvalds if (!page_has_buffers(page)) 20851da177e4SLinus Torvalds create_empty_buffers(page, blocksize, 0); 20861da177e4SLinus Torvalds head = page_buffers(page); 20871da177e4SLinus Torvalds 20881da177e4SLinus Torvalds iblock = (sector_t)page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits); 20891da177e4SLinus Torvalds lblock = (i_size_read(inode)+blocksize-1) >> inode->i_blkbits; 20901da177e4SLinus Torvalds bh = head; 20911da177e4SLinus Torvalds nr = 0; 20921da177e4SLinus Torvalds i = 0; 20931da177e4SLinus Torvalds 20941da177e4SLinus Torvalds do { 20951da177e4SLinus Torvalds if (buffer_uptodate(bh)) 20961da177e4SLinus Torvalds continue; 20971da177e4SLinus Torvalds 20981da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 20991da177e4SLinus Torvalds fully_mapped = 0; 21001da177e4SLinus Torvalds if (iblock < lblock) { 21011da177e4SLinus Torvalds if (get_block(inode, iblock, bh, 0)) 21021da177e4SLinus Torvalds SetPageError(page); 21031da177e4SLinus Torvalds } 21041da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 21051da177e4SLinus Torvalds void *kaddr = kmap_atomic(page, KM_USER0); 21061da177e4SLinus Torvalds memset(kaddr + i * blocksize, 0, blocksize); 21071da177e4SLinus Torvalds flush_dcache_page(page); 21081da177e4SLinus Torvalds kunmap_atomic(kaddr, KM_USER0); 21091da177e4SLinus Torvalds set_buffer_uptodate(bh); 21101da177e4SLinus Torvalds continue; 21111da177e4SLinus Torvalds } 21121da177e4SLinus Torvalds /* 21131da177e4SLinus Torvalds * get_block() might have updated the buffer 21141da177e4SLinus Torvalds * synchronously 21151da177e4SLinus Torvalds */ 21161da177e4SLinus Torvalds if (buffer_uptodate(bh)) 21171da177e4SLinus Torvalds continue; 21181da177e4SLinus Torvalds } 21191da177e4SLinus Torvalds arr[nr++] = bh; 21201da177e4SLinus Torvalds } while (i++, iblock++, (bh = bh->b_this_page) != head); 21211da177e4SLinus Torvalds 21221da177e4SLinus Torvalds if (fully_mapped) 21231da177e4SLinus Torvalds SetPageMappedToDisk(page); 21241da177e4SLinus Torvalds 21251da177e4SLinus Torvalds if (!nr) { 21261da177e4SLinus Torvalds /* 21271da177e4SLinus Torvalds * All buffers are uptodate - we can set the page uptodate 21281da177e4SLinus Torvalds * as well. But not if get_block() returned an error. 21291da177e4SLinus Torvalds */ 21301da177e4SLinus Torvalds if (!PageError(page)) 21311da177e4SLinus Torvalds SetPageUptodate(page); 21321da177e4SLinus Torvalds unlock_page(page); 21331da177e4SLinus Torvalds return 0; 21341da177e4SLinus Torvalds } 21351da177e4SLinus Torvalds 21361da177e4SLinus Torvalds /* Stage two: lock the buffers */ 21371da177e4SLinus Torvalds for (i = 0; i < nr; i++) { 21381da177e4SLinus Torvalds bh = arr[i]; 21391da177e4SLinus Torvalds lock_buffer(bh); 21401da177e4SLinus Torvalds mark_buffer_async_read(bh); 21411da177e4SLinus Torvalds } 21421da177e4SLinus Torvalds 21431da177e4SLinus Torvalds /* 21441da177e4SLinus Torvalds * Stage 3: start the IO. Check for uptodateness 21451da177e4SLinus Torvalds * inside the buffer lock in case another process reading 21461da177e4SLinus Torvalds * the underlying blockdev brought it uptodate (the sct fix). 21471da177e4SLinus Torvalds */ 21481da177e4SLinus Torvalds for (i = 0; i < nr; i++) { 21491da177e4SLinus Torvalds bh = arr[i]; 21501da177e4SLinus Torvalds if (buffer_uptodate(bh)) 21511da177e4SLinus Torvalds end_buffer_async_read(bh, 1); 21521da177e4SLinus Torvalds else 21531da177e4SLinus Torvalds submit_bh(READ, bh); 21541da177e4SLinus Torvalds } 21551da177e4SLinus Torvalds return 0; 21561da177e4SLinus Torvalds } 21571da177e4SLinus Torvalds 21581da177e4SLinus Torvalds /* utility function for filesystems that need to do work on expanding 21591da177e4SLinus Torvalds * truncates. Uses prepare/commit_write to allow the filesystem to 21601da177e4SLinus Torvalds * deal with the hole. 21611da177e4SLinus Torvalds */ 21621da177e4SLinus Torvalds int generic_cont_expand(struct inode *inode, loff_t size) 21631da177e4SLinus Torvalds { 21641da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 21651da177e4SLinus Torvalds struct page *page; 21661da177e4SLinus Torvalds unsigned long index, offset, limit; 21671da177e4SLinus Torvalds int err; 21681da177e4SLinus Torvalds 21691da177e4SLinus Torvalds err = -EFBIG; 21701da177e4SLinus Torvalds limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur; 21711da177e4SLinus Torvalds if (limit != RLIM_INFINITY && size > (loff_t)limit) { 21721da177e4SLinus Torvalds send_sig(SIGXFSZ, current, 0); 21731da177e4SLinus Torvalds goto out; 21741da177e4SLinus Torvalds } 21751da177e4SLinus Torvalds if (size > inode->i_sb->s_maxbytes) 21761da177e4SLinus Torvalds goto out; 21771da177e4SLinus Torvalds 21781da177e4SLinus Torvalds offset = (size & (PAGE_CACHE_SIZE-1)); /* Within page */ 21791da177e4SLinus Torvalds 21801da177e4SLinus Torvalds /* ugh. in prepare/commit_write, if from==to==start of block, we 21811da177e4SLinus Torvalds ** skip the prepare. make sure we never send an offset for the start 21821da177e4SLinus Torvalds ** of a block 21831da177e4SLinus Torvalds */ 21841da177e4SLinus Torvalds if ((offset & (inode->i_sb->s_blocksize - 1)) == 0) { 21851da177e4SLinus Torvalds offset++; 21861da177e4SLinus Torvalds } 21871da177e4SLinus Torvalds index = size >> PAGE_CACHE_SHIFT; 21881da177e4SLinus Torvalds err = -ENOMEM; 21891da177e4SLinus Torvalds page = grab_cache_page(mapping, index); 21901da177e4SLinus Torvalds if (!page) 21911da177e4SLinus Torvalds goto out; 21921da177e4SLinus Torvalds err = mapping->a_ops->prepare_write(NULL, page, offset, offset); 21931da177e4SLinus Torvalds if (!err) { 21941da177e4SLinus Torvalds err = mapping->a_ops->commit_write(NULL, page, offset, offset); 21951da177e4SLinus Torvalds } 21961da177e4SLinus Torvalds unlock_page(page); 21971da177e4SLinus Torvalds page_cache_release(page); 21981da177e4SLinus Torvalds if (err > 0) 21991da177e4SLinus Torvalds err = 0; 22001da177e4SLinus Torvalds out: 22011da177e4SLinus Torvalds return err; 22021da177e4SLinus Torvalds } 22031da177e4SLinus Torvalds 22041da177e4SLinus Torvalds /* 22051da177e4SLinus Torvalds * For moronic filesystems that do not allow holes in file. 22061da177e4SLinus Torvalds * We may have to extend the file. 22071da177e4SLinus Torvalds */ 22081da177e4SLinus Torvalds 22091da177e4SLinus Torvalds int cont_prepare_write(struct page *page, unsigned offset, 22101da177e4SLinus Torvalds unsigned to, get_block_t *get_block, loff_t *bytes) 22111da177e4SLinus Torvalds { 22121da177e4SLinus Torvalds struct address_space *mapping = page->mapping; 22131da177e4SLinus Torvalds struct inode *inode = mapping->host; 22141da177e4SLinus Torvalds struct page *new_page; 22151da177e4SLinus Torvalds pgoff_t pgpos; 22161da177e4SLinus Torvalds long status; 22171da177e4SLinus Torvalds unsigned zerofrom; 22181da177e4SLinus Torvalds unsigned blocksize = 1 << inode->i_blkbits; 22191da177e4SLinus Torvalds void *kaddr; 22201da177e4SLinus Torvalds 22211da177e4SLinus Torvalds while(page->index > (pgpos = *bytes>>PAGE_CACHE_SHIFT)) { 22221da177e4SLinus Torvalds