1b2441318SGreg Kroah-Hartman // SPDX-License-Identifier: GPL-2.0 2ac27a0ecSDave Kleikamp /* 3617ba13bSMingming Cao * linux/fs/ext4/ialloc.c 4ac27a0ecSDave Kleikamp * 5ac27a0ecSDave Kleikamp * Copyright (C) 1992, 1993, 1994, 1995 6ac27a0ecSDave Kleikamp * Remy Card (card@masi.ibp.fr) 7ac27a0ecSDave Kleikamp * Laboratoire MASI - Institut Blaise Pascal 8ac27a0ecSDave Kleikamp * Universite Pierre et Marie Curie (Paris VI) 9ac27a0ecSDave Kleikamp * 10ac27a0ecSDave Kleikamp * BSD ufs-inspired inode and directory allocation by 11ac27a0ecSDave Kleikamp * Stephen Tweedie (sct@redhat.com), 1993 12ac27a0ecSDave Kleikamp * Big-endian to little-endian byte-swapping/bitmaps by 13ac27a0ecSDave Kleikamp * David S. Miller (davem@caip.rutgers.edu), 1995 14ac27a0ecSDave Kleikamp */ 15ac27a0ecSDave Kleikamp 16ac27a0ecSDave Kleikamp #include <linux/time.h> 17ac27a0ecSDave Kleikamp #include <linux/fs.h> 18ac27a0ecSDave Kleikamp #include <linux/stat.h> 19ac27a0ecSDave Kleikamp #include <linux/string.h> 20ac27a0ecSDave Kleikamp #include <linux/quotaops.h> 21ac27a0ecSDave Kleikamp #include <linux/buffer_head.h> 22ac27a0ecSDave Kleikamp #include <linux/random.h> 23ac27a0ecSDave Kleikamp #include <linux/bitops.h> 243a5b2ecdSMingming Cao #include <linux/blkdev.h> 255b825c3aSIngo Molnar #include <linux/cred.h> 265b825c3aSIngo Molnar 27ac27a0ecSDave Kleikamp #include <asm/byteorder.h> 289bffad1eSTheodore Ts'o 293dcf5451SChristoph Hellwig #include "ext4.h" 303dcf5451SChristoph Hellwig #include "ext4_jbd2.h" 31ac27a0ecSDave Kleikamp #include "xattr.h" 32ac27a0ecSDave Kleikamp #include "acl.h" 33ac27a0ecSDave Kleikamp 349bffad1eSTheodore Ts'o #include <trace/events/ext4.h> 359bffad1eSTheodore Ts'o 36ac27a0ecSDave Kleikamp /* 37ac27a0ecSDave Kleikamp * ialloc.c contains the inodes allocation and deallocation routines 38ac27a0ecSDave Kleikamp */ 39ac27a0ecSDave Kleikamp 40ac27a0ecSDave Kleikamp /* 41ac27a0ecSDave Kleikamp * The free inodes are managed by bitmaps. A file system contains several 42ac27a0ecSDave Kleikamp * blocks groups. Each group contains 1 bitmap block for blocks, 1 bitmap 43ac27a0ecSDave Kleikamp * block for inodes, N blocks for the inode table and data blocks. 44ac27a0ecSDave Kleikamp * 45ac27a0ecSDave Kleikamp * The file system contains group descriptors which are located after the 46ac27a0ecSDave Kleikamp * super block. Each descriptor contains the number of the bitmap block and 47ac27a0ecSDave Kleikamp * the free blocks count in the block. 48ac27a0ecSDave Kleikamp */ 49ac27a0ecSDave Kleikamp 50717d50e4SAndreas Dilger /* 51717d50e4SAndreas Dilger * To avoid calling the atomic setbit hundreds or thousands of times, we only 52717d50e4SAndreas Dilger * need to use it within a single byte (to ensure we get endianness right). 53717d50e4SAndreas Dilger * We can use memset for the rest of the bitmap as there are no other users. 54717d50e4SAndreas Dilger */ 5561d08673STheodore Ts'o void ext4_mark_bitmap_end(int start_bit, int end_bit, char *bitmap) 56717d50e4SAndreas Dilger { 57717d50e4SAndreas Dilger int i; 58717d50e4SAndreas Dilger 59717d50e4SAndreas Dilger if (start_bit >= end_bit) 60717d50e4SAndreas Dilger return; 61717d50e4SAndreas Dilger 62717d50e4SAndreas Dilger ext4_debug("mark end bits +%d through +%d used\n", start_bit, end_bit); 63717d50e4SAndreas Dilger for (i = start_bit; i < ((start_bit + 7) & ~7UL); i++) 64717d50e4SAndreas Dilger ext4_set_bit(i, bitmap); 65717d50e4SAndreas Dilger if (i < end_bit) 66717d50e4SAndreas Dilger memset(bitmap + (i >> 3), 0xff, (end_bit - i) >> 3); 67717d50e4SAndreas Dilger } 68717d50e4SAndreas Dilger 69813e5727STheodore Ts'o void ext4_end_bitmap_read(struct buffer_head *bh, int uptodate) 70813e5727STheodore Ts'o { 71813e5727STheodore Ts'o if (uptodate) { 72813e5727STheodore Ts'o set_buffer_uptodate(bh); 73813e5727STheodore Ts'o set_bitmap_uptodate(bh); 74813e5727STheodore Ts'o } 75813e5727STheodore Ts'o unlock_buffer(bh); 76813e5727STheodore Ts'o put_bh(bh); 77813e5727STheodore Ts'o } 78813e5727STheodore Ts'o 799008a58eSDarrick J. Wong static int ext4_validate_inode_bitmap(struct super_block *sb, 809008a58eSDarrick J. Wong struct ext4_group_desc *desc, 819008a58eSDarrick J. Wong ext4_group_t block_group, 829008a58eSDarrick J. Wong struct buffer_head *bh) 839008a58eSDarrick J. Wong { 849008a58eSDarrick J. Wong ext4_fsblk_t blk; 859008a58eSDarrick J. Wong struct ext4_group_info *grp = ext4_get_group_info(sb, block_group); 869008a58eSDarrick J. Wong 879008a58eSDarrick J. Wong if (buffer_verified(bh)) 889008a58eSDarrick J. Wong return 0; 899008a58eSDarrick J. Wong if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) 909008a58eSDarrick J. Wong return -EFSCORRUPTED; 919008a58eSDarrick J. Wong 929008a58eSDarrick J. Wong ext4_lock_group(sb, block_group); 938d5a803cSTheodore Ts'o if (buffer_verified(bh)) 948d5a803cSTheodore Ts'o goto verified; 959008a58eSDarrick J. Wong blk = ext4_inode_bitmap(sb, desc); 969008a58eSDarrick J. Wong if (!ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh, 979008a58eSDarrick J. Wong EXT4_INODES_PER_GROUP(sb) / 8)) { 989008a58eSDarrick J. Wong ext4_unlock_group(sb, block_group); 999008a58eSDarrick J. Wong ext4_error(sb, "Corrupt inode bitmap - block_group = %u, " 1009008a58eSDarrick J. Wong "inode_bitmap = %llu", block_group, blk); 101db79e6d1SWang Shilong ext4_mark_group_bitmap_corrupted(sb, block_group, 102db79e6d1SWang Shilong EXT4_GROUP_INFO_IBITMAP_CORRUPT); 1039008a58eSDarrick J. Wong return -EFSBADCRC; 1049008a58eSDarrick J. Wong } 1059008a58eSDarrick J. Wong set_buffer_verified(bh); 1068d5a803cSTheodore Ts'o verified: 1079008a58eSDarrick J. Wong ext4_unlock_group(sb, block_group); 1089008a58eSDarrick J. Wong return 0; 1099008a58eSDarrick J. Wong } 1109008a58eSDarrick J. Wong 111ac27a0ecSDave Kleikamp /* 112ac27a0ecSDave Kleikamp * Read the inode allocation bitmap for a given block_group, reading 113ac27a0ecSDave Kleikamp * into the specified slot in the superblock's bitmap cache. 114ac27a0ecSDave Kleikamp * 115ac27a0ecSDave Kleikamp * Return buffer_head of bitmap on success or NULL. 116ac27a0ecSDave Kleikamp */ 117ac27a0ecSDave Kleikamp static struct buffer_head * 118e29d1cdeSEric Sandeen ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group) 119ac27a0ecSDave Kleikamp { 120617ba13bSMingming Cao struct ext4_group_desc *desc; 1217dac4a17STheodore Ts'o struct ext4_sb_info *sbi = EXT4_SB(sb); 122ac27a0ecSDave Kleikamp struct buffer_head *bh = NULL; 123e29d1cdeSEric Sandeen ext4_fsblk_t bitmap_blk; 1249008a58eSDarrick J. Wong int err; 125ac27a0ecSDave Kleikamp 126617ba13bSMingming Cao desc = ext4_get_group_desc(sb, block_group, NULL); 127ac27a0ecSDave Kleikamp if (!desc) 1289008a58eSDarrick J. Wong return ERR_PTR(-EFSCORRUPTED); 129bfff6873SLukas Czerner 130e29d1cdeSEric Sandeen bitmap_blk = ext4_inode_bitmap(sb, desc); 1317dac4a17STheodore Ts'o if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) || 1327dac4a17STheodore Ts'o (bitmap_blk >= ext4_blocks_count(sbi->s_es))) { 1337dac4a17STheodore Ts'o ext4_error(sb, "Invalid inode bitmap blk %llu in " 1347dac4a17STheodore Ts'o "block_group %u", bitmap_blk, block_group); 135206f6d55SWang Shilong ext4_mark_group_bitmap_corrupted(sb, block_group, 136206f6d55SWang Shilong EXT4_GROUP_INFO_IBITMAP_CORRUPT); 1377dac4a17STheodore Ts'o return ERR_PTR(-EFSCORRUPTED); 1387dac4a17STheodore Ts'o } 139e29d1cdeSEric Sandeen bh = sb_getblk(sb, bitmap_blk); 140e29d1cdeSEric Sandeen if (unlikely(!bh)) { 1415ef2a699SWang Shilong ext4_warning(sb, "Cannot read inode bitmap - " 142a9df9a49STheodore Ts'o "block_group = %u, inode_bitmap = %llu", 143e29d1cdeSEric Sandeen block_group, bitmap_blk); 1440db9fdebSWang Shilong return ERR_PTR(-ENOMEM); 145e29d1cdeSEric Sandeen } 1462ccb5fb9SAneesh Kumar K.V if (bitmap_uptodate(bh)) 14741a246d1SDarrick J. Wong goto verify; 148e29d1cdeSEric Sandeen 149c806e68fSFrederic Bohe lock_buffer(bh); 1502ccb5fb9SAneesh Kumar K.V if (bitmap_uptodate(bh)) { 1512ccb5fb9SAneesh Kumar K.V unlock_buffer(bh); 15241a246d1SDarrick J. Wong goto verify; 1532ccb5fb9SAneesh Kumar K.V } 154bfff6873SLukas Czerner 155955ce5f5SAneesh Kumar K.V ext4_lock_group(sb, block_group); 1568844618dSTheodore Ts'o if (ext4_has_group_desc_csum(sb) && 1578844618dSTheodore Ts'o (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) { 1588844618dSTheodore Ts'o if (block_group == 0) { 1598844618dSTheodore Ts'o ext4_unlock_group(sb, block_group); 1608844618dSTheodore Ts'o unlock_buffer(bh); 1618844618dSTheodore Ts'o ext4_error(sb, "Inode bitmap for bg 0 marked " 1628844618dSTheodore Ts'o "uninitialized"); 1638844618dSTheodore Ts'o err = -EFSCORRUPTED; 1648844618dSTheodore Ts'o goto out; 1658844618dSTheodore Ts'o } 166044e6e3dSTheodore Ts'o memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8); 167044e6e3dSTheodore Ts'o ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), 168044e6e3dSTheodore Ts'o sb->s_blocksize * 8, bh->b_data); 1692ccb5fb9SAneesh Kumar K.V set_bitmap_uptodate(bh); 170e29d1cdeSEric Sandeen set_buffer_uptodate(bh); 17141a246d1SDarrick J. Wong set_buffer_verified(bh); 172955ce5f5SAneesh Kumar K.V ext4_unlock_group(sb, block_group); 1733300bedaSAneesh Kumar K.V unlock_buffer(bh); 174e29d1cdeSEric Sandeen return bh; 175e29d1cdeSEric Sandeen } 176955ce5f5SAneesh Kumar K.V ext4_unlock_group(sb, block_group); 177bfff6873SLukas Czerner 1782ccb5fb9SAneesh Kumar K.V if (buffer_uptodate(bh)) { 1792ccb5fb9SAneesh Kumar K.V /* 1802ccb5fb9SAneesh Kumar K.V * if not uninit if bh is uptodate, 1812ccb5fb9SAneesh Kumar K.V * bitmap