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); 939008a58eSDarrick J. Wong blk = ext4_inode_bitmap(sb, desc); 949008a58eSDarrick J. Wong if (!ext4_inode_bitmap_csum_verify(sb, block_group, desc, bh, 959008a58eSDarrick J. Wong EXT4_INODES_PER_GROUP(sb) / 8)) { 969008a58eSDarrick J. Wong ext4_unlock_group(sb, block_group); 979008a58eSDarrick J. Wong ext4_error(sb, "Corrupt inode bitmap - block_group = %u, " 989008a58eSDarrick J. Wong "inode_bitmap = %llu", block_group, blk); 99db79e6d1SWang Shilong ext4_mark_group_bitmap_corrupted(sb, block_group, 100db79e6d1SWang Shilong EXT4_GROUP_INFO_IBITMAP_CORRUPT); 1019008a58eSDarrick J. Wong return -EFSBADCRC; 1029008a58eSDarrick J. Wong } 1039008a58eSDarrick J. Wong set_buffer_verified(bh); 1049008a58eSDarrick J. Wong ext4_unlock_group(sb, block_group); 1059008a58eSDarrick J. Wong return 0; 1069008a58eSDarrick J. Wong } 1079008a58eSDarrick J. Wong 108ac27a0ecSDave Kleikamp /* 109ac27a0ecSDave Kleikamp * Read the inode allocation bitmap for a given block_group, reading 110ac27a0ecSDave Kleikamp * into the specified slot in the superblock's bitmap cache. 111ac27a0ecSDave Kleikamp * 112ac27a0ecSDave Kleikamp * Return buffer_head of bitmap on success or NULL. 113ac27a0ecSDave Kleikamp */ 114ac27a0ecSDave Kleikamp static struct buffer_head * 115e29d1cdeSEric Sandeen ext4_read_inode_bitmap(struct super_block *sb, ext4_group_t block_group) 116ac27a0ecSDave Kleikamp { 117617ba13bSMingming Cao struct ext4_group_desc *desc; 1187dac4a17STheodore Ts'o struct ext4_sb_info *sbi = EXT4_SB(sb); 119ac27a0ecSDave Kleikamp struct buffer_head *bh = NULL; 120e29d1cdeSEric Sandeen ext4_fsblk_t bitmap_blk; 1219008a58eSDarrick J. Wong int err; 122ac27a0ecSDave Kleikamp 123617ba13bSMingming Cao desc = ext4_get_group_desc(sb, block_group, NULL); 124ac27a0ecSDave Kleikamp if (!desc) 1259008a58eSDarrick J. Wong return ERR_PTR(-EFSCORRUPTED); 126bfff6873SLukas Czerner 127e29d1cdeSEric Sandeen bitmap_blk = ext4_inode_bitmap(sb, desc); 1287dac4a17STheodore Ts'o if ((bitmap_blk <= le32_to_cpu(sbi->s_es->s_first_data_block)) || 1297dac4a17STheodore Ts'o (bitmap_blk >= ext4_blocks_count(sbi->s_es))) { 1307dac4a17STheodore Ts'o ext4_error(sb, "Invalid inode bitmap blk %llu in " 1317dac4a17STheodore Ts'o "block_group %u", bitmap_blk, block_group); 132*206f6d55SWang Shilong ext4_mark_group_bitmap_corrupted(sb, block_group, 133*206f6d55SWang Shilong EXT4_GROUP_INFO_IBITMAP_CORRUPT); 1347dac4a17STheodore Ts'o return ERR_PTR(-EFSCORRUPTED); 1357dac4a17STheodore Ts'o } 136e29d1cdeSEric Sandeen bh = sb_getblk(sb, bitmap_blk); 137e29d1cdeSEric Sandeen if (unlikely(!bh)) { 13812062dddSEric Sandeen ext4_error(sb, "Cannot read inode bitmap - " 139a9df9a49STheodore Ts'o "block_group = %u, inode_bitmap = %llu", 140e29d1cdeSEric Sandeen block_group, bitmap_blk); 1410db9fdebSWang Shilong return ERR_PTR(-ENOMEM); 142e29d1cdeSEric Sandeen } 1432ccb5fb9SAneesh Kumar K.V if (bitmap_uptodate(bh)) 14441a246d1SDarrick J. Wong goto verify; 145e29d1cdeSEric Sandeen 146c806e68fSFrederic Bohe lock_buffer(bh); 1472ccb5fb9SAneesh Kumar K.V if (bitmap_uptodate(bh)) { 1482ccb5fb9SAneesh Kumar K.V unlock_buffer(bh); 14941a246d1SDarrick J. Wong goto verify; 1502ccb5fb9SAneesh Kumar K.V } 151bfff6873SLukas Czerner 152955ce5f5SAneesh Kumar K.V ext4_lock_group(sb, block_group); 153e29d1cdeSEric Sandeen if (desc->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { 154044e6e3dSTheodore Ts'o memset(bh->b_data, 0, (EXT4_INODES_PER_GROUP(sb) + 7) / 8); 155044e6e3dSTheodore Ts'o ext4_mark_bitmap_end(EXT4_INODES_PER_GROUP(sb), 156044e6e3dSTheodore Ts'o sb->s_blocksize * 8, bh->b_data); 1572ccb5fb9SAneesh Kumar K.V set_bitmap_uptodate(bh); 158e29d1cdeSEric Sandeen set_buffer_uptodate(bh); 15941a246d1SDarrick J. Wong set_buffer_verified(bh); 160955ce5f5SAneesh Kumar K.V ext4_unlock_group(sb, block_group); 1613300bedaSAneesh Kumar K.V unlock_buffer(bh); 162e29d1cdeSEric Sandeen return bh; 163e29d1cdeSEric Sandeen } 164955ce5f5SAneesh Kumar K.V ext4_unlock_group(sb, block_group); 165bfff6873SLukas Czerner 1662ccb5fb9SAneesh Kumar K.V if (buffer_uptodate(bh)) { 1672ccb5fb9SAneesh Kumar K.V /* 1682ccb5fb9SAneesh Kumar K.V * if not uninit if bh is uptodate, 1692ccb5fb9SAneesh Kumar K.V * bitmap is also uptodate 1702ccb5fb9SAneesh Kumar K.V */ 1712ccb5fb9SAneesh Kumar K.V set_bitmap_uptodate(bh); 1722ccb5fb9SAneesh Kumar K.V unlock_buffer(bh); 17341a246d1SDarrick J. Wong goto verify; 1742ccb5fb9SAneesh Kumar K.V } 1752ccb5fb9SAneesh Kumar K.V /* 176813e5727STheodore Ts'o * submit the buffer_head for reading 1772ccb5fb9SAneesh Kumar K.V */ 1780562e0baSJiaying Zhang trace_ext4_load_inode_bitmap(sb, block_group); 179813e5727STheodore Ts'o bh->b_end_io = ext4_end_bitmap_read; 180813e5727STheodore Ts'o get_bh(bh); 1812a222ca9SMike Christie submit_bh(REQ_OP_READ, REQ_META | REQ_PRIO, bh); 182813e5727STheodore Ts'o wait_on_buffer(bh); 183813e5727STheodore Ts'o if (!buffer_uptodate(bh)) { 184e29d1cdeSEric Sandeen put_bh(bh); 18512062dddSEric Sandeen ext4_error(sb, "Cannot read inode bitmap - " 186a9df9a49STheodore Ts'o "block_group = %u, inode_bitmap = %llu", 187e29d1cdeSEric Sandeen block_group, bitmap_blk); 188*206f6d55SWang Shilong ext4_mark_group_bitmap_corrupted(sb, block_group, 189*206f6d55SWang Shilong EXT4_GROUP_INFO_IBITMAP_CORRUPT); 1909008a58eSDarrick J. Wong return ERR_PTR(-EIO); 191e29d1cdeSEric Sandeen } 19241a246d1SDarrick J. Wong 19341a246d1SDarrick J. Wong verify: 1949008a58eSDarrick J. Wong err = ext4_validate_inode_bitmap(sb, desc, block_group, bh); 1959008a58eSDarrick J. Wong if (err) 1969008a58eSDarrick J. Wong goto out; 197ac27a0ecSDave Kleikamp return bh; 1989008a58eSDarrick J. Wong out: 1999008a58eSDarrick J. Wong put_bh(bh); 2009008a58eSDarrick J. Wong return ERR_PTR(err); 201ac27a0ecSDave Kleikamp } 202ac27a0ecSDave Kleikamp 203ac27a0ecSDave Kleikamp /* 204ac27a0ecSDave Kleikamp * NOTE! When we get the inode, we're the only people 205ac27a0ecSDave Kleikamp * that have access to it, and as such there are no 206ac27a0ecSDave Kleikamp * race conditions we have to worry about. The inode 207ac27a0ecSDave Kleikamp * is not on the hash-lists, and it cannot be reached 208ac27a0ecSDave Kleikamp * through the filesystem because the directory entry 209ac27a0ecSDave Kleikamp * has been deleted earlier. 210ac27a0ecSDave Kleikamp * 211ac27a0ecSDave Kleikamp * HOWEVER: we must make sure that we get no aliases, 212ac27a0ecSDave Kleikamp * which means that we have to call "clear_inode()" 213ac27a0ecSDave Kleikamp * _before_ we mark the inode not in use in the inode 214ac27a0ecSDave Kleikamp * bitmaps. Otherwise a newly created file might use 215ac27a0ecSDave Kleikamp * the same inode number (not actually the same pointer 216ac27a0ecSDave Kleikamp * though), and then we'd have two inodes sharing the 217ac27a0ecSDave Kleikamp * same inode number and space on the harddisk. 