status = -ENOMEM; 22231da177e4SLinus Torvalds new_page = grab_cache_page(mapping, pgpos); 22241da177e4SLinus Torvalds if (!new_page) 22251da177e4SLinus Torvalds goto out; 22261da177e4SLinus Torvalds /* we might sleep */ 22271da177e4SLinus Torvalds if (*bytes>>PAGE_CACHE_SHIFT != pgpos) { 22281da177e4SLinus Torvalds unlock_page(new_page); 22291da177e4SLinus Torvalds page_cache_release(new_page); 22301da177e4SLinus Torvalds continue; 22311da177e4SLinus Torvalds } 22321da177e4SLinus Torvalds zerofrom = *bytes & ~PAGE_CACHE_MASK; 22331da177e4SLinus Torvalds if (zerofrom & (blocksize-1)) { 22341da177e4SLinus Torvalds *bytes |= (blocksize-1); 22351da177e4SLinus Torvalds (*bytes)++; 22361da177e4SLinus Torvalds } 22371da177e4SLinus Torvalds status = __block_prepare_write(inode, new_page, zerofrom, 22381da177e4SLinus Torvalds PAGE_CACHE_SIZE, get_block); 22391da177e4SLinus Torvalds if (status) 22401da177e4SLinus Torvalds goto out_unmap; 22411da177e4SLinus Torvalds kaddr = kmap_atomic(new_page, KM_USER0); 22421da177e4SLinus Torvalds memset(kaddr+zerofrom, 0, PAGE_CACHE_SIZE-zerofrom); 22431da177e4SLinus Torvalds flush_dcache_page(new_page); 22441da177e4SLinus Torvalds kunmap_atomic(kaddr, KM_USER0); 22451da177e4SLinus Torvalds generic_commit_write(NULL, new_page, zerofrom, PAGE_CACHE_SIZE); 22461da177e4SLinus Torvalds unlock_page(new_page); 22471da177e4SLinus Torvalds page_cache_release(new_page); 22481da177e4SLinus Torvalds } 22491da177e4SLinus Torvalds 22501da177e4SLinus Torvalds if (page->index < pgpos) { 22511da177e4SLinus Torvalds /* completely inside the area */ 22521da177e4SLinus Torvalds zerofrom = offset; 22531da177e4SLinus Torvalds } else { 22541da177e4SLinus Torvalds /* page covers the boundary, find the boundary offset */ 22551da177e4SLinus Torvalds zerofrom = *bytes & ~PAGE_CACHE_MASK; 22561da177e4SLinus Torvalds 22571da177e4SLinus Torvalds /* if we will expand the thing last block will be filled */ 22581da177e4SLinus Torvalds if (to > zerofrom && (zerofrom & (blocksize-1))) { 22591da177e4SLinus Torvalds *bytes |= (blocksize-1); 22601da177e4SLinus Torvalds (*bytes)++; 22611da177e4SLinus Torvalds } 22621da177e4SLinus Torvalds 22631da177e4SLinus Torvalds /* starting below the boundary? Nothing to zero out */ 22641da177e4SLinus Torvalds if (offset <= zerofrom) 22651da177e4SLinus Torvalds zerofrom = offset; 22661da177e4SLinus Torvalds } 22671da177e4SLinus Torvalds status = __block_prepare_write(inode, page, zerofrom, to, get_block); 22681da177e4SLinus Torvalds if (status) 22691da177e4SLinus Torvalds goto out1; 22701da177e4SLinus Torvalds if (zerofrom < offset) { 22711da177e4SLinus Torvalds kaddr = kmap_atomic(page, KM_USER0); 22721da177e4SLinus Torvalds memset(kaddr+zerofrom, 0, offset-zerofrom); 22731da177e4SLinus Torvalds flush_dcache_page(page); 22741da177e4SLinus Torvalds kunmap_atomic(kaddr, KM_USER0); 22751da177e4SLinus Torvalds __block_commit_write(inode, page, zerofrom, offset); 22761da177e4SLinus Torvalds } 22771da177e4SLinus Torvalds return 0; 22781da177e4SLinus Torvalds out1: 22791da177e4SLinus Torvalds ClearPageUptodate(page); 22801da177e4SLinus Torvalds return status; 22811da177e4SLinus Torvalds 22821da177e4SLinus Torvalds out_unmap: 22831da177e4SLinus Torvalds ClearPageUptodate(new_page); 22841da177e4SLinus Torvalds unlock_page(new_page); 22851da177e4SLinus Torvalds page_cache_release(new_page); 22861da177e4SLinus Torvalds out: 22871da177e4SLinus Torvalds return status; 22881da177e4SLinus Torvalds } 22891da177e4SLinus Torvalds 22901da177e4SLinus Torvalds int block_prepare_write(struct page *page, unsigned from, unsigned to, 22911da177e4SLinus Torvalds get_block_t *get_block) 22921da177e4SLinus Torvalds { 22931da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 22941da177e4SLinus Torvalds int err = __block_prepare_write(inode, page, from, to, get_block); 22951da177e4SLinus Torvalds if (err) 22961da177e4SLinus Torvalds ClearPageUptodate(page); 22971da177e4SLinus Torvalds return err; 22981da177e4SLinus Torvalds } 22991da177e4SLinus Torvalds 23001da177e4SLinus Torvalds int block_commit_write(struct page *page, unsigned from, unsigned to) 23011da177e4SLinus Torvalds { 23021da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 23031da177e4SLinus Torvalds __block_commit_write(inode,page,from,to); 23041da177e4SLinus Torvalds return 0; 23051da177e4SLinus Torvalds } 23061da177e4SLinus Torvalds 23071da177e4SLinus Torvalds int generic_commit_write(struct file *file, struct page *page, 23081da177e4SLinus Torvalds unsigned from, unsigned to) 23091da177e4SLinus Torvalds { 23101da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 23111da177e4SLinus Torvalds loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to; 23121da177e4SLinus Torvalds __block_commit_write(inode,page,from,to); 23131da177e4SLinus Torvalds /* 23141da177e4SLinus Torvalds * No need to use i_size_read() here, the i_size 23151da177e4SLinus Torvalds * cannot change under us because we hold i_sem. 23161da177e4SLinus Torvalds */ 23171da177e4SLinus Torvalds if (pos > inode->i_size) { 23181da177e4SLinus Torvalds i_size_write(inode, pos); 23191da177e4SLinus Torvalds mark_inode_dirty(inode); 23201da177e4SLinus Torvalds } 23211da177e4SLinus Torvalds return 0; 23221da177e4SLinus Torvalds } 23231da177e4SLinus Torvalds 23241da177e4SLinus Torvalds 23251da177e4SLinus Torvalds /* 23261da177e4SLinus Torvalds * nobh_prepare_write()'s prereads are special: the buffer_heads are freed 23271da177e4SLinus Torvalds * immediately, while under the page lock. So it needs a special end_io 23281da177e4SLinus Torvalds * handler which does not touch the bh after unlocking it. 23291da177e4SLinus Torvalds * 23301da177e4SLinus Torvalds * Note: unlock_buffer() sort-of does touch the bh after unlocking it, but 23311da177e4SLinus Torvalds * a race there is benign: unlock_buffer() only use the bh's address for 23321da177e4SLinus Torvalds * hashing after unlocking the buffer, so it doesn't actually touch the bh 23331da177e4SLinus Torvalds * itself. 23341da177e4SLinus Torvalds */ 23351da177e4SLinus Torvalds static void end_buffer_read_nobh(struct buffer_head *bh, int uptodate) 23361da177e4SLinus Torvalds { 23371da177e4SLinus Torvalds if (uptodate) { 23381da177e4SLinus Torvalds set_buffer_uptodate(bh); 23391da177e4SLinus Torvalds } else { 23401da177e4SLinus Torvalds /* This happens, due to failed READA attempts. */ 23411da177e4SLinus Torvalds clear_buffer_uptodate(bh); 23421da177e4SLinus Torvalds } 23431da177e4SLinus Torvalds unlock_buffer(bh); 23441da177e4SLinus Torvalds } 23451da177e4SLinus Torvalds 23461da177e4SLinus Torvalds /* 23471da177e4SLinus Torvalds * On entry, the page is fully not uptodate. 