is also uptodate 1822ccb5fb9SAneesh Kumar K.V */ 1832ccb5fb9SAneesh Kumar K.V set_bitmap_uptodate(bh); 1842ccb5fb9SAneesh Kumar K.V unlock_buffer(bh); 18541a246d1SDarrick J. Wong goto verify; 1862ccb5fb9SAneesh Kumar K.V } 1872ccb5fb9SAneesh Kumar K.V /* 188813e5727STheodore Ts'o * submit the buffer_head for reading 1892ccb5fb9SAneesh Kumar K.V */ 1900562e0baSJiaying Zhang trace_ext4_load_inode_bitmap(sb, block_group); 191813e5727STheodore Ts'o bh->b_end_io = ext4_end_bitmap_read; 192813e5727STheodore Ts'o get_bh(bh); 1932a222ca9SMike Christie submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh); 194813e5727STheodore Ts'o wait_on_buffer(bh); 195813e5727STheodore Ts'o if (!buffer_uptodate(bh)) { 196e29d1cdeSEric Sandeen put_bh(bh); 19712062dddSEric Sandeen ext4_error(sb, "Cannot read inode bitmap - " 198a9df9a49STheodore Ts'o "block_group = %u, inode_bitmap = %llu", 199e29d1cdeSEric Sandeen block_group, bitmap_blk); 200206f6d55SWang Shilong ext4_mark_group_bitmap_corrupted(sb, block_group, 201206f6d55SWang Shilong EXT4_GROUP_INFO_IBITMAP_CORRUPT); 2029008a58eSDarrick J. Wong return ERR_PTR(-EIO); 203e29d1cdeSEric Sandeen } 20441a246d1SDarrick J. Wong 20541a246d1SDarrick J. Wong verify: 2069008a58eSDarrick J. Wong err = ext4_validate_inode_bitmap(sb, desc, block_group, bh); 2079008a58eSDarrick J. Wong if (err) 2089008a58eSDarrick J. Wong goto out; 209ac27a0ecSDave Kleikamp return bh; 2109008a58eSDarrick J. Wong out: 2119008a58eSDarrick J. Wong put_bh(bh); 2129008a58eSDarrick J. Wong return ERR_PTR(err); 213ac27a0ecSDave Kleikamp } 214ac27a0ecSDave Kleikamp 215ac27a0ecSDave Kleikamp /* 216ac27a0ecSDave Kleikamp * NOTE! When we get the inode, we're the only people 217ac27a0ecSDave Kleikamp * that have access to it, and as such there are no 218ac27a0ecSDave Kleikamp * race conditions we have to worry about. The inode 219ac27a0ecSDave Kleikamp * is not on the hash-lists, and it cannot be reached 220ac27a0ecSDave Kleikamp * through the filesystem because the directory entry 221ac27a0ecSDave Kleikamp * has been deleted earlier. 222ac27a0ecSDave Kleikamp * 223ac27a0ecSDave Kleikamp * HOWEVER: we must make sure that we get no aliases, 224ac27a0ecSDave Kleikamp * which means that we have to call "clear_inode()" 225ac27a0ecSDave Kleikamp * _before_ we mark the inode not in use in the inode 226ac27a0ecSDave Kleikamp * bitmaps. Otherwise a newly created file might use 227ac27a0ecSDave Kleikamp * the same inode number (not actually the same pointer 228ac27a0ecSDave Kleikamp * though), and then we'd have two inodes sharing the 229ac27a0ecSDave Kleikamp * same inode number and space on the harddisk. 230ac27a0ecSDave Kleikamp */ 231617ba13bSMingming Cao void ext4_free_inode(handle_t *handle, struct inode *inode) 232ac27a0ecSDave Kleikamp { 233ac27a0ecSDave Kleikamp struct super_block *sb = inode->i_sb; 234ac27a0ecSDave Kleikamp int is_directory; 235ac27a0ecSDave Kleikamp unsigned long ino; 236ac27a0ecSDave Kleikamp struct buffer_head *bitmap_bh = NULL; 237ac27a0ecSDave Kleikamp struct buffer_head *bh2; 238fd2d4291SAvantika Mathur ext4_group_t block_group; 239ac27a0ecSDave Kleikamp unsigned long bit; 240617ba13bSMingming Cao struct ext4_group_desc *gdp; 241617ba13bSMingming Cao struct ext4_super_block *es; 242617ba13bSMingming Cao struct ext4_sb_info *sbi; 2437ce9d5d1SEric Sandeen int fatal = 0, err, count, cleared; 24487a39389SDarrick J. Wong struct ext4_group_info *grp; 245ac27a0ecSDave Kleikamp 24692b97816STheodore Ts'o if (!sb) { 24792b97816STheodore Ts'o printk(KERN_ERR "EXT4-fs: %s:%d: inode on " 24892b97816STheodore Ts'o "nonexistent device\n", __func__, __LINE__); 24992b97816STheodore Ts'o return; 25092b97816STheodore Ts'o } 251ac27a0ecSDave Kleikamp if (atomic_read(&inode->i_count) > 1) { 25292b97816STheodore Ts'o ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d", 25392b97816STheodore Ts'o __func__, __LINE__, inode->i_ino, 254ac27a0ecSDave Kleikamp atomic_read(&inode->i_count)); 255ac27a0ecSDave Kleikamp return; 256ac27a0ecSDave Kleikamp } 257ac27a0ecSDave Kleikamp if (inode->i_nlink) { 25892b97816STheodore Ts'o ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n", 25992b97816STheodore Ts'o __func__, __LINE__, inode->i_ino, inode->i_nlink); 260ac27a0ecSDave Kleikamp return; 261ac27a0ecSDave Kleikamp } 262617ba13bSMingming Cao sbi = EXT4_SB(sb); 263ac27a0ecSDave Kleikamp 264ac27a0ecSDave Kleikamp ino = inode->i_ino; 265617ba13bSMingming Cao ext4_debug("freeing inode %lu\n", ino); 2669bffad1eSTheodore Ts'o trace_ext4_free_inode(inode); 267ac27a0ecSDave Kleikamp 268ac27a0ecSDave Kleikamp /* 269ac27a0ecSDave Kleikamp * Note: we must free any quota before locking the superblock, 270ac27a0ecSDave Kleikamp * as writing the quota to disk may need the lock as well. 271ac27a0ecSDave Kleikamp */ 272871a2931SChristoph Hellwig dquot_initialize(inode); 27363936ddaSChristoph Hellwig dquot_free_inode(inode); 2749f754758SChristoph Hellwig dquot_drop(inode); 275ac27a0ecSDave Kleikamp 276ac27a0ecSDave Kleikamp is_directory = S_ISDIR(inode->i_mode); 277ac27a0ecSDave Kleikamp 278ac27a0ecSDave Kleikamp /* Do this BEFORE marking the inode not in use or returning an error */ 2790930fcc1SAl Viro ext4_clear_inode(inode); 280ac27a0ecSDave Kleikamp 28149598e04SJun Piao es = sbi->s_es; 282617ba13bSMingming Cao if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) { 28312062dddSEric Sandeen ext4_error(sb, "reserved or nonexistent inode %lu", ino); 284ac27a0ecSDave Kleikamp goto error_return; 285ac27a0ecSDave Kleikamp } 286617ba13bSMingming Cao block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); 287617ba13bSMingming Cao bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); 288e29d1cdeSEric Sandeen bitmap_bh = ext4_read_inode_bitmap(sb, block_group); 28987a39389SDarrick J. Wong /* Don't bother if the inode bitmap is corrupt. */ 29087a39389SDarrick J. Wong grp = ext4_get_group_info(sb, block_group); 2919008a58eSDarrick J. Wong if (IS_ERR(bitmap_bh)) { 2929008a58eSDarrick J. Wong fatal = PTR_ERR(bitmap_bh); 2939008a58eSDarrick J. Wong bitmap_bh = NULL; 294ac27a0ecSDave Kleikamp goto error_return; 2959008a58eSDarrick J. Wong } 2969008a58eSDarrick J. Wong if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) { 2979008a58eSDarrick J. Wong fatal = -EFSCORRUPTED; 2989008a58eSDarrick J. Wong goto error_return; 2999008a58eSDarrick J. Wong } 300ac27a0ecSDave Kleikamp 301ac27a0ecSDave Kleikamp BUFFER_TRACE(bitmap_bh, "get_write_access"); 302617ba13bSMingming Cao fatal = ext4_journal_get_write_access(handle, bitmap_bh); 303ac27a0ecSDave Kleikamp if (fatal) 304ac27a0ecSDave Kleikamp goto error_return; 305ac27a0ecSDave Kleikamp 306d17413c0SDmitry Monakhov fatal = -ESRCH; 307617ba13bSMingming Cao gdp = ext4_get_group_desc(sb, block_group, &bh2); 308d17413c0SDmitry Monakhov if (gdp) { 309ac27a0ecSDave Kleikamp BUFFER_TRACE(bh2, "get_write_access"); 310617ba13bSMingming Cao fatal = ext4_journal_get_write_access(handle, bh2); 311d17413c0SDmitry Monakhov } 312955ce5f5SAneesh Kumar K.V ext4_lock_group(sb, block_group); 313597d508cSAkinobu Mita cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data); 314d17413c0SDmitry Monakhov if (fatal || !cleared) { 315d17413c0SDmitry Monakhov ext4_unlock_group(sb, block_group); 316d17413c0SDmitry Monakhov goto out; 317d17413c0SDmitry Monakhov } 318d17413c0SDmitry Monakhov 319560671a0SAneesh Kumar K.V count = ext4_free_inodes_count(sb, gdp) + 1; 320560671a0SAneesh Kumar K.V ext4_free_inodes_set(sb, gdp, count); 321560671a0SAneesh Kumar K.V if (is_directory) { 322560671a0SAneesh Kumar K.V count = ext4_used_dirs_count(sb, gdp) - 1; 323560671a0SAneesh Kumar K.V ext4_used_dirs_set(sb, gdp, count); 324d17413c0SDmitry Monakhov percpu_counter_dec(&sbi->s_dirs_counter); 325d17413c0SDmitry Monakhov } 32641a246d1SDarrick J. Wong ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh, 32741a246d1SDarrick J. Wong EXT4_INODES_PER_GROUP(sb) / 8); 328feb0ab32SDarrick J. Wong ext4_group_desc_csum_set(sb, block_group, gdp); 329d17413c0SDmitry Monakhov ext4_unlock_group(sb, block_group); 3307d39db14STheodore Ts'o 331d17413c0SDmitry Monakhov percpu_counter_inc(&sbi->s_freeinodes_counter); 332d17413c0SDmitry Monakhov if (sbi->s_log_groups_per_flex) { 333d17413c0SDmitry Monakhov ext4_group_t f = ext4_flex_group(sbi, block_group); 334d17413c0SDmitry Monakhov 335d17413c0SDmitry Monakhov atomic_inc(&sbi->s_flex_groups[f].free_inodes); 336d17413c0SDmitry Monakhov if (is_directory) 337c4caae25SEric Sandeen atomic_dec(&sbi->s_flex_groups[f].used_dirs); 3387d39db14STheodore Ts'o } 3390390131bSFrank Mayhar BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata"); 340d17413c0SDmitry Monakhov fatal = ext4_handle_dirty_metadata(handle, NULL, bh2); 341d17413c0SDmitry Monakhov out: 342d17413c0SDmitry Monakhov if (cleared) { 3430390131bSFrank Mayhar BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata"); 3440390131bSFrank Mayhar err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh); 345ac27a0ecSDave Kleikamp if (!fatal) 346ac27a0ecSDave Kleikamp fatal = err; 34787a39389SDarrick J. Wong } else { 348d17413c0SDmitry Monakhov ext4_error(sb, "bit already cleared for inode %lu", ino); 349db79e6d1SWang Shilong ext4_mark_group_bitmap_corrupted(sb, block_group, 350db79e6d1SWang Shilong EXT4_GROUP_INFO_IBITMAP_CORRUPT); 35187a39389SDarrick J. Wong } 352d17413c0SDmitry Monakhov 353ac27a0ecSDave Kleikamp error_return: 354ac27a0ecSDave Kleikamp brelse(bitmap_bh); 355617ba13bSMingming Cao ext4_std_error(sb, fatal); 356ac27a0ecSDave Kleikamp } 357ac27a0ecSDave Kleikamp 358a4912123STheodore Ts'o struct orlov_stats { 35990ba983fSTheodore Ts'o __u64 free_clusters; 360a4912123STheodore Ts'o __u32 free_inodes; 361a4912123STheodore Ts'o __u32 used_dirs; 362a4912123STheodore Ts'o }; 363a4912123STheodore Ts'o 364a4912123STheodore Ts'o /* 365a4912123STheodore Ts'o * Helper function for Orlov's allocator; returns critical information 366a4912123STheodore Ts'o * for a particular block group or flex_bg. If flex_size is 1, then g 367a4912123STheodore Ts'o * is a block group number; otherwise it is flex_bg number. 