218ac27a0ecSDave Kleikamp */ 219617ba13bSMingming Cao void ext4_free_inode(handle_t *handle, struct inode *inode) 220ac27a0ecSDave Kleikamp { 221ac27a0ecSDave Kleikamp struct super_block *sb = inode->i_sb; 222ac27a0ecSDave Kleikamp int is_directory; 223ac27a0ecSDave Kleikamp unsigned long ino; 224ac27a0ecSDave Kleikamp struct buffer_head *bitmap_bh = NULL; 225ac27a0ecSDave Kleikamp struct buffer_head *bh2; 226fd2d4291SAvantika Mathur ext4_group_t block_group; 227ac27a0ecSDave Kleikamp unsigned long bit; 228617ba13bSMingming Cao struct ext4_group_desc *gdp; 229617ba13bSMingming Cao struct ext4_super_block *es; 230617ba13bSMingming Cao struct ext4_sb_info *sbi; 2317ce9d5d1SEric Sandeen int fatal = 0, err, count, cleared; 23287a39389SDarrick J. Wong struct ext4_group_info *grp; 233ac27a0ecSDave Kleikamp 23492b97816STheodore Ts'o if (!sb) { 23592b97816STheodore Ts'o printk(KERN_ERR "EXT4-fs: %s:%d: inode on " 23692b97816STheodore Ts'o "nonexistent device\n", __func__, __LINE__); 23792b97816STheodore Ts'o return; 23892b97816STheodore Ts'o } 239ac27a0ecSDave Kleikamp if (atomic_read(&inode->i_count) > 1) { 24092b97816STheodore Ts'o ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: count=%d", 24192b97816STheodore Ts'o __func__, __LINE__, inode->i_ino, 242ac27a0ecSDave Kleikamp atomic_read(&inode->i_count)); 243ac27a0ecSDave Kleikamp return; 244ac27a0ecSDave Kleikamp } 245ac27a0ecSDave Kleikamp if (inode->i_nlink) { 24692b97816STheodore Ts'o ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: nlink=%d\n", 24792b97816STheodore Ts'o __func__, __LINE__, inode->i_ino, inode->i_nlink); 248ac27a0ecSDave Kleikamp return; 249ac27a0ecSDave Kleikamp } 250617ba13bSMingming Cao sbi = EXT4_SB(sb); 251ac27a0ecSDave Kleikamp 252ac27a0ecSDave Kleikamp ino = inode->i_ino; 253617ba13bSMingming Cao ext4_debug("freeing inode %lu\n", ino); 2549bffad1eSTheodore Ts'o trace_ext4_free_inode(inode); 255ac27a0ecSDave Kleikamp 256ac27a0ecSDave Kleikamp /* 257ac27a0ecSDave Kleikamp * Note: we must free any quota before locking the superblock, 258ac27a0ecSDave Kleikamp * as writing the quota to disk may need the lock as well. 259ac27a0ecSDave Kleikamp */ 260871a2931SChristoph Hellwig dquot_initialize(inode); 26163936ddaSChristoph Hellwig dquot_free_inode(inode); 2629f754758SChristoph Hellwig dquot_drop(inode); 263ac27a0ecSDave Kleikamp 264ac27a0ecSDave Kleikamp is_directory = S_ISDIR(inode->i_mode); 265ac27a0ecSDave Kleikamp 266ac27a0ecSDave Kleikamp /* Do this BEFORE marking the inode not in use or returning an error */ 2670930fcc1SAl Viro ext4_clear_inode(inode); 268ac27a0ecSDave Kleikamp 26949598e04SJun Piao es = sbi->s_es; 270617ba13bSMingming Cao if (ino < EXT4_FIRST_INO(sb) || ino > le32_to_cpu(es->s_inodes_count)) { 27112062dddSEric Sandeen ext4_error(sb, "reserved or nonexistent inode %lu", ino); 272ac27a0ecSDave Kleikamp goto error_return; 273ac27a0ecSDave Kleikamp } 274617ba13bSMingming Cao block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); 275617ba13bSMingming Cao bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); 276e29d1cdeSEric Sandeen bitmap_bh = ext4_read_inode_bitmap(sb, block_group); 27787a39389SDarrick J. Wong /* Don't bother if the inode bitmap is corrupt. */ 27887a39389SDarrick J. Wong grp = ext4_get_group_info(sb, block_group); 2799008a58eSDarrick J. Wong if (IS_ERR(bitmap_bh)) { 2809008a58eSDarrick J. Wong fatal = PTR_ERR(bitmap_bh); 2819008a58eSDarrick J. Wong bitmap_bh = NULL; 282ac27a0ecSDave Kleikamp goto error_return; 2839008a58eSDarrick J. Wong } 2849008a58eSDarrick J. Wong if (unlikely(EXT4_MB_GRP_IBITMAP_CORRUPT(grp))) { 2859008a58eSDarrick J. Wong fatal = -EFSCORRUPTED; 2869008a58eSDarrick J. Wong goto error_return; 2879008a58eSDarrick J. Wong } 288ac27a0ecSDave Kleikamp 289ac27a0ecSDave Kleikamp BUFFER_TRACE(bitmap_bh, "get_write_access"); 290617ba13bSMingming Cao fatal = ext4_journal_get_write_access(handle, bitmap_bh); 291ac27a0ecSDave Kleikamp if (fatal) 292ac27a0ecSDave Kleikamp goto error_return; 293ac27a0ecSDave Kleikamp 294d17413c0SDmitry Monakhov fatal = -ESRCH; 295617ba13bSMingming Cao gdp = ext4_get_group_desc(sb, block_group, &bh2); 296d17413c0SDmitry Monakhov if (gdp) { 297ac27a0ecSDave Kleikamp BUFFER_TRACE(bh2, "get_write_access"); 298617ba13bSMingming Cao fatal = ext4_journal_get_write_access(handle, bh2); 299d17413c0SDmitry Monakhov } 300955ce5f5SAneesh Kumar K.V ext4_lock_group(sb, block_group); 301597d508cSAkinobu Mita cleared = ext4_test_and_clear_bit(bit, bitmap_bh->b_data); 302d17413c0SDmitry Monakhov if (fatal || !cleared) { 303d17413c0SDmitry Monakhov ext4_unlock_group(sb, block_group); 304d17413c0SDmitry Monakhov goto out; 305d17413c0SDmitry Monakhov } 306d17413c0SDmitry Monakhov 307560671a0SAneesh Kumar K.V count = ext4_free_inodes_count(sb, gdp) + 1; 308560671a0SAneesh Kumar K.V ext4_free_inodes_set(sb, gdp, count); 309560671a0SAneesh Kumar K.V if (is_directory) { 310560671a0SAneesh Kumar K.V count = ext4_used_dirs_count(sb, gdp) - 1; 311560671a0SAneesh Kumar K.V ext4_used_dirs_set(sb, gdp, count); 312d17413c0SDmitry Monakhov percpu_counter_dec(&sbi->s_dirs_counter); 313d17413c0SDmitry Monakhov } 31441a246d1SDarrick J. Wong ext4_inode_bitmap_csum_set(sb, block_group, gdp, bitmap_bh, 31541a246d1SDarrick J. Wong EXT4_INODES_PER_GROUP(sb) / 8); 316feb0ab32SDarrick J. Wong ext4_group_desc_csum_set(sb, block_group, gdp); 317d17413c0SDmitry Monakhov ext4_unlock_group(sb, block_group); 3187d39db14STheodore Ts'o 319d17413c0SDmitry Monakhov percpu_counter_inc(&sbi->s_freeinodes_counter); 320d17413c0SDmitry Monakhov if (sbi->s_log_groups_per_flex) { 321d17413c0SDmitry Monakhov ext4_group_t f = ext4_flex_group(sbi, block_group); 322d17413c0SDmitry Monakhov 323d17413c0SDmitry Monakhov atomic_inc(&sbi->s_flex_groups[f].free_inodes); 324d17413c0SDmitry Monakhov if (is_directory) 325c4caae25SEric Sandeen atomic_dec(&sbi->s_flex_groups[f].used_dirs); 3267d39db14STheodore Ts'o } 3270390131bSFrank Mayhar BUFFER_TRACE(bh2, "call ext4_handle_dirty_metadata"); 328d17413c0SDmitry Monakhov fatal = ext4_handle_dirty_metadata(handle, NULL, bh2); 329d17413c0SDmitry Monakhov out: 330d17413c0SDmitry Monakhov if (cleared) { 3310390131bSFrank Mayhar BUFFER_TRACE(bitmap_bh, "call ext4_handle_dirty_metadata"); 3320390131bSFrank Mayhar err = ext4_handle_dirty_metadata(handle, NULL, bitmap_bh); 333ac27a0ecSDave Kleikamp if (!fatal) 334ac27a0ecSDave Kleikamp fatal = err; 33587a39389SDarrick J. Wong } else { 336d17413c0SDmitry Monakhov ext4_error(sb, "bit already cleared for inode %lu", ino); 337db79e6d1SWang Shilong ext4_mark_group_bitmap_corrupted(sb, block_group, 338db79e6d1SWang Shilong EXT4_GROUP_INFO_IBITMAP_CORRUPT); 33987a39389SDarrick J. Wong } 340d17413c0SDmitry Monakhov 341ac27a0ecSDave Kleikamp error_return: 342ac27a0ecSDave Kleikamp brelse(bitmap_bh); 343617ba13bSMingming Cao ext4_std_error(sb, fatal); 344ac27a0ecSDave Kleikamp } 345ac27a0ecSDave Kleikamp 346a4912123STheodore Ts'o struct orlov_stats { 34790ba983fSTheodore Ts'o __u64 free_clusters; 348a4912123STheodore Ts'o __u32 free_inodes; 349a4912123STheodore Ts'o __u32 used_dirs; 350a4912123STheodore Ts'o }; 351a4912123STheodore Ts'o 352a4912123STheodore Ts'o /* 353a4912123STheodore Ts'o * Helper function for Orlov's allocator; returns critical information 354a4912123STheodore Ts'o * for a particular block group or flex_bg. If flex_size is 1, then g 355a4912123STheodore Ts'o * is a block group number; otherwise it is flex_bg number. 