23481da177e4SLinus Torvalds * On exit the page is fully uptodate in the areas outside (from,to) 23491da177e4SLinus Torvalds */ 23501da177e4SLinus Torvalds int nobh_prepare_write(struct page *page, unsigned from, unsigned to, 23511da177e4SLinus Torvalds get_block_t *get_block) 23521da177e4SLinus Torvalds { 23531da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 23541da177e4SLinus Torvalds const unsigned blkbits = inode->i_blkbits; 23551da177e4SLinus Torvalds const unsigned blocksize = 1 << blkbits; 23561da177e4SLinus Torvalds struct buffer_head map_bh; 23571da177e4SLinus Torvalds struct buffer_head *read_bh[MAX_BUF_PER_PAGE]; 23581da177e4SLinus Torvalds unsigned block_in_page; 23591da177e4SLinus Torvalds unsigned block_start; 23601da177e4SLinus Torvalds sector_t block_in_file; 23611da177e4SLinus Torvalds char *kaddr; 23621da177e4SLinus Torvalds int nr_reads = 0; 23631da177e4SLinus Torvalds int i; 23641da177e4SLinus Torvalds int ret = 0; 23651da177e4SLinus Torvalds int is_mapped_to_disk = 1; 23661da177e4SLinus Torvalds int dirtied_it = 0; 23671da177e4SLinus Torvalds 23681da177e4SLinus Torvalds if (PageMappedToDisk(page)) 23691da177e4SLinus Torvalds return 0; 23701da177e4SLinus Torvalds 23711da177e4SLinus Torvalds block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits); 23721da177e4SLinus Torvalds map_bh.b_page = page; 23731da177e4SLinus Torvalds 23741da177e4SLinus Torvalds /* 23751da177e4SLinus Torvalds * We loop across all blocks in the page, whether or not they are 23761da177e4SLinus Torvalds * part of the affected region. This is so we can discover if the 23771da177e4SLinus Torvalds * page is fully mapped-to-disk. 23781da177e4SLinus Torvalds */ 23791da177e4SLinus Torvalds for (block_start = 0, block_in_page = 0; 23801da177e4SLinus Torvalds block_start < PAGE_CACHE_SIZE; 23811da177e4SLinus Torvalds block_in_page++, block_start += blocksize) { 23821da177e4SLinus Torvalds unsigned block_end = block_start + blocksize; 23831da177e4SLinus Torvalds int create; 23841da177e4SLinus Torvalds 23851da177e4SLinus Torvalds map_bh.b_state = 0; 23861da177e4SLinus Torvalds create = 1; 23871da177e4SLinus Torvalds if (block_start >= to) 23881da177e4SLinus Torvalds create = 0; 23891da177e4SLinus Torvalds ret = get_block(inode, block_in_file + block_in_page, 23901da177e4SLinus Torvalds &map_bh, create); 23911da177e4SLinus Torvalds if (ret) 23921da177e4SLinus Torvalds goto failed; 23931da177e4SLinus Torvalds if (!buffer_mapped(&map_bh)) 23941da177e4SLinus Torvalds is_mapped_to_disk = 0; 23951da177e4SLinus Torvalds if (buffer_new(&map_bh)) 23961da177e4SLinus Torvalds unmap_underlying_metadata(map_bh.b_bdev, 23971da177e4SLinus Torvalds map_bh.b_blocknr); 23981da177e4SLinus Torvalds if (PageUptodate(page)) 23991da177e4SLinus Torvalds continue; 24001da177e4SLinus Torvalds if (buffer_new(&map_bh) || !buffer_mapped(&map_bh)) { 24011da177e4SLinus Torvalds kaddr = kmap_atomic(page, KM_USER0); 24021da177e4SLinus Torvalds if (block_start < from) { 24031da177e4SLinus Torvalds memset(kaddr+block_start, 0, from-block_start); 24041da177e4SLinus Torvalds dirtied_it = 1; 24051da177e4SLinus Torvalds } 24061da177e4SLinus Torvalds if (block_end > to) { 24071da177e4SLinus Torvalds memset(kaddr + to, 0, block_end - to); 24081da177e4SLinus Torvalds dirtied_it = 1; 24091da177e4SLinus Torvalds } 24101da177e4SLinus Torvalds flush_dcache_page(page); 24111da177e4SLinus Torvalds kunmap_atomic(kaddr, KM_USER0); 24121da177e4SLinus Torvalds continue; 24131da177e4SLinus Torvalds } 24141da177e4SLinus Torvalds if (buffer_uptodate(&map_bh)) 24151da177e4SLinus Torvalds continue; /* reiserfs does this */ 24161da177e4SLinus Torvalds if (block_start < from || block_end > to) { 24171da177e4SLinus Torvalds struct buffer_head *bh = alloc_buffer_head(GFP_NOFS); 24181da177e4SLinus Torvalds 24191da177e4SLinus Torvalds if (!bh) { 24201da177e4SLinus Torvalds ret = -ENOMEM; 24211da177e4SLinus Torvalds goto failed; 24221da177e4SLinus Torvalds } 24231da177e4SLinus Torvalds bh->b_state = map_bh.b_state; 24241da177e4SLinus Torvalds atomic_set(&bh->b_count, 0); 24251da177e4SLinus Torvalds bh->b_this_page = NULL; 24261da177e4SLinus Torvalds bh->b_page = page; 24271da177e4SLinus Torvalds bh->b_blocknr = map_bh.b_blocknr; 24281da177e4SLinus Torvalds bh->b_size = blocksize; 24291da177e4SLinus Torvalds bh->b_data = (char *)(long)block_start; 24301da177e4SLinus Torvalds bh->b_bdev = map_bh.b_bdev; 24311da177e4SLinus Torvalds bh->b_private = NULL; 24321da177e4SLinus Torvalds read_bh[nr_reads++] = bh; 24331da177e4SLinus Torvalds } 24341da177e4SLinus Torvalds } 24351da177e4SLinus Torvalds 24361da177e4SLinus Torvalds if (nr_reads) { 24371da177e4SLinus Torvalds struct buffer_head *bh; 24381da177e4SLinus Torvalds 24391da177e4SLinus Torvalds /* 24401da177e4SLinus Torvalds * The page is locked, so these buffers are protected from 24411da177e4SLinus Torvalds * any VM or truncate activity. Hence we don't need to care 24421da177e4SLinus Torvalds * for the buffer_head refcounts. 24431da177e4SLinus Torvalds */ 24441da177e4SLinus Torvalds for (i = 0; i < nr_reads; i++) { 24451da177e4SLinus Torvalds bh = read_bh[i]; 24461da177e4SLinus Torvalds lock_buffer(bh); 24471da177e4SLinus Torvalds bh->b_end_io = end_buffer_read_nobh; 24481da177e4SLinus Torvalds submit_bh(READ, bh); 24491da177e4SLinus Torvalds } 24501da177e4SLinus Torvalds for (i = 0; i < nr_reads; i++) { 24511da177e4SLinus Torvalds bh = read_bh[i]; 24521da177e4SLinus Torvalds wait_on_buffer(bh); 24531da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 24541da177e4SLinus Torvalds ret = -EIO; 24551da177e4SLinus Torvalds free_buffer_head(bh); 24561da177e4SLinus Torvalds read_bh[i] = NULL; 24571da177e4SLinus Torvalds } 24581da177e4SLinus Torvalds if (ret) 24591da177e4SLinus Torvalds goto failed; 24601da177e4SLinus Torvalds } 24611da177e4SLinus Torvalds 24621da177e4SLinus Torvalds if (is_mapped_to_disk) 24631da177e4SLinus Torvalds SetPageMappedToDisk(page); 24641da177e4SLinus Torvalds SetPageUptodate(page); 24651da177e4SLinus Torvalds 24661da177e4SLinus Torvalds /* 24671da177e4SLinus Torvalds * Setting the page dirty here isn't necessary for the prepare_write 24681da177e4SLinus Torvalds * function - commit_write will do that. But if/when this function is 24691da177e4SLinus Torvalds * used within the pagefault handler to ensure that all mmapped pages 24701da177e4SLinus Torvalds * have backing space in the filesystem, we will need to dirty the page 24711da177e4SLinus Torvalds * if its contents were altered. 24721da177e4SLinus Torvalds */ 24731da177e4SLinus Torvalds if (dirtied_it) 24741da177e4SLinus Torvalds set_page_dirty(page); 24751da177e4SLinus Torvalds 24761da177e4SLinus Torvalds return 0; 24771da177e4SLinus Torvalds 24781da177e4SLinus Torvalds failed: 24791da177e4SLinus Torvalds for (i = 0; i < nr_reads; i++) { 24801da177e4SLinus Torvalds if (read_bh[i]) 24811da177e4SLinus Torvalds free_buffer_head(read_bh[i]); 24821da177e4SLinus Torvalds } 24831da177e4SLinus Torvalds 24841da177e4SLinus Torvalds /* 24851da177e4SLinus Torvalds * Error recovery is pretty slack. Clear the page and mark it dirty 24861da177e4SLinus Torvalds * so we'll later zero out any blocks which _were_ allocated. 