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, 9746f870d6STheodore Ts'o EXT4_INODES_PER_GROUP(sb) / 8) || 9846f870d6STheodore Ts'o ext4_simulate_fail(sb, EXT4_SIM_IBITMAP_CRC)) { 999008a58eSDarrick J. Wong ext4_unlock_group(sb, block_group); 1009008a58eSDarrick J. Wong ext4_error(sb, "Corrupt inode bitmap - block_group = %u, " 1019008a58eSDarrick J. Wong "inode_bitmap = %llu", block_group, blk); 102db79e6d1SWang Shilong ext4_mark_group_bitmap_corrupted(sb, block_group, 103db79e6d1SWang Shilong EXT4_GROUP_INFO_IBITMAP_CORRUPT); 1049008a58eSDarrick J. Wong return -EFSBADCRC; 1059008a58eSDarrick J. Wong } 1069008a58eSDarrick J. Wong set_buffer_verified(bh); 1078d5a803cSTheodore Ts'o verified: 1089008a58eSDarrick J. Wong ext4_unlock_group(sb, block_group); 1099008a58eSDarrick J. Wong return 0; 1109008a58eSDarrick J. Wong } 1119008a58eSDarrick J. Wong 112ac27a0ecSDave Kleikamp /* 113ac27a0ecSDave Kleikamp * Read the inode allocation bitmap for a given block_group, reading 114ac27a0ecSDave Kleikamp * into the specified slot in the superblock's bitmap cache. 115ac27a0ecSDave Kleikamp * 1169033783cSJosh Triplett * Return buffer_head of bitmap on success, or an ERR_PTR on error. 117ac27a0ecSDave Kleikamp */ 118ac27a0ecSDave Kleikamp static struct buffer_head * 119e29d1cdeSEric Sandeen ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group) 120ac27a0ecSDave Kleikamp { 121617ba13bSMingming Cao struct ext4_group_desc *desc; 1227dac4a17STheodore Ts'o struct ext4_sb_info *sbi = EXT4_SB(sb); 123ac27a0ecSDave Kleikamp struct buffer_head *bh = NULL; 124e29d1cdeSEric Sandeen ext4_fsblk_t bitmap_blk; 1259008a58eSDarrick J. Wong int err; 126ac27a0ecSDave Kleikamp 127617ba13bSMingming Cao desc = ext4_get_group_desc(sb, block_group, NULL); 128ac27a0ecSDave Kleikamp if (!desc) 1299008a58eSDarrick J. Wong return ERR_PTR(-EFSCORRUPTED); 130bfff6873SLukas Czerner 131e29d1cdeSEric Sandeen bitmap_blk = ext4_inode_bitmap(sb, desc); 1327dac4a17STheodore Ts'o if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) || 1337dac4a17STheodore Ts'o (bitmap_blk >= ext4_blocks_count(sbi->s_es))) { 1347dac4a17STheodore Ts'o ext4_error(sb, "Invalid inode bitmap blk %llu in " 1357dac4a17STheodore Ts'o "block_group %u", bitmap_blk, block_group); 136206f6d55SWang Shilong ext4_mark_group_bitmap_corrupted(sb, block_group, 137206f6d55SWang Shilong EXT4_GROUP_INFO_IBITMAP_CORRUPT); 1387dac4a17STheodore Ts'o return ERR_PTR(-EFSCORRUPTED); 1397dac4a17STheodore Ts'o } 140e29d1cdeSEric Sandeen bh = sb_getblk(sb, bitmap_blk); 141e29d1cdeSEric Sandeen if (unlikely(!bh)) { 1425ef2a699SWang Shilong ext4_warning(sb, "Cannot read inode bitmap - " 143a9df9a49STheodore Ts'o "block_group = %u, inode_bitmap = %llu", 144e29d1cdeSEric Sandeen block_group, bitmap_blk); 1450db9fdebSWang Shilong return ERR_PTR(-ENOMEM); 146e29d1cdeSEric Sandeen } 1472ccb5fb9SAneesh Kumar K.V if (bitmap_uptodate(bh)) 14841a246d1SDarrick J. Wong goto verify; 149e29d1cdeSEric Sandeen 150c806e68fSFrederic Bohe lock_buffer(bh); 1512ccb5fb9SAneesh Kumar K.V if (bitmap_uptodate(bh)) { 1522ccb5fb9SAneesh Kumar K.V unlock_buffer(bh); 15341a246d1SDarrick J. Wong goto verify; 1542ccb5fb9SAneesh Kumar K.V } 155bfff6873SLukas Czerner 156955ce5f5SAneesh Kumar K.V ext4_lock_group(sb, block_group); 1578844618dSTheodore Ts'o if (ext4_has_group_desc_csum(sb) && 1588844618dSTheodore Ts'o (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) { 1598844618dSTheodore Ts'o if (block_group == 0) { 1608844618dSTheodore Ts'o ext4_unlock_group(sb, block_group); 1618844618dSTheodore Ts'o unlock_buffer(bh); 1628844618dSTheodore Ts'o ext4_error(sb, "Inode bitmap for bg 0 marked " 1638844618dSTheodore Ts'o "uninitialized"); 1648844618dSTheodore Ts'o err = -EFSCORRUPTED; 1658844618dSTheodore Ts'o goto out; 1668844618dSTheodore Ts'o } 167044e6e3dSTheodore Ts'o memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8); 168044e6e3dSTheodore Ts'o ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), 169044e6e3dSTheodore Ts'o sb->s_blocksize * 8, bh->b_data); 1702ccb5fb9SAneesh Kumar K.V set_bitmap_uptodate(bh); 171e29d1cdeSEric Sandeen set_buffer_uptodate(bh); 17241a246d1SDarrick J. Wong set_buffer_verified(bh); 173955ce5f5SAneesh Kumar K.V ext4_unlock_group(sb, block_group); 1743300bedaSAneesh Kumar K.V unlock_buffer(bh); 175e29d1cdeSEric Sandeen return bh; 176e29d1cdeSEric Sandeen } 177955ce5f5SAneesh Kumar K.V ext4_unlock_group(sb, block_group); 178bfff6873SLukas Czerner 1792ccb5fb9SAneesh Kumar K.V if (buffer_uptodate(bh)) { 1802ccb5fb9SAneesh Kumar K.V /* 1812ccb5fb9SAneesh Kumar K.V * if not uninit if bh is uptodate, 1822ccb5fb9SAneesh Kumar K.V * bitmap is also uptodate 1832ccb5fb9SAneesh Kumar K.V */ 1842ccb5fb9SAneesh Kumar K.V set_bitmap_uptodate(bh); 1852ccb5fb9SAneesh Kumar K.V unlock_buffer(bh); 18641a246d1SDarrick J. Wong goto verify; 1872ccb5fb9SAneesh Kumar K.V } 1882ccb5fb9SAneesh Kumar K.V /* 189813e5727STheodore Ts'o * submit the buffer_head for reading 1902ccb5fb9SAneesh Kumar K.V */ 1910562e0baSJiaying Zhang trace_ext4_load_inode_bitmap(sb, block_group); 192813e5727STheodore Ts'o bh->b_end_io = ext4_end_bitmap_read; 193813e5727STheodore Ts'o get_bh(bh); 1942a222ca9SMike Christie submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh); 195813e5727STheodore Ts'o wait_on_buffer(bh); 19646f870d6STheodore Ts'o ext4_simulate_fail_bh(sb, bh, EXT4_SIM_IBITMAP_EIO); 197813e5727STheodore Ts'o if (!buffer_uptodate(bh)) { 198e29d1cdeSEric Sandeen put_bh(bh); 19954d3adbcSTheodore Ts'o ext4_error_err(sb, EIO, "Cannot read inode bitmap - " 200a9df9a49STheodore Ts'o "block_group = %u, inode_bitmap = %llu", 201e29d1cdeSEric Sandeen block_group, bitmap_blk); 202206f6d55SWang Shilong ext4_mark_group_bitmap_corrupted(sb, block_group, 203206f6d55SWang Shilong EXT4_GROUP_INFO_IBITMAP_CORRUPT); 2049008a58eSDarrick J. Wong return ERR_PTR(-EIO); 205e29d1cdeSEric Sandeen } 20641a246d1SDarrick J. Wong 20741a246d1SDarrick J. Wong verify: 2089008a58eSDarrick J. Wong err = ext4_validate_inode_bitmap(sb, desc, block_group, bh); 2099008a58eSDarrick J. Wong if (err) 2109008a58eSDarrick J. Wong goto out; 211ac27a0ecSDave Kleikamp return bh; 2129008a58eSDarrick J. Wong out: 2139008a58eSDarrick J. Wong put_bh(bh); 2149008a58eSDarrick J. Wong return ERR_PTR(err); 215ac27a0ecSDave Kleikamp } 216ac27a0ecSDave Kleikamp 217ac27a0ecSDave Kleikamp /* 218ac27a0ecSDave Kleikamp * NOTE! When we get the inode, we're the only people 219ac27a0ecSDave Kleikamp * that have access to it, and as such there are no 220ac27a0ecSDave Kleikamp * race conditions we have to worry about. The inode 221ac27a0ecSDave Kleikamp * is not on the hash-lists, and it cannot be reached 222ac27a0ecSDave Kleikamp * through the filesystem because the directory entry 223ac27a0ecSDave Kleikamp * has been deleted earlier. 