368a4912123STheodore Ts'o */ 3691f109d5aSTheodore Ts'o static void get_orlov_stats(struct super_block *sb, ext4_group_t g, 370a4912123STheodore Ts'o int flex_size, struct orlov_stats *stats) 371a4912123STheodore Ts'o { 372a4912123STheodore Ts'o struct ext4_group_desc *desc; 3737d39db14STheodore Ts'o struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups; 374a4912123STheodore Ts'o 3757d39db14STheodore Ts'o if (flex_size > 1) { 3767d39db14STheodore Ts'o stats->free_inodes = atomic_read(&flex_group[g].free_inodes); 37790ba983fSTheodore Ts'o stats->free_clusters = atomic64_read(&flex_group[g].free_clusters); 3787d39db14STheodore Ts'o stats->used_dirs = atomic_read(&flex_group[g].used_dirs); 3797d39db14STheodore Ts'o return; 3807d39db14STheodore Ts'o } 3817d39db14STheodore Ts'o 3827d39db14STheodore Ts'o desc = ext4_get_group_desc(sb, g, NULL); 3837d39db14STheodore Ts'o if (desc) { 3847d39db14STheodore Ts'o stats->free_inodes = ext4_free_inodes_count(sb, desc); 385021b65bbSTheodore Ts'o stats->free_clusters = ext4_free_group_clusters(sb, desc); 3867d39db14STheodore Ts'o stats->used_dirs = ext4_used_dirs_count(sb, desc); 3877d39db14STheodore Ts'o } else { 388a4912123STheodore Ts'o stats->free_inodes = 0; 38924aaa8efSTheodore Ts'o stats->free_clusters = 0; 390a4912123STheodore Ts'o stats->used_dirs = 0; 391a4912123STheodore Ts'o } 392a4912123STheodore Ts'o } 393a4912123STheodore Ts'o 394ac27a0ecSDave Kleikamp /* 395ac27a0ecSDave Kleikamp * Orlov's allocator for directories. 396ac27a0ecSDave Kleikamp * 397ac27a0ecSDave Kleikamp * We always try to spread first-level directories. 398ac27a0ecSDave Kleikamp * 399ac27a0ecSDave Kleikamp * If there are blockgroups with both free inodes and free blocks counts 400ac27a0ecSDave Kleikamp * not worse than average we return one with smallest directory count. 401ac27a0ecSDave Kleikamp * Otherwise we simply return a random group. 402ac27a0ecSDave Kleikamp * 403ac27a0ecSDave Kleikamp * For the rest rules look so: 404ac27a0ecSDave Kleikamp * 405ac27a0ecSDave Kleikamp * It's OK to put directory into a group unless 406ac27a0ecSDave Kleikamp * it has too many directories already (max_dirs) or 407ac27a0ecSDave Kleikamp * it has too few free inodes left (min_inodes) or 408ac27a0ecSDave Kleikamp * it has too few free blocks left (min_blocks) or 4091cc8dcf5SBenoit Boissinot * Parent's group is preferred, if it doesn't satisfy these 410ac27a0ecSDave Kleikamp * conditions we search cyclically through the rest. If none 411ac27a0ecSDave Kleikamp * of the groups look good we just look for a group with more 412ac27a0ecSDave Kleikamp * free inodes than average (starting at parent's group). 413ac27a0ecSDave Kleikamp */ 414ac27a0ecSDave Kleikamp 4152aa9fc4cSAvantika Mathur static int find_group_orlov(struct super_block *sb, struct inode *parent, 416dcca3fecSAl Viro ext4_group_t *group, umode_t mode, 417f157a4aaSTheodore Ts'o const struct qstr *qstr) 418ac27a0ecSDave Kleikamp { 419fd2d4291SAvantika Mathur ext4_group_t parent_group = EXT4_I(parent)->i_block_group; 420617ba13bSMingming Cao struct ext4_sb_info *sbi = EXT4_SB(sb); 4218df9675fSTheodore Ts'o ext4_group_t real_ngroups = ext4_get_groups_count(sb); 422617ba13bSMingming Cao int inodes_per_group = EXT4_INODES_PER_GROUP(sb); 42314c83c9fSTheodore Ts'o unsigned int freei, avefreei, grp_free; 42424aaa8efSTheodore Ts'o ext4_fsblk_t freeb, avefreec; 425ac27a0ecSDave Kleikamp unsigned int ndirs; 426a4912123STheodore Ts'o int max_dirs, min_inodes; 42724aaa8efSTheodore Ts'o ext4_grpblk_t min_clusters; 4288df9675fSTheodore Ts'o ext4_group_t i, grp, g, ngroups; 429617ba13bSMingming Cao struct ext4_group_desc *desc; 430a4912123STheodore Ts'o struct orlov_stats stats; 431a4912123STheodore Ts'o int flex_size = ext4_flex_bg_size(sbi); 432f157a4aaSTheodore Ts'o struct dx_hash_info hinfo; 433a4912123STheodore Ts'o 4348df9675fSTheodore Ts'o ngroups = real_ngroups; 435a4912123STheodore Ts'o if (flex_size > 1) { 4368df9675fSTheodore Ts'o ngroups = (real_ngroups + flex_size - 1) >> 437a4912123STheodore Ts'o sbi->s_log_groups_per_flex; 438a4912123STheodore Ts'o parent_group >>= sbi->s_log_groups_per_flex; 439a4912123STheodore Ts'o } 440ac27a0ecSDave Kleikamp 441ac27a0ecSDave Kleikamp freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter); 442ac27a0ecSDave Kleikamp avefreei = freei / ngroups; 44357042651STheodore Ts'o freeb = EXT4_C2B(sbi, 44457042651STheodore Ts'o percpu_counter_read_positive(&sbi->s_freeclusters_counter)); 44524aaa8efSTheodore Ts'o avefreec = freeb; 44624aaa8efSTheodore Ts'o do_div(avefreec, ngroups); 447ac27a0ecSDave Kleikamp ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter); 448ac27a0ecSDave Kleikamp 449a4912123STheodore Ts'o if (S_ISDIR(mode) && 4502b0143b5SDavid Howells ((parent == d_inode(sb->s_root)) || 45112e9b892SDmitry Monakhov (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) { 452ac27a0ecSDave Kleikamp int best_ndir = inodes_per_group; 4532aa9fc4cSAvantika Mathur int ret = -1; 454ac27a0ecSDave Kleikamp 455f157a4aaSTheodore Ts'o if (qstr) { 456f157a4aaSTheodore Ts'o hinfo.hash_version = DX_HASH_HALF_MD4; 457f157a4aaSTheodore Ts'o hinfo.seed = sbi->s_hash_seed; 458b886ee3eSGabriel Krisman Bertazi ext4fs_dirhash(parent, qstr->name, qstr->len, &hinfo); 459f157a4aaSTheodore Ts'o grp = hinfo.hash; 460f157a4aaSTheodore Ts'o } else 461dd1f723bSTheodore Ts'o grp = prandom_u32(); 4622aa9fc4cSAvantika Mathur parent_group = (unsigned)grp % ngroups; 463ac27a0ecSDave Kleikamp for (i = 0; i < ngroups; i++) { 464a4912123STheodore Ts'o g = (parent_group + i) % ngroups; 465a4912123STheodore Ts'o get_orlov_stats(sb, g, flex_size, &stats); 466a4912123STheodore Ts'o if (!stats.free_inodes) 467ac27a0ecSDave Kleikamp continue; 468a4912123STheodore Ts'o if (stats.used_dirs >= best_ndir) 469ac27a0ecSDave Kleikamp continue; 470a4912123STheodore Ts'o if (stats.free_inodes < avefreei) 471ac27a0ecSDave Kleikamp continue; 47224aaa8efSTheodore Ts'o if (stats.free_clusters < avefreec) 473ac27a0ecSDave Kleikamp continue; 474a4912123STheodore Ts'o grp = g; 4752aa9fc4cSAvantika Mathur ret = 0; 476a4912123STheodore Ts'o best_ndir = stats.used_dirs; 477ac27a0ecSDave Kleikamp } 478a4912123STheodore Ts'o if (ret) 479a4912123STheodore Ts'o goto fallback; 480a4912123STheodore Ts'o found_flex_bg: 481a4912123STheodore Ts'o if (flex_size == 1) { 482a4912123STheodore Ts'o *group = grp; 483a4912123STheodore Ts'o return 0; 484a4912123STheodore Ts'o } 485a4912123STheodore Ts'o 486a4912123STheodore Ts'o /* 487a4912123STheodore Ts'o * We pack inodes at the beginning of the flexgroup's 488a4912123STheodore Ts'o * inode tables. Block allocation decisions will do 489a4912123STheodore Ts'o * something similar, although regular files will 490a4912123STheodore Ts'o * start at 2nd block group of the flexgroup. See 491a4912123STheodore Ts'o * ext4_ext_find_goal() and ext4_find_near(). 492a4912123STheodore Ts'o */ 493a4912123STheodore Ts'o grp *= flex_size; 494a4912123STheodore Ts'o for (i = 0; i < flex_size; i++) { 4958df9675fSTheodore Ts'o if (grp+i >= real_ngroups) 496a4912123STheodore Ts'o break; 497a4912123STheodore Ts'o desc = ext4_get_group_desc(sb, grp+i, NULL); 498a4912123STheodore Ts'o if (desc && ext4_free_inodes_count(sb, desc)) { 499a4912123STheodore Ts'o *group = grp+i; 500a4912123STheodore Ts'o return 0; 501a4912123STheodore Ts'o } 502a4912123STheodore Ts'o } 503ac27a0ecSDave Kleikamp goto fallback; 504ac27a0ecSDave Kleikamp } 505ac27a0ecSDave Kleikamp 506ac27a0ecSDave Kleikamp max_dirs = ndirs / ngroups + inodes_per_group / 16; 507a4912123STheodore Ts'o min_inodes = avefreei - inodes_per_group*flex_size / 4; 508a4912123STheodore Ts'o if (min_inodes < 1) 509a4912123STheodore Ts'o min_inodes = 1; 51024aaa8efSTheodore Ts'o min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4; 511ac27a0ecSDave Kleikamp 512a4912123STheodore Ts'o /* 513a4912123STheodore Ts'o * Start looking in the flex group where we last allocated an 514a4912123STheodore Ts'o * inode for this parent directory 515a4912123STheodore Ts'o */ 516a4912123STheodore Ts'o if (EXT4_I(parent)->i_last_alloc_group != ~0) { 517a4912123STheodore Ts'o parent_group = EXT4_I(parent)->i_last_alloc_group; 518a4912123STheodore Ts'o if (flex_size > 1) 519a4912123STheodore Ts'o parent_group >>= sbi->s_log_groups_per_flex; 520a4912123STheodore Ts'o } 521ac27a0ecSDave Kleikamp 522ac27a0ecSDave Kleikamp for (i = 0; i < ngroups; i++) { 523a4912123STheodore Ts'o grp = (parent_group + i) % ngroups; 524a4912123STheodore Ts'o get_orlov_stats(sb, grp, flex_size, &stats); 525a4912123STheodore Ts'o if (stats.used_dirs >= max_dirs) 526ac27a0ecSDave Kleikamp continue; 527a4912123STheodore Ts'o if (stats.free_inodes < min_inodes) 528ac27a0ecSDave Kleikamp continue; 52924aaa8efSTheodore Ts'o if (stats.free_clusters < min_clusters) 530ac27a0ecSDave Kleikamp continue; 531a4912123STheodore Ts'o goto found_flex_bg; 532ac27a0ecSDave Kleikamp } 533ac27a0ecSDave Kleikamp 