24871da177e4SLinus Torvalds */ 24881da177e4SLinus Torvalds kaddr = kmap_atomic(page, KM_USER0); 24891da177e4SLinus Torvalds memset(kaddr, 0, PAGE_CACHE_SIZE); 24901da177e4SLinus Torvalds kunmap_atomic(kaddr, KM_USER0); 24911da177e4SLinus Torvalds SetPageUptodate(page); 24921da177e4SLinus Torvalds set_page_dirty(page); 24931da177e4SLinus Torvalds return ret; 24941da177e4SLinus Torvalds } 24951da177e4SLinus Torvalds EXPORT_SYMBOL(nobh_prepare_write); 24961da177e4SLinus Torvalds 24971da177e4SLinus Torvalds int nobh_commit_write(struct file *file, struct page *page, 24981da177e4SLinus Torvalds unsigned from, unsigned to) 24991da177e4SLinus Torvalds { 25001da177e4SLinus Torvalds struct inode *inode = page->mapping->host; 25011da177e4SLinus Torvalds loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to; 25021da177e4SLinus Torvalds 25031da177e4SLinus Torvalds set_page_dirty(page); 25041da177e4SLinus Torvalds if (pos > inode->i_size) { 25051da177e4SLinus Torvalds i_size_write(inode, pos); 25061da177e4SLinus Torvalds mark_inode_dirty(inode); 25071da177e4SLinus Torvalds } 25081da177e4SLinus Torvalds return 0; 25091da177e4SLinus Torvalds } 25101da177e4SLinus Torvalds EXPORT_SYMBOL(nobh_commit_write); 25111da177e4SLinus Torvalds 25121da177e4SLinus Torvalds /* 25131da177e4SLinus Torvalds * nobh_writepage() - based on block_full_write_page() except 25141da177e4SLinus Torvalds * that it tries to operate without attaching bufferheads to 25151da177e4SLinus Torvalds * the page. 25161da177e4SLinus Torvalds */ 25171da177e4SLinus Torvalds int nobh_writepage(struct page *page, get_block_t *get_block, 25181da177e4SLinus Torvalds struct writeback_control *wbc) 25191da177e4SLinus Torvalds { 25201da177e4SLinus Torvalds struct inode * const inode = page->mapping->host; 25211da177e4SLinus Torvalds loff_t i_size = i_size_read(inode); 25221da177e4SLinus Torvalds const pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; 25231da177e4SLinus Torvalds unsigned offset; 25241da177e4SLinus Torvalds void *kaddr; 25251da177e4SLinus Torvalds int ret; 25261da177e4SLinus Torvalds 25271da177e4SLinus Torvalds /* Is the page fully inside i_size? */ 25281da177e4SLinus Torvalds if (page->index < end_index) 25291da177e4SLinus Torvalds goto out; 25301da177e4SLinus Torvalds 25311da177e4SLinus Torvalds /* Is the page fully outside i_size? (truncate in progress) */ 25321da177e4SLinus Torvalds offset = i_size & (PAGE_CACHE_SIZE-1); 25331da177e4SLinus Torvalds if (page->index >= end_index+1 || !offset) { 25341da177e4SLinus Torvalds /* 25351da177e4SLinus Torvalds * The page may have dirty, unmapped buffers. For example, 25361da177e4SLinus Torvalds * they may have been added in ext3_writepage(). Make them 25371da177e4SLinus Torvalds * freeable here, so the page does not leak. 25381da177e4SLinus Torvalds */ 25391da177e4SLinus Torvalds #if 0 25401da177e4SLinus Torvalds /* Not really sure about this - do we need this ? */ 25411da177e4SLinus Torvalds if (page->mapping->a_ops->invalidatepage) 25421da177e4SLinus Torvalds page->mapping->a_ops->invalidatepage(page, offset); 25431da177e4SLinus Torvalds #endif 25441da177e4SLinus Torvalds unlock_page(page); 25451da177e4SLinus Torvalds return 0; /* don't care */ 25461da177e4SLinus Torvalds } 25471da177e4SLinus Torvalds 25481da177e4SLinus Torvalds /* 25491da177e4SLinus Torvalds * The page straddles i_size. It must be zeroed out on each and every 25501da177e4SLinus Torvalds * writepage invocation because it may be mmapped. "A file is mapped 25511da177e4SLinus Torvalds * in multiples of the page size. For a file that is not a multiple of 25521da177e4SLinus Torvalds * the page size, the remaining memory is zeroed when mapped, and 25531da177e4SLinus Torvalds * writes to that region are not written out to the file." 25541da177e4SLinus Torvalds */ 25551da177e4SLinus Torvalds kaddr = kmap_atomic(page, KM_USER0); 25561da177e4SLinus Torvalds memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset); 25571da177e4SLinus Torvalds flush_dcache_page(page); 25581da177e4SLinus Torvalds kunmap_atomic(kaddr, KM_USER0); 25591da177e4SLinus Torvalds out: 25601da177e4SLinus Torvalds ret = mpage_writepage(page, get_block, wbc); 25611da177e4SLinus Torvalds if (ret == -EAGAIN) 25621da177e4SLinus Torvalds ret = __block_write_full_page(inode, page, get_block, wbc); 25631da177e4SLinus Torvalds return ret; 25641da177e4SLinus Torvalds } 25651da177e4SLinus Torvalds EXPORT_SYMBOL(nobh_writepage); 25661da177e4SLinus Torvalds 25671da177e4SLinus Torvalds /* 25681da177e4SLinus Torvalds * This function assumes that ->prepare_write() uses nobh_prepare_write(). 25691da177e4SLinus Torvalds */ 25701da177e4SLinus Torvalds int nobh_truncate_page(struct address_space *mapping, loff_t from) 25711da177e4SLinus Torvalds { 25721da177e4SLinus Torvalds struct inode *inode = mapping->host; 25731da177e4SLinus Torvalds unsigned blocksize = 1 << inode->i_blkbits; 25741da177e4SLinus Torvalds pgoff_t index = from >> PAGE_CACHE_SHIFT; 25751da177e4SLinus Torvalds unsigned offset = from & (PAGE_CACHE_SIZE-1); 25761da177e4SLinus Torvalds unsigned to; 25771da177e4SLinus Torvalds struct page *page; 25781da177e4SLinus Torvalds struct address_space_operations *a_ops = mapping->a_ops; 25791da177e4SLinus Torvalds char *kaddr; 25801da177e4SLinus Torvalds int ret = 0; 25811da177e4SLinus Torvalds 25821da177e4SLinus Torvalds if ((offset & (blocksize - 1)) == 0) 25831da177e4SLinus Torvalds goto out; 25841da177e4SLinus Torvalds 25851da177e4SLinus Torvalds ret = -ENOMEM; 25861da177e4SLinus Torvalds page = grab_cache_page(mapping, index); 25871da177e4SLinus Torvalds if (!page) 25881da177e4SLinus Torvalds goto out; 25891da177e4SLinus Torvalds 25901da177e4SLinus Torvalds to = (offset + blocksize) & ~(blocksize - 1); 25911da177e4SLinus Torvalds ret = a_ops->prepare_write(NULL, page, offset, to); 25921da177e4SLinus Torvalds if (ret == 0) { 25931da177e4SLinus Torvalds kaddr = kmap_atomic(page, KM_USER0); 25941da177e4SLinus Torvalds memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset); 25951da177e4SLinus Torvalds flush_dcache_page(page); 25961da177e4SLinus Torvalds kunmap_atomic(kaddr, KM_USER0); 25971da177e4SLinus Torvalds set_page_dirty(page); 25981da177e4SLinus Torvalds } 25991da177e4SLinus Torvalds unlock_page(page); 26001da177e4SLinus Torvalds page_cache_release(page); 26011da177e4SLinus Torvalds out: 26021da177e4SLinus Torvalds return ret; 26031da177e4SLinus Torvalds } 26041da177e4SLinus Torvalds EXPORT_SYMBOL(nobh_truncate_page); 26051da177e4SLinus Torvalds 26061da177e4SLinus Torvalds int block_truncate_page(struct address_space *mapping, 26071da177e4SLinus Torvalds loff_t from, get_block_t *get_block) 26081da177e4SLinus Torvalds { 26091da177e4SLinus Torvalds pgoff_t index = from >> PAGE_CACHE_SHIFT; 26101da177e4SLinus Torvalds unsigned offset = from & (PAGE_CACHE_SIZE-1); 26111da177e4SLinus Torvalds unsigned blocksize; 26121da177e4SLinus Torvalds pgoff_t iblock; 26131da177e4SLinus Torvalds unsigned length, pos; 26141da177e4SLinus Torvalds struct inode *inode = mapping->host; 26151da177e4SLinus Torvalds struct page *page; 26161da177e4SLinus Torvalds struct buffer_head *bh; 26171da177e4SLinus