224ac27a0ecSDave Kleikamp * 225ac27a0ecSDave Kleikamp * HOWEVER: we must make sure that we get no aliases, 226ac27a0ecSDave Kleikamp * which means that we have to call "clear_inode()" 227ac27a0ecSDave Kleikamp * _before_ we mark the inode not in use in the inode 228ac27a0ecSDave Kleikamp * bitmaps. Otherwise a newly created file might use 229ac27a0ecSDave Kleikamp * the same inode number (not actually the same pointer 230ac27a0ecSDave Kleikamp * though), and then we'd have two inodes sharing the 231ac27a0ecSDave Kleikamp * same inode number and space on the harddisk. 232ac27a0ecSDave Kleikamp */ 233617ba13bSMingming Cao void ext4_free_inode(handle_t *handle, struct inode *inode) 234ac27a0ecSDave Kleikamp { 235ac27a0ecSDave Kleikamp struct super_block *sb = inode->i_sb; 236ac27a0ecSDave Kleikamp int is_directory; 237ac27a0ecSDave Kleikamp unsigned long ino; 238ac27a0ecSDave Kleikamp struct buffer_head *bitmap_bh = NULL; 239ac27a0ecSDave Kleikamp struct buffer_head *bh2; 240fd2d4291SAvantika Mathur ext4_group_t block_group; 241ac27a0ecSDave Kleikamp unsigned long bit; 242617ba13bSMingming Cao struct ext4_group_desc *gdp; 243617ba13bSMingming Cao struct ext4_super_block *es; 244617ba13bSMingming Cao struct ext4_sb_info *sbi; 2457ce9d5d1SEric Sandeen int fatal = 0, err, count, cleared; 24687a39389SDarrick J. Wong struct ext4_group_info *grp; 247ac27a0ecSDave Kleikamp 24892b97816STheodore Ts'o if (!sb) { 24992b97816STheodore Ts'o printk(KERN_ERR "EXT4-fs: %s:%d: inode on " 25092b97816STheodore Ts'o "nonexistent device\n", __func__, __LINE__); 25192b97816STheodore Ts'o return; 25292b97816STheodore Ts'o } 253ac27a0ecSDave Kleikamp if (atomic_read(&inode->i_count) > 1) { 25492b97816STheodore Ts'o ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d", 25592b97816STheodore Ts'o __func__, __LINE__, inode->i_ino, 256ac27a0ecSDave Kleikamp atomic_read(&inode->i_count)); 257ac27a0ecSDave Kleikamp return; 258ac27a0ecSDave Kleikamp } 259ac27a0ecSDave Kleikamp if (inode->i_nlink) { 26092b97816STheodore Ts'o ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n", 26192b97816STheodore Ts'o __func__, __LINE__, inode->i_ino, inode->i_nlink); 262ac27a0ecSDave Kleikamp return; 263ac27a0ecSDave Kleikamp } 264617ba13bSMingming Cao sbi = EXT4_SB(sb); 265ac27a0ecSDave Kleikamp 266ac27a0ecSDave Kleikamp ino = inode->i_ino; 267617ba13bSMingming Cao ext4_debug("freeing inode %lu\n", ino); 2689bffad1eSTheodore Ts'o trace_ext4_free_inode(inode); 269ac27a0ecSDave Kleikamp 270871a2931SChristoph Hellwig dquot_initialize(inode); 27163936ddaSChristoph Hellwig dquot_free_inode(inode); 272ac27a0ecSDave Kleikamp 273ac27a0ecSDave Kleikamp is_directory = S_ISDIR(inode->i_mode); 274ac27a0ecSDave Kleikamp 275ac27a0ecSDave Kleikamp /* Do this BEFORE marking the inode not in use or returning an error */ 2760930fcc1SAl Viro ext4_clear_inode(inode); 277ac27a0ecSDave Kleikamp 27849598e04SJun Piao es = sbi->s_es; 279617ba13bSMingming Cao if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) { 28012062dddSEric Sandeen ext4_error(sb, "reserved or nonexistent inode %lu", ino); 281ac27a0ecSDave Kleikamp goto error_return; 282ac27a0ecSDave Kleikamp } 283617ba13bSMingming Cao block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); 284617ba13bSMingming Cao bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); 285e29d1cdeSEric Sandeen bitmap_bh = ext4_read_inode_bitmap(sb, block_group); 28687a39389SDarrick J. Wong /* Don't bother if the inode bitmap is corrupt. */ 28787a39389SDarrick J. Wong grp = ext4_get_group_info(sb, block_group); 2889008a58eSDarrick J. Wong if (IS_ERR(bitmap_bh)) { 2899008a58eSDarrick J. Wong fatal = PTR_ERR(bitmap_bh); 2909008a58eSDarrick J. Wong bitmap_bh = NULL; 291ac27a0ecSDave Kleikamp goto error_return; 2929008a58eSDarrick J. Wong } 2939008a58eSDarrick J. Wong if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) { 2949008a58eSDarrick J. Wong fatal = -EFSCORRUPTED; 2959008a58eSDarrick J. Wong goto error_return; 2969008a58eSDarrick J. Wong } 297ac27a0ecSDave Kleikamp 298ac27a0ecSDave Kleikamp BUFFER_TRACE(bitmap_bh, "get_write_access"); 299617ba13bSMingming Cao fatal = ext4_journal_get_write_access(handle, bitmap_bh); 300ac27a0ecSDave Kleikamp if (fatal) 301ac27a0ecSDave Kleikamp goto error_return; 302ac27a0ecSDave Kleikamp 303d17413c0SDmitry Monakhov fatal = -ESRCH; 304617ba13bSMingming Cao gdp = ext4_get_group_desc(sb, block_group, &bh2); 305d17413c0SDmitry Monakhov if (gdp) { 306ac27a0ecSDave Kleikamp BUFFER_TRACE(bh2, "get_write_access"); 307617ba13bSMingming Cao fatal = ext4_journal_get_write_access(handle, bh2); 308d17413c0SDmitry Monakhov } 309955ce5f5SAneesh Kumar K.V ext4_lock_group(sb, block_group); 310597d508cSAkinobu Mita cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data); 311d17413c0SDmitry Monakhov if (fatal || !cleared) { 312d17413c0SDmitry Monakhov ext4_unlock_group(sb, block_group); 313d17413c0SDmitry Monakhov goto out; 314d17413c0SDmitry Monakhov } 315d17413c0SDmitry Monakhov 316560671a0SAneesh Kumar K.V count = ext4_free_inodes_count(sb, gdp) + 1; 317560671a0SAneesh Kumar K.V ext4_free_inodes_set(sb, gdp, count); 318560671a0SAneesh Kumar K.V if (is_directory) { 319560671a0SAneesh Kumar K.V count = ext4_used_dirs_count(sb, gdp) - 1; 320560671a0SAneesh Kumar K.V ext4_used_dirs_set(sb, gdp, count); 321d17413c0SDmitry Monakhov percpu_counter_dec(&sbi->s_dirs_counter); 322d17413c0SDmitry Monakhov } 32341a246d1SDarrick J. Wong ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh, 32441a246d1SDarrick J. Wong EXT4_INODES_PER_GROUP(sb) / 8); 325feb0ab32SDarrick J. Wong ext4_group_desc_csum_set(sb, block_group, gdp); 326d17413c0SDmitry Monakhov ext4_unlock_group(sb, block_group); 3277d39db14STheodore Ts'o 328d17413c0SDmitry Monakhov percpu_counter_inc(&sbi->s_freeinodes_counter); 329d17413c0SDmitry Monakhov if (sbi->s_log_groups_per_flex) { 3307c990728SSuraj Jitindar Singh struct flex_groups *fg; 331d17413c0SDmitry Monakhov 3327c990728SSuraj Jitindar Singh fg = sbi_array_rcu_deref(sbi, s_flex_groups, 3337c990728SSuraj Jitindar Singh ext4_flex_group(sbi, block_group)); 3347c990728SSuraj Jitindar Singh atomic_inc(&fg->free_inodes); 335d17413c0SDmitry Monakhov if (is_directory) 3367c990728SSuraj Jitindar Singh atomic_dec(&fg->used_dirs); 3377d39db14STheodore Ts'o } 3380390131bSFrank Mayhar BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata"); 339d17413c0SDmitry Monakhov fatal = ext4_handle_dirty_metadata(handle, NULL, bh2); 340d17413c0SDmitry Monakhov out: 341d17413c0SDmitry Monakhov if (cleared) { 3420390131bSFrank Mayhar BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata"); 3430390131bSFrank Mayhar err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh); 344ac27a0ecSDave Kleikamp if (!fatal) 345ac27a0ecSDave Kleikamp fatal = err; 34687a39389SDarrick J. Wong } else { 347d17413c0SDmitry Monakhov ext4_error(sb, "bit already cleared for inode %lu", ino); 348db79e6d1SWang Shilong ext4_mark_group_bitmap_corrupted(sb, block_group, 349db79e6d1SWang Shilong EXT4_GROUP_INFO_IBITMAP_CORRUPT); 35087a39389SDarrick J. Wong } 351d17413c0SDmitry Monakhov 352ac27a0ecSDave Kleikamp error_return: 353ac27a0ecSDave Kleikamp brelse(bitmap_bh); 354617ba13bSMingming Cao ext4_std_error(sb, fatal); 355ac27a0ecSDave Kleikamp } 356ac27a0ecSDave Kleikamp 357a4912123STheodore Ts'o struct orlov_stats { 35890ba983fSTheodore Ts'o __u64 free_clusters; 359a4912123STheodore Ts'o __u32 free_inodes; 360a4912123STheodore Ts'o __u32 used_dirs; 361a4912123STheodore Ts'o }; 362a4912123STheodore Ts'o 363a4912123STheodore Ts'o /* 364a4912123STheodore Ts'o * Helper function for Orlov's allocator; returns critical information 365a4912123STheodore Ts'o * for a particular block group or flex_bg. If flex_size is 1, then g 366a4912123STheodore Ts'o * is a block group number; otherwise it is flex_bg number. 