534ac27a0ecSDave Kleikamp fallback: 5358df9675fSTheodore Ts'o ngroups = real_ngroups; 536a4912123STheodore Ts'o avefreei = freei / ngroups; 537b5451f7bSTheodore Ts'o fallback_retry: 538a4912123STheodore Ts'o parent_group = EXT4_I(parent)->i_block_group; 539ac27a0ecSDave Kleikamp for (i = 0; i < ngroups; i++) { 540a4912123STheodore Ts'o grp = (parent_group + i) % ngroups; 541a4912123STheodore Ts'o desc = ext4_get_group_desc(sb, grp, NULL); 542bb3d132aSDan Carpenter if (desc) { 54314c83c9fSTheodore Ts'o grp_free = ext4_free_inodes_count(sb, desc); 544bb3d132aSDan Carpenter if (grp_free && grp_free >= avefreei) { 545a4912123STheodore Ts'o *group = grp; 5462aa9fc4cSAvantika Mathur return 0; 547ac27a0ecSDave Kleikamp } 548a4912123STheodore Ts'o } 549bb3d132aSDan Carpenter } 550ac27a0ecSDave Kleikamp 551ac27a0ecSDave Kleikamp if (avefreei) { 552ac27a0ecSDave Kleikamp /* 553ac27a0ecSDave Kleikamp * The free-inodes counter is approximate, and for really small 554ac27a0ecSDave Kleikamp * filesystems the above test can fail to find any blockgroups 555ac27a0ecSDave Kleikamp */ 556ac27a0ecSDave Kleikamp avefreei = 0; 557b5451f7bSTheodore Ts'o goto fallback_retry; 558ac27a0ecSDave Kleikamp } 559ac27a0ecSDave Kleikamp 560ac27a0ecSDave Kleikamp return -1; 561ac27a0ecSDave Kleikamp } 562ac27a0ecSDave Kleikamp 5632aa9fc4cSAvantika Mathur static int find_group_other(struct super_block *sb, struct inode *parent, 564dcca3fecSAl Viro ext4_group_t *group, umode_t mode) 565ac27a0ecSDave Kleikamp { 566fd2d4291SAvantika Mathur ext4_group_t parent_group = EXT4_I(parent)->i_block_group; 5678df9675fSTheodore Ts'o ext4_group_t i, last, ngroups = ext4_get_groups_count(sb); 568617ba13bSMingming Cao struct ext4_group_desc *desc; 569a4912123STheodore Ts'o int flex_size = ext4_flex_bg_size(EXT4_SB(sb)); 570a4912123STheodore Ts'o 571a4912123STheodore Ts'o /* 572a4912123STheodore Ts'o * Try to place the inode is the same flex group as its 573a4912123STheodore Ts'o * parent. If we can't find space, use the Orlov algorithm to 574a4912123STheodore Ts'o * find another flex group, and store that information in the 575a4912123STheodore Ts'o * parent directory's inode information so that use that flex 576a4912123STheodore Ts'o * group for future allocations. 577a4912123STheodore Ts'o */ 578a4912123STheodore Ts'o if (flex_size > 1) { 579a4912123STheodore Ts'o int retry = 0; 580a4912123STheodore Ts'o 581a4912123STheodore Ts'o try_again: 582a4912123STheodore Ts'o parent_group &= ~(flex_size-1); 583a4912123STheodore Ts'o last = parent_group + flex_size; 584a4912123STheodore Ts'o if (last > ngroups) 585a4912123STheodore Ts'o last = ngroups; 586a4912123STheodore Ts'o for (i = parent_group; i < last; i++) { 587a4912123STheodore Ts'o desc = ext4_get_group_desc(sb, i, NULL); 588a4912123STheodore Ts'o if (desc && ext4_free_inodes_count(sb, desc)) { 589a4912123STheodore Ts'o *group = i; 590a4912123STheodore Ts'o return 0; 591a4912123STheodore Ts'o } 592a4912123STheodore Ts'o } 593a4912123STheodore Ts'o if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) { 594a4912123STheodore Ts'o retry = 1; 595a4912123STheodore Ts'o parent_group = EXT4_I(parent)->i_last_alloc_group; 596a4912123STheodore Ts'o goto try_again; 597a4912123STheodore Ts'o } 598a4912123STheodore Ts'o /* 599a4912123STheodore Ts'o * If this didn't work, use the Orlov search algorithm 600a4912123STheodore Ts'o * to find a new flex group; we pass in the mode to 601a4912123STheodore Ts'o * avoid the topdir algorithms. 602a4912123STheodore Ts'o */ 603a4912123STheodore Ts'o *group = parent_group + flex_size; 604a4912123STheodore Ts'o if (*group > ngroups) 605a4912123STheodore Ts'o *group = 0; 6067dc57615SPeter Huewe return find_group_orlov(sb, parent, group, mode, NULL); 607a4912123STheodore Ts'o } 608ac27a0ecSDave Kleikamp 609ac27a0ecSDave Kleikamp /* 610ac27a0ecSDave Kleikamp * Try to place the inode in its parent directory 611ac27a0ecSDave Kleikamp */ 6122aa9fc4cSAvantika Mathur *group = parent_group; 6132aa9fc4cSAvantika Mathur desc = ext4_get_group_desc(sb, *group, NULL); 614560671a0SAneesh Kumar K.V if (desc && ext4_free_inodes_count(sb, desc) && 615021b65bbSTheodore Ts'o ext4_free_group_clusters(sb, desc)) 6162aa9fc4cSAvantika Mathur return 0; 617ac27a0ecSDave Kleikamp 618ac27a0ecSDave Kleikamp /* 619ac27a0ecSDave Kleikamp * We're going to place this inode in a different blockgroup from its 620ac27a0ecSDave Kleikamp * parent. We want to cause files in a common directory to all land in 621ac27a0ecSDave Kleikamp * the same blockgroup. But we want files which are in a different 622ac27a0ecSDave Kleikamp * directory which shares a blockgroup with our parent to land in a 623ac27a0ecSDave Kleikamp * different blockgroup. 624ac27a0ecSDave Kleikamp * 625ac27a0ecSDave Kleikamp * So add our directory's i_ino into the starting point for the hash. 626ac27a0ecSDave Kleikamp */ 6272aa9fc4cSAvantika Mathur *group = (*group + parent->i_ino) % ngroups; 628ac27a0ecSDave Kleikamp 629ac27a0ecSDave Kleikamp /* 630ac27a0ecSDave Kleikamp * Use a quadratic hash to find a group with a free inode and some free 631ac27a0ecSDave Kleikamp * blocks. 632ac27a0ecSDave Kleikamp */ 633ac27a0ecSDave Kleikamp for (i = 1; i < ngroups; i <<= 1) { 6342aa9fc4cSAvantika Mathur *group += i; 6352aa9fc4cSAvantika Mathur if (*group >= ngroups) 6362aa9fc4cSAvantika Mathur *group -= ngroups; 6372aa9fc4cSAvantika Mathur desc = ext4_get_group_desc(sb, *group, NULL); 638560671a0SAneesh Kumar K.V if (desc && ext4_free_inodes_count(sb, desc) && 639021b65bbSTheodore Ts'o ext4_free_group_clusters(sb, desc)) 6402aa9fc4cSAvantika Mathur return 0; 641ac27a0ecSDave Kleikamp } 642ac27a0ecSDave Kleikamp 643ac27a0ecSDave Kleikamp /* 644ac27a0ecSDave Kleikamp * That failed: try linear search for a free inode, even if that group 645ac27a0ecSDave Kleikamp * has no free blocks. 646ac27a0ecSDave Kleikamp */ 6472aa9fc4cSAvantika Mathur *group = parent_group; 648ac27a0ecSDave Kleikamp for (i = 0; i < ngroups; i++) { 6492aa9fc4cSAvantika Mathur if (++*group >= ngroups) 6502aa9fc4cSAvantika Mathur *group = 0; 6512aa9fc4cSAvantika Mathur desc = ext4_get_group_desc(sb, *group, NULL); 652560671a0SAneesh Kumar K.V if (desc && ext4_free_inodes_count(sb, desc)) 6532aa9fc4cSAvantika Mathur return 0; 654ac27a0ecSDave Kleikamp } 655ac27a0ecSDave Kleikamp 656ac27a0ecSDave Kleikamp return -1; 657ac27a0ecSDave Kleikamp } 658ac27a0ecSDave Kleikamp 659ac27a0ecSDave Kleikamp /* 66019883bd9STheodore Ts'o * In no journal mode, if an inode has recently been deleted, we want 66119883bd9STheodore Ts'o * to avoid reusing it until we're reasonably sure the inode table 66219883bd9STheodore Ts'o * block has been written back to disk. (Yes, these values are 66319883bd9STheodore Ts'o * somewhat arbitrary...) 66419883bd9STheodore Ts'o */ 66519883bd9STheodore Ts'o #define RECENTCY_MIN 5 666b5f51573SAndreas Dilger #define RECENTCY_DIRTY 300 66719883bd9STheodore Ts'o 66819883bd9STheodore Ts'o static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino) 66919883bd9STheodore Ts'o { 67019883bd9STheodore Ts'o struct ext4_group_desc *gdp; 67119883bd9STheodore Ts'o struct ext4_inode *raw_inode; 67219883bd9STheodore Ts'o struct buffer_head *bh; 67319883bd9STheodore Ts'o int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; 674b5f51573SAndreas Dilger int offset, ret = 0; 675b5f51573SAndreas Dilger int recentcy = RECENTCY_MIN; 676b5f51573SAndreas Dilger u32 dtime, now; 67719883bd9STheodore Ts'o 67819883bd9STheodore Ts'o gdp = ext4_get_group_desc(sb, group, NULL); 67919883bd9STheodore Ts'o if (unlikely(!gdp)) 68019883bd9STheodore Ts'o return 0; 68119883bd9STheodore Ts'o 6824f9d956dSJan Kara bh = sb_find_get_block(sb, ext4_inode_table(sb, gdp) + 68319883bd9STheodore Ts'o (ino / inodes_per_block)); 6844f9d956dSJan Kara if (!bh || !buffer_uptodate(bh)) 68519883bd9STheodore Ts'o /* 68619883bd9STheodore Ts'o * If the block is not in the buffer cache, then it 68719883bd9STheodore Ts'o * must have been written out. 68819883bd9STheodore Ts'o */ 68919883bd9STheodore Ts'o goto out; 69019883bd9STheodore Ts'o 69119883bd9STheodore Ts'o offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb); 69219883bd9STheodore Ts'o raw_inode = (struct ext4_inode *) (bh->b_data + offset); 693b5f51573SAndreas Dilger 694b5f51573SAndreas Dilger /* i_dtime is only 32 bits on disk, but we only care about relative 695b5f51573SAndreas Dilger * times in the range of a few minutes (i.e. long enough to sync a 696b5f51573SAndreas Dilger * recently-deleted inode to disk), so using the low 32 bits of the 697b5f51573SAndreas Dilger * clock (a 68 year range) is enough, see time_before32() */ 69819883bd9STheodore Ts'o dtime = le32_to_cpu(raw_inode->i_dtime); 699b5f51573SAndreas Dilger now = ktime_get_real_seconds(); 70019883bd9STheodore Ts'o if (buffer_dirty(bh)) 70119883bd9STheodore Ts'o recentcy += RECENTCY_DIRTY; 70219883bd9STheodore Ts'o 703b5f51573SAndreas Dilger if (dtime && time_before32(dtime, now) && 704b5f51573SAndreas Dilger time_before32(now, dtime + recentcy)) 70519883bd9STheodore Ts'o ret = 1; 70619883bd9STheodore Ts'o out: 70719883bd9STheodore Ts'o brelse(bh); 70819883bd9STheodore Ts'o return ret; 70919883bd9STheodore Ts'o } 71019883bd9STheodore Ts'o 711901ed070SWang Shilong static int find_inode_bit(struct super_block *sb, ext4_group_t group, 712901ed070SWang Shilong struct buffer_head *bitmap, unsigned long *ino) 713901ed070SWang Shilong { 714901ed070SWang Shilong next: 715901ed070SWang Shilong *ino = ext4_find_next_zero_bit((unsigned long *) 716901ed070SWang Shilong bitmap->b_data, 717901ed070SWang Shilong EXT4_INODES_PER_GROUP(sb), *ino); 718901ed070SWang Shilong if (*ino >= EXT4_INODES_PER_GROUP(sb)) 719901ed070SWang Shilong return 0; 720901ed070SWang Shilong 721901ed070SWang Shilong if ((EXT4_SB(sb)->s_journal == NULL) && 722901ed070SWang Shilong recently_deleted(sb, group, *ino)) { 723901ed070SWang Shilong *ino = *ino + 1; 724901ed070SWang Shilong if (*ino < EXT4_INODES_PER_GROUP(sb)) 725901ed070SWang Shilong goto next; 726901ed070SWang Shilong return 0; 727901ed070SWang Shilong } 728901ed070SWang Shilong 729901ed070SWang Shilong return 1; 730901ed070SWang Shilong } 731901ed070SWang Shilong 73219883bd9STheodore Ts'o /* 733ac27a0ecSDave Kleikamp * There are two policies for allocating an inode. If the new inode is 734ac27a0ecSDave Kleikamp * a directory, then a forward search is made for a block group with both 735ac27a0ecSDave Kleikamp * free space and a low directory-to-inode ratio; if that fails, then of 736ac27a0ecSDave Kleikamp * the groups with above-average free space, that group with the fewest 737ac27a0ecSDave Kleikamp * directories already is chosen. 