356a4912123STheodore Ts'o */ 3571f109d5aSTheodore Ts'o static void get_orlov_stats(struct super_block *sb, ext4_group_t g, 358a4912123STheodore Ts'o int flex_size, struct orlov_stats *stats) 359a4912123STheodore Ts'o { 360a4912123STheodore Ts'o struct ext4_group_desc *desc; 3617d39db14STheodore Ts'o struct flex_groups *flex_group = EXT4_SB(sb)->s_flex_groups; 362a4912123STheodore Ts'o 3637d39db14STheodore Ts'o if (flex_size > 1) { 3647d39db14STheodore Ts'o stats->free_inodes = atomic_read(&flex_group[g].free_inodes); 36590ba983fSTheodore Ts'o stats->free_clusters = atomic64_read(&flex_group[g].free_clusters); 3667d39db14STheodore Ts'o stats->used_dirs = atomic_read(&flex_group[g].used_dirs); 3677d39db14STheodore Ts'o return; 3687d39db14STheodore Ts'o } 3697d39db14STheodore Ts'o 3707d39db14STheodore Ts'o desc = ext4_get_group_desc(sb, g, NULL); 3717d39db14STheodore Ts'o if (desc) { 3727d39db14STheodore Ts'o stats->free_inodes = ext4_free_inodes_count(sb, desc); 373021b65bbSTheodore Ts'o stats->free_clusters = ext4_free_group_clusters(sb, desc); 3747d39db14STheodore Ts'o stats->used_dirs = ext4_used_dirs_count(sb, desc); 3757d39db14STheodore Ts'o } else { 376a4912123STheodore Ts'o stats->free_inodes = 0; 37724aaa8efSTheodore Ts'o stats->free_clusters = 0; 378a4912123STheodore Ts'o stats->used_dirs = 0; 379a4912123STheodore Ts'o } 380a4912123STheodore Ts'o } 381a4912123STheodore Ts'o 382ac27a0ecSDave Kleikamp /* 383ac27a0ecSDave Kleikamp * Orlov's allocator for directories. 384ac27a0ecSDave Kleikamp * 385ac27a0ecSDave Kleikamp * We always try to spread first-level directories. 386ac27a0ecSDave Kleikamp * 387ac27a0ecSDave Kleikamp * If there are blockgroups with both free inodes and free blocks counts 388ac27a0ecSDave Kleikamp * not worse than average we return one with smallest directory count. 389ac27a0ecSDave Kleikamp * Otherwise we simply return a random group. 390ac27a0ecSDave Kleikamp * 391ac27a0ecSDave Kleikamp * For the rest rules look so: 392ac27a0ecSDave Kleikamp * 393ac27a0ecSDave Kleikamp * It's OK to put directory into a group unless 394ac27a0ecSDave Kleikamp * it has too many directories already (max_dirs) or 395ac27a0ecSDave Kleikamp * it has too few free inodes left (min_inodes) or 396ac27a0ecSDave Kleikamp * it has too few free blocks left (min_blocks) or 3971cc8dcf5SBenoit Boissinot * Parent's group is preferred, if it doesn't satisfy these 398ac27a0ecSDave Kleikamp * conditions we search cyclically through the rest. If none 399ac27a0ecSDave Kleikamp * of the groups look good we just look for a group with more 400ac27a0ecSDave Kleikamp * free inodes than average (starting at parent's group). 401ac27a0ecSDave Kleikamp */ 402ac27a0ecSDave Kleikamp 4032aa9fc4cSAvantika Mathur static int find_group_orlov(struct super_block *sb, struct inode *parent, 404dcca3fecSAl Viro ext4_group_t *group, umode_t mode, 405f157a4aaSTheodore Ts'o const struct qstr *qstr) 406ac27a0ecSDave Kleikamp { 407fd2d4291SAvantika Mathur ext4_group_t parent_group = EXT4_I(parent)->i_block_group; 408617ba13bSMingming Cao struct ext4_sb_info *sbi = EXT4_SB(sb); 4098df9675fSTheodore Ts'o ext4_group_t real_ngroups = ext4_get_groups_count(sb); 410617ba13bSMingming Cao int inodes_per_group = EXT4_INODES_PER_GROUP(sb); 41114c83c9fSTheodore Ts'o unsigned int freei, avefreei, grp_free; 41224aaa8efSTheodore Ts'o ext4_fsblk_t freeb, avefreec; 413ac27a0ecSDave Kleikamp unsigned int ndirs; 414a4912123STheodore Ts'o int max_dirs, min_inodes; 41524aaa8efSTheodore Ts'o ext4_grpblk_t min_clusters; 4168df9675fSTheodore Ts'o ext4_group_t i, grp, g, ngroups; 417617ba13bSMingming Cao struct ext4_group_desc *desc; 418a4912123STheodore Ts'o struct orlov_stats stats; 419a4912123STheodore Ts'o int flex_size = ext4_flex_bg_size(sbi); 420f157a4aaSTheodore Ts'o struct dx_hash_info hinfo; 421a4912123STheodore Ts'o 4228df9675fSTheodore Ts'o ngroups = real_ngroups; 423a4912123STheodore Ts'o if (flex_size > 1) { 4248df9675fSTheodore Ts'o ngroups = (real_ngroups + flex_size - 1) >> 425a4912123STheodore Ts'o sbi->s_log_groups_per_flex; 426a4912123STheodore Ts'o parent_group >>= sbi->s_log_groups_per_flex; 427a4912123STheodore Ts'o } 428ac27a0ecSDave Kleikamp 429ac27a0ecSDave Kleikamp freei = percpu_counter_read_positive(&sbi->s_freeinodes_counter); 430ac27a0ecSDave Kleikamp avefreei = freei / ngroups; 43157042651STheodore Ts'o freeb = EXT4_C2B(sbi, 43257042651STheodore Ts'o percpu_counter_read_positive(&sbi->s_freeclusters_counter)); 43324aaa8efSTheodore Ts'o avefreec = freeb; 43424aaa8efSTheodore Ts'o do_div(avefreec, ngroups); 435ac27a0ecSDave Kleikamp ndirs = percpu_counter_read_positive(&sbi->s_dirs_counter); 436ac27a0ecSDave Kleikamp 437a4912123STheodore Ts'o if (S_ISDIR(mode) && 4382b0143b5SDavid Howells ((parent == d_inode(sb->s_root)) || 43912e9b892SDmitry Monakhov (ext4_test_inode_flag(parent, EXT4_INODE_TOPDIR)))) { 440ac27a0ecSDave Kleikamp int best_ndir = inodes_per_group; 4412aa9fc4cSAvantika Mathur int ret = -1; 442ac27a0ecSDave Kleikamp 443f157a4aaSTheodore Ts'o if (qstr) { 444f157a4aaSTheodore Ts'o hinfo.hash_version = DX_HASH_HALF_MD4; 445f157a4aaSTheodore Ts'o hinfo.seed = sbi->s_hash_seed; 446f157a4aaSTheodore Ts'o ext4fs_dirhash(qstr->name, qstr->len, &hinfo); 447f157a4aaSTheodore Ts'o grp = hinfo.hash; 448f157a4aaSTheodore Ts'o } else 449dd1f723bSTheodore Ts'o grp = prandom_u32(); 4502aa9fc4cSAvantika Mathur parent_group = (unsigned)grp % ngroups; 451ac27a0ecSDave Kleikamp for (i = 0; i < ngroups; i++) { 452a4912123STheodore Ts'o g = (parent_group + i) % ngroups; 453a4912123STheodore Ts'o get_orlov_stats(sb, g, flex_size, &stats); 454a4912123STheodore Ts'o if (!stats.free_inodes) 455ac27a0ecSDave Kleikamp continue; 456a4912123STheodore Ts'o if (stats.used_dirs >= best_ndir) 457ac27a0ecSDave Kleikamp continue; 458a4912123STheodore Ts'o if (stats.free_inodes < avefreei) 459ac27a0ecSDave Kleikamp continue; 46024aaa8efSTheodore Ts'o if (stats.free_clusters < avefreec) 461ac27a0ecSDave Kleikamp continue; 462a4912123STheodore Ts'o grp = g; 4632aa9fc4cSAvantika Mathur ret = 0; 464a4912123STheodore Ts'o best_ndir = stats.used_dirs; 465ac27a0ecSDave Kleikamp } 466a4912123STheodore Ts'o if (ret) 467a4912123STheodore Ts'o goto fallback; 468a4912123STheodore Ts'o found_flex_bg: 469a4912123STheodore Ts'o if (flex_size == 1) { 470a4912123STheodore Ts'o *group = grp; 471a4912123STheodore Ts'o return 0; 472a4912123STheodore Ts'o } 473a4912123STheodore Ts'o 474a4912123STheodore Ts'o /* 475a4912123STheodore Ts'o * We pack inodes at the beginning of the flexgroup's 476a4912123STheodore Ts'o * inode tables. Block allocation decisions will do 477a4912123STheodore Ts'o * something similar, although regular files will 478a4912123STheodore Ts'o * start at 2nd block group of the flexgroup. See 479a4912123STheodore Ts'o * ext4_ext_find_goal() and ext4_find_near(). 480a4912123STheodore Ts'o */ 481a4912123STheodore Ts'o grp *= flex_size; 482a4912123STheodore Ts'o for (i = 0; i < flex_size; i++) { 4838df9675fSTheodore Ts'o if (grp+i >= real_ngroups) 484a4912123STheodore Ts'o break; 485a4912123STheodore Ts'o desc = ext4_get_group_desc(sb, grp+i, NULL); 486a4912123STheodore Ts'o if (desc && ext4_free_inodes_count(sb, desc)) { 487a4912123STheodore Ts'o *group = grp+i; 488a4912123STheodore Ts'o return 0; 489a4912123STheodore Ts'o } 490a4912123STheodore Ts'o } 491ac27a0ecSDave Kleikamp goto fallback; 492ac27a0ecSDave Kleikamp } 493ac27a0ecSDave Kleikamp 494ac27a0ecSDave Kleikamp max_dirs = ndirs / ngroups + inodes_per_group / 16; 495a4912123STheodore Ts'o min_inodes = avefreei - inodes_per_group*flex_size / 4; 496a4912123STheodore Ts'o if (min_inodes < 1) 497a4912123STheodore Ts'o min_inodes = 1; 49824aaa8efSTheodore Ts'o min_clusters = avefreec - EXT4_CLUSTERS_PER_GROUP(sb)*flex_size / 4; 499ac27a0ecSDave Kleikamp 500a4912123STheodore Ts'o /* 501a4912123STheodore Ts'o * Start looking in the flex group where we last allocated an 502a4912123STheodore Ts'o * inode for this parent directory 503a4912123STheodore Ts'o */ 504a4912123STheodore Ts'o if (EXT4_I(parent)->i_last_alloc_group != ~0) { 505a4912123STheodore Ts'o parent_group = EXT4_I(parent)->i_last_alloc_group; 