367a4912123STheodore Ts'o */ 3681f109d5aSTheodore Ts'o static void get_orlov_stats(struct super_block *sb, ext4_group_t g, 369a4912123STheodore Ts'o int flex_size, struct orlov_stats *stats) 370a4912123STheodore Ts'o { 371a4912123STheodore Ts'o struct ext4_group_desc *desc; 372a4912123STheodore Ts'o 3737d39db14STheodore Ts'o if (flex_size > 1) { 3747c990728SSuraj Jitindar Singh struct flex_groups *fg = sbi_array_rcu_deref(EXT4_SB(sb), 3757c990728SSuraj Jitindar Singh s_flex_groups, g); 3767c990728SSuraj Jitindar Singh stats->free_inodes = atomic_read(&fg->free_inodes); 3777c990728SSuraj Jitindar Singh stats->free_clusters = atomic64_read(&fg->free_clusters); 3787c990728SSuraj Jitindar Singh stats->used_dirs = atomic_read(&fg->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 */ 665a17a9d93STheodore Ts'o #define RECENTCY_MIN 60 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 { 714d05466b2SJan Kara bool check_recently_deleted = EXT4_SB(sb)->s_journal == NULL; 715d05466b2SJan Kara unsigned long recently_deleted_ino = EXT4_INODES_PER_GROUP(sb); 716d05466b2SJan Kara 717901ed070SWang Shilong next: 718901ed070SWang Shilong *ino = ext4_find_next_zero_bit((unsigned long *) 719901ed070SWang Shilong bitmap->b_data, 720901ed070SWang Shilong EXT4_INODES_PER_GROUP(sb), *ino); 721901ed070SWang Shilong if (*ino >= EXT4_INODES_PER_GROUP(sb)) 722d05466b2SJan Kara goto not_found; 723901ed070SWang Shilong 724d05466b2SJan Kara if (check_recently_deleted && recently_deleted(sb, group, *ino)) { 725d05466b2SJan Kara recently_deleted_ino = *ino; 726901ed070SWang Shilong *ino = *ino + 1; 727901ed070SWang Shilong if (*ino < EXT4_INODES_PER_GROUP(sb)) 728901ed070SWang Shilong goto next; 729d05466b2SJan Kara goto not_found; 730901ed070SWang Shilong } 731d05466b2SJan Kara return 1; 732d05466b2SJan Kara not_found: 733d05466b2SJan Kara if (recently_deleted_ino >= EXT4_INODES_PER_GROUP(sb)) 734d05466b2SJan Kara return 0; 735d05466b2SJan Kara /* 736d05466b2SJan Kara * Not reusing recently deleted inodes is mostly a preference. We don't 737d05466b2SJan Kara * want to report ENOSPC or skew allocation patterns because of that. 738d05466b2SJan Kara * So return even recently deleted inode if we could find better in the 739d05466b2SJan Kara * given range. 740d05466b2SJan Kara */ 741d05466b2SJan Kara *ino = recently_deleted_ino; 742901ed070SWang Shilong return 1; 743901ed070SWang Shilong } 744901ed070SWang Shilong 74519883bd9STheodore Ts'o /* 746ac27a0ecSDave Kleikamp * There are two policies for allocating an inode. If the new inode is 747ac27a0ecSDave Kleikamp * a directory, then a forward search is made for a block group with both 748ac27a0ecSDave Kleikamp * free space and a low directory-to-inode ratio; if that fails, then of 749ac27a0ecSDave Kleikamp * the groups with above-average free space, that group with the fewest 750ac27a0ecSDave Kleikamp * directories already is chosen. 751ac27a0ecSDave Kleikamp * 752ac27a0ecSDave Kleikamp * For other inodes, search forward from the parent directory's block 753ac27a0ecSDave Kleikamp * group to find a free inode. 754ac27a0ecSDave Kleikamp */ 7551139575aSTheodore Ts'o struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir, 7561139575aSTheodore Ts'o umode_t mode, const struct qstr *qstr, 7571b917ed8STahsin Erdogan __u32 goal, uid_t *owner, __u32 i_flags, 7581b917ed8STahsin Erdogan int handle_type, unsigned int line_no, 7591b917ed8STahsin Erdogan int nblocks) 760ac27a0ecSDave Kleikamp { 761ac27a0ecSDave Kleikamp struct super_block *sb; 7623300bedaSAneesh Kumar K.V struct buffer_head *inode_bitmap_bh = NULL; 7633300bedaSAneesh Kumar K.V struct buffer_head *group_desc_bh; 7648df9675fSTheodore Ts'o ext4_group_t ngroups, group = 0; 765ac27a0ecSDave Kleikamp unsigned long ino = 0; 766ac27a0ecSDave Kleikamp struct inode *inode; 767617ba13bSMingming Cao struct ext4_group_desc *gdp = NULL; 768617ba13bSMingming Cao struct ext4_inode_info *ei; 769617ba13bSMingming Cao struct ext4_sb_info *sbi; 770a7cdadeeSJan Kara int ret2, err; 771ac27a0ecSDave Kleikamp struct inode *ret; 7722aa9fc4cSAvantika Mathur ext4_group_t i; 773772cb7c8SJose R. Santos ext4_group_t flex_group; 77487a39389SDarrick J. Wong struct ext4_group_info *grp; 775e709e9dfSTheodore Ts'o int encrypt = 0; 776ac27a0ecSDave Kleikamp 777ac27a0ecSDave Kleikamp /* Cannot create files in a deleted directory */ 778ac27a0ecSDave Kleikamp if (!dir || !dir->i_nlink) 779ac27a0ecSDave Kleikamp return ERR_PTR(-EPERM); 780ac27a0ecSDave Kleikamp 781af65207cSTahsin Erdogan sb = dir->i_sb; 782af65207cSTahsin Erdogan sbi = EXT4_SB(sb); 783af65207cSTahsin Erdogan 784af65207cSTahsin Erdogan if (unlikely(ext4_forced_shutdown(sbi))) 7850db1ff22STheodore Ts'o return ERR_PTR(-EIO); 7860db1ff22STheodore Ts'o 787592ddec7SChandan Rajendra if ((IS_ENCRYPTED(dir) || DUMMY_ENCRYPTION_ENABLED(sbi)) && 788ad47f953STahsin Erdogan (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) && 789ad47f953STahsin Erdogan !(i_flags & EXT4_EA_INODE_FL)) { 790a7550b30SJaegeuk Kim err = fscrypt_get_encryption_info(dir); 791e709e9dfSTheodore Ts'o if (err) 792e709e9dfSTheodore Ts'o return ERR_PTR(err); 793a7550b30SJaegeuk Kim if (!fscrypt_has_encryption_key(dir)) 79454475f53SEric Biggers return ERR_PTR(-ENOKEY); 795e709e9dfSTheodore Ts'o encrypt = 1; 796e709e9dfSTheodore Ts'o } 797e709e9dfSTheodore Ts'o 798af65207cSTahsin Erdogan if (!handle && sbi->s_journal && !(i_flags & EXT4_EA_INODE_FL)) { 799af65207cSTahsin Erdogan #ifdef CONFIG_EXT4_FS_POSIX_ACL 800af65207cSTahsin Erdogan struct posix_acl *p = get_acl(dir, ACL_TYPE_DEFAULT); 801af65207cSTahsin Erdogan 802996fc447STheodore Ts'o if (IS_ERR(p)) 803996fc447STheodore Ts'o return ERR_CAST(p); 804af65207cSTahsin Erdogan if (p) { 805af65207cSTahsin Erdogan int acl_size = p->a_count * sizeof(ext4_acl_entry); 806af65207cSTahsin Erdogan 807af65207cSTahsin Erdogan nblocks += (S_ISDIR(mode) ? 