Torvalds void *kaddr; 26181da177e4SLinus Torvalds int err; 26191da177e4SLinus Torvalds 26201da177e4SLinus Torvalds blocksize = 1 << inode->i_blkbits; 26211da177e4SLinus Torvalds length = offset & (blocksize - 1); 26221da177e4SLinus Torvalds 26231da177e4SLinus Torvalds /* Block boundary? Nothing to do */ 26241da177e4SLinus Torvalds if (!length) 26251da177e4SLinus Torvalds return 0; 26261da177e4SLinus Torvalds 26271da177e4SLinus Torvalds length = blocksize - length; 26281da177e4SLinus Torvalds iblock = index << (PAGE_CACHE_SHIFT - inode->i_blkbits); 26291da177e4SLinus Torvalds 26301da177e4SLinus Torvalds page = grab_cache_page(mapping, index); 26311da177e4SLinus Torvalds err = -ENOMEM; 26321da177e4SLinus Torvalds if (!page) 26331da177e4SLinus Torvalds goto out; 26341da177e4SLinus Torvalds 26351da177e4SLinus Torvalds if (!page_has_buffers(page)) 26361da177e4SLinus Torvalds create_empty_buffers(page, blocksize, 0); 26371da177e4SLinus Torvalds 26381da177e4SLinus Torvalds /* Find the buffer that contains "offset" */ 26391da177e4SLinus Torvalds bh = page_buffers(page); 26401da177e4SLinus Torvalds pos = blocksize; 26411da177e4SLinus Torvalds while (offset >= pos) { 26421da177e4SLinus Torvalds bh = bh->b_this_page; 26431da177e4SLinus Torvalds iblock++; 26441da177e4SLinus Torvalds pos += blocksize; 26451da177e4SLinus Torvalds } 26461da177e4SLinus Torvalds 26471da177e4SLinus Torvalds err = 0; 26481da177e4SLinus Torvalds if (!buffer_mapped(bh)) { 26491da177e4SLinus Torvalds err = get_block(inode, iblock, bh, 0); 26501da177e4SLinus Torvalds if (err) 26511da177e4SLinus Torvalds goto unlock; 26521da177e4SLinus Torvalds /* unmapped? It's a hole - nothing to do */ 26531da177e4SLinus Torvalds if (!buffer_mapped(bh)) 26541da177e4SLinus Torvalds goto unlock; 26551da177e4SLinus Torvalds } 26561da177e4SLinus Torvalds 26571da177e4SLinus Torvalds /* Ok, it's mapped. Make sure it's up-to-date */ 26581da177e4SLinus Torvalds if (PageUptodate(page)) 26591da177e4SLinus Torvalds set_buffer_uptodate(bh); 26601da177e4SLinus Torvalds 26611da177e4SLinus Torvalds if (!buffer_uptodate(bh) && !buffer_delay(bh)) { 26621da177e4SLinus Torvalds err = -EIO; 26631da177e4SLinus Torvalds ll_rw_block(READ, 1, &bh); 26641da177e4SLinus Torvalds wait_on_buffer(bh); 26651da177e4SLinus Torvalds /* Uhhuh. Read error. Complain and punt. */ 26661da177e4SLinus Torvalds if (!buffer_uptodate(bh)) 26671da177e4SLinus Torvalds goto unlock; 26681da177e4SLinus Torvalds } 26691da177e4SLinus Torvalds 26701da177e4SLinus Torvalds kaddr = kmap_atomic(page, KM_USER0); 26711da177e4SLinus Torvalds memset(kaddr + offset, 0, length); 26721da177e4SLinus Torvalds flush_dcache_page(page); 26731da177e4SLinus Torvalds kunmap_atomic(kaddr, KM_USER0); 26741da177e4SLinus Torvalds 26751da177e4SLinus Torvalds mark_buffer_dirty(bh); 26761da177e4SLinus Torvalds err = 0; 26771da177e4SLinus Torvalds 26781da177e4SLinus Torvalds unlock: 26791da177e4SLinus Torvalds unlock_page(page); 26801da177e4SLinus Torvalds page_cache_release(page); 26811da177e4SLinus Torvalds out: 26821da177e4SLinus Torvalds return err; 26831da177e4SLinus Torvalds } 26841da177e4SLinus Torvalds 26851da177e4SLinus Torvalds /* 26861da177e4SLinus Torvalds * The generic ->writepage function for buffer-backed address_spaces 26871da177e4SLinus Torvalds */ 26881da177e4SLinus Torvalds int block_write_full_page(struct page *page, get_block_t *get_block, 26891da177e4SLinus Torvalds struct writeback_control *wbc) 26901da177e4SLinus Torvalds { 26911da177e4SLinus Torvalds struct inode * const inode = page->mapping->host; 26921da177e4SLinus Torvalds loff_t i_size = i_size_read(inode); 26931da177e4SLinus Torvalds const pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; 26941da177e4SLinus Torvalds unsigned offset; 26951da177e4SLinus Torvalds void *kaddr; 26961da177e4SLinus Torvalds 26971da177e4SLinus Torvalds /* Is the page fully inside i_size? */ 26981da177e4SLinus Torvalds if (page->index < end_index) 26991da177e4SLinus Torvalds return __block_write_full_page(inode, page, get_block, wbc); 27001da177e4SLinus Torvalds 27011da177e4SLinus Torvalds /* Is the page fully outside i_size? (truncate in progress) */ 27021da177e4SLinus Torvalds offset = i_size & (PAGE_CACHE_SIZE-1); 27031da177e4SLinus Torvalds if (page->index >= end_index+1 || !offset) { 27041da177e4SLinus Torvalds /* 27051da177e4SLinus Torvalds * The page may have dirty, unmapped buffers. For example, 27061da177e4SLinus Torvalds * they may have been added in ext3_writepage(). Make them 27071da177e4SLinus Torvalds * freeable here, so the page does not leak. 27081da177e4SLinus Torvalds */ 27091da177e4SLinus Torvalds block_invalidatepage(page, 0); 27101da177e4SLinus Torvalds unlock_page(page); 27111da177e4SLinus Torvalds return 0; /* don't care */ 27121da177e4SLinus Torvalds } 27131da177e4SLinus Torvalds 27141da177e4SLinus Torvalds /* 27151da177e4SLinus Torvalds * The page straddles i_size. It must be zeroed out on each and every 27161da177e4SLinus Torvalds * writepage invokation because it may be mmapped. "A file is mapped 27171da177e4SLinus Torvalds * in multiples of the page size. For a file that is not a multiple of 27181da177e4SLinus Torvalds * the page size, the remaining memory is zeroed when mapped, and 27191da177e4SLinus Torvalds * writes to that region are not written out to the file." 27201da177e4SLinus Torvalds */ 27211da177e4SLinus Torvalds kaddr = kmap_atomic(page, KM_USER0); 27221da177e4SLinus Torvalds memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset); 27231da177e4SLinus Torvalds flush_dcache_page(page); 27241da177e4SLinus Torvalds kunmap_atomic(kaddr, KM_USER0); 27251da177e4SLinus Torvalds return __block_write_full_page(inode, page, get_block, wbc); 27261da177e4SLinus Torvalds } 27271da177e4SLinus Torvalds 27281da177e4SLinus Torvalds sector_t generic_block_bmap(struct address_space *mapping, sector_t block, 27291da177e4SLinus Torvalds get_block_t *get_block) 27301da177e4SLinus Torvalds { 27311da177e4SLinus Torvalds struct buffer_head tmp; 27321da177e4SLinus Torvalds struct inode *inode = mapping->host; 27331da177e4SLinus Torvalds tmp.b_state = 0; 27341da177e4SLinus Torvalds tmp.b_blocknr = 0; 27351da177e4SLinus Torvalds get_block(inode, block, &tmp, 0); 27361da177e4SLinus Torvalds return tmp.b_blocknr; 27371da177e4SLinus Torvalds } 27381da177e4SLinus Torvalds 27391da177e4SLinus Torvalds static int end_bio_bh_io_sync(struct bio *bio, unsigned int bytes_done, int err) 27401da177e4SLinus Torvalds { 27411da177e4SLinus Torvalds struct buffer_head *bh = bio->bi_private; 27421da177e4SLinus Torvalds 27431da177e4SLinus Torvalds if (bio->bi_size) 27441da177e4SLinus Torvalds return 1; 27451da177e4SLinus Torvalds 27461da177e4SLinus Torvalds if (err == -EOPNOTSUPP) { 27471da177e4SLinus Torvalds set_bit(BIO_EOPNOTSUPP, &bio->bi_flags); 27481da177e4SLinus Torvalds set_bit(BH_Eopnotsupp, &bh->b_state); 27491da177e4SLinus Torvalds } 27501da177e4SLinus Torvalds 27511da177e4SLinus Torvalds bh->b_end_io(bh, test_bit(BIO_UPTODATE, &bio->bi_flags)); 