506a4912123STheodore Ts'o if (flex_size > 1) 507a4912123STheodore Ts'o parent_group >>= sbi->s_log_groups_per_flex; 508a4912123STheodore Ts'o } 509ac27a0ecSDave Kleikamp 510ac27a0ecSDave Kleikamp for (i = 0; i < ngroups; i++) { 511a4912123STheodore Ts'o grp = (parent_group + i) % ngroups; 512a4912123STheodore Ts'o get_orlov_stats(sb, grp, flex_size, &stats); 513a4912123STheodore Ts'o if (stats.used_dirs >= max_dirs) 514ac27a0ecSDave Kleikamp continue; 515a4912123STheodore Ts'o if (stats.free_inodes < min_inodes) 516ac27a0ecSDave Kleikamp continue; 51724aaa8efSTheodore Ts'o if (stats.free_clusters < min_clusters) 518ac27a0ecSDave Kleikamp continue; 519a4912123STheodore Ts'o goto found_flex_bg; 520ac27a0ecSDave Kleikamp } 521ac27a0ecSDave Kleikamp 522ac27a0ecSDave Kleikamp fallback: 5238df9675fSTheodore Ts'o ngroups = real_ngroups; 524a4912123STheodore Ts'o avefreei = freei / ngroups; 525b5451f7bSTheodore Ts'o fallback_retry: 526a4912123STheodore Ts'o parent_group = EXT4_I(parent)->i_block_group; 527ac27a0ecSDave Kleikamp for (i = 0; i < ngroups; i++) { 528a4912123STheodore Ts'o grp = (parent_group + i) % ngroups; 529a4912123STheodore Ts'o desc = ext4_get_group_desc(sb, grp, NULL); 530bb3d132aSDan Carpenter if (desc) { 53114c83c9fSTheodore Ts'o grp_free = ext4_free_inodes_count(sb, desc); 532bb3d132aSDan Carpenter if (grp_free && grp_free >= avefreei) { 533a4912123STheodore Ts'o *group = grp; 5342aa9fc4cSAvantika Mathur return 0; 535ac27a0ecSDave Kleikamp } 536a4912123STheodore Ts'o } 537bb3d132aSDan Carpenter } 538ac27a0ecSDave Kleikamp 539ac27a0ecSDave Kleikamp if (avefreei) { 540ac27a0ecSDave Kleikamp /* 541ac27a0ecSDave Kleikamp * The free-inodes counter is approximate, and for really small 542ac27a0ecSDave Kleikamp * filesystems the above test can fail to find any blockgroups 543ac27a0ecSDave Kleikamp */ 544ac27a0ecSDave Kleikamp avefreei = 0; 545b5451f7bSTheodore Ts'o goto fallback_retry; 546ac27a0ecSDave Kleikamp } 547ac27a0ecSDave Kleikamp 548ac27a0ecSDave Kleikamp return -1; 549ac27a0ecSDave Kleikamp } 550ac27a0ecSDave Kleikamp 5512aa9fc4cSAvantika Mathur static int find_group_other(struct super_block *sb, struct inode *parent, 552dcca3fecSAl Viro ext4_group_t *group, umode_t mode) 553ac27a0ecSDave Kleikamp { 554fd2d4291SAvantika Mathur ext4_group_t parent_group = EXT4_I(parent)->i_block_group; 5558df9675fSTheodore Ts'o ext4_group_t i, last, ngroups = ext4_get_groups_count(sb); 556617ba13bSMingming Cao struct ext4_group_desc *desc; 557a4912123STheodore Ts'o int flex_size = ext4_flex_bg_size(EXT4_SB(sb)); 558a4912123STheodore Ts'o 559a4912123STheodore Ts'o /* 560a4912123STheodore Ts'o * Try to place the inode is the same flex group as its 561a4912123STheodore Ts'o * parent. If we can't find space, use the Orlov algorithm to 562a4912123STheodore Ts'o * find another flex group, and store that information in the 563a4912123STheodore Ts'o * parent directory's inode information so that use that flex 564a4912123STheodore Ts'o * group for future allocations. 565a4912123STheodore Ts'o */ 566a4912123STheodore Ts'o if (flex_size > 1) { 567a4912123STheodore Ts'o int retry = 0; 568a4912123STheodore Ts'o 569a4912123STheodore Ts'o try_again: 570a4912123STheodore Ts'o parent_group &= ~(flex_size-1); 571a4912123STheodore Ts'o last = parent_group + flex_size; 572a4912123STheodore Ts'o if (last > ngroups) 573a4912123STheodore Ts'o last = ngroups; 574a4912123STheodore Ts'o for (i = parent_group; i < last; i++) { 575a4912123STheodore Ts'o desc = ext4_get_group_desc(sb, i, NULL); 576a4912123STheodore Ts'o if (desc && ext4_free_inodes_count(sb, desc)) { 577a4912123STheodore Ts'o *group = i; 578a4912123STheodore Ts'o return 0; 579a4912123STheodore Ts'o } 580a4912123STheodore Ts'o } 581a4912123STheodore Ts'o if (!retry && EXT4_I(parent)->i_last_alloc_group != ~0) { 582a4912123STheodore Ts'o retry = 1; 583a4912123STheodore Ts'o parent_group = EXT4_I(parent)->i_last_alloc_group; 584a4912123STheodore Ts'o goto try_again; 585a4912123STheodore Ts'o } 586a4912123STheodore Ts'o /* 587a4912123STheodore Ts'o * If this didn't work, use the Orlov search algorithm 588a4912123STheodore Ts'o * to find a new flex group; we pass in the mode to 589a4912123STheodore Ts'o * avoid the topdir algorithms. 590a4912123STheodore Ts'o */ 591a4912123STheodore Ts'o *group = parent_group + flex_size; 592a4912123STheodore Ts'o if (*group > ngroups) 593a4912123STheodore Ts'o *group = 0; 5947dc57615SPeter Huewe return find_group_orlov(sb, parent, group, mode, NULL); 595a4912123STheodore Ts'o } 596ac27a0ecSDave Kleikamp 597ac27a0ecSDave Kleikamp /* 598ac27a0ecSDave Kleikamp * Try to place the inode in its parent directory 599ac27a0ecSDave Kleikamp */ 6002aa9fc4cSAvantika Mathur *group = parent_group; 6012aa9fc4cSAvantika Mathur desc = ext4_get_group_desc(sb, *group, NULL); 602560671a0SAneesh Kumar K.V if (desc && ext4_free_inodes_count(sb, desc) && 603021b65bbSTheodore Ts'o ext4_free_group_clusters(sb, desc)) 6042aa9fc4cSAvantika Mathur return 0; 605ac27a0ecSDave Kleikamp 606ac27a0ecSDave Kleikamp /* 607ac27a0ecSDave Kleikamp * We're going to place this inode in a different blockgroup from its 608ac27a0ecSDave Kleikamp * parent. We want to cause files in a common directory to all land in 609ac27a0ecSDave Kleikamp * the same blockgroup. But we want files which are in a different 610ac27a0ecSDave Kleikamp * directory which shares a blockgroup with our parent to land in a 611ac27a0ecSDave Kleikamp * different blockgroup. 612ac27a0ecSDave Kleikamp * 613ac27a0ecSDave Kleikamp * So add our directory's i_ino into the starting point for the hash. 614ac27a0ecSDave Kleikamp */ 6152aa9fc4cSAvantika Mathur *group = (*group + parent->i_ino) % ngroups; 616ac27a0ecSDave Kleikamp 617ac27a0ecSDave Kleikamp /* 618ac27a0ecSDave Kleikamp * Use a quadratic hash to find a group with a free inode and some free 619ac27a0ecSDave Kleikamp * blocks. 620ac27a0ecSDave Kleikamp */ 621ac27a0ecSDave Kleikamp for (i = 1; i < ngroups; i <<= 1) { 6222aa9fc4cSAvantika Mathur *group += i; 6232aa9fc4cSAvantika Mathur if (*group >= ngroups) 6242aa9fc4cSAvantika Mathur *group -= ngroups; 6252aa9fc4cSAvantika Mathur desc = ext4_get_group_desc(sb, *group, NULL); 626560671a0SAneesh Kumar K.V if (desc && ext4_free_inodes_count(sb, desc) && 627021b65bbSTheodore Ts'o ext4_free_group_clusters(sb, desc)) 6282aa9fc4cSAvantika Mathur return 0; 629ac27a0ecSDave Kleikamp } 630ac27a0ecSDave Kleikamp 631ac27a0ecSDave Kleikamp /* 632ac27a0ecSDave Kleikamp * That failed: try linear search for a free inode, even if that group 633ac27a0ecSDave Kleikamp * has no free blocks. 634ac27a0ecSDave Kleikamp */ 6352aa9fc4cSAvantika Mathur *group = parent_group; 636ac27a0ecSDave Kleikamp for (i = 0; i < ngroups; i++) { 6372aa9fc4cSAvantika Mathur if (++*group >= ngroups) 6382aa9fc4cSAvantika Mathur *group = 0; 6392aa9fc4cSAvantika Mathur desc = ext4_get_group_desc(sb, *group, NULL); 640560671a0SAneesh Kumar K.V if (desc && ext4_free_inodes_count(sb, desc)) 6412aa9fc4cSAvantika Mathur return 0; 642ac27a0ecSDave Kleikamp } 643ac27a0ecSDave Kleikamp 644ac27a0ecSDave Kleikamp return -1; 645ac27a0ecSDave Kleikamp } 646ac27a0ecSDave Kleikamp 647ac27a0ecSDave Kleikamp /* 64819883bd9STheodore Ts'o * In no journal mode, if an inode has recently been deleted, we want 64919883bd9STheodore Ts'o * to avoid reusing it until we're reasonably sure the inode table 65019883bd9STheodore Ts'o * block has been written back to disk. (Yes, these values are 65119883bd9STheodore Ts'o * somewhat arbitrary...) 