2 : 1) * 808af65207cSTahsin Erdogan __ext4_xattr_set_credits(sb, NULL /* inode */, 809af65207cSTahsin Erdogan NULL /* block_bh */, acl_size, 810af65207cSTahsin Erdogan true /* is_create */); 811af65207cSTahsin Erdogan posix_acl_release(p); 812af65207cSTahsin Erdogan } 813af65207cSTahsin Erdogan #endif 814af65207cSTahsin Erdogan 815af65207cSTahsin Erdogan #ifdef CONFIG_SECURITY 816af65207cSTahsin Erdogan { 817af65207cSTahsin Erdogan int num_security_xattrs = 1; 818af65207cSTahsin Erdogan 819af65207cSTahsin Erdogan #ifdef CONFIG_INTEGRITY 820af65207cSTahsin Erdogan num_security_xattrs++; 821af65207cSTahsin Erdogan #endif 822af65207cSTahsin Erdogan /* 823af65207cSTahsin Erdogan * We assume that security xattrs are never 824af65207cSTahsin Erdogan * more than 1k. In practice they are under 825af65207cSTahsin Erdogan * 128 bytes. 826af65207cSTahsin Erdogan */ 827af65207cSTahsin Erdogan nblocks += num_security_xattrs * 828af65207cSTahsin Erdogan __ext4_xattr_set_credits(sb, NULL /* inode */, 829af65207cSTahsin Erdogan NULL /* block_bh */, 1024, 830af65207cSTahsin Erdogan true /* is_create */); 831af65207cSTahsin Erdogan } 832af65207cSTahsin Erdogan #endif 833af65207cSTahsin Erdogan if (encrypt) 834af65207cSTahsin Erdogan nblocks += __ext4_xattr_set_credits(sb, 835af65207cSTahsin Erdogan NULL /* inode */, NULL /* block_bh */, 836af65207cSTahsin Erdogan FSCRYPT_SET_CONTEXT_MAX_SIZE, 837af65207cSTahsin Erdogan true /* is_create */); 838af65207cSTahsin Erdogan } 839af65207cSTahsin Erdogan 8408df9675fSTheodore Ts'o ngroups = ext4_get_groups_count(sb); 8419bffad1eSTheodore Ts'o trace_ext4_request_inode(dir, mode); 842ac27a0ecSDave Kleikamp inode = new_inode(sb); 843ac27a0ecSDave Kleikamp if (!inode) 844ac27a0ecSDave Kleikamp return ERR_PTR(-ENOMEM); 845617ba13bSMingming Cao ei = EXT4_I(inode); 846772cb7c8SJose R. Santos 847eb9cc7e1SJan Kara /* 848b8a07463SAdam Buchbinder * Initialize owners and quota early so that we don't have to account 849eb9cc7e1SJan Kara * for quota initialization worst case in standard inode creating 850eb9cc7e1SJan Kara * transaction 851eb9cc7e1SJan Kara */ 852eb9cc7e1SJan Kara if (owner) { 853eb9cc7e1SJan Kara inode->i_mode = mode; 854eb9cc7e1SJan Kara i_uid_write(inode, owner[0]); 855eb9cc7e1SJan Kara i_gid_write(inode, owner[1]); 856eb9cc7e1SJan Kara } else if (test_opt(sb, GRPID)) { 857eb9cc7e1SJan Kara inode->i_mode = mode; 858eb9cc7e1SJan Kara inode->i_uid = current_fsuid(); 859eb9cc7e1SJan Kara inode->i_gid = dir->i_gid; 860eb9cc7e1SJan Kara } else 861eb9cc7e1SJan Kara inode_init_owner(inode, dir, mode); 862040cb378SLi Xi 8630b7b7779SKaho Ng if (ext4_has_feature_project(sb) && 864040cb378SLi Xi ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) 865040cb378SLi Xi ei->i_projid = EXT4_I(dir)->i_projid; 866040cb378SLi Xi else 867040cb378SLi Xi ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID); 868040cb378SLi Xi 869a7cdadeeSJan Kara err = dquot_initialize(inode); 870a7cdadeeSJan Kara if (err) 871a7cdadeeSJan Kara goto out; 872eb9cc7e1SJan Kara 87311013911SAndreas Dilger if (!goal) 87411013911SAndreas Dilger goal = sbi->s_inode_goal; 87511013911SAndreas Dilger 876e6462869SJohann Lombardi if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) { 87711013911SAndreas Dilger group = (goal - 1) / EXT4_INODES_PER_GROUP(sb); 87811013911SAndreas Dilger ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb); 87911013911SAndreas Dilger ret2 = 0; 88011013911SAndreas Dilger goto got_group; 88111013911SAndreas Dilger } 88211013911SAndreas Dilger 8834113c4caSLukas Czerner if (S_ISDIR(mode)) 884f157a4aaSTheodore Ts'o ret2 = find_group_orlov(sb, dir, &group, mode, qstr); 8854113c4caSLukas Czerner else 886a4912123STheodore Ts'o ret2 = find_group_other(sb, dir, &group, mode); 887ac27a0ecSDave Kleikamp 888772cb7c8SJose R. Santos got_group: 889a4912123STheodore Ts'o EXT4_I(dir)->i_last_alloc_group = group; 890ac27a0ecSDave Kleikamp err = -ENOSPC; 8912aa9fc4cSAvantika Mathur if (ret2 == -1) 892ac27a0ecSDave Kleikamp goto out; 893ac27a0ecSDave Kleikamp 894119c0d44STheodore Ts'o /* 895119c0d44STheodore Ts'o * Normally we will only go through one pass of this loop, 896119c0d44STheodore Ts'o * unless we get unlucky and it turns out the group we selected 897119c0d44STheodore Ts'o * had its last inode grabbed by someone else. 898119c0d44STheodore Ts'o */ 89911013911SAndreas Dilger for (i = 0; i < ngroups; i++, ino = 0) { 900ac27a0ecSDave Kleikamp err = -EIO; 901ac27a0ecSDave Kleikamp 9023300bedaSAneesh Kumar K.V gdp = ext4_get_group_desc(sb, group, &group_desc_bh); 903ac27a0ecSDave Kleikamp if (!gdp) 904eb9cc7e1SJan Kara goto out; 905ac27a0ecSDave Kleikamp 906f2a09af6SYongqiang Yang /* 907f2a09af6SYongqiang Yang * Check free inodes count before loading bitmap. 908f2a09af6SYongqiang Yang */ 9092fe435d8SWang Shilong if (ext4_free_inodes_count(sb, gdp) == 0) 9102fe435d8SWang Shilong goto next_group; 911f2a09af6SYongqiang Yang 91287a39389SDarrick J. Wong grp = ext4_get_group_info(sb, group); 91387a39389SDarrick J. Wong /* Skip groups with already-known suspicious inode tables */ 9142fe435d8SWang Shilong if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) 9152fe435d8SWang Shilong goto next_group; 91687a39389SDarrick J. Wong 9173300bedaSAneesh Kumar K.V brelse(inode_bitmap_bh); 9183300bedaSAneesh Kumar K.V inode_bitmap_bh = ext4_read_inode_bitmap(sb, group); 91987a39389SDarrick J. Wong /* Skip groups with suspicious inode tables */ 9209008a58eSDarrick J. Wong if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) || 9219008a58eSDarrick J. Wong IS_ERR(inode_bitmap_bh)) { 9229008a58eSDarrick J. Wong inode_bitmap_bh = NULL; 9232fe435d8SWang Shilong goto next_group; 92487a39389SDarrick J. Wong } 925ac27a0ecSDave Kleikamp 926ac27a0ecSDave Kleikamp repeat_in_this_group: 927901ed070SWang Shilong ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino); 928901ed070SWang Shilong if (!ret2) 929a34eb503STheodore Ts'o goto next_group; 930901ed070SWang Shilong 931119c0d44STheodore Ts'o if (group == 0 && (ino + 1) < EXT4_FIRST_INO(sb)) { 932119c0d44STheodore Ts'o ext4_error(sb, "reserved inode found cleared - " 933119c0d44STheodore Ts'o "inode=%lu", ino + 1); 934206f6d55SWang Shilong ext4_mark_group_bitmap_corrupted(sb, group, 935206f6d55SWang Shilong EXT4_GROUP_INFO_IBITMAP_CORRUPT); 9362fe435d8SWang Shilong goto next_group; 937119c0d44STheodore Ts'o } 938901ed070SWang Shilong 9391139575aSTheodore Ts'o if (!handle) { 9401139575aSTheodore Ts'o BUG_ON(nblocks <= 0); 9411139575aSTheodore Ts'o handle = __ext4_journal_start_sb(dir->i_sb, line_no, 942a70fd5acSyangerkun handle_type, nblocks, 0, 943a70fd5acSyangerkun ext4_trans_default_revoke_credits(sb)); 9441139575aSTheodore Ts'o if (IS_ERR(handle)) { 9451139575aSTheodore Ts'o err = PTR_ERR(handle); 946eb9cc7e1SJan Kara ext4_std_error(sb, err); 947eb9cc7e1SJan