738ac27a0ecSDave Kleikamp * 739ac27a0ecSDave Kleikamp * For other inodes, search forward from the parent directory's block 740ac27a0ecSDave Kleikamp * group to find a free inode. 741ac27a0ecSDave Kleikamp */ 7421139575aSTheodore Ts'o struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir, 7431139575aSTheodore Ts'o umode_t mode, const struct qstr *qstr, 7441b917ed8STahsin Erdogan __u32 goal, uid_t *owner, __u32 i_flags, 7451b917ed8STahsin Erdogan int handle_type, unsigned int line_no, 7461b917ed8STahsin Erdogan int nblocks) 747ac27a0ecSDave Kleikamp { 748ac27a0ecSDave Kleikamp struct super_block *sb; 7493300bedaSAneesh Kumar K.V struct buffer_head *inode_bitmap_bh = NULL; 7503300bedaSAneesh Kumar K.V struct buffer_head *group_desc_bh; 7518df9675fSTheodore Ts'o ext4_group_t ngroups, group = 0; 752ac27a0ecSDave Kleikamp unsigned long ino = 0; 753ac27a0ecSDave Kleikamp struct inode *inode; 754617ba13bSMingming Cao struct ext4_group_desc *gdp = NULL; 755617ba13bSMingming Cao struct ext4_inode_info *ei; 756617ba13bSMingming Cao struct ext4_sb_info *sbi; 757a7cdadeeSJan Kara int ret2, err; 758ac27a0ecSDave Kleikamp struct inode *ret; 7592aa9fc4cSAvantika Mathur ext4_group_t i; 760772cb7c8SJose R. Santos ext4_group_t flex_group; 76187a39389SDarrick J. Wong struct ext4_group_info *grp; 762e709e9dfSTheodore Ts'o int encrypt = 0; 763ac27a0ecSDave Kleikamp 764ac27a0ecSDave Kleikamp /* Cannot create files in a deleted directory */ 765ac27a0ecSDave Kleikamp if (!dir || !dir->i_nlink) 766ac27a0ecSDave Kleikamp return ERR_PTR(-EPERM); 767ac27a0ecSDave Kleikamp 768af65207cSTahsin Erdogan sb = dir->i_sb; 769af65207cSTahsin Erdogan sbi = EXT4_SB(sb); 770af65207cSTahsin Erdogan 771af65207cSTahsin Erdogan if (unlikely(ext4_forced_shutdown(sbi))) 7720db1ff22STheodore Ts'o return ERR_PTR(-EIO); 7730db1ff22STheodore Ts'o 774592ddec7SChandan Rajendra if ((IS_ENCRYPTED(dir) || DUMMY_ENCRYPTION_ENABLED(sbi)) && 775ad47f953STahsin Erdogan (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) && 776ad47f953STahsin Erdogan !(i_flags & EXT4_EA_INODE_FL)) { 777a7550b30SJaegeuk Kim err = fscrypt_get_encryption_info(dir); 778e709e9dfSTheodore Ts'o if (err) 779e709e9dfSTheodore Ts'o return ERR_PTR(err); 780a7550b30SJaegeuk Kim if (!fscrypt_has_encryption_key(dir)) 78154475f53SEric Biggers return ERR_PTR(-ENOKEY); 782e709e9dfSTheodore Ts'o encrypt = 1; 783e709e9dfSTheodore Ts'o } 784e709e9dfSTheodore Ts'o 785af65207cSTahsin Erdogan if (!handle && sbi->s_journal && !(i_flags & EXT4_EA_INODE_FL)) { 786af65207cSTahsin Erdogan #ifdef CONFIG_EXT4_FS_POSIX_ACL 787af65207cSTahsin Erdogan struct posix_acl *p = get_acl(dir, ACL_TYPE_DEFAULT); 788af65207cSTahsin Erdogan 789996fc447STheodore Ts'o if (IS_ERR(p)) 790996fc447STheodore Ts'o return ERR_CAST(p); 791af65207cSTahsin Erdogan if (p) { 792af65207cSTahsin Erdogan int acl_size = p->a_count * sizeof(ext4_acl_entry); 793af65207cSTahsin Erdogan 794af65207cSTahsin Erdogan nblocks += (S_ISDIR(mode) ? 2 : 1) * 795af65207cSTahsin Erdogan __ext4_xattr_set_credits(sb, NULL /* inode */, 796af65207cSTahsin Erdogan NULL /* block_bh */, acl_size, 797af65207cSTahsin Erdogan true /* is_create */); 798af65207cSTahsin Erdogan posix_acl_release(p); 799af65207cSTahsin Erdogan } 800af65207cSTahsin Erdogan #endif 801af65207cSTahsin Erdogan 802af65207cSTahsin Erdogan #ifdef CONFIG_SECURITY 803af65207cSTahsin Erdogan { 804af65207cSTahsin Erdogan int num_security_xattrs = 1; 805af65207cSTahsin Erdogan 806af65207cSTahsin Erdogan #ifdef CONFIG_INTEGRITY 807af65207cSTahsin Erdogan num_security_xattrs++; 808af65207cSTahsin Erdogan #endif 809af65207cSTahsin Erdogan /* 810af65207cSTahsin Erdogan * We assume that security xattrs are never 811af65207cSTahsin Erdogan * more than 1k. In practice they are under 812af65207cSTahsin Erdogan * 128 bytes. 813af65207cSTahsin Erdogan */ 814af65207cSTahsin Erdogan nblocks += num_security_xattrs * 815af65207cSTahsin Erdogan __ext4_xattr_set_credits(sb, NULL /* inode */, 816af65207cSTahsin Erdogan NULL /* block_bh */, 1024, 817af65207cSTahsin Erdogan true /* is_create */); 818af65207cSTahsin Erdogan } 819af65207cSTahsin Erdogan #endif 820af65207cSTahsin Erdogan if (encrypt) 821af65207cSTahsin Erdogan nblocks += __ext4_xattr_set_credits(sb, 822af65207cSTahsin Erdogan NULL /* inode */, NULL /* block_bh */, 823af65207cSTahsin Erdogan FSCRYPT_SET_CONTEXT_MAX_SIZE, 824af65207cSTahsin Erdogan true /* is_create */); 825af65207cSTahsin Erdogan } 826af65207cSTahsin Erdogan 8278df9675fSTheodore Ts'o ngroups = ext4_get_groups_count(sb); 8289bffad1eSTheodore Ts'o trace_ext4_request_inode(dir, mode); 829ac27a0ecSDave Kleikamp inode = new_inode(sb); 830ac27a0ecSDave Kleikamp if (!inode) 831ac27a0ecSDave Kleikamp return ERR_PTR(-ENOMEM); 832617ba13bSMingming Cao ei = EXT4_I(inode); 833772cb7c8SJose R. Santos 834eb9cc7e1SJan Kara /* 835b8a07463SAdam Buchbinder * Initialize owners and quota early so that we don't have to account 836eb9cc7e1SJan Kara * for quota initialization worst case in standard inode creating 837eb9cc7e1SJan Kara * transaction 838eb9cc7e1SJan Kara */ 839eb9cc7e1SJan Kara if (owner) { 840eb9cc7e1SJan Kara inode->i_mode = mode; 841eb9cc7e1SJan Kara i_uid_write(inode, owner[0]); 842eb9cc7e1SJan Kara i_gid_write(inode, owner[1]); 843eb9cc7e1SJan Kara } else if (test_opt(sb, GRPID)) { 844eb9cc7e1SJan Kara inode->i_mode = mode; 845eb9cc7e1SJan Kara inode->i_uid = current_fsuid(); 846eb9cc7e1SJan Kara inode->i_gid = dir->i_gid; 847eb9cc7e1SJan Kara } else 848eb9cc7e1SJan Kara inode_init_owner(inode, dir, mode); 849040cb378SLi Xi 8500b7b7779SKaho Ng if (ext4_has_feature_project(sb) && 851040cb378SLi Xi ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) 852040cb378SLi Xi ei->i_projid = EXT4_I(dir)->i_projid; 853040cb378SLi Xi else 854040cb378SLi Xi ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID); 855040cb378SLi Xi 856a7cdadeeSJan Kara err = dquot_initialize(inode); 857a7cdadeeSJan Kara if (err) 858a7cdadeeSJan Kara goto out; 859eb9cc7e1SJan Kara 86011013911SAndreas Dilger if (!goal) 86111013911SAndreas Dilger goal = sbi->s_inode_goal; 86211013911SAndreas Dilger 863e6462869SJohann Lombardi if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) { 86411013911SAndreas Dilger group = (goal - 1) / EXT4_INODES_PER_GROUP(sb); 86511013911SAndreas Dilger ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb); 86611013911SAndreas Dilger ret2 = 0; 86711013911SAndreas Dilger goto got_group; 86811013911SAndreas Dilger } 86911013911SAndreas Dilger 8704113c4caSLukas Czerner if (S_ISDIR(mode)) 871f157a4aaSTheodore Ts'o ret2 = find_group_orlov(sb, dir, &group, mode, qstr); 8724113c4caSLukas Czerner else 873a4912123STheodore Ts'o ret2 = find_group_other(sb, dir, &group, mode); 874ac27a0ecSDave Kleikamp 875772cb7c8SJose R. Santos got_group: 876a4912123STheodore Ts'o EXT4_I(dir)->i_last_alloc_group = group; 877ac27a0ecSDave Kleikamp err = -ENOSPC; 8782aa9fc4cSAvantika Mathur if (ret2 == -1) 879ac27a0ecSDave Kleikamp goto out; 880ac27a0ecSDave Kleikamp 881119c0d44STheodore Ts'o /* 882119c0d44STheodore Ts'o * Normally we will only go through one pass of this loop, 883119c0d44STheodore Ts'o * unless we get unlucky and it turns out the group we selected 884119c0d44STheodore Ts'o * had its last inode grabbed by someone else. 885119c0d44STheodore Ts'o */ 88611013911SAndreas Dilger for (i = 0; i < ngroups; i++, ino = 0) { 887ac27a0ecSDave Kleikamp err = -EIO; 888ac27a0ecSDave Kleikamp 8893300bedaSAneesh Kumar K.V gdp = ext4_get_group_desc(sb, group, &group_desc_bh); 890ac27a0ecSDave Kleikamp if (!gdp) 891eb9cc7e1SJan Kara goto out; 892ac27a0ecSDave Kleikamp 893f2a09af6SYongqiang Yang /* 894f2a09af6SYongqiang Yang * Check free inodes count before loading bitmap. 