65219883bd9STheodore Ts'o */ 65319883bd9STheodore Ts'o #define RECENTCY_MIN 5 654b5f51573SAndreas Dilger #define RECENTCY_DIRTY 300 65519883bd9STheodore Ts'o 65619883bd9STheodore Ts'o static int recently_deleted(struct super_block *sb, ext4_group_t group, int ino) 65719883bd9STheodore Ts'o { 65819883bd9STheodore Ts'o struct ext4_group_desc *gdp; 65919883bd9STheodore Ts'o struct ext4_inode *raw_inode; 66019883bd9STheodore Ts'o struct buffer_head *bh; 66119883bd9STheodore Ts'o int inodes_per_block = EXT4_SB(sb)->s_inodes_per_block; 662b5f51573SAndreas Dilger int offset, ret = 0; 663b5f51573SAndreas Dilger int recentcy = RECENTCY_MIN; 664b5f51573SAndreas Dilger u32 dtime, now; 66519883bd9STheodore Ts'o 66619883bd9STheodore Ts'o gdp = ext4_get_group_desc(sb, group, NULL); 66719883bd9STheodore Ts'o if (unlikely(!gdp)) 66819883bd9STheodore Ts'o return 0; 66919883bd9STheodore Ts'o 6704f9d956dSJan Kara bh = sb_find_get_block(sb, ext4_inode_table(sb, gdp) + 67119883bd9STheodore Ts'o (ino / inodes_per_block)); 6724f9d956dSJan Kara if (!bh || !buffer_uptodate(bh)) 67319883bd9STheodore Ts'o /* 67419883bd9STheodore Ts'o * If the block is not in the buffer cache, then it 67519883bd9STheodore Ts'o * must have been written out. 67619883bd9STheodore Ts'o */ 67719883bd9STheodore Ts'o goto out; 67819883bd9STheodore Ts'o 67919883bd9STheodore Ts'o offset = (ino % inodes_per_block) * EXT4_INODE_SIZE(sb); 68019883bd9STheodore Ts'o raw_inode = (struct ext4_inode *) (bh->b_data + offset); 681b5f51573SAndreas Dilger 682b5f51573SAndreas Dilger /* i_dtime is only 32 bits on disk, but we only care about relative 683b5f51573SAndreas Dilger * times in the range of a few minutes (i.e. long enough to sync a 684b5f51573SAndreas Dilger * recently-deleted inode to disk), so using the low 32 bits of the 685b5f51573SAndreas Dilger * clock (a 68 year range) is enough, see time_before32() */ 68619883bd9STheodore Ts'o dtime = le32_to_cpu(raw_inode->i_dtime); 687b5f51573SAndreas Dilger now = ktime_get_real_seconds(); 68819883bd9STheodore Ts'o if (buffer_dirty(bh)) 68919883bd9STheodore Ts'o recentcy += RECENTCY_DIRTY; 69019883bd9STheodore Ts'o 691b5f51573SAndreas Dilger if (dtime && time_before32(dtime, now) && 692b5f51573SAndreas Dilger time_before32(now, dtime + recentcy)) 69319883bd9STheodore Ts'o ret = 1; 69419883bd9STheodore Ts'o out: 69519883bd9STheodore Ts'o brelse(bh); 69619883bd9STheodore Ts'o return ret; 69719883bd9STheodore Ts'o } 69819883bd9STheodore Ts'o 699901ed070SWang Shilong static int find_inode_bit(struct super_block *sb, ext4_group_t group, 700901ed070SWang Shilong struct buffer_head *bitmap, unsigned long *ino) 701901ed070SWang Shilong { 702901ed070SWang Shilong next: 703901ed070SWang Shilong *ino = ext4_find_next_zero_bit((unsigned long *) 704901ed070SWang Shilong bitmap->b_data, 705901ed070SWang Shilong EXT4_INODES_PER_GROUP(sb), *ino); 706901ed070SWang Shilong if (*ino >= EXT4_INODES_PER_GROUP(sb)) 707901ed070SWang Shilong return 0; 708901ed070SWang Shilong 709901ed070SWang Shilong if ((EXT4_SB(sb)->s_journal == NULL) && 710901ed070SWang Shilong recently_deleted(sb, group, *ino)) { 711901ed070SWang Shilong *ino = *ino + 1; 712901ed070SWang Shilong if (*ino < EXT4_INODES_PER_GROUP(sb)) 713901ed070SWang Shilong goto next; 714901ed070SWang Shilong return 0; 715901ed070SWang Shilong } 716901ed070SWang Shilong 717901ed070SWang Shilong return 1; 718901ed070SWang Shilong } 719901ed070SWang Shilong 72019883bd9STheodore Ts'o /* 721ac27a0ecSDave Kleikamp * There are two policies for allocating an inode. If the new inode is 722ac27a0ecSDave Kleikamp * a directory, then a forward search is made for a block group with both 723ac27a0ecSDave Kleikamp * free space and a low directory-to-inode ratio; if that fails, then of 724ac27a0ecSDave Kleikamp * the groups with above-average free space, that group with the fewest 725ac27a0ecSDave Kleikamp * directories already is chosen. 726ac27a0ecSDave Kleikamp * 727ac27a0ecSDave Kleikamp * For other inodes, search forward from the parent directory's block 728ac27a0ecSDave Kleikamp * group to find a free inode. 729ac27a0ecSDave Kleikamp */ 7301139575aSTheodore Ts'o struct inode *__ext4_new_inode(handle_t *handle, struct inode *dir, 7311139575aSTheodore Ts'o umode_t mode, const struct qstr *qstr, 7321b917ed8STahsin Erdogan __u32 goal, uid_t *owner, __u32 i_flags, 7331b917ed8STahsin Erdogan int handle_type, unsigned int line_no, 7341b917ed8STahsin Erdogan int nblocks) 735ac27a0ecSDave Kleikamp { 736ac27a0ecSDave Kleikamp struct super_block *sb; 7373300bedaSAneesh Kumar K.V struct buffer_head *inode_bitmap_bh = NULL; 7383300bedaSAneesh Kumar K.V struct buffer_head *group_desc_bh; 7398df9675fSTheodore Ts'o ext4_group_t ngroups, group = 0; 740ac27a0ecSDave Kleikamp unsigned long ino = 0; 741ac27a0ecSDave Kleikamp struct inode *inode; 742617ba13bSMingming Cao struct ext4_group_desc *gdp = NULL; 743617ba13bSMingming Cao struct ext4_inode_info *ei; 744617ba13bSMingming Cao struct ext4_sb_info *sbi; 745a7cdadeeSJan Kara int ret2, err; 746ac27a0ecSDave Kleikamp struct inode *ret; 7472aa9fc4cSAvantika Mathur ext4_group_t i; 748772cb7c8SJose R. Santos ext4_group_t flex_group; 74987a39389SDarrick J. Wong struct ext4_group_info *grp; 750e709e9dfSTheodore Ts'o int encrypt = 0; 751ac27a0ecSDave Kleikamp 752ac27a0ecSDave Kleikamp /* Cannot create files in a deleted directory */ 753ac27a0ecSDave Kleikamp if (!dir || !dir->i_nlink) 754ac27a0ecSDave Kleikamp return ERR_PTR(-EPERM); 755ac27a0ecSDave Kleikamp 756af65207cSTahsin Erdogan sb = dir->i_sb; 757af65207cSTahsin Erdogan sbi = EXT4_SB(sb); 758af65207cSTahsin Erdogan 759af65207cSTahsin Erdogan if (unlikely(ext4_forced_shutdown(sbi))) 7600db1ff22STheodore Ts'o return ERR_PTR(-EIO); 7610db1ff22STheodore Ts'o 762af65207cSTahsin Erdogan if ((ext4_encrypted_inode(dir) || DUMMY_ENCRYPTION_ENABLED(sbi)) && 763ad47f953STahsin Erdogan (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) && 764ad47f953STahsin Erdogan !(i_flags & EXT4_EA_INODE_FL)) { 765a7550b30SJaegeuk Kim err = fscrypt_get_encryption_info(dir); 766e709e9dfSTheodore Ts'o if (err) 767e709e9dfSTheodore Ts'o return ERR_PTR(err); 768a7550b30SJaegeuk Kim if (!fscrypt_has_encryption_key(dir)) 76954475f53SEric Biggers return ERR_PTR(-ENOKEY); 770e709e9dfSTheodore Ts'o encrypt = 1; 771e709e9dfSTheodore Ts'o } 772e709e9dfSTheodore Ts'o 773af65207cSTahsin Erdogan if (!handle && sbi->s_journal && !(i_flags & EXT4_EA_INODE_FL)) { 774af65207cSTahsin Erdogan #ifdef CONFIG_EXT4_FS_POSIX_ACL 775af65207cSTahsin Erdogan struct posix_acl *p = get_acl(dir, ACL_TYPE_DEFAULT); 776af65207cSTahsin Erdogan 777996fc447STheodore Ts'o if (IS_ERR(p)) 778996fc447STheodore Ts'o return ERR_CAST(p); 779af65207cSTahsin Erdogan if (p) { 780af65207cSTahsin Erdogan int acl_size = p->a_count * sizeof(ext4_acl_entry); 781af65207cSTahsin Erdogan 782af65207cSTahsin Erdogan nblocks += (S_ISDIR(mode) ? 2 : 1) * 783af65207cSTahsin Erdogan __ext4_xattr_set_credits(sb, NULL /* inode */, 784af65207cSTahsin Erdogan NULL /* block_bh */, acl_size, 785af65207cSTahsin Erdogan true /* is_create */); 786af65207cSTahsin Erdogan posix_acl_release(p); 787af65207cSTahsin Erdogan } 788af65207cSTahsin Erdogan #endif 789af65207cSTahsin Erdogan 790af65207cSTahsin Erdogan #ifdef CONFIG_SECURITY 791af65207cSTahsin Erdogan { 792af65207cSTahsin Erdogan int num_security_xattrs = 1; 793af65207cSTahsin Erdogan 794af65207cSTahsin Erdogan #ifdef CONFIG_INTEGRITY 795af65207cSTahsin Erdogan num_security_xattrs++; 796af65207cSTahsin Erdogan #endif 797af65207cSTahsin Erdogan /* 798af65207cSTahsin Erdogan * We assume that security xattrs are never 799af65207cSTahsin Erdogan * more than 1k. In practice they are under 800af65207cSTahsin Erdogan * 128 bytes. 