Kara goto out; 9481139575aSTheodore Ts'o } 9491139575aSTheodore Ts'o } 950ffb5387eSEric Sandeen BUFFER_TRACE(inode_bitmap_bh, "get_write_access"); 951ffb5387eSEric Sandeen err = ext4_journal_get_write_access(handle, inode_bitmap_bh); 952eb9cc7e1SJan Kara if (err) { 953eb9cc7e1SJan Kara ext4_std_error(sb, err); 954eb9cc7e1SJan Kara goto out; 955eb9cc7e1SJan Kara } 956119c0d44STheodore Ts'o ext4_lock_group(sb, group); 957119c0d44STheodore Ts'o ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data); 958901ed070SWang Shilong if (ret2) { 959901ed070SWang Shilong /* Someone already took the bit. Repeat the search 960901ed070SWang Shilong * with lock held. 961901ed070SWang Shilong */ 962901ed070SWang Shilong ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino); 963901ed070SWang Shilong if (ret2) { 964901ed070SWang Shilong ext4_set_bit(ino, inode_bitmap_bh->b_data); 965901ed070SWang Shilong ret2 = 0; 966901ed070SWang Shilong } else { 967901ed070SWang Shilong ret2 = 1; /* we didn't grab the inode */ 968901ed070SWang Shilong } 969901ed070SWang Shilong } 970119c0d44STheodore Ts'o ext4_unlock_group(sb, group); 971119c0d44STheodore Ts'o ino++; /* the inode bitmap is zero-based */ 972119c0d44STheodore Ts'o if (!ret2) 973119c0d44STheodore Ts'o goto got; /* we grabbed the inode! */ 974901ed070SWang Shilong 975119c0d44STheodore Ts'o if (ino < EXT4_INODES_PER_GROUP(sb)) 976119c0d44STheodore Ts'o goto repeat_in_this_group; 977a34eb503STheodore Ts'o next_group: 978a34eb503STheodore Ts'o if (++group == ngroups) 979a34eb503STheodore Ts'o group = 0; 980ac27a0ecSDave Kleikamp } 981ac27a0ecSDave Kleikamp err = -ENOSPC; 982ac27a0ecSDave Kleikamp goto out; 983ac27a0ecSDave Kleikamp 984ac27a0ecSDave Kleikamp got: 985ffb5387eSEric Sandeen BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata"); 986ffb5387eSEric Sandeen err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh); 987eb9cc7e1SJan Kara if (err) { 988eb9cc7e1SJan Kara ext4_std_error(sb, err); 989eb9cc7e1SJan Kara goto out; 990eb9cc7e1SJan Kara } 991ffb5387eSEric Sandeen 99261c219f5STheodore Ts'o BUFFER_TRACE(group_desc_bh, "get_write_access"); 99361c219f5STheodore Ts'o err = ext4_journal_get_write_access(handle, group_desc_bh); 99461c219f5STheodore Ts'o if (err) { 99561c219f5STheodore Ts'o ext4_std_error(sb, err); 99661c219f5STheodore Ts'o goto out; 99761c219f5STheodore Ts'o } 99861c219f5STheodore Ts'o 999717d50e4SAndreas Dilger /* We may have to initialize the block bitmap if it isn't already */ 1000feb0ab32SDarrick J. Wong if (ext4_has_group_desc_csum(sb) && 1001717d50e4SAndreas Dilger gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { 10023300bedaSAneesh Kumar K.V struct buffer_head *block_bitmap_bh; 1003717d50e4SAndreas Dilger 10043300bedaSAneesh Kumar K.V block_bitmap_bh = ext4_read_block_bitmap(sb, group); 10059008a58eSDarrick J. Wong if (IS_ERR(block_bitmap_bh)) { 10069008a58eSDarrick J. Wong err = PTR_ERR(block_bitmap_bh); 1007599a9b77SJan Kara goto out; 1008599a9b77SJan Kara } 10093300bedaSAneesh Kumar K.V BUFFER_TRACE(block_bitmap_bh, "get block bitmap access"); 10103300bedaSAneesh Kumar K.V err = ext4_journal_get_write_access(handle, block_bitmap_bh); 1011717d50e4SAndreas Dilger if (err) { 10123300bedaSAneesh Kumar K.V brelse(block_bitmap_bh); 1013eb9cc7e1SJan Kara ext4_std_error(sb, err); 1014eb9cc7e1SJan Kara goto out; 1015717d50e4SAndreas Dilger } 1016717d50e4SAndreas Dilger 1017fd034a84STheodore Ts'o BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap"); 1018fd034a84STheodore Ts'o err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh); 1019fd034a84STheodore Ts'o 1020717d50e4SAndreas Dilger /* recheck and clear flag under lock if we still need to */ 1021fd034a84STheodore Ts'o ext4_lock_group(sb, group); 10228844618dSTheodore Ts'o if (ext4_has_group_desc_csum(sb) && 10238844618dSTheodore Ts'o (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT))) { 10243300bedaSAneesh Kumar K.V gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT); 1025021b65bbSTheodore Ts'o ext4_free_group_clusters_set(sb, gdp, 1026cff1dfd7STheodore Ts'o ext4_free_clusters_after_init(sb, group, gdp)); 1027fa77dcfaSDarrick J. Wong ext4_block_bitmap_csum_set(sb, group, gdp, 102879f1ba49STao Ma block_bitmap_bh); 1029feb0ab32SDarrick J. Wong ext4_group_desc_csum_set(sb, group, gdp); 1030717d50e4SAndreas Dilger } 1031955ce5f5SAneesh Kumar K.V ext4_unlock_group(sb, group); 1032aeb1e5d6STheodore Ts'o brelse(block_bitmap_bh); 1033717d50e4SAndreas Dilger 1034eb9cc7e1SJan Kara if (err) { 1035eb9cc7e1SJan Kara ext4_std_error(sb, err); 1036eb9cc7e1SJan Kara goto out; 1037eb9cc7e1SJan Kara } 1038717d50e4SAndreas Dilger } 1039119c0d44STheodore Ts'o 1040119c0d44STheodore Ts'o /* Update the relevant bg descriptor fields */ 104141a246d1SDarrick J. Wong if (ext4_has_group_desc_csum(sb)) { 1042119c0d44STheodore Ts'o int free; 1043119c0d44STheodore Ts'o struct ext4_group_info *grp = ext4_get_group_info(sb, group); 1044119c0d44STheodore Ts'o 1045119c0d44STheodore Ts'o down_read(&grp->alloc_sem); /* protect vs itable lazyinit */ 1046119c0d44STheodore Ts'o ext4_lock_group(sb, group); /* while we modify the bg desc */ 1047119c0d44STheodore Ts'o free = EXT4_INODES_PER_GROUP(sb) - 1048119c0d44STheodore Ts'o ext4_itable_unused_count(sb, gdp); 1049119c0d44STheodore Ts'o if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { 1050119c0d44STheodore Ts'o gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT); 1051119c0d44STheodore Ts'o free = 0; 1052119c0d44STheodore Ts'o } 1053119c0d44STheodore Ts'o /* 1054119c0d44STheodore Ts'o * Check the relative inode number against the last used 1055119c0d44STheodore Ts'o * relative inode number in this group. if it is greater 1056119c0d44STheodore Ts'o * we need to update the bg_itable_unused count 1057119c0d44STheodore Ts'o */ 1058119c0d44STheodore Ts'o if (ino > free) 1059119c0d44STheodore Ts'o ext4_itable_unused_set(sb, gdp, 1060119c0d44STheodore Ts'o (EXT4_INODES_PER_GROUP(sb) - ino)); 1061119c0d44STheodore Ts'o up_read(&grp->alloc_sem); 10626f2e9f0eSTao Ma } else { 10636f2e9f0eSTao Ma ext4_lock_group(sb, group); 1064119c0d44STheodore Ts'o } 10656f2e9f0eSTao Ma 1066119c0d44STheodore Ts'o ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1); 1067119c0d44STheodore Ts'o if (S_ISDIR(mode)) { 1068119c0d44STheodore Ts'o ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1); 1069119c0d44STheodore Ts'o if (sbi->s_log_groups_per_flex) { 1070119c0d44STheodore Ts'o ext4_group_t f = ext4_flex_group(sbi, group); 1071119c0d44STheodore Ts'o 10727c990728SSuraj Jitindar Singh atomic_inc(&sbi_array_rcu_deref(sbi, s_flex_groups, 10737c990728SSuraj Jitindar Singh f)->used_dirs); 1074119c0d44STheodore Ts'o } 1075119c0d44STheodore Ts'o } 107641a246d1SDarrick J. Wong if (ext4_has_group_desc_csum(sb)) { 107741a246d1SDarrick J. Wong ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh, 107841a246d1SDarrick J. Wong EXT4_INODES_PER_GROUP(sb) / 8); 1079feb0ab32SDarrick J. Wong ext4_group_desc_csum_set(sb, group, gdp); 1080119c0d44STheodore Ts'o } 10816f2e9f0eSTao Ma ext4_unlock_group(sb, group); 1082119c0d44STheodore Ts'o 10833300bedaSAneesh Kumar K.V BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata"); 10843300bedaSAneesh Kumar K.V err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh); 1085eb9cc7e1SJan Kara if (err) { 1086eb9cc7e1SJan Kara ext4_std_error(sb, err); 1087eb9cc7e1SJan Kara goto out; 1088eb9cc7e1SJan Kara } 1089ac27a0ecSDave Kleikamp 1090ac27a0ecSDave Kleikamp percpu_counter_dec(&sbi->s_freeinodes_counter); 1091ac27a0ecSDave Kleikamp if (S_ISDIR(mode)) 1092ac27a0ecSDave Kleikamp percpu_counter_inc(&sbi->s_dirs_counter); 1093ac27a0ecSDave Kleikamp 1094772cb7c8SJose R. Santos if (sbi->s_log_groups_per_flex) { 1095772cb7c8SJose R. Santos flex_group = ext4_flex_group(sbi, group); 10967c990728SSuraj Jitindar Singh atomic_dec(&sbi_array_rcu_deref(sbi, s_flex_groups, 10977c990728SSuraj Jitindar Singh flex_group)->free_inodes); 1098772cb7c8SJose R. Santos } 1099ac27a0ecSDave Kleikamp 1100717d50e4SAndreas Dilger inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb); 1101ac27a0ecSDave Kleikamp /* This is the optimal IO size (for stat), not the fs block size */ 1102ac27a0ecSDave Kleikamp inode->i_blocks = 0; 110395582b00SDeepa Dinamani inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode); 11047b62b293SArnd Bergmann ei->i_crtime = inode->i_mtime; 1105ac27a0ecSDave Kleikamp 1106ac27a0ecSDave Kleikamp memset(ei->i_data, 0, sizeof(ei->i_data)); 1107ac27a0ecSDave Kleikamp ei->i_dir_start_lookup = 0; 1108ac27a0ecSDave Kleikamp ei->i_disksize = 0; 1109ac27a0ecSDave Kleikamp 11104af83508SEryu Guan /* Don't inherit extent flag from directory, amongst others. */ 11112dc6b0d4SDuane Griffin ei->i_flags = 11122dc6b0d4SDuane Griffin ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED); 11131b917ed8STahsin Erdogan ei->i_flags |= i_flags; 1114ac27a0ecSDave Kleikamp ei->i_file_acl = 0; 1115ac27a0ecSDave Kleikamp ei->i_dtime = 0; 1116ac27a0ecSDave Kleikamp ei->i_block_group = group; 1117a4912123STheodore Ts'o ei->i_last_alloc_group = ~0; 1118ac27a0ecSDave Kleikamp 1119617ba13bSMingming Cao ext4_set_inode_flags(inode); 1120ac27a0ecSDave Kleikamp if (IS_DIRSYNC(inode)) 11210390131bSFrank Mayhar ext4_handle_sync(handle); 11226b38e842SAl Viro if (insert_inode_locked(inode) < 0) { 1123acd6ad83SJan Kara /* 1124acd6ad83SJan Kara * Likely a bitmap corruption causing inode to be allocated 1125acd6ad83SJan Kara * twice. 1126acd6ad83SJan Kara */ 1127acd6ad83SJan Kara err = -EIO; 1128eb9cc7e1SJan Kara ext4_error(sb, "failed to insert inode %lu: doubly allocated?", 1129eb9cc7e1SJan Kara inode->i_ino); 1130206f6d55SWang Shilong ext4_mark_group_bitmap_corrupted(sb, group, 1131206f6d55SWang Shilong EXT4_GROUP_INFO_IBITMAP_CORRUPT); 1132eb9cc7e1SJan Kara goto out; 11336b38e842SAl Viro } 113423253068STheodore Ts'o inode->i_generation = prandom_u32(); 1135ac27a0ecSDave Kleikamp 1136814525f4SDarrick J. Wong /* Precompute checksum seed for inode metadata */ 11379aa5d32bSDmitry Monakhov if (ext4_has_metadata_csum(sb)) { 1138814525f4SDarrick J. Wong __u32 csum; 1139814525f4SDarrick J. Wong __le32 inum = cpu_to_le32(inode->i_ino); 1140814525f4SDarrick J. Wong __le32 gen = cpu_to_le32(inode->i_generation); 1141814525f4SDarrick J. Wong csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, 1142814525f4SDarrick J. Wong sizeof(inum)); 1143814525f4SDarrick J. Wong ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, 1144814525f4SDarrick J. Wong sizeof(gen)); 1145814525f4SDarrick J. Wong } 1146814525f4SDarrick J. Wong 1147353eb83cSTheodore Ts'o ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ 114819f5fb7aSTheodore Ts'o ext4_set_inode_state(inode, EXT4_STATE_NEW); 1149ef7f3835SKalpak Shah 115049598e04SJun Piao ei->i_extra_isize = sbi->s_want_extra_isize; 1151f08225d1STao Ma ei->i_inline_off = 0; 1152e2b911c5SDarrick J. Wong if (ext4_has_feature_inline_data(sb)) 1153f08225d1STao Ma ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); 1154ac27a0ecSDave Kleikamp ret = inode; 115563936ddaSChristoph Hellwig err = dquot_alloc_inode(inode); 115663936ddaSChristoph Hellwig if (err) 1157ac27a0ecSDave Kleikamp goto fail_drop; 1158ac27a0ecSDave Kleikamp 1159aa1dca3bSEric Biggers /* 1160aa1dca3bSEric Biggers * Since the encryption xattr will always be unique, create it first so 1161aa1dca3bSEric Biggers * that it's less likely to end up in an external xattr block and 1162aa1dca3bSEric Biggers * prevent its deduplication. 1163aa1dca3bSEric Biggers */ 1164aa1dca3bSEric Biggers if (encrypt) { 1165aa1dca3bSEric Biggers err = fscrypt_inherit_context(dir, inode, handle, true); 1166aa1dca3bSEric Biggers if (err) 1167aa1dca3bSEric Biggers goto fail_free_drop; 1168aa1dca3bSEric Biggers } 1169aa1dca3bSEric Biggers 11701b917ed8STahsin Erdogan if (!(ei->i_flags & EXT4_EA_INODE_FL)) { 1171617ba13bSMingming Cao err = ext4_init_acl(handle, inode, dir); 1172ac27a0ecSDave Kleikamp if (err) 1173ac27a0ecSDave Kleikamp goto fail_free_drop; 1174ac27a0ecSDave Kleikamp 11752a7dba39SEric Paris err = ext4_init_security(handle, inode, dir, qstr); 1176ac27a0ecSDave Kleikamp if (err) 1177ac27a0ecSDave Kleikamp goto fail_free_drop; 1178ad47f953STahsin Erdogan } 1179ac27a0ecSDave Kleikamp 1180e2b911c5SDarrick J. Wong if (ext4_has_feature_extents(sb)) { 1181e4079a11SEric Sandeen /* set extent flag only for directory, file and normal symlink*/ 1182e65187e6SAneesh Kumar K.V if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) { 118312e9b892SDmitry Monakhov ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS); 1184a86c6181SAlex Tomas ext4_ext_tree_init(handle, inode); 1185a86c6181SAlex Tomas } 118642bf0383SAneesh Kumar K.V } 1187ac27a0ecSDave Kleikamp 1188688f869cSTheodore Ts'o if (ext4_handle_valid(handle)) { 1189688f869cSTheodore Ts'o ei->i_sync_tid = handle->h_transaction->t_tid; 1190688f869cSTheodore Ts'o ei->i_datasync_tid = handle->h_transaction->t_tid; 1191688f869cSTheodore Ts'o } 1192688f869cSTheodore Ts'o 11938753e88fSAneesh Kumar K.V err = ext4_mark_inode_dirty(handle, inode); 11948753e88fSAneesh Kumar K.V if (err) { 11958753e88fSAneesh Kumar K.V ext4_std_error(sb, err); 11968753e88fSAneesh Kumar K.V goto fail_free_drop; 11978753e88fSAneesh Kumar K.V } 11988753e88fSAneesh Kumar K.V 1199617ba13bSMingming Cao ext4_debug("allocating inode %lu\n", inode->i_ino); 12009bffad1eSTheodore Ts'o trace_ext4_allocate_inode(inode, dir, mode); 12013300bedaSAneesh Kumar K.V brelse(inode_bitmap_bh); 1202ac27a0ecSDave Kleikamp return ret; 1203ac27a0ecSDave Kleikamp 1204ac27a0ecSDave Kleikamp fail_free_drop: 120563936ddaSChristoph Hellwig dquot_free_inode(inode); 1206ac27a0ecSDave Kleikamp fail_drop: 