895f2a09af6SYongqiang Yang */ 8962fe435d8SWang Shilong if (ext4_free_inodes_count(sb, gdp) == 0) 8972fe435d8SWang Shilong goto next_group; 898f2a09af6SYongqiang Yang 89987a39389SDarrick J. Wong grp = ext4_get_group_info(sb, group); 90087a39389SDarrick J. Wong /* Skip groups with already-known suspicious inode tables */ 9012fe435d8SWang Shilong if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) 9022fe435d8SWang Shilong goto next_group; 90387a39389SDarrick J. Wong 9043300bedaSAneesh Kumar K.V brelse(inode_bitmap_bh); 9053300bedaSAneesh Kumar K.V inode_bitmap_bh = ext4_read_inode_bitmap(sb, group); 90687a39389SDarrick J. Wong /* Skip groups with suspicious inode tables */ 9079008a58eSDarrick J. Wong if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) || 9089008a58eSDarrick J. Wong IS_ERR(inode_bitmap_bh)) { 9099008a58eSDarrick J. Wong inode_bitmap_bh = NULL; 9102fe435d8SWang Shilong goto next_group; 91187a39389SDarrick J. Wong } 912ac27a0ecSDave Kleikamp 913ac27a0ecSDave Kleikamp repeat_in_this_group: 914901ed070SWang Shilong ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino); 915901ed070SWang Shilong if (!ret2) 916a34eb503STheodore Ts'o goto next_group; 917901ed070SWang Shilong 918119c0d44STheodore Ts'o if (group == 0 && (ino + 1) < EXT4_FIRST_INO(sb)) { 919119c0d44STheodore Ts'o ext4_error(sb, "reserved inode found cleared - " 920119c0d44STheodore Ts'o "inode=%lu", ino + 1); 921206f6d55SWang Shilong ext4_mark_group_bitmap_corrupted(sb, group, 922206f6d55SWang Shilong EXT4_GROUP_INFO_IBITMAP_CORRUPT); 9232fe435d8SWang Shilong goto next_group; 924119c0d44STheodore Ts'o } 925901ed070SWang Shilong 9261139575aSTheodore Ts'o if (!handle) { 9271139575aSTheodore Ts'o BUG_ON(nblocks <= 0); 9281139575aSTheodore Ts'o handle = __ext4_journal_start_sb(dir->i_sb, line_no, 9295fe2fe89SJan Kara handle_type, nblocks, 9305fe2fe89SJan Kara 0); 9311139575aSTheodore Ts'o if (IS_ERR(handle)) { 9321139575aSTheodore Ts'o err = PTR_ERR(handle); 933eb9cc7e1SJan Kara ext4_std_error(sb, err); 934eb9cc7e1SJan Kara goto out; 9351139575aSTheodore Ts'o } 9361139575aSTheodore Ts'o } 937ffb5387eSEric Sandeen BUFFER_TRACE(inode_bitmap_bh, "get_write_access"); 938ffb5387eSEric Sandeen err = ext4_journal_get_write_access(handle, inode_bitmap_bh); 939eb9cc7e1SJan Kara if (err) { 940eb9cc7e1SJan Kara ext4_std_error(sb, err); 941eb9cc7e1SJan Kara goto out; 942eb9cc7e1SJan Kara } 943119c0d44STheodore Ts'o ext4_lock_group(sb, group); 944119c0d44STheodore Ts'o ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data); 945901ed070SWang Shilong if (ret2) { 946901ed070SWang Shilong /* Someone already took the bit. Repeat the search 947901ed070SWang Shilong * with lock held. 948901ed070SWang Shilong */ 949901ed070SWang Shilong ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino); 950901ed070SWang Shilong if (ret2) { 951901ed070SWang Shilong ext4_set_bit(ino, inode_bitmap_bh->b_data); 952901ed070SWang Shilong ret2 = 0; 953901ed070SWang Shilong } else { 954901ed070SWang Shilong ret2 = 1; /* we didn't grab the inode */ 955901ed070SWang Shilong } 956901ed070SWang Shilong } 957119c0d44STheodore Ts'o ext4_unlock_group(sb, group); 958119c0d44STheodore Ts'o ino++; /* the inode bitmap is zero-based */ 959119c0d44STheodore Ts'o if (!ret2) 960119c0d44STheodore Ts'o goto got; /* we grabbed the inode! */ 961901ed070SWang Shilong 962119c0d44STheodore Ts'o if (ino < EXT4_INODES_PER_GROUP(sb)) 963119c0d44STheodore Ts'o goto repeat_in_this_group; 964a34eb503STheodore Ts'o next_group: 965a34eb503STheodore Ts'o if (++group == ngroups) 966a34eb503STheodore Ts'o group = 0; 967ac27a0ecSDave Kleikamp } 968ac27a0ecSDave Kleikamp err = -ENOSPC; 969ac27a0ecSDave Kleikamp goto out; 970ac27a0ecSDave Kleikamp 971ac27a0ecSDave Kleikamp got: 972ffb5387eSEric Sandeen BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata"); 973ffb5387eSEric Sandeen err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh); 974eb9cc7e1SJan Kara if (err) { 975eb9cc7e1SJan Kara ext4_std_error(sb, err); 976eb9cc7e1SJan Kara goto out; 977eb9cc7e1SJan Kara } 978ffb5387eSEric Sandeen 97961c219f5STheodore Ts'o BUFFER_TRACE(group_desc_bh, "get_write_access"); 98061c219f5STheodore Ts'o err = ext4_journal_get_write_access(handle, group_desc_bh); 98161c219f5STheodore Ts'o if (err) { 98261c219f5STheodore Ts'o ext4_std_error(sb, err); 98361c219f5STheodore Ts'o goto out; 98461c219f5STheodore Ts'o } 98561c219f5STheodore Ts'o 986717d50e4SAndreas Dilger /* We may have to initialize the block bitmap if it isn't already */ 987feb0ab32SDarrick J. Wong if (ext4_has_group_desc_csum(sb) && 988717d50e4SAndreas Dilger gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { 9893300bedaSAneesh Kumar K.V struct buffer_head *block_bitmap_bh; 990717d50e4SAndreas Dilger 9913300bedaSAneesh Kumar K.V block_bitmap_bh = ext4_read_block_bitmap(sb, group); 9929008a58eSDarrick J. Wong if (IS_ERR(block_bitmap_bh)) { 9939008a58eSDarrick J. Wong err = PTR_ERR(block_bitmap_bh); 994599a9b77SJan Kara goto out; 995599a9b77SJan Kara } 9963300bedaSAneesh Kumar K.V BUFFER_TRACE(block_bitmap_bh, "get block bitmap access"); 9973300bedaSAneesh Kumar K.V err = ext4_journal_get_write_access(handle, block_bitmap_bh); 998717d50e4SAndreas Dilger if (err) { 9993300bedaSAneesh Kumar K.V brelse(block_bitmap_bh); 1000eb9cc7e1SJan Kara ext4_std_error(sb, err); 1001eb9cc7e1SJan Kara goto out; 1002717d50e4SAndreas Dilger } 1003717d50e4SAndreas Dilger 1004fd034a84STheodore Ts'o BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap"); 1005fd034a84STheodore Ts'o err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh); 1006fd034a84STheodore Ts'o 1007717d50e4SAndreas Dilger /* recheck and clear flag under lock if we still need to */ 1008fd034a84STheodore Ts'o ext4_lock_group(sb, group); 10098844618dSTheodore Ts'o if (ext4_has_group_desc_csum(sb) && 10108844618dSTheodore Ts'o (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) { 10113300bedaSAneesh Kumar K.V gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT); 1012021b65bbSTheodore Ts'o ext4_free_group_clusters_set(sb, gdp, 1013cff1dfd7STheodore Ts'o ext4_free_clusters_after_init(sb, group, gdp)); 1014fa77dcfaSDarrick J. Wong ext4_block_bitmap_csum_set(sb, group, gdp, 101579f1ba49STao Ma block_bitmap_bh); 1016feb0ab32SDarrick J. Wong ext4_group_desc_csum_set(sb, group, gdp); 1017717d50e4SAndreas Dilger } 1018955ce5f5SAneesh Kumar K.V ext4_unlock_group(sb, group); 1019aeb1e5d6STheodore Ts'o brelse(block_bitmap_bh); 1020717d50e4SAndreas Dilger 1021eb9cc7e1SJan Kara if (err) { 1022eb9cc7e1SJan Kara ext4_std_error(sb, err); 1023eb9cc7e1SJan Kara goto out; 1024eb9cc7e1SJan Kara } 1025717d50e4SAndreas Dilger } 1026119c0d44STheodore Ts'o 1027119c0d44STheodore Ts'o /* Update the relevant bg descriptor fields */ 102841a246d1SDarrick J. Wong if (ext4_has_group_desc_csum(sb)) { 1029119c0d44STheodore Ts'o int free; 1030119c0d44STheodore Ts'o struct ext4_group_info *grp = ext4_get_group_info(sb, group); 1031119c0d44STheodore Ts'o 1032119c0d44STheodore Ts'o down_read(&grp->alloc_sem); /* protect vs itable lazyinit */ 1033119c0d44STheodore Ts'o ext4_lock_group(sb, group); /* while we modify the bg desc */ 1034119c0d44STheodore Ts'o free = EXT4_INODES_PER_GROUP(sb) - 1035119c0d44STheodore Ts'o ext4_itable_unused_count(sb, gdp); 1036119c0d44STheodore Ts'o if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { 1037119c0d44STheodore Ts'o gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT); 1038119c0d44STheodore Ts'o free = 0; 1039119c0d44STheodore Ts'o } 1040119c0d44STheodore Ts'o /* 1041119c0d44STheodore Ts'o * Check the relative inode number against the last used 1042119c0d44STheodore Ts'o * relative inode number in this group. if it is greater 1043119c0d44STheodore Ts'o * we need to update the bg_itable_unused count 1044119c0d44STheodore Ts'o */ 1045119c0d44STheodore Ts'o if (ino > free) 1046119c0d44STheodore Ts'o ext4_itable_unused_set(sb, gdp, 1047119c0d44STheodore Ts'o (EXT4_INODES_PER_GROUP(sb) - ino)); 1048119c0d44STheodore Ts'o up_read(&grp->alloc_sem); 10496f2e9f0eSTao Ma } else { 10506f2e9f0eSTao Ma ext4_lock_group(sb, group); 1051119c0d44STheodore Ts'o } 10526f2e9f0eSTao Ma 1053119c0d44STheodore Ts'o ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1); 1054119c0d44STheodore Ts'o if (S_ISDIR(mode)) { 1055119c0d44STheodore Ts'o ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1); 1056119c0d44STheodore Ts'o if (sbi->s_log_groups_per_flex) { 1057119c0d44STheodore Ts'o ext4_group_t f = ext4_flex_group(sbi, group); 1058119c0d44STheodore Ts'o 1059119c0d44STheodore Ts'o atomic_inc(&sbi->s_flex_groups[f].used_dirs); 1060119c0d44STheodore Ts'o } 1061119c0d44STheodore Ts'o } 106241a246d1SDarrick J. Wong if (ext4_has_group_desc_csum(sb)) { 106341a246d1SDarrick J. Wong ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh, 106441a246d1SDarrick J. Wong EXT4_INODES_PER_GROUP(sb) / 8); 1065feb0ab32SDarrick J. Wong ext4_group_desc_csum_set(sb, group, gdp); 1066119c0d44STheodore Ts'o } 10676f2e9f0eSTao Ma ext4_unlock_group(sb, group); 1068119c0d44STheodore Ts'o 10693300bedaSAneesh Kumar K.V BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata"); 10703300bedaSAneesh Kumar K.V err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh); 1071eb9cc7e1SJan Kara if (err) { 1072eb9cc7e1SJan Kara ext4_std_error(sb, err); 1073eb9cc7e1SJan Kara goto out; 1074eb9cc7e1SJan Kara } 1075ac27a0ecSDave Kleikamp 1076ac27a0ecSDave Kleikamp percpu_counter_dec(&sbi->s_freeinodes_counter); 1077ac27a0ecSDave Kleikamp if (S_ISDIR(mode)) 1078ac27a0ecSDave Kleikamp percpu_counter_inc(&sbi->s_dirs_counter); 1079ac27a0ecSDave Kleikamp 1080772cb7c8SJose R. Santos if (sbi->s_log_groups_per_flex) { 1081772cb7c8SJose R. Santos flex_group = ext4_flex_group(sbi, group); 10829f24e420STheodore Ts'o atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes); 1083772cb7c8SJose R. Santos } 1084ac27a0ecSDave Kleikamp 1085717d50e4SAndreas Dilger inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb); 1086ac27a0ecSDave Kleikamp /* This is the optimal IO size (for stat), not the fs block size */ 1087ac27a0ecSDave Kleikamp inode->i_blocks = 0; 108895582b00SDeepa Dinamani inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode); 10897b62b293SArnd Bergmann ei->i_crtime = inode->i_mtime; 1090ac27a0ecSDave Kleikamp 1091ac27a0ecSDave Kleikamp memset(ei->i_data, 0, sizeof(ei->i_data)); 1092ac27a0ecSDave Kleikamp ei->i_dir_start_lookup = 0; 1093ac27a0ecSDave Kleikamp ei->i_disksize = 0; 1094ac27a0ecSDave Kleikamp 10954af83508SEryu Guan /* Don't inherit extent flag from directory, amongst others. */ 10962dc6b0d4SDuane Griffin ei->i_flags = 10972dc6b0d4SDuane Griffin ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED); 10981b917ed8STahsin Erdogan ei->i_flags |= i_flags; 1099ac27a0ecSDave Kleikamp ei->i_file_acl = 0; 1100ac27a0ecSDave Kleikamp ei->i_dtime = 0; 1101ac27a0ecSDave Kleikamp ei->i_block_group = group; 1102a4912123STheodore Ts'o ei->i_last_alloc_group = ~0; 1103ac27a0ecSDave Kleikamp 1104617ba13bSMingming Cao ext4_set_inode_flags(inode); 1105ac27a0ecSDave Kleikamp if (IS_DIRSYNC(inode)) 11060390131bSFrank Mayhar ext4_handle_sync(handle); 11076b38e842SAl Viro if (insert_inode_locked(inode) < 0) { 1108acd6ad83SJan Kara /* 1109acd6ad83SJan Kara * Likely a bitmap corruption causing inode to be allocated 1110acd6ad83SJan Kara * twice. 