27521da177e4SLinus Torvalds bio_put(bio); 27531da177e4SLinus Torvalds return 0; 27541da177e4SLinus Torvalds } 27551da177e4SLinus Torvalds 27561da177e4SLinus Torvalds int submit_bh(int rw, struct buffer_head * bh) 27571da177e4SLinus Torvalds { 27581da177e4SLinus Torvalds struct bio *bio; 27591da177e4SLinus Torvalds int ret = 0; 27601da177e4SLinus Torvalds 27611da177e4SLinus Torvalds BUG_ON(!buffer_locked(bh)); 27621da177e4SLinus Torvalds BUG_ON(!buffer_mapped(bh)); 27631da177e4SLinus Torvalds BUG_ON(!bh->b_end_io); 27641da177e4SLinus Torvalds 27651da177e4SLinus Torvalds if (buffer_ordered(bh) && (rw == WRITE)) 27661da177e4SLinus Torvalds rw = WRITE_BARRIER; 27671da177e4SLinus Torvalds 27681da177e4SLinus Torvalds /* 27691da177e4SLinus Torvalds * Only clear out a write error when rewriting, should this 27701da177e4SLinus Torvalds * include WRITE_SYNC as well? 27711da177e4SLinus Torvalds */ 27721da177e4SLinus Torvalds if (test_set_buffer_req(bh) && (rw == WRITE || rw == WRITE_BARRIER)) 27731da177e4SLinus Torvalds clear_buffer_write_io_error(bh); 27741da177e4SLinus Torvalds 27751da177e4SLinus Torvalds /* 27761da177e4SLinus Torvalds * from here on down, it's all bio -- do the initial mapping, 27771da177e4SLinus Torvalds * submit_bio -> generic_make_request may further map this bio around 27781da177e4SLinus Torvalds */ 27791da177e4SLinus Torvalds bio = bio_alloc(GFP_NOIO, 1); 27801da177e4SLinus Torvalds 27811da177e4SLinus Torvalds bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9); 27821da177e4SLinus Torvalds bio->bi_bdev = bh->b_bdev; 27831da177e4SLinus Torvalds bio->bi_io_vec[0].bv_page = bh->b_page; 27841da177e4SLinus Torvalds bio->bi_io_vec[0].bv_len = bh->b_size; 27851da177e4SLinus Torvalds bio->bi_io_vec[0].bv_offset = bh_offset(bh); 27861da177e4SLinus Torvalds 27871da177e4SLinus Torvalds bio->bi_vcnt = 1; 27881da177e4SLinus Torvalds bio->bi_idx = 0; 27891da177e4SLinus Torvalds bio->bi_size = bh->b_size; 27901da177e4SLinus Torvalds 27911da177e4SLinus Torvalds bio->bi_end_io = end_bio_bh_io_sync; 27921da177e4SLinus Torvalds bio->bi_private = bh; 27931da177e4SLinus Torvalds 27941da177e4SLinus Torvalds bio_get(bio); 27951da177e4SLinus Torvalds submit_bio(rw, bio); 27961da177e4SLinus Torvalds 27971da177e4SLinus Torvalds if (bio_flagged(bio, BIO_EOPNOTSUPP)) 27981da177e4SLinus Torvalds ret = -EOPNOTSUPP; 27991da177e4SLinus Torvalds 28001da177e4SLinus Torvalds bio_put(bio); 28011da177e4SLinus Torvalds return ret; 28021da177e4SLinus Torvalds } 28031da177e4SLinus Torvalds 28041da177e4SLinus Torvalds /** 28051da177e4SLinus Torvalds * ll_rw_block: low-level access to block devices (DEPRECATED) 28061da177e4SLinus Torvalds * @rw: whether to %READ or %WRITE or maybe %READA (readahead) 28071da177e4SLinus Torvalds * @nr: number of &struct buffer_heads in the array 28081da177e4SLinus Torvalds * @bhs: array of pointers to &struct buffer_head 28091da177e4SLinus Torvalds * 28101da177e4SLinus Torvalds * ll_rw_block() takes an array of pointers to &struct buffer_heads, 28111da177e4SLinus Torvalds * and requests an I/O operation on them, either a %READ or a %WRITE. 28121da177e4SLinus Torvalds * The third %READA option is described in the documentation for 28131da177e4SLinus Torvalds * generic_make_request() which ll_rw_block() calls. 28141da177e4SLinus Torvalds * 28151da177e4SLinus Torvalds * This function drops any buffer that it cannot get a lock on (with the 28161da177e4SLinus Torvalds * BH_Lock state bit), any buffer that appears to be clean when doing a 28171da177e4SLinus Torvalds * write request, and any buffer that appears to be up-to-date when doing 28181da177e4SLinus Torvalds * read request. Further it marks as clean buffers that are processed for 28191da177e4SLinus Torvalds * writing (the buffer cache won't assume that they are actually clean until 28201da177e4SLinus Torvalds * the buffer gets unlocked). 28211da177e4SLinus Torvalds * 28221da177e4SLinus Torvalds * ll_rw_block sets b_end_io to simple completion handler that marks 28231da177e4SLinus Torvalds * the buffer up-to-date (if approriate), unlocks the buffer and wakes 28241da177e4SLinus Torvalds * any waiters. 28251da177e4SLinus Torvalds * 28261da177e4SLinus Torvalds * All of the buffers must be for the same device, and must also be a 28271da177e4SLinus Torvalds * multiple of the current approved size for the device. 28281da177e4SLinus Torvalds */ 28291da177e4SLinus Torvalds void ll_rw_block(int rw, int nr, struct buffer_head *bhs[]) 28301da177e4SLinus Torvalds { 28311da177e4SLinus Torvalds int i; 28321da177e4SLinus Torvalds 28331da177e4SLinus Torvalds for (i = 0; i < nr; i++) { 28341da177e4SLinus Torvalds struct buffer_head *bh = bhs[i]; 28351da177e4SLinus Torvalds 28361da177e4SLinus Torvalds if (test_set_buffer_locked(bh)) 28371da177e4SLinus Torvalds continue; 28381da177e4SLinus Torvalds 28391da177e4SLinus Torvalds get_bh(bh); 28401da177e4SLinus Torvalds if (rw == WRITE) { 28411da177e4SLinus Torvalds if (test_clear_buffer_dirty(bh)) { 284276c3073aSakpm@osdl.org bh->b_end_io = end_buffer_write_sync; 28431da177e4SLinus Torvalds submit_bh(WRITE, bh); 28441da177e4SLinus Torvalds continue; 28451da177e4SLinus Torvalds } 28461da177e4SLinus Torvalds } else { 28471da177e4SLinus Torvalds if (!buffer_uptodate(bh)) { 284876c3073aSakpm@osdl.org bh->b_end_io = end_buffer_read_sync; 28491da177e4SLinus Torvalds submit_bh(rw, bh); 28501da177e4SLinus Torvalds continue; 28511da177e4SLinus Torvalds } 28521da177e4SLinus Torvalds } 28531da177e4SLinus Torvalds unlock_buffer(bh); 28541da177e4SLinus Torvalds put_bh(bh); 28551da177e4SLinus Torvalds } 28561da177e4SLinus Torvalds } 28571da177e4SLinus Torvalds 28581da177e4SLinus Torvalds /* 28591da177e4SLinus Torvalds * For a data-integrity writeout, we need to wait upon any in-progress I/O 28601da177e4SLinus Torvalds * and then start new I/O and then wait upon it. The caller must have a ref on 28611da177e4SLinus Torvalds * the buffer_head. 28621da177e4SLinus Torvalds */ 28631da177e4SLinus Torvalds int sync_dirty_buffer(struct buffer_head *bh) 28641da177e4SLinus Torvalds { 28651da177e4SLinus Torvalds int ret = 0; 28661da177e4SLinus Torvalds 28671da177e4SLinus Torvalds WARN_ON(atomic_read(&bh->b_count) < 1); 28681da177e4SLinus Torvalds lock_buffer(bh); 28691da177e4SLinus Torvalds if (test_clear_buffer_dirty(bh)) { 28701da177e4SLinus Torvalds get_bh(bh); 28711da177e4SLinus Torvalds bh->b_end_io = end_buffer_write_sync; 28721da177e4SLinus Torvalds ret = submit_bh(WRITE, bh); 28731da177e4SLinus Torvalds wait_on_buffer(bh); 28741da177e4SLinus Torvalds if (buffer_eopnotsupp(bh)) { 28751da177e4SLinus Torvalds clear_buffer_eopnotsupp(bh); 28761da177e4SLinus Torvalds ret = -EOPNOTSUPP; 28771da177e4SLinus Torvalds } 28781da177e4SLinus Torvalds if (!ret && !buffer_uptodate(bh)) 28791da177e4SLinus Torvalds ret = -EIO; 28801da177e4SLinus Torvalds } else { 28811da177e4SLinus Torvalds unlock_buffer(bh); 28821da177e4SLinus Torvalds } 28831da177e4SLinus Torvalds return ret; 28841da177e4SLinus Torvalds } 28851da177e4SLinus Torvalds 28861da177e4SLinus Torvalds /* 28871da177e4SLinus Torvalds * try_to_free_buffers() checks if all the buffers on this particular page 28881da177e4SLinus Torvalds * are unused, and releases them if so. 28891da177e4SLinus Torvalds * 28901da177e4SLinus Torvalds * Exclusion against try_to_free_buffers may be obtained by either 28911da177e4SLinus Torvalds * locking the page or by holding its mapping's private_lock. 28921da177e4SLinus Torvalds * 28931da177e4SLinus Torvalds * If the page is dirty but all the buffers are clean then we need to 28941da177e4SLinus Torvalds * be sure to mark the page clean as well. This is because the page 28951da177e4SLinus Torvalds * may be against a block device, and a later reattachment of buffers 28961da177e4SLinus Torvalds * to a dirty page will set *all* buffers dirty. Which would corrupt 28971da177e4SLinus Torvalds * filesystem data on the same device. 