801af65207cSTahsin Erdogan */ 802af65207cSTahsin Erdogan nblocks += num_security_xattrs * 803af65207cSTahsin Erdogan __ext4_xattr_set_credits(sb, NULL /* inode */, 804af65207cSTahsin Erdogan NULL /* block_bh */, 1024, 805af65207cSTahsin Erdogan true /* is_create */); 806af65207cSTahsin Erdogan } 807af65207cSTahsin Erdogan #endif 808af65207cSTahsin Erdogan if (encrypt) 809af65207cSTahsin Erdogan nblocks += __ext4_xattr_set_credits(sb, 810af65207cSTahsin Erdogan NULL /* inode */, NULL /* block_bh */, 811af65207cSTahsin Erdogan FSCRYPT_SET_CONTEXT_MAX_SIZE, 812af65207cSTahsin Erdogan true /* is_create */); 813af65207cSTahsin Erdogan } 814af65207cSTahsin Erdogan 8158df9675fSTheodore Ts'o ngroups = ext4_get_groups_count(sb); 8169bffad1eSTheodore Ts'o trace_ext4_request_inode(dir, mode); 817ac27a0ecSDave Kleikamp inode = new_inode(sb); 818ac27a0ecSDave Kleikamp if (!inode) 819ac27a0ecSDave Kleikamp return ERR_PTR(-ENOMEM); 820617ba13bSMingming Cao ei = EXT4_I(inode); 821772cb7c8SJose R. Santos 822eb9cc7e1SJan Kara /* 823b8a07463SAdam Buchbinder * Initialize owners and quota early so that we don't have to account 824eb9cc7e1SJan Kara * for quota initialization worst case in standard inode creating 825eb9cc7e1SJan Kara * transaction 826eb9cc7e1SJan Kara */ 827eb9cc7e1SJan Kara if (owner) { 828eb9cc7e1SJan Kara inode->i_mode = mode; 829eb9cc7e1SJan Kara i_uid_write(inode, owner[0]); 830eb9cc7e1SJan Kara i_gid_write(inode, owner[1]); 831eb9cc7e1SJan Kara } else if (test_opt(sb, GRPID)) { 832eb9cc7e1SJan Kara inode->i_mode = mode; 833eb9cc7e1SJan Kara inode->i_uid = current_fsuid(); 834eb9cc7e1SJan Kara inode->i_gid = dir->i_gid; 835eb9cc7e1SJan Kara } else 836eb9cc7e1SJan Kara inode_init_owner(inode, dir, mode); 837040cb378SLi Xi 8380b7b7779SKaho Ng if (ext4_has_feature_project(sb) && 839040cb378SLi Xi ext4_test_inode_flag(dir, EXT4_INODE_PROJINHERIT)) 840040cb378SLi Xi ei->i_projid = EXT4_I(dir)->i_projid; 841040cb378SLi Xi else 842040cb378SLi Xi ei->i_projid = make_kprojid(&init_user_ns, EXT4_DEF_PROJID); 843040cb378SLi Xi 844a7cdadeeSJan Kara err = dquot_initialize(inode); 845a7cdadeeSJan Kara if (err) 846a7cdadeeSJan Kara goto out; 847eb9cc7e1SJan Kara 84811013911SAndreas Dilger if (!goal) 84911013911SAndreas Dilger goal = sbi->s_inode_goal; 85011013911SAndreas Dilger 851e6462869SJohann Lombardi if (goal && goal <= le32_to_cpu(sbi->s_es->s_inodes_count)) { 85211013911SAndreas Dilger group = (goal - 1) / EXT4_INODES_PER_GROUP(sb); 85311013911SAndreas Dilger ino = (goal - 1) % EXT4_INODES_PER_GROUP(sb); 85411013911SAndreas Dilger ret2 = 0; 85511013911SAndreas Dilger goto got_group; 85611013911SAndreas Dilger } 85711013911SAndreas Dilger 8584113c4caSLukas Czerner if (S_ISDIR(mode)) 859f157a4aaSTheodore Ts'o ret2 = find_group_orlov(sb, dir, &group, mode, qstr); 8604113c4caSLukas Czerner else 861a4912123STheodore Ts'o ret2 = find_group_other(sb, dir, &group, mode); 862ac27a0ecSDave Kleikamp 863772cb7c8SJose R. Santos got_group: 864a4912123STheodore Ts'o EXT4_I(dir)->i_last_alloc_group = group; 865ac27a0ecSDave Kleikamp err = -ENOSPC; 8662aa9fc4cSAvantika Mathur if (ret2 == -1) 867ac27a0ecSDave Kleikamp goto out; 868ac27a0ecSDave Kleikamp 869119c0d44STheodore Ts'o /* 870119c0d44STheodore Ts'o * Normally we will only go through one pass of this loop, 871119c0d44STheodore Ts'o * unless we get unlucky and it turns out the group we selected 872119c0d44STheodore Ts'o * had its last inode grabbed by someone else. 873119c0d44STheodore Ts'o */ 87411013911SAndreas Dilger for (i = 0; i < ngroups; i++, ino = 0) { 875ac27a0ecSDave Kleikamp err = -EIO; 876ac27a0ecSDave Kleikamp 8773300bedaSAneesh Kumar K.V gdp = ext4_get_group_desc(sb, group, &group_desc_bh); 878ac27a0ecSDave Kleikamp if (!gdp) 879eb9cc7e1SJan Kara goto out; 880ac27a0ecSDave Kleikamp 881f2a09af6SYongqiang Yang /* 882f2a09af6SYongqiang Yang * Check free inodes count before loading bitmap. 883f2a09af6SYongqiang Yang */ 8842fe435d8SWang Shilong if (ext4_free_inodes_count(sb, gdp) == 0) 8852fe435d8SWang Shilong goto next_group; 886f2a09af6SYongqiang Yang 88787a39389SDarrick J. Wong grp = ext4_get_group_info(sb, group); 88887a39389SDarrick J. Wong /* Skip groups with already-known suspicious inode tables */ 8892fe435d8SWang Shilong if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp)) 8902fe435d8SWang Shilong goto next_group; 89187a39389SDarrick J. Wong 8923300bedaSAneesh Kumar K.V brelse(inode_bitmap_bh); 8933300bedaSAneesh Kumar K.V inode_bitmap_bh = ext4_read_inode_bitmap(sb, group); 89487a39389SDarrick J. Wong /* Skip groups with suspicious inode tables */ 8959008a58eSDarrick J. Wong if (EXT4_MB_GRP_IBITMAP_CORRUPT(grp) || 8969008a58eSDarrick J. Wong IS_ERR(inode_bitmap_bh)) { 8979008a58eSDarrick J. Wong inode_bitmap_bh = NULL; 8982fe435d8SWang Shilong goto next_group; 89987a39389SDarrick J. Wong } 900ac27a0ecSDave Kleikamp 901ac27a0ecSDave Kleikamp repeat_in_this_group: 902901ed070SWang Shilong ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino); 903901ed070SWang Shilong if (!ret2) 904a34eb503STheodore Ts'o goto next_group; 905901ed070SWang Shilong 906119c0d44STheodore Ts'o if (group == 0 && (ino + 1) < EXT4_FIRST_INO(sb)) { 907119c0d44STheodore Ts'o ext4_error(sb, "reserved inode found cleared - " 908119c0d44STheodore Ts'o "inode=%lu", ino + 1); 909*206f6d55SWang Shilong ext4_mark_group_bitmap_corrupted(sb, group, 910*206f6d55SWang Shilong EXT4_GROUP_INFO_IBITMAP_CORRUPT); 9112fe435d8SWang Shilong goto next_group; 912119c0d44STheodore Ts'o } 913901ed070SWang Shilong 9141139575aSTheodore Ts'o if (!handle) { 9151139575aSTheodore Ts'o BUG_ON(nblocks <= 0); 9161139575aSTheodore Ts'o handle = __ext4_journal_start_sb(dir->i_sb, line_no, 9175fe2fe89SJan Kara handle_type, nblocks, 9185fe2fe89SJan Kara 0); 9191139575aSTheodore Ts'o if (IS_ERR(handle)) { 9201139575aSTheodore Ts'o err = PTR_ERR(handle); 921eb9cc7e1SJan Kara ext4_std_error(sb, err); 922eb9cc7e1SJan Kara goto out; 9231139575aSTheodore Ts'o } 9241139575aSTheodore Ts'o } 925ffb5387eSEric Sandeen BUFFER_TRACE(inode_bitmap_bh, "get_write_access"); 926ffb5387eSEric Sandeen err = ext4_journal_get_write_access(handle, inode_bitmap_bh); 927eb9cc7e1SJan Kara if (err) { 928eb9cc7e1SJan Kara ext4_std_error(sb, err); 929eb9cc7e1SJan Kara goto out; 930eb9cc7e1SJan Kara } 931119c0d44STheodore Ts'o ext4_lock_group(sb, group); 932119c0d44STheodore Ts'o ret2 = ext4_test_and_set_bit(ino, inode_bitmap_bh->b_data); 933901ed070SWang Shilong if (ret2) { 934901ed070SWang Shilong /* Someone already took the bit. Repeat the search 935901ed070SWang Shilong * with lock held. 936901ed070SWang Shilong */ 937901ed070SWang Shilong ret2 = find_inode_bit(sb, group, inode_bitmap_bh, &ino); 938901ed070SWang Shilong if (ret2) { 939901ed070SWang Shilong ext4_set_bit(ino, inode_bitmap_bh->b_data); 940901ed070SWang Shilong ret2 = 0; 941901ed070SWang Shilong } else { 942901ed070SWang Shilong ret2 = 1; /* we didn't grab the inode */ 943901ed070SWang Shilong } 944901ed070SWang Shilong } 945119c0d44STheodore Ts'o ext4_unlock_group(sb, group); 946119c0d44STheodore Ts'o ino++; /* the inode bitmap is zero-based */ 947119c0d44STheodore Ts'o if (!ret2) 948119c0d44STheodore Ts'o goto got; /* we grabbed the inode! */ 949901ed070SWang