1111acd6ad83SJan Kara */ 1112acd6ad83SJan Kara err = -EIO; 1113eb9cc7e1SJan Kara ext4_error(sb, "failed to insert inode %lu: doubly allocated?", 1114eb9cc7e1SJan Kara inode->i_ino); 1115206f6d55SWang Shilong ext4_mark_group_bitmap_corrupted(sb, group, 1116206f6d55SWang Shilong EXT4_GROUP_INFO_IBITMAP_CORRUPT); 1117eb9cc7e1SJan Kara goto out; 11186b38e842SAl Viro } 111923253068STheodore Ts'o inode->i_generation = prandom_u32(); 1120ac27a0ecSDave Kleikamp 1121814525f4SDarrick J. Wong /* Precompute checksum seed for inode metadata */ 11229aa5d32bSDmitry Monakhov if (ext4_has_metadata_csum(sb)) { 1123814525f4SDarrick J. Wong __u32 csum; 1124814525f4SDarrick J. Wong __le32 inum = cpu_to_le32(inode->i_ino); 1125814525f4SDarrick J. Wong __le32 gen = cpu_to_le32(inode->i_generation); 1126814525f4SDarrick J. Wong csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, 1127814525f4SDarrick J. Wong sizeof(inum)); 1128814525f4SDarrick J. Wong ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, 1129814525f4SDarrick J. Wong sizeof(gen)); 1130814525f4SDarrick J. Wong } 1131814525f4SDarrick J. Wong 1132353eb83cSTheodore Ts'o ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ 113319f5fb7aSTheodore Ts'o ext4_set_inode_state(inode, EXT4_STATE_NEW); 1134ef7f3835SKalpak Shah 113549598e04SJun Piao ei->i_extra_isize = sbi->s_want_extra_isize; 1136f08225d1STao Ma ei->i_inline_off = 0; 1137e2b911c5SDarrick J. Wong if (ext4_has_feature_inline_data(sb)) 1138f08225d1STao Ma ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); 1139ac27a0ecSDave Kleikamp ret = inode; 114063936ddaSChristoph Hellwig err = dquot_alloc_inode(inode); 114163936ddaSChristoph Hellwig if (err) 1142ac27a0ecSDave Kleikamp goto fail_drop; 1143ac27a0ecSDave Kleikamp 1144aa1dca3bSEric Biggers /* 1145aa1dca3bSEric Biggers * Since the encryption xattr will always be unique, create it first so 1146aa1dca3bSEric Biggers * that it's less likely to end up in an external xattr block and 1147aa1dca3bSEric Biggers * prevent its deduplication. 1148aa1dca3bSEric Biggers */ 1149aa1dca3bSEric Biggers if (encrypt) { 1150aa1dca3bSEric Biggers err = fscrypt_inherit_context(dir, inode, handle, true); 1151aa1dca3bSEric Biggers if (err) 1152aa1dca3bSEric Biggers goto fail_free_drop; 1153aa1dca3bSEric Biggers } 1154aa1dca3bSEric Biggers 11551b917ed8STahsin Erdogan if (!(ei->i_flags & EXT4_EA_INODE_FL)) { 1156617ba13bSMingming Cao err = ext4_init_acl(handle, inode, dir); 1157ac27a0ecSDave Kleikamp if (err) 1158ac27a0ecSDave Kleikamp goto fail_free_drop; 1159ac27a0ecSDave Kleikamp 11602a7dba39SEric Paris err = ext4_init_security(handle, inode, dir, qstr); 1161ac27a0ecSDave Kleikamp if (err) 1162ac27a0ecSDave Kleikamp goto fail_free_drop; 1163ad47f953STahsin Erdogan } 1164ac27a0ecSDave Kleikamp 1165e2b911c5SDarrick J. Wong if (ext4_has_feature_extents(sb)) { 1166e4079a11SEric Sandeen /* set extent flag only for directory, file and normal symlink*/ 1167e65187e6SAneesh Kumar K.V if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) { 116812e9b892SDmitry Monakhov ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS); 1169a86c6181SAlex Tomas ext4_ext_tree_init(handle, inode); 1170a86c6181SAlex Tomas } 117142bf0383SAneesh Kumar K.V } 1172ac27a0ecSDave Kleikamp 1173688f869cSTheodore Ts'o if (ext4_handle_valid(handle)) { 1174688f869cSTheodore Ts'o ei->i_sync_tid = handle->h_transaction->t_tid; 1175688f869cSTheodore Ts'o ei->i_datasync_tid = handle->h_transaction->t_tid; 1176688f869cSTheodore Ts'o } 1177688f869cSTheodore Ts'o 11788753e88fSAneesh Kumar K.V err = ext4_mark_inode_dirty(handle, inode); 11798753e88fSAneesh Kumar K.V if (err) { 11808753e88fSAneesh Kumar K.V ext4_std_error(sb, err); 11818753e88fSAneesh Kumar K.V goto fail_free_drop; 11828753e88fSAneesh Kumar K.V } 11838753e88fSAneesh Kumar K.V 1184617ba13bSMingming Cao ext4_debug("allocating inode %lu\n", inode->i_ino); 11859bffad1eSTheodore Ts'o trace_ext4_allocate_inode(inode, dir, mode); 11863300bedaSAneesh Kumar K.V brelse(inode_bitmap_bh); 1187ac27a0ecSDave Kleikamp return ret; 1188ac27a0ecSDave Kleikamp 1189ac27a0ecSDave Kleikamp fail_free_drop: 119063936ddaSChristoph Hellwig dquot_free_inode(inode); 1191ac27a0ecSDave Kleikamp fail_drop: 11926d6b77f1SMiklos Szeredi clear_nlink(inode); 11936b38e842SAl Viro unlock_new_inode(inode); 1194eb9cc7e1SJan Kara out: 1195eb9cc7e1SJan Kara dquot_drop(inode); 1196eb9cc7e1SJan Kara inode->i_flags |= S_NOQUOTA; 1197ac27a0ecSDave Kleikamp iput(inode); 11983300bedaSAneesh Kumar K.V brelse(inode_bitmap_bh); 1199ac27a0ecSDave Kleikamp return ERR_PTR(err); 1200ac27a0ecSDave Kleikamp } 1201ac27a0ecSDave Kleikamp 1202ac27a0ecSDave Kleikamp /* Verify that we are loading a valid orphan from disk */ 1203617ba13bSMingming Cao struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino) 1204ac27a0ecSDave Kleikamp { 1205617ba13bSMingming Cao unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count); 1206fd2d4291SAvantika Mathur ext4_group_t block_group; 1207ac27a0ecSDave Kleikamp int bit; 12087827a7f6STheodore Ts'o struct buffer_head *bitmap_bh = NULL; 1209ac27a0ecSDave Kleikamp struct inode *inode = NULL; 12107827a7f6STheodore Ts'o int err = -EFSCORRUPTED; 1211ac27a0ecSDave Kleikamp 12127827a7f6STheodore Ts'o if (ino < EXT4_FIRST_INO(sb) || ino > max_ino) 12137827a7f6STheodore Ts'o goto bad_orphan; 1214ac27a0ecSDave Kleikamp 1215617ba13bSMingming Cao block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); 1216617ba13bSMingming Cao bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); 1217e29d1cdeSEric Sandeen bitmap_bh = ext4_read_inode_bitmap(sb, block_group); 1218206f6d55SWang Shilong if (IS_ERR(bitmap_bh)) 1219e884bce1SAl Viro return ERR_CAST(bitmap_bh); 1220ac27a0ecSDave Kleikamp 1221ac27a0ecSDave Kleikamp /* Having the inode bit set should be a 100% indicator that this 1222ac27a0ecSDave Kleikamp * is a valid orphan (no e2fsck run on fs). Orphans also include 1223ac27a0ecSDave Kleikamp * inodes that were being truncated, so we can't check i_nlink==0. 1224ac27a0ecSDave Kleikamp */ 12251d1fe1eeSDavid Howells if (!ext4_test_bit(bit, bitmap_bh->b_data)) 12261d1fe1eeSDavid Howells goto bad_orphan; 12271d1fe1eeSDavid Howells 12288a363970STheodore Ts'o inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL); 12297827a7f6STheodore Ts'o if (IS_ERR(inode)) { 12307827a7f6STheodore Ts'o err = PTR_ERR(inode); 12317827a7f6STheodore Ts'o ext4_error(sb, "couldn't read orphan inode %lu (err %d)", 12327827a7f6STheodore Ts'o ino, err); 12337827a7f6STheodore Ts'o return inode; 12347827a7f6STheodore Ts'o } 12351d1fe1eeSDavid Howells 123691ef4cafSDuane Griffin /* 1237c9eb13a9STheodore Ts'o * If the orphans has i_nlinks > 0 then it should be able to 1238c9eb13a9STheodore Ts'o * be truncated, otherwise it won't be removed from the orphan 1239c9eb13a9STheodore Ts'o * list during processing and an infinite loop will result. 1240c9eb13a9STheodore Ts'o * Similarly, it must not be a bad inode. 