28981da177e4SLinus Torvalds * 28991da177e4SLinus Torvalds * The same applies to regular filesystem pages: if all the buffers are 29001da177e4SLinus Torvalds * clean then we set the page clean and proceed. To do that, we require 29011da177e4SLinus Torvalds * total exclusion from __set_page_dirty_buffers(). That is obtained with 29021da177e4SLinus Torvalds * private_lock. 29031da177e4SLinus Torvalds * 29041da177e4SLinus Torvalds * try_to_free_buffers() is non-blocking. 29051da177e4SLinus Torvalds */ 29061da177e4SLinus Torvalds static inline int buffer_busy(struct buffer_head *bh) 29071da177e4SLinus Torvalds { 29081da177e4SLinus Torvalds return atomic_read(&bh->b_count) | 29091da177e4SLinus Torvalds (bh->b_state & ((1 << BH_Dirty) | (1 << BH_Lock))); 29101da177e4SLinus Torvalds } 29111da177e4SLinus Torvalds 29121da177e4SLinus Torvalds static int 29131da177e4SLinus Torvalds drop_buffers(struct page *page, struct buffer_head **buffers_to_free) 29141da177e4SLinus Torvalds { 29151da177e4SLinus Torvalds struct buffer_head *head = page_buffers(page); 29161da177e4SLinus Torvalds struct buffer_head *bh; 29171da177e4SLinus Torvalds 29181da177e4SLinus Torvalds bh = head; 29191da177e4SLinus Torvalds do { 2920de7d5a3bSakpm@osdl.org if (buffer_write_io_error(bh) && page->mapping) 29211da177e4SLinus Torvalds set_bit(AS_EIO, &page->mapping->flags); 29221da177e4SLinus Torvalds if (buffer_busy(bh)) 29231da177e4SLinus Torvalds goto failed; 29241da177e4SLinus Torvalds bh = bh->b_this_page; 29251da177e4SLinus Torvalds } while (bh != head); 29261da177e4SLinus Torvalds 29271da177e4SLinus Torvalds do { 29281da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 29291da177e4SLinus Torvalds 29301da177e4SLinus Torvalds if (!list_empty(&bh->b_assoc_buffers)) 29311da177e4SLinus Torvalds __remove_assoc_queue(bh); 29321da177e4SLinus Torvalds bh = next; 29331da177e4SLinus Torvalds } while (bh != head); 29341da177e4SLinus Torvalds *buffers_to_free = head; 29351da177e4SLinus Torvalds __clear_page_buffers(page); 29361da177e4SLinus Torvalds return 1; 29371da177e4SLinus Torvalds failed: 29381da177e4SLinus Torvalds return 0; 29391da177e4SLinus Torvalds } 29401da177e4SLinus Torvalds 29411da177e4SLinus Torvalds int try_to_free_buffers(struct page *page) 29421da177e4SLinus Torvalds { 29431da177e4SLinus Torvalds struct address_space * const mapping = page->mapping; 29441da177e4SLinus Torvalds struct buffer_head *buffers_to_free = NULL; 29451da177e4SLinus Torvalds int ret = 0; 29461da177e4SLinus Torvalds 29471da177e4SLinus Torvalds BUG_ON(!PageLocked(page)); 29481da177e4SLinus Torvalds if (PageWriteback(page)) 29491da177e4SLinus Torvalds return 0; 29501da177e4SLinus Torvalds 29511da177e4SLinus Torvalds if (mapping == NULL) { /* can this still happen? */ 29521da177e4SLinus Torvalds ret = drop_buffers(page, &buffers_to_free); 29531da177e4SLinus Torvalds goto out; 29541da177e4SLinus Torvalds } 29551da177e4SLinus Torvalds 29561da177e4SLinus Torvalds spin_lock(&mapping->private_lock); 29571da177e4SLinus Torvalds ret = drop_buffers(page, &buffers_to_free); 29581da177e4SLinus Torvalds if (ret) { 29591da177e4SLinus Torvalds /* 29601da177e4SLinus Torvalds * If the filesystem writes its buffers by hand (eg ext3) 29611da177e4SLinus Torvalds * then we can have clean buffers against a dirty page. We 29621da177e4SLinus Torvalds * clean the page here; otherwise later reattachment of buffers 29631da177e4SLinus Torvalds * could encounter a non-uptodate page, which is unresolvable. 29641da177e4SLinus Torvalds * This only applies in the rare case where try_to_free_buffers 29651da177e4SLinus Torvalds * succeeds but the page is not freed. 29661da177e4SLinus Torvalds */ 29671da177e4SLinus Torvalds clear_page_dirty(page); 29681da177e4SLinus Torvalds } 29691da177e4SLinus Torvalds spin_unlock(&mapping->private_lock); 29701da177e4SLinus Torvalds out: 29711da177e4SLinus Torvalds if (buffers_to_free) { 29721da177e4SLinus Torvalds struct buffer_head *bh = buffers_to_free; 29731da177e4SLinus Torvalds 29741da177e4SLinus Torvalds do { 29751da177e4SLinus Torvalds struct buffer_head *next = bh->b_this_page; 29761da177e4SLinus Torvalds free_buffer_head(bh); 29771da177e4SLinus Torvalds bh = next; 29781da177e4SLinus Torvalds } while (bh != buffers_to_free); 29791da177e4SLinus Torvalds } 29801da177e4SLinus Torvalds return ret; 29811da177e4SLinus Torvalds } 29821da177e4SLinus Torvalds EXPORT_SYMBOL(try_to_free_buffers); 29831da177e4SLinus Torvalds 29841da177e4SLinus Torvalds int block_sync_page(struct page *page) 29851da177e4SLinus Torvalds { 29861da177e4SLinus Torvalds struct address_space *mapping; 29871da177e4SLinus Torvalds 29881da177e4SLinus Torvalds smp_mb(); 29891da177e4SLinus Torvalds mapping = page_mapping(page); 29901da177e4SLinus Torvalds if (mapping) 29911da177e4SLinus Torvalds blk_run_backing_dev(mapping->backing_dev_info, page); 29921da177e4SLinus Torvalds return 0; 29931da177e4SLinus Torvalds } 29941da177e4SLinus Torvalds 29951da177e4SLinus Torvalds /* 29961da177e4SLinus Torvalds * There are no bdflush tunables left. But distributions are 29971da177e4SLinus Torvalds * still running obsolete flush daemons, so we terminate them here. 29981da177e4SLinus Torvalds * 29991da177e4SLinus Torvalds * Use of bdflush() is deprecated and will be removed in a future kernel. 30001da177e4SLinus Torvalds * The `pdflush' kernel threads fully replace bdflush daemons and this call. 30011da177e4SLinus Torvalds */ 30021da177e4SLinus Torvalds asmlinkage long sys_bdflush(int func, long data) 30031da177e4SLinus Torvalds { 30041da177e4SLinus Torvalds static int msg_count; 30051da177e4SLinus Torvalds 30061da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 30071da177e4SLinus Torvalds return -EPERM; 30081da177e4SLinus Torvalds 30091da177e4SLinus Torvalds if (msg_count < 5) { 30101da177e4SLinus Torvalds msg_count++; 30111da177e4SLinus Torvalds printk(KERN_INFO 30121da177e4SLinus Torvalds "warning: process `%s' used the obsolete bdflush" 30131da177e4SLinus Torvalds " system call\n", current->comm); 30141da177e4SLinus Torvalds printk(KERN_INFO "Fix your initscripts?