12076d6b77f1SMiklos Szeredi clear_nlink(inode); 12086b38e842SAl Viro unlock_new_inode(inode); 1209eb9cc7e1SJan Kara out: 1210eb9cc7e1SJan Kara dquot_drop(inode); 1211eb9cc7e1SJan Kara inode->i_flags |= S_NOQUOTA; 1212ac27a0ecSDave Kleikamp iput(inode); 12133300bedaSAneesh Kumar K.V brelse(inode_bitmap_bh); 1214ac27a0ecSDave Kleikamp return ERR_PTR(err); 1215ac27a0ecSDave Kleikamp } 1216ac27a0ecSDave Kleikamp 1217ac27a0ecSDave Kleikamp /* Verify that we are loading a valid orphan from disk */ 1218617ba13bSMingming Cao struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino) 1219ac27a0ecSDave Kleikamp { 1220617ba13bSMingming Cao unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count); 1221fd2d4291SAvantika Mathur ext4_group_t block_group; 1222ac27a0ecSDave Kleikamp int bit; 12237827a7f6STheodore Ts'o struct buffer_head *bitmap_bh = NULL; 1224ac27a0ecSDave Kleikamp struct inode *inode = NULL; 12257827a7f6STheodore Ts'o int err = -EFSCORRUPTED; 1226ac27a0ecSDave Kleikamp 12277827a7f6STheodore Ts'o if (ino < EXT4_FIRST_INO(sb) || ino > max_ino) 12287827a7f6STheodore Ts'o goto bad_orphan; 1229ac27a0ecSDave Kleikamp 1230617ba13bSMingming Cao block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); 1231617ba13bSMingming Cao bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); 1232e29d1cdeSEric Sandeen bitmap_bh = ext4_read_inode_bitmap(sb, block_group); 1233206f6d55SWang Shilong if (IS_ERR(bitmap_bh)) 1234e884bce1SAl Viro return ERR_CAST(bitmap_bh); 1235ac27a0ecSDave Kleikamp 1236ac27a0ecSDave Kleikamp /* Having the inode bit set should be a 100% indicator that this 1237ac27a0ecSDave Kleikamp * is a valid orphan (no e2fsck run on fs). Orphans also include 1238ac27a0ecSDave Kleikamp * inodes that were being truncated, so we can't check i_nlink==0. 1239ac27a0ecSDave Kleikamp */ 12401d1fe1eeSDavid Howells if (!ext4_test_bit(bit, bitmap_bh->b_data)) 12411d1fe1eeSDavid Howells goto bad_orphan; 12421d1fe1eeSDavid Howells 12438a363970STheodore Ts'o inode = ext4_iget(sb, ino, EXT4_IGET_NORMAL); 12447827a7f6STheodore Ts'o if (IS_ERR(inode)) { 12457827a7f6STheodore Ts'o err = PTR_ERR(inode); 124654d3adbcSTheodore Ts'o ext4_error_err(sb, -err, 124754d3adbcSTheodore Ts'o "couldn't read orphan inode %lu (err %d)", 12487827a7f6STheodore Ts'o ino, err); 12493bbd0ef2SXiyu Yang brelse(bitmap_bh); 12507827a7f6STheodore Ts'o return inode; 12517827a7f6STheodore Ts'o } 12521d1fe1eeSDavid Howells 125391ef4cafSDuane Griffin /* 1254c9eb13a9STheodore Ts'o * If the orphans has i_nlinks > 0 then it should be able to 1255c9eb13a9STheodore Ts'o * be truncated, otherwise it won't be removed from the orphan 1256c9eb13a9STheodore Ts'o * list during processing and an infinite loop will result. 1257c9eb13a9STheodore Ts'o * Similarly, it must not be a bad inode. 125891ef4cafSDuane Griffin */ 1259c9eb13a9STheodore Ts'o if ((inode->i_nlink && !ext4_can_truncate(inode)) || 1260c9eb13a9STheodore Ts'o is_bad_inode(inode)) 126191ef4cafSDuane Griffin goto bad_orphan; 126291ef4cafSDuane Griffin 12631d1fe1eeSDavid Howells if (NEXT_ORPHAN(inode) > max_ino) 12641d1fe1eeSDavid Howells goto bad_orphan; 12651d1fe1eeSDavid Howells brelse(bitmap_bh); 12661d1fe1eeSDavid Howells return inode; 12671d1fe1eeSDavid Howells 12681d1fe1eeSDavid Howells bad_orphan: 12697827a7f6STheodore Ts'o ext4_error(sb, "bad orphan inode %lu", ino); 12707827a7f6STheodore Ts'o if (bitmap_bh) 12717827a7f6STheodore Ts'o printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n", 1272ac27a0ecSDave Kleikamp bit, (unsigned long long)bitmap_bh->b_blocknr, 1273617ba13bSMingming Cao ext4_test_bit(bit, bitmap_bh->b_data)); 1274ac27a0ecSDave Kleikamp if (inode) { 12757827a7f6STheodore Ts'o printk(KERN_ERR "is_bad_inode(inode)=%d\n", 1276ac27a0ecSDave Kleikamp is_bad_inode(inode)); 12777827a7f6STheodore Ts'o printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n", 1278ac27a0ecSDave Kleikamp NEXT_ORPHAN(inode)); 12797827a7f6STheodore Ts'o printk(KERN_ERR "max_ino=%lu\n", max_ino); 12807827a7f6STheodore Ts'o printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink); 1281ac27a0ecSDave Kleikamp /* Avoid freeing blocks if we got a bad deleted inode */ 12821d1fe1eeSDavid Howells if (inode->i_nlink == 0) 1283ac27a0ecSDave Kleikamp inode->i_blocks = 0; 1284ac27a0ecSDave Kleikamp iput(inode); 1285ac27a0ecSDave Kleikamp } 1286ac27a0ecSDave Kleikamp brelse(bitmap_bh); 12871d1fe1eeSDavid Howells return ERR_PTR(err); 1288ac27a0ecSDave Kleikamp } 1289ac27a0ecSDave Kleikamp 1290617ba13bSMingming Cao unsigned long ext4_count_free_inodes(struct super_block *sb) 1291ac27a0ecSDave Kleikamp { 1292ac27a0ecSDave Kleikamp unsigned long desc_count; 1293617ba13bSMingming Cao struct ext4_group_desc *gdp; 12948df9675fSTheodore Ts'o ext4_group_t i, ngroups = ext4_get_groups_count(sb); 1295617ba13bSMingming Cao #ifdef EXT4FS_DEBUG 1296617ba13bSMingming Cao struct ext4_super_block *es; 1297ac27a0ecSDave Kleikamp unsigned long bitmap_count, x; 1298ac27a0ecSDave Kleikamp struct buffer_head *bitmap_bh = NULL; 1299ac27a0ecSDave Kleikamp 1300617ba13bSMingming Cao es = EXT4_SB(sb)->s_es; 1301ac27a0ecSDave Kleikamp desc_count = 0; 1302ac27a0ecSDave Kleikamp bitmap_count = 0; 1303ac27a0ecSDave Kleikamp gdp = NULL; 13048df9675fSTheodore Ts'o for (i = 0; i < ngroups; i++) { 1305617ba13bSMingming Cao gdp = ext4_get_group_desc(sb, i, NULL); 1306ac27a0ecSDave Kleikamp if (!gdp) 1307ac27a0ecSDave Kleikamp continue; 1308560671a0SAneesh Kumar K.V desc_count += ext4_free_inodes_count(sb, gdp); 1309ac27a0ecSDave Kleikamp brelse(bitmap_bh); 1310e29d1cdeSEric Sandeen bitmap_bh = ext4_read_inode_bitmap(sb, i); 13119008a58eSDarrick J. Wong if (IS_ERR(bitmap_bh)) { 13129008a58eSDarrick J. Wong bitmap_bh = NULL; 1313ac27a0ecSDave Kleikamp continue; 13149008a58eSDarrick J. Wong } 1315ac27a0ecSDave Kleikamp 1316f6fb99caSTheodore Ts'o x = ext4_count_free(bitmap_bh->b_data, 1317f6fb99caSTheodore Ts'o EXT4_INODES_PER_GROUP(sb) / 8); 1318c549a95dSEric Sandeen printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n", 1319785b4b3aSPeng Tao (unsigned long) i, ext4_free_inodes_count(sb, gdp), x); 1320ac27a0ecSDave Kleikamp bitmap_count += x; 1321ac27a0ecSDave Kleikamp } 1322ac27a0ecSDave Kleikamp brelse(bitmap_bh); 13234776004fSTheodore Ts'o printk(KERN_DEBUG "ext4_count_free_inodes: " 13244776004fSTheodore Ts'o "stored = %u, computed = %lu, %lu\n", 1325ac27a0ecSDave Kleikamp le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count); 1326ac27a0ecSDave Kleikamp return desc_count; 1327ac27a0ecSDave Kleikamp #else 1328ac27a0ecSDave Kleikamp desc_count = 0; 13298df9675fSTheodore Ts'o for (i = 0; i < ngroups; i++) { 1330617ba13bSMingming Cao gdp = ext4_get_group_desc(sb, i, NULL); 1331ac27a0ecSDave Kleikamp if (!gdp) 1332ac27a0ecSDave Kleikamp continue; 1333560671a0SAneesh Kumar K.V desc_count += ext4_free_inodes_count(sb, gdp); 