124191ef4cafSDuane Griffin */ 1242c9eb13a9STheodore Ts'o if ((inode->i_nlink && !ext4_can_truncate(inode)) || 1243c9eb13a9STheodore Ts'o is_bad_inode(inode)) 124491ef4cafSDuane Griffin goto bad_orphan; 124591ef4cafSDuane Griffin 12461d1fe1eeSDavid Howells if (NEXT_ORPHAN(inode) > max_ino) 12471d1fe1eeSDavid Howells goto bad_orphan; 12481d1fe1eeSDavid Howells brelse(bitmap_bh); 12491d1fe1eeSDavid Howells return inode; 12501d1fe1eeSDavid Howells 12511d1fe1eeSDavid Howells bad_orphan: 12527827a7f6STheodore Ts'o ext4_error(sb, "bad orphan inode %lu", ino); 12537827a7f6STheodore Ts'o if (bitmap_bh) 12547827a7f6STheodore Ts'o printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n", 1255ac27a0ecSDave Kleikamp bit, (unsigned long long)bitmap_bh->b_blocknr, 1256617ba13bSMingming Cao ext4_test_bit(bit, bitmap_bh->b_data)); 1257ac27a0ecSDave Kleikamp if (inode) { 12587827a7f6STheodore Ts'o printk(KERN_ERR "is_bad_inode(inode)=%d\n", 1259ac27a0ecSDave Kleikamp is_bad_inode(inode)); 12607827a7f6STheodore Ts'o printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n", 1261ac27a0ecSDave Kleikamp NEXT_ORPHAN(inode)); 12627827a7f6STheodore Ts'o printk(KERN_ERR "max_ino=%lu\n", max_ino); 12637827a7f6STheodore Ts'o printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink); 1264ac27a0ecSDave Kleikamp /* Avoid freeing blocks if we got a bad deleted inode */ 12651d1fe1eeSDavid Howells if (inode->i_nlink == 0) 1266ac27a0ecSDave Kleikamp inode->i_blocks = 0; 1267ac27a0ecSDave Kleikamp iput(inode); 1268ac27a0ecSDave Kleikamp } 1269ac27a0ecSDave Kleikamp brelse(bitmap_bh); 12701d1fe1eeSDavid Howells return ERR_PTR(err); 1271ac27a0ecSDave Kleikamp } 1272ac27a0ecSDave Kleikamp 1273617ba13bSMingming Cao unsigned long ext4_count_free_inodes(struct super_block *sb) 1274ac27a0ecSDave Kleikamp { 1275ac27a0ecSDave Kleikamp unsigned long desc_count; 1276617ba13bSMingming Cao struct ext4_group_desc *gdp; 12778df9675fSTheodore Ts'o ext4_group_t i, ngroups = ext4_get_groups_count(sb); 1278617ba13bSMingming Cao #ifdef EXT4FS_DEBUG 1279617ba13bSMingming Cao struct ext4_super_block *es; 1280ac27a0ecSDave Kleikamp unsigned long bitmap_count, x; 1281ac27a0ecSDave Kleikamp struct buffer_head *bitmap_bh = NULL; 1282ac27a0ecSDave Kleikamp 1283617ba13bSMingming Cao es = EXT4_SB(sb)->s_es; 1284ac27a0ecSDave Kleikamp desc_count = 0; 1285ac27a0ecSDave Kleikamp bitmap_count = 0; 1286ac27a0ecSDave Kleikamp gdp = NULL; 12878df9675fSTheodore Ts'o for (i = 0; i < ngroups; i++) { 1288617ba13bSMingming Cao gdp = ext4_get_group_desc(sb, i, NULL); 1289ac27a0ecSDave Kleikamp if (!gdp) 1290ac27a0ecSDave Kleikamp continue; 1291560671a0SAneesh Kumar K.V desc_count += ext4_free_inodes_count(sb, gdp); 1292ac27a0ecSDave Kleikamp brelse(bitmap_bh); 1293e29d1cdeSEric Sandeen bitmap_bh = ext4_read_inode_bitmap(sb, i); 12949008a58eSDarrick J. Wong if (IS_ERR(bitmap_bh)) { 12959008a58eSDarrick J. Wong bitmap_bh = NULL; 1296ac27a0ecSDave Kleikamp continue; 12979008a58eSDarrick J. Wong } 1298ac27a0ecSDave Kleikamp 1299f6fb99caSTheodore Ts'o x = ext4_count_free(bitmap_bh->b_data, 1300f6fb99caSTheodore Ts'o EXT4_INODES_PER_GROUP(sb) / 8); 1301c549a95dSEric Sandeen printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n", 1302785b4b3aSPeng Tao (unsigned long) i, ext4_free_inodes_count(sb, gdp), x); 1303ac27a0ecSDave Kleikamp bitmap_count += x; 1304ac27a0ecSDave Kleikamp } 1305ac27a0ecSDave Kleikamp brelse(bitmap_bh); 13064776004fSTheodore Ts'o printk(KERN_DEBUG "ext4_count_free_inodes: " 13074776004fSTheodore Ts'o "stored = %u, computed = %lu, %lu\n", 1308ac27a0ecSDave Kleikamp le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count); 1309ac27a0ecSDave Kleikamp return desc_count; 1310ac27a0ecSDave Kleikamp #else 1311ac27a0ecSDave Kleikamp desc_count = 0; 13128df9675fSTheodore Ts'o for (i = 0; i < ngroups; i++) { 1313617ba13bSMingming Cao gdp = ext4_get_group_desc(sb, i, NULL); 1314ac27a0ecSDave Kleikamp if (!gdp) 1315ac27a0ecSDave Kleikamp continue; 1316560671a0SAneesh Kumar K.V desc_count += ext4_free_inodes_count(sb, gdp); 1317ac27a0ecSDave Kleikamp cond_resched(); 1318ac27a0ecSDave Kleikamp } 1319ac27a0ecSDave Kleikamp return desc_count; 1320ac27a0ecSDave Kleikamp #endif 1321ac27a0ecSDave Kleikamp } 1322ac27a0ecSDave Kleikamp 1323ac27a0ecSDave Kleikamp /* Called at mount-time, super-block is locked */ 1324617ba13bSMingming Cao unsigned long ext4_count_dirs(struct super_block * sb) 1325ac27a0ecSDave Kleikamp { 1326ac27a0ecSDave Kleikamp unsigned long count = 0; 13278df9675fSTheodore Ts'o ext4_group_t i, ngroups = ext4_get_groups_count(sb); 1328ac27a0ecSDave Kleikamp 13298df9675fSTheodore Ts'o for (i = 0; i < ngroups; i++) { 1330617ba13bSMingming Cao struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); 1331ac27a0ecSDave Kleikamp if (!gdp) 1332ac27a0ecSDave Kleikamp continue; 1333560671a0SAneesh Kumar K.V count += ext4_used_dirs_count(sb, gdp); 1334ac27a0ecSDave Kleikamp } 1335ac27a0ecSDave Kleikamp return count; 1336ac27a0ecSDave Kleikamp } 1337bfff6873SLukas Czerner 1338bfff6873SLukas Czerner /* 1339bfff6873SLukas Czerner * Zeroes not yet zeroed inode table - just write zeroes through the whole 1340bfff6873SLukas Czerner * inode table. Must be called without any spinlock held. The only place 1341bfff6873SLukas Czerner * where it is called from on active part of filesystem is ext4lazyinit 1342bfff6873SLukas Czerner * thread, so we do not need any special locks, however we have to prevent 1343bfff6873SLukas Czerner * inode allocation from the current group, so we take alloc_sem lock, to 1344119c0d44STheodore Ts'o * block ext4_new_inode() until we are finished. 1345bfff6873SLukas Czerner */ 1346e0cbee3eSH Hartley Sweeten int ext4_init_inode_table(struct super_block *sb, ext4_group_t group, 1347bfff6873SLukas Czerner int barrier) 1348bfff6873SLukas Czerner { 1349bfff6873SLukas Czerner struct ext4_group_info *grp = ext4_get_group_info(sb, group); 1350bfff6873SLukas Czerner struct ext4_sb_info *sbi = EXT4_SB(sb); 1351bfff6873SLukas Czerner struct ext4_group_desc *gdp = NULL; 1352bfff6873SLukas Czerner struct buffer_head *group_desc_bh; 1353bfff6873SLukas Czerner handle_t *handle; 1354bfff6873SLukas Czerner ext4_fsblk_t blk; 1355bfff6873SLukas Czerner int num, ret = 0, used_blks = 0; 1356bfff6873SLukas Czerner 1357bfff6873SLukas Czerner /* This should not happen, but just to be sure check this */ 1358bc98a42cSDavid Howells if (sb_rdonly(sb)) { 1359bfff6873SLukas Czerner ret = 1; 1360bfff6873SLukas Czerner goto out; 1361bfff6873SLukas Czerner } 1362bfff6873SLukas Czerner 1363bfff6873SLukas Czerner gdp = ext4_get_group_desc(sb, group, &group_desc_bh); 1364bfff6873SLukas Czerner if (!gdp) 1365bfff6873SLukas Czerner goto out; 1366bfff6873SLukas Czerner 1367bfff6873SLukas Czerner /* 1368bfff6873SLukas Czerner * We do not need to lock this, because we are the only one 1369bfff6873SLukas Czerner * handling this flag. 1370bfff6873SLukas Czerner */ 1371bfff6873SLukas Czerner if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)) 1372bfff6873SLukas Czerner goto out; 1373bfff6873SLukas Czerner 13749924a92aSTheodore Ts'o handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1); 1375bfff6873SLukas Czerner if (IS_ERR(handle)) { 1376bfff6873SLukas Czerner ret = PTR_ERR(handle); 1377bfff6873SLukas Czerner goto out; 1378bfff6873SLukas Czerner } 1379bfff6873SLukas Czerner 1380bfff6873SLukas Czerner down_write(&grp->alloc_sem); 1381bfff6873SLukas Czerner /* 1382bfff6873SLukas Czerner * If inode bitmap was already initialized there may be some 1383bfff6873SLukas Czerner * used inodes so we need to skip blocks with used inodes in 1384bfff6873SLukas Czerner * inode table. 1385bfff6873SLukas Czerner */ 1386bfff6873SLukas Czerner if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) 1387bfff6873SLukas Czerner used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) - 1388bfff6873SLukas Czerner ext4_itable_unused_count(sb, gdp)), 1389bfff6873SLukas Czerner sbi->s_inodes_per_block); 1390bfff6873SLukas Czerner 139150122847STheodore Ts'o if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group) || 139250122847STheodore Ts'o ((group == 0) && ((EXT4_INODES_PER_GROUP(sb) - 139350122847STheodore Ts'o ext4_itable_unused_count(sb, gdp)) < 139450122847STheodore Ts'o EXT4_FIRST_INO(sb)))) { 13951084f252STheodore Ts'o ext4_error(sb, "Something is wrong with group %u: " 13961084f252STheodore Ts'o "used itable blocks: %d; " 13971084f252STheodore Ts'o "itable unused count: %u", 1398857ac889SLukas Czerner group, used_blks, 1399857ac889SLukas Czerner ext4_itable_unused_count(sb, gdp)); 1400857ac889SLukas Czerner ret = 1; 140133853a0dSYongqiang Yang goto err_out; 1402857ac889SLukas Czerner } 1403857ac889SLukas Czerner 1404bfff6873SLukas Czerner blk = ext4_inode_table(sb, gdp) + used_blks; 1405bfff6873SLukas Czerner num = sbi->s_itb_per_group - used_blks; 1406bfff6873SLukas Czerner 1407bfff6873SLukas Czerner BUFFER_TRACE(group_desc_bh, "get_write_access"); 1408bfff6873SLukas Czerner ret = ext4_journal_get_write_access(handle, 1409bfff6873SLukas Czerner group_desc_bh); 1410bfff6873SLukas Czerner if (ret) 1411bfff6873SLukas Czerner goto err_out; 1412bfff6873SLukas Czerner 1413bfff6873SLukas Czerner /* 1414bfff6873SLukas Czerner * Skip zeroout if the inode table is full. But we set the ZEROED 1415bfff6873SLukas Czerner * flag anyway, because obviously, when it is full it does not need 1416bfff6873SLukas Czerner * further zeroing. 1417bfff6873SLukas Czerner */ 1418bfff6873SLukas Czerner if (unlikely(num == 0)) 1419bfff6873SLukas Czerner goto skip_zeroout; 1420bfff6873SLukas Czerner 1421bfff6873SLukas Czerner ext4_debug("going to zero out inode table in group %d\n", 1422bfff6873SLukas Czerner group); 1423a107e5a3STheodore Ts'o ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS); 1424bfff6873SLukas Czerner if (ret < 0) 1425bfff6873SLukas Czerner goto err_out; 1426a107e5a3STheodore Ts'o if (barrier) 1427a107e5a3STheodore Ts'o blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL); 1428bfff6873SLukas Czerner 1429bfff6873SLukas Czerner skip_zeroout: 1430bfff6873SLukas Czerner ext4_lock_group(sb, group); 1431bfff6873SLukas Czerner gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED); 1432feb0ab32SDarrick J. Wong ext4_group_desc_csum_set(sb, group, gdp); 1433bfff6873SLukas Czerner ext4_unlock_group(sb, group); 1434bfff6873SLukas Czerner 1435bfff6873SLukas Czerner BUFFER_TRACE(group_desc_bh, 1436bfff6873SLukas Czerner "call ext4_handle_dirty_metadata"); 1437bfff6873SLukas Czerner ret = ext4_handle_dirty_metadata(handle, NULL, 1438bfff6873SLukas Czerner group_desc_bh); 1439bfff6873SLukas Czerner 1440bfff6873SLukas Czerner err_out: 1441bfff6873SLukas Czerner up_write(&grp->alloc_sem); 1442bfff6873SLukas Czerner ext4_journal_stop(handle); 1443bfff6873SLukas Czerner out: 1444bfff6873SLukas Czerner return ret; 1445bfff6873SLukas Czerner } 1446