Shilong 950119c0d44STheodore Ts'o if (ino < EXT4_INODES_PER_GROUP(sb)) 951119c0d44STheodore Ts'o goto repeat_in_this_group; 952a34eb503STheodore Ts'o next_group: 953a34eb503STheodore Ts'o if (++group == ngroups) 954a34eb503STheodore Ts'o group = 0; 955ac27a0ecSDave Kleikamp } 956ac27a0ecSDave Kleikamp err = -ENOSPC; 957ac27a0ecSDave Kleikamp goto out; 958ac27a0ecSDave Kleikamp 959ac27a0ecSDave Kleikamp got: 960ffb5387eSEric Sandeen BUFFER_TRACE(inode_bitmap_bh, "call ext4_handle_dirty_metadata"); 961ffb5387eSEric Sandeen err = ext4_handle_dirty_metadata(handle, NULL, inode_bitmap_bh); 962eb9cc7e1SJan Kara if (err) { 963eb9cc7e1SJan Kara ext4_std_error(sb, err); 964eb9cc7e1SJan Kara goto out; 965eb9cc7e1SJan Kara } 966ffb5387eSEric Sandeen 96761c219f5STheodore Ts'o BUFFER_TRACE(group_desc_bh, "get_write_access"); 96861c219f5STheodore Ts'o err = ext4_journal_get_write_access(handle, group_desc_bh); 96961c219f5STheodore Ts'o if (err) { 97061c219f5STheodore Ts'o ext4_std_error(sb, err); 97161c219f5STheodore Ts'o goto out; 97261c219f5STheodore Ts'o } 97361c219f5STheodore Ts'o 974717d50e4SAndreas Dilger /* We may have to initialize the block bitmap if it isn't already */ 975feb0ab32SDarrick J. Wong if (ext4_has_group_desc_csum(sb) && 976717d50e4SAndreas Dilger gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { 9773300bedaSAneesh Kumar K.V struct buffer_head *block_bitmap_bh; 978717d50e4SAndreas Dilger 9793300bedaSAneesh Kumar K.V block_bitmap_bh = ext4_read_block_bitmap(sb, group); 9809008a58eSDarrick J. Wong if (IS_ERR(block_bitmap_bh)) { 9819008a58eSDarrick J. Wong err = PTR_ERR(block_bitmap_bh); 982599a9b77SJan Kara goto out; 983599a9b77SJan Kara } 9843300bedaSAneesh Kumar K.V BUFFER_TRACE(block_bitmap_bh, "get block bitmap access"); 9853300bedaSAneesh Kumar K.V err = ext4_journal_get_write_access(handle, block_bitmap_bh); 986717d50e4SAndreas Dilger if (err) { 9873300bedaSAneesh Kumar K.V brelse(block_bitmap_bh); 988eb9cc7e1SJan Kara ext4_std_error(sb, err); 989eb9cc7e1SJan Kara goto out; 990717d50e4SAndreas Dilger } 991717d50e4SAndreas Dilger 992fd034a84STheodore Ts'o BUFFER_TRACE(block_bitmap_bh, "dirty block bitmap"); 993fd034a84STheodore Ts'o err = ext4_handle_dirty_metadata(handle, NULL, block_bitmap_bh); 994fd034a84STheodore Ts'o 995717d50e4SAndreas Dilger /* recheck and clear flag under lock if we still need to */ 996fd034a84STheodore Ts'o ext4_lock_group(sb, group); 997717d50e4SAndreas Dilger if (gdp->bg_flags & cpu_to_le16(EXT4_BG_BLOCK_UNINIT)) { 9983300bedaSAneesh Kumar K.V gdp->bg_flags &= cpu_to_le16(~EXT4_BG_BLOCK_UNINIT); 999021b65bbSTheodore Ts'o ext4_free_group_clusters_set(sb, gdp, 1000cff1dfd7STheodore Ts'o ext4_free_clusters_after_init(sb, group, gdp)); 1001fa77dcfaSDarrick J. Wong ext4_block_bitmap_csum_set(sb, group, gdp, 100279f1ba49STao Ma block_bitmap_bh); 1003feb0ab32SDarrick J. Wong ext4_group_desc_csum_set(sb, group, gdp); 1004717d50e4SAndreas Dilger } 1005955ce5f5SAneesh Kumar K.V ext4_unlock_group(sb, group); 1006aeb1e5d6STheodore Ts'o brelse(block_bitmap_bh); 1007717d50e4SAndreas Dilger 1008eb9cc7e1SJan Kara if (err) { 1009eb9cc7e1SJan Kara ext4_std_error(sb, err); 1010eb9cc7e1SJan Kara goto out; 1011eb9cc7e1SJan Kara } 1012717d50e4SAndreas Dilger } 1013119c0d44STheodore Ts'o 1014119c0d44STheodore Ts'o /* Update the relevant bg descriptor fields */ 101541a246d1SDarrick J. Wong if (ext4_has_group_desc_csum(sb)) { 1016119c0d44STheodore Ts'o int free; 1017119c0d44STheodore Ts'o struct ext4_group_info *grp = ext4_get_group_info(sb, group); 1018119c0d44STheodore Ts'o 1019119c0d44STheodore Ts'o down_read(&grp->alloc_sem); /* protect vs itable lazyinit */ 1020119c0d44STheodore Ts'o ext4_lock_group(sb, group); /* while we modify the bg desc */ 1021119c0d44STheodore Ts'o free = EXT4_INODES_PER_GROUP(sb) - 1022119c0d44STheodore Ts'o ext4_itable_unused_count(sb, gdp); 1023119c0d44STheodore Ts'o if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT)) { 1024119c0d44STheodore Ts'o gdp->bg_flags &= cpu_to_le16(~EXT4_BG_INODE_UNINIT); 1025119c0d44STheodore Ts'o free = 0; 1026119c0d44STheodore Ts'o } 1027119c0d44STheodore Ts'o /* 1028119c0d44STheodore Ts'o * Check the relative inode number against the last used 1029119c0d44STheodore Ts'o * relative inode number in this group. if it is greater 1030119c0d44STheodore Ts'o * we need to update the bg_itable_unused count 1031119c0d44STheodore Ts'o */ 1032119c0d44STheodore Ts'o if (ino > free) 1033119c0d44STheodore Ts'o ext4_itable_unused_set(sb, gdp, 1034119c0d44STheodore Ts'o (EXT4_INODES_PER_GROUP(sb) - ino)); 1035119c0d44STheodore Ts'o up_read(&grp->alloc_sem); 10366f2e9f0eSTao Ma } else { 10376f2e9f0eSTao Ma ext4_lock_group(sb, group); 1038119c0d44STheodore Ts'o } 10396f2e9f0eSTao Ma 1040119c0d44STheodore Ts'o ext4_free_inodes_set(sb, gdp, ext4_free_inodes_count(sb, gdp) - 1); 1041119c0d44STheodore Ts'o if (S_ISDIR(mode)) { 1042119c0d44STheodore Ts'o ext4_used_dirs_set(sb, gdp, ext4_used_dirs_count(sb, gdp) + 1); 1043119c0d44STheodore Ts'o if (sbi->s_log_groups_per_flex) { 1044119c0d44STheodore Ts'o ext4_group_t f = ext4_flex_group(sbi, group); 1045119c0d44STheodore Ts'o 1046119c0d44STheodore Ts'o atomic_inc(&sbi->s_flex_groups[f].used_dirs); 1047119c0d44STheodore Ts'o } 1048119c0d44STheodore Ts'o } 104941a246d1SDarrick J. Wong if (ext4_has_group_desc_csum(sb)) { 105041a246d1SDarrick J. Wong ext4_inode_bitmap_csum_set(sb, group, gdp, inode_bitmap_bh, 105141a246d1SDarrick J. Wong EXT4_INODES_PER_GROUP(sb) / 8); 1052feb0ab32SDarrick J. Wong ext4_group_desc_csum_set(sb, group, gdp); 1053119c0d44STheodore Ts'o } 10546f2e9f0eSTao Ma ext4_unlock_group(sb, group); 1055119c0d44STheodore Ts'o 10563300bedaSAneesh Kumar K.V BUFFER_TRACE(group_desc_bh, "call ext4_handle_dirty_metadata"); 10573300bedaSAneesh Kumar K.V err = ext4_handle_dirty_metadata(handle, NULL, group_desc_bh); 1058eb9cc7e1SJan Kara if (err) { 1059eb9cc7e1SJan Kara ext4_std_error(sb, err); 1060eb9cc7e1SJan Kara goto out; 1061eb9cc7e1SJan Kara } 1062ac27a0ecSDave Kleikamp 1063ac27a0ecSDave Kleikamp percpu_counter_dec(&sbi->s_freeinodes_counter); 1064ac27a0ecSDave Kleikamp if (S_ISDIR(mode)) 1065ac27a0ecSDave Kleikamp percpu_counter_inc(&sbi->s_dirs_counter); 1066ac27a0ecSDave Kleikamp 1067772cb7c8SJose R. Santos if (sbi->s_log_groups_per_flex) { 1068772cb7c8SJose R. Santos flex_group = ext4_flex_group(sbi, group); 10699f24e420STheodore Ts'o atomic_dec(&sbi->s_flex_groups[flex_group].free_inodes); 1070772cb7c8SJose R. Santos } 1071ac27a0ecSDave Kleikamp 1072717d50e4SAndreas Dilger inode->i_ino = ino + group * EXT4_INODES_PER_GROUP(sb); 1073ac27a0ecSDave Kleikamp /* This is the optimal IO size (for stat), not the fs block size */ 1074ac27a0ecSDave Kleikamp inode->i_blocks = 0; 1075ef7f3835SKalpak Shah inode->i_mtime = inode->i_atime = inode->i_ctime = ei->i_crtime = 1076eeca7ea1SDeepa Dinamani current_time(inode); 1077ac27a0ecSDave Kleikamp 1078ac27a0ecSDave Kleikamp memset(ei->i_data, 0, sizeof(ei->i_data)); 1079ac27a0ecSDave Kleikamp ei->i_dir_start_lookup = 0; 1080ac27a0ecSDave Kleikamp ei->i_disksize = 0; 1081ac27a0ecSDave Kleikamp 10824af83508SEryu Guan /* Don't inherit extent flag from directory, amongst others. */ 10832dc6b0d4SDuane Griffin ei->i_flags = 10842dc6b0d4SDuane Griffin ext4_mask_flags(mode, EXT4_I(dir)->i_flags & EXT4_FL_INHERITED); 10851b917ed8STahsin Erdogan ei->i_flags |= i_flags; 1086ac27a0ecSDave Kleikamp ei->i_file_acl = 0; 1087ac27a0ecSDave Kleikamp ei->i_dtime = 0; 1088ac27a0ecSDave Kleikamp ei->i_block_group = group; 1089a4912123STheodore Ts'o ei->i_last_alloc_group = ~0; 1090ac27a0ecSDave Kleikamp 1091617ba13bSMingming Cao ext4_set_inode_flags(inode); 1092ac27a0ecSDave Kleikamp if (IS_DIRSYNC(inode)) 10930390131bSFrank Mayhar ext4_handle_sync(handle); 10946b38e842SAl Viro if (insert_inode_locked(inode) < 0) { 1095acd6ad83SJan Kara /* 1096acd6ad83SJan Kara * Likely a bitmap corruption causing inode to be allocated 1097acd6ad83SJan Kara * twice. 