1334ac27a0ecSDave Kleikamp cond_resched(); 1335ac27a0ecSDave Kleikamp } 1336ac27a0ecSDave Kleikamp return desc_count; 1337ac27a0ecSDave Kleikamp #endif 1338ac27a0ecSDave Kleikamp } 1339ac27a0ecSDave Kleikamp 1340ac27a0ecSDave Kleikamp /* Called at mount-time, super-block is locked */ 1341617ba13bSMingming Cao unsigned long ext4_count_dirs(struct super_block * sb) 1342ac27a0ecSDave Kleikamp { 1343ac27a0ecSDave Kleikamp unsigned long count = 0; 13448df9675fSTheodore Ts'o ext4_group_t i, ngroups = ext4_get_groups_count(sb); 1345ac27a0ecSDave Kleikamp 13468df9675fSTheodore Ts'o for (i = 0; i < ngroups; i++) { 1347617ba13bSMingming Cao struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); 1348ac27a0ecSDave Kleikamp if (!gdp) 1349ac27a0ecSDave Kleikamp continue; 1350560671a0SAneesh Kumar K.V count += ext4_used_dirs_count(sb, gdp); 1351ac27a0ecSDave Kleikamp } 1352ac27a0ecSDave Kleikamp return count; 1353ac27a0ecSDave Kleikamp } 1354bfff6873SLukas Czerner 1355bfff6873SLukas Czerner /* 1356bfff6873SLukas Czerner * Zeroes not yet zeroed inode table - just write zeroes through the whole 1357bfff6873SLukas Czerner * inode table. Must be called without any spinlock held. The only place 1358bfff6873SLukas Czerner * where it is called from on active part of filesystem is ext4lazyinit 1359bfff6873SLukas Czerner * thread, so we do not need any special locks, however we have to prevent 1360bfff6873SLukas Czerner * inode allocation from the current group, so we take alloc_sem lock, to 1361119c0d44STheodore Ts'o * block ext4_new_inode() until we are finished. 1362bfff6873SLukas Czerner */ 1363e0cbee3eSH Hartley Sweeten int ext4_init_inode_table(struct super_block *sb, ext4_group_t group, 1364bfff6873SLukas Czerner int barrier) 1365bfff6873SLukas Czerner { 1366bfff6873SLukas Czerner struct ext4_group_info *grp = ext4_get_group_info(sb, group); 1367bfff6873SLukas Czerner struct ext4_sb_info *sbi = EXT4_SB(sb); 1368bfff6873SLukas Czerner struct ext4_group_desc *gdp = NULL; 1369bfff6873SLukas Czerner struct buffer_head *group_desc_bh; 1370bfff6873SLukas Czerner handle_t *handle; 1371bfff6873SLukas Czerner ext4_fsblk_t blk; 1372bfff6873SLukas Czerner int num, ret = 0, used_blks = 0; 1373bfff6873SLukas Czerner 1374bfff6873SLukas Czerner /* This should not happen, but just to be sure check this */ 1375bc98a42cSDavid Howells if (sb_rdonly(sb)) { 1376bfff6873SLukas Czerner ret = 1; 1377bfff6873SLukas Czerner goto out; 1378bfff6873SLukas Czerner } 1379bfff6873SLukas Czerner 1380bfff6873SLukas Czerner gdp = ext4_get_group_desc(sb, group, &group_desc_bh); 1381bfff6873SLukas Czerner if (!gdp) 1382bfff6873SLukas Czerner goto out; 1383bfff6873SLukas Czerner 1384bfff6873SLukas Czerner /* 1385bfff6873SLukas Czerner * We do not need to lock this, because we are the only one 1386bfff6873SLukas Czerner * handling this flag. 1387bfff6873SLukas Czerner */ 1388bfff6873SLukas Czerner if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)) 1389bfff6873SLukas Czerner goto out; 1390bfff6873SLukas Czerner 13919924a92aSTheodore Ts'o handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1); 1392bfff6873SLukas Czerner if (IS_ERR(handle)) { 1393bfff6873SLukas Czerner ret = PTR_ERR(handle); 1394bfff6873SLukas Czerner goto out; 1395bfff6873SLukas Czerner } 1396bfff6873SLukas Czerner 1397bfff6873SLukas Czerner down_write(&grp->alloc_sem); 1398bfff6873SLukas Czerner /* 1399bfff6873SLukas Czerner * If inode bitmap was already initialized there may be some 1400bfff6873SLukas Czerner * used inodes so we need to skip blocks with used inodes in 1401bfff6873SLukas Czerner * inode table. 1402bfff6873SLukas Czerner */ 1403bfff6873SLukas Czerner if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) 1404bfff6873SLukas Czerner used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) - 1405bfff6873SLukas Czerner ext4_itable_unused_count(sb, gdp)), 1406bfff6873SLukas Czerner sbi->s_inodes_per_block); 1407bfff6873SLukas Czerner 140850122847STheodore Ts'o if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group) || 140950122847STheodore Ts'o ((group == 0) && ((EXT4_INODES_PER_GROUP(sb) - 141050122847STheodore Ts'o ext4_itable_unused_count(sb, gdp)) < 141150122847STheodore Ts'o EXT4_FIRST_INO(sb)))) { 14121084f252STheodore Ts'o ext4_error(sb, "Something is wrong with group %u: " 14131084f252STheodore Ts'o "used itable blocks: %d; " 14141084f252STheodore Ts'o "itable unused count: %u", 1415857ac889SLukas Czerner group, used_blks, 1416857ac889SLukas Czerner ext4_itable_unused_count(sb, gdp)); 1417857ac889SLukas Czerner ret = 1; 141833853a0dSYongqiang Yang goto err_out; 1419857ac889SLukas Czerner } 1420857ac889SLukas Czerner 1421bfff6873SLukas Czerner blk = ext4_inode_table(sb, gdp) + used_blks; 1422bfff6873SLukas Czerner num = sbi->s_itb_per_group - used_blks; 1423bfff6873SLukas Czerner 1424bfff6873SLukas Czerner BUFFER_TRACE(group_desc_bh, "get_write_access"); 1425bfff6873SLukas Czerner ret = ext4_journal_get_write_access(handle, 1426bfff6873SLukas Czerner group_desc_bh); 1427bfff6873SLukas Czerner if (ret) 1428bfff6873SLukas Czerner goto err_out; 1429bfff6873SLukas Czerner 1430bfff6873SLukas Czerner /* 1431bfff6873SLukas Czerner * Skip zeroout if the inode table is full. But we set the ZEROED 1432bfff6873SLukas Czerner * flag anyway, because obviously, when it is full it does not need 1433bfff6873SLukas Czerner * further zeroing. 1434bfff6873SLukas Czerner */ 1435bfff6873SLukas Czerner if (unlikely(num == 0)) 1436bfff6873SLukas Czerner goto skip_zeroout; 1437bfff6873SLukas Czerner 1438bfff6873SLukas Czerner ext4_debug("going to zero out inode table in group %d\n", 1439bfff6873SLukas Czerner group); 1440a107e5a3STheodore Ts'o ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS); 1441bfff6873SLukas Czerner if (ret < 0) 1442bfff6873SLukas Czerner goto err_out; 1443a107e5a3STheodore Ts'o if (barrier) 1444a107e5a3STheodore Ts'o blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL); 1445bfff6873SLukas Czerner 1446bfff6873SLukas Czerner skip_zeroout: 1447bfff6873SLukas Czerner ext4_lock_group(sb, group); 1448bfff6873SLukas Czerner gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED); 1449feb0ab32SDarrick J. Wong ext4_group_desc_csum_set(sb, group, gdp); 1450bfff6873SLukas Czerner ext4_unlock_group(sb, group); 1451bfff6873SLukas Czerner 1452bfff6873SLukas Czerner BUFFER_TRACE(group_desc_bh, 1453bfff6873SLukas Czerner "call ext4_handle_dirty_metadata"); 1454bfff6873SLukas Czerner ret = ext4_handle_dirty_metadata(handle, NULL, 1455bfff6873SLukas Czerner group_desc_bh); 1456bfff6873SLukas Czerner 1457bfff6873SLukas Czerner err_out: 1458bfff6873SLukas Czerner up_write(&grp->alloc_sem); 1459bfff6873SLukas Czerner ext4_journal_stop(handle); 1460bfff6873SLukas Czerner out: 1461bfff6873SLukas Czerner return ret; 1462bfff6873SLukas Czerner } 1463