\n"); 30151da177e4SLinus Torvalds } 30161da177e4SLinus Torvalds 30171da177e4SLinus Torvalds if (func == 1) 30181da177e4SLinus Torvalds do_exit(0); 30191da177e4SLinus Torvalds return 0; 30201da177e4SLinus Torvalds } 30211da177e4SLinus Torvalds 30221da177e4SLinus Torvalds /* 30231da177e4SLinus Torvalds * Buffer-head allocation 30241da177e4SLinus Torvalds */ 30251da177e4SLinus Torvalds static kmem_cache_t *bh_cachep; 30261da177e4SLinus Torvalds 30271da177e4SLinus Torvalds /* 30281da177e4SLinus Torvalds * Once the number of bh's in the machine exceeds this level, we start 30291da177e4SLinus Torvalds * stripping them in writeback. 30301da177e4SLinus Torvalds */ 30311da177e4SLinus Torvalds static int max_buffer_heads; 30321da177e4SLinus Torvalds 30331da177e4SLinus Torvalds int buffer_heads_over_limit; 30341da177e4SLinus Torvalds 30351da177e4SLinus Torvalds struct bh_accounting { 30361da177e4SLinus Torvalds int nr; /* Number of live bh's */ 30371da177e4SLinus Torvalds int ratelimit; /* Limit cacheline bouncing */ 30381da177e4SLinus Torvalds }; 30391da177e4SLinus Torvalds 30401da177e4SLinus Torvalds static DEFINE_PER_CPU(struct bh_accounting, bh_accounting) = {0, 0}; 30411da177e4SLinus Torvalds 30421da177e4SLinus Torvalds static void recalc_bh_state(void) 30431da177e4SLinus Torvalds { 30441da177e4SLinus Torvalds int i; 30451da177e4SLinus Torvalds int tot = 0; 30461da177e4SLinus Torvalds 30471da177e4SLinus Torvalds if (__get_cpu_var(bh_accounting).ratelimit++ < 4096) 30481da177e4SLinus Torvalds return; 30491da177e4SLinus Torvalds __get_cpu_var(bh_accounting).ratelimit = 0; 30501da177e4SLinus Torvalds for_each_cpu(i) 30511da177e4SLinus Torvalds tot += per_cpu(bh_accounting, i).nr; 30521da177e4SLinus Torvalds buffer_heads_over_limit = (tot > max_buffer_heads); 30531da177e4SLinus Torvalds } 30541da177e4SLinus Torvalds 30551da177e4SLinus Torvalds struct buffer_head *alloc_buffer_head(unsigned int __nocast gfp_flags) 30561da177e4SLinus Torvalds { 30571da177e4SLinus Torvalds struct buffer_head *ret = kmem_cache_alloc(bh_cachep, gfp_flags); 30581da177e4SLinus Torvalds if (ret) { 30591da177e4SLinus Torvalds preempt_disable(); 30601da177e4SLinus Torvalds __get_cpu_var(bh_accounting).nr++; 30611da177e4SLinus Torvalds recalc_bh_state(); 30621da177e4SLinus Torvalds preempt_enable(); 30631da177e4SLinus Torvalds } 30641da177e4SLinus Torvalds return ret; 30651da177e4SLinus Torvalds } 30661da177e4SLinus Torvalds EXPORT_SYMBOL(alloc_buffer_head); 30671da177e4SLinus Torvalds 30681da177e4SLinus Torvalds void free_buffer_head(struct buffer_head *bh) 30691da177e4SLinus Torvalds { 30701da177e4SLinus Torvalds BUG_ON(!list_empty(&bh->b_assoc_buffers)); 30711da177e4SLinus Torvalds kmem_cache_free(bh_cachep, bh); 30721da177e4SLinus Torvalds preempt_disable(); 30731da177e4SLinus Torvalds __get_cpu_var(bh_accounting).nr--; 30741da177e4SLinus Torvalds recalc_bh_state(); 30751da177e4SLinus Torvalds preempt_enable(); 30761da177e4SLinus Torvalds } 30771da177e4SLinus Torvalds EXPORT_SYMBOL(free_buffer_head); 30781da177e4SLinus Torvalds 30791da177e4SLinus Torvalds static void 30801da177e4SLinus Torvalds init_buffer_head(void *data, kmem_cache_t *cachep, unsigned long flags) 30811da177e4SLinus Torvalds { 30821da177e4SLinus Torvalds if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) == 30831da177e4SLinus Torvalds SLAB_CTOR_CONSTRUCTOR) { 30841da177e4SLinus Torvalds struct buffer_head * bh = (struct buffer_head *)data; 30851da177e4SLinus Torvalds 30861da177e4SLinus Torvalds memset(bh, 0, sizeof(*bh)); 30871da177e4SLinus Torvalds INIT_LIST_HEAD(&bh->b_assoc_buffers); 30881da177e4SLinus Torvalds } 30891da177e4SLinus Torvalds } 30901da177e4SLinus Torvalds 30911da177e4SLinus Torvalds #ifdef CONFIG_HOTPLUG_CPU 30921da177e4SLinus Torvalds static void buffer_exit_cpu(int cpu) 30931da177e4SLinus Torvalds { 30941da177e4SLinus Torvalds int i; 30951da177e4SLinus Torvalds struct bh_lru *b = &per_cpu(bh_lrus, cpu); 30961da177e4SLinus Torvalds 30971da177e4SLinus Torvalds for (i = 0; i < BH_LRU_SIZE; i++) { 30981da177e4SLinus Torvalds brelse(b->bhs[i]); 30991da177e4SLinus Torvalds b->bhs[i] = NULL; 31001da177e4SLinus Torvalds } 31011da177e4SLinus Torvalds } 31021da177e4SLinus Torvalds 31031da177e4SLinus Torvalds static int buffer_cpu_notify(struct notifier_block *self, 31041da177e4SLinus Torvalds unsigned long action, void *hcpu) 31051da177e4SLinus Torvalds { 31061da177e4SLinus Torvalds if (action == CPU_DEAD) 31071da177e4SLinus Torvalds buffer_exit_cpu((unsigned long)hcpu); 31081da177e4SLinus Torvalds return NOTIFY_OK; 31091da177e4SLinus Torvalds } 31101da177e4SLinus Torvalds #endif /* CONFIG_HOTPLUG_CPU */ 31111da177e4SLinus Torvalds 31121da177e4SLinus Torvalds void __init buffer_init(void) 31131da177e4SLinus Torvalds { 31141da177e4SLinus Torvalds int nrpages; 31151da177e4SLinus Torvalds 31161da177e4SLinus Torvalds bh_cachep = kmem_cache_create("buffer_head", 31171da177e4SLinus Torvalds sizeof(struct buffer_head), 0, 31181da177e4SLinus Torvalds SLAB_PANIC, init_buffer_head, NULL); 31191da177e4SLinus Torvalds 31201da177e4SLinus Torvalds /* 31211da177e4SLinus Torvalds * Limit the bh occupancy to 10% of ZONE_NORMAL 31221da177e4SLinus Torvalds */ 31231da177e4SLinus Torvalds nrpages = (nr_free_buffer_pages() * 10) / 100; 31241da177e4SLinus Torvalds max_buffer_heads = nrpages * (PAGE_SIZE / sizeof(struct buffer_head)); 31251da177e4SLinus Torvalds hotcpu_notifier(buffer_cpu_notify, 0); 31261da177e4SLinus Torvalds } 31271da177e4SLinus Torvalds 31281da177e4SLinus Torvalds EXPORT_SYMBOL(__bforget); 31291da177e4SLinus Torvalds EXPORT_SYMBOL(__brelse); 31301da177e4SLinus Torvalds EXPORT_SYMBOL(__wait_on_buffer); 31311da177e4SLinus Torvalds EXPORT_SYMBOL(block_commit_write); 31321da177e4SLinus Torvalds EXPORT_SYMBOL(block_prepare_write); 31331da177e4SLinus Torvalds EXPORT_SYMBOL(block_read_full_page); 31341da177e4SLinus Torvalds EXPORT_SYMBOL(block_sync_page); 31351da177e4SLinus Torvalds EXPORT_SYMBOL(block_truncate_page); 31361da177e4SLinus Torvalds EXPORT_SYMBOL(block_write_full_page); 31371da177e4SLinus Torvalds EXPORT_SYMBOL(cont_prepare_write); 31381da177e4SLinus Torvalds EXPORT_SYMBOL(end_buffer_async_write); 31391da177e4SLinus Torvalds EXPORT_SYMBOL(end_buffer_read_sync); 31401da177e4SLinus Torvalds EXPORT_SYMBOL(end_buffer_write_sync); 31411da177e4SLinus Torvalds EXPORT_SYMBOL(file_fsync); 31421da177e4SLinus Torvalds EXPORT_SYMBOL(fsync_bdev); 31431da177e4SLinus Torvalds EXPORT_SYMBOL(generic_block_bmap); 31441da177e4SLinus Torvalds EXPORT_SYMBOL(generic_commit_write); 31451da177e4SLinus Torvalds EXPORT_SYMBOL(generic_cont_expand); 31461da177e4SLinus Torvalds EXPORT_SYMBOL(init_buffer); 31471da177e4SLinus Torvalds EXPORT_SYMBOL(invalidate_bdev); 31481da177e4SLinus Torvalds EXPORT_SYMBOL(ll_rw_block); 31491da177e4SLinus Torvalds EXPORT_SYMBOL(mark_buffer_dirty); 31501da177e4SLinus Torvalds EXPORT_SYMBOL(submit_bh); 31511da177e4SLinus Torvalds EXPORT_SYMBOL(sync_dirty_buffer); 31521da177e4SLinus Torvalds EXPORT_SYMBOL(unlock_buffer); 3153