1098acd6ad83SJan Kara */ 1099acd6ad83SJan Kara err = -EIO; 1100eb9cc7e1SJan Kara ext4_error(sb, "failed to insert inode %lu: doubly allocated?", 1101eb9cc7e1SJan Kara inode->i_ino); 1102*206f6d55SWang Shilong ext4_mark_group_bitmap_corrupted(sb, group, 1103*206f6d55SWang Shilong EXT4_GROUP_INFO_IBITMAP_CORRUPT); 1104eb9cc7e1SJan Kara goto out; 11056b38e842SAl Viro } 110623253068STheodore Ts'o inode->i_generation = prandom_u32(); 1107ac27a0ecSDave Kleikamp 1108814525f4SDarrick J. Wong /* Precompute checksum seed for inode metadata */ 11099aa5d32bSDmitry Monakhov if (ext4_has_metadata_csum(sb)) { 1110814525f4SDarrick J. Wong __u32 csum; 1111814525f4SDarrick J. Wong __le32 inum = cpu_to_le32(inode->i_ino); 1112814525f4SDarrick J. Wong __le32 gen = cpu_to_le32(inode->i_generation); 1113814525f4SDarrick J. Wong csum = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&inum, 1114814525f4SDarrick J. Wong sizeof(inum)); 1115814525f4SDarrick J. Wong ei->i_csum_seed = ext4_chksum(sbi, csum, (__u8 *)&gen, 1116814525f4SDarrick J. Wong sizeof(gen)); 1117814525f4SDarrick J. Wong } 1118814525f4SDarrick J. Wong 1119353eb83cSTheodore Ts'o ext4_clear_state_flags(ei); /* Only relevant on 32-bit archs */ 112019f5fb7aSTheodore Ts'o ext4_set_inode_state(inode, EXT4_STATE_NEW); 1121ef7f3835SKalpak Shah 112249598e04SJun Piao ei->i_extra_isize = sbi->s_want_extra_isize; 1123f08225d1STao Ma ei->i_inline_off = 0; 1124e2b911c5SDarrick J. Wong if (ext4_has_feature_inline_data(sb)) 1125f08225d1STao Ma ext4_set_inode_state(inode, EXT4_STATE_MAY_INLINE_DATA); 1126ac27a0ecSDave Kleikamp ret = inode; 112763936ddaSChristoph Hellwig err = dquot_alloc_inode(inode); 112863936ddaSChristoph Hellwig if (err) 1129ac27a0ecSDave Kleikamp goto fail_drop; 1130ac27a0ecSDave Kleikamp 1131aa1dca3bSEric Biggers /* 1132aa1dca3bSEric Biggers * Since the encryption xattr will always be unique, create it first so 1133aa1dca3bSEric Biggers * that it's less likely to end up in an external xattr block and 1134aa1dca3bSEric Biggers * prevent its deduplication. 1135aa1dca3bSEric Biggers */ 1136aa1dca3bSEric Biggers if (encrypt) { 1137aa1dca3bSEric Biggers err = fscrypt_inherit_context(dir, inode, handle, true); 1138aa1dca3bSEric Biggers if (err) 1139aa1dca3bSEric Biggers goto fail_free_drop; 1140aa1dca3bSEric Biggers } 1141aa1dca3bSEric Biggers 11421b917ed8STahsin Erdogan if (!(ei->i_flags & EXT4_EA_INODE_FL)) { 1143617ba13bSMingming Cao err = ext4_init_acl(handle, inode, dir); 1144ac27a0ecSDave Kleikamp if (err) 1145ac27a0ecSDave Kleikamp goto fail_free_drop; 1146ac27a0ecSDave Kleikamp 11472a7dba39SEric Paris err = ext4_init_security(handle, inode, dir, qstr); 1148ac27a0ecSDave Kleikamp if (err) 1149ac27a0ecSDave Kleikamp goto fail_free_drop; 1150ad47f953STahsin Erdogan } 1151ac27a0ecSDave Kleikamp 1152e2b911c5SDarrick J. Wong if (ext4_has_feature_extents(sb)) { 1153e4079a11SEric Sandeen /* set extent flag only for directory, file and normal symlink*/ 1154e65187e6SAneesh Kumar K.V if (S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode)) { 115512e9b892SDmitry Monakhov ext4_set_inode_flag(inode, EXT4_INODE_EXTENTS); 1156a86c6181SAlex Tomas ext4_ext_tree_init(handle, inode); 1157a86c6181SAlex Tomas } 115842bf0383SAneesh Kumar K.V } 1159ac27a0ecSDave Kleikamp 1160688f869cSTheodore Ts'o if (ext4_handle_valid(handle)) { 1161688f869cSTheodore Ts'o ei->i_sync_tid = handle->h_transaction->t_tid; 1162688f869cSTheodore Ts'o ei->i_datasync_tid = handle->h_transaction->t_tid; 1163688f869cSTheodore Ts'o } 1164688f869cSTheodore Ts'o 11658753e88fSAneesh Kumar K.V err = ext4_mark_inode_dirty(handle, inode); 11668753e88fSAneesh Kumar K.V if (err) { 11678753e88fSAneesh Kumar K.V ext4_std_error(sb, err); 11688753e88fSAneesh Kumar K.V goto fail_free_drop; 11698753e88fSAneesh Kumar K.V } 11708753e88fSAneesh Kumar K.V 1171617ba13bSMingming Cao ext4_debug("allocating inode %lu\n", inode->i_ino); 11729bffad1eSTheodore Ts'o trace_ext4_allocate_inode(inode, dir, mode); 11733300bedaSAneesh Kumar K.V brelse(inode_bitmap_bh); 1174ac27a0ecSDave Kleikamp return ret; 1175ac27a0ecSDave Kleikamp 1176ac27a0ecSDave Kleikamp fail_free_drop: 117763936ddaSChristoph Hellwig dquot_free_inode(inode); 1178ac27a0ecSDave Kleikamp fail_drop: 11796d6b77f1SMiklos Szeredi clear_nlink(inode); 11806b38e842SAl Viro unlock_new_inode(inode); 1181eb9cc7e1SJan Kara out: 1182eb9cc7e1SJan Kara dquot_drop(inode); 1183eb9cc7e1SJan Kara inode->i_flags |= S_NOQUOTA; 1184ac27a0ecSDave Kleikamp iput(inode); 11853300bedaSAneesh Kumar K.V brelse(inode_bitmap_bh); 1186ac27a0ecSDave Kleikamp return ERR_PTR(err); 1187ac27a0ecSDave Kleikamp } 1188ac27a0ecSDave Kleikamp 1189ac27a0ecSDave Kleikamp /* Verify that we are loading a valid orphan from disk */ 1190617ba13bSMingming Cao struct inode *ext4_orphan_get(struct super_block *sb, unsigned long ino) 1191ac27a0ecSDave Kleikamp { 1192617ba13bSMingming Cao unsigned long max_ino = le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count); 1193fd2d4291SAvantika Mathur ext4_group_t block_group; 1194ac27a0ecSDave Kleikamp int bit; 11957827a7f6STheodore Ts'o struct buffer_head *bitmap_bh = NULL; 1196ac27a0ecSDave Kleikamp struct inode *inode = NULL; 11977827a7f6STheodore Ts'o int err = -EFSCORRUPTED; 1198ac27a0ecSDave Kleikamp 11997827a7f6STheodore Ts'o if (ino < EXT4_FIRST_INO(sb) || ino > max_ino) 12007827a7f6STheodore Ts'o goto bad_orphan; 1201ac27a0ecSDave Kleikamp 1202617ba13bSMingming Cao block_group = (ino - 1) / EXT4_INODES_PER_GROUP(sb); 1203617ba13bSMingming Cao bit = (ino - 1) % EXT4_INODES_PER_GROUP(sb); 1204e29d1cdeSEric Sandeen bitmap_bh = ext4_read_inode_bitmap(sb, block_group); 1205*206f6d55SWang Shilong if (IS_ERR(bitmap_bh)) 12067827a7f6STheodore Ts'o return (struct inode *) bitmap_bh; 1207ac27a0ecSDave Kleikamp 1208ac27a0ecSDave Kleikamp /* Having the inode bit set should be a 100% indicator that this 1209ac27a0ecSDave Kleikamp * is a valid orphan (no e2fsck run on fs). Orphans also include 1210ac27a0ecSDave Kleikamp * inodes that were being truncated, so we can't check i_nlink==0. 1211ac27a0ecSDave Kleikamp */ 12121d1fe1eeSDavid Howells if (!ext4_test_bit(bit, bitmap_bh->b_data)) 12131d1fe1eeSDavid Howells goto bad_orphan; 12141d1fe1eeSDavid Howells 12151d1fe1eeSDavid Howells inode = ext4_iget(sb, ino); 12167827a7f6STheodore Ts'o if (IS_ERR(inode)) { 12177827a7f6STheodore Ts'o err = PTR_ERR(inode); 12187827a7f6STheodore Ts'o ext4_error(sb, "couldn't read orphan inode %lu (err %d)", 12197827a7f6STheodore Ts'o ino, err); 12207827a7f6STheodore Ts'o return inode; 12217827a7f6STheodore Ts'o } 12221d1fe1eeSDavid Howells 122391ef4cafSDuane Griffin /* 1224c9eb13a9STheodore Ts'o * If the orphans has i_nlinks > 0 then it should be able to 1225c9eb13a9STheodore Ts'o * be truncated, otherwise it won't be removed from the orphan 1226c9eb13a9STheodore Ts'o * list during processing and an infinite loop will result. 1227c9eb13a9STheodore Ts'o * Similarly, it must not be a bad inode. 122891ef4cafSDuane Griffin */ 1229c9eb13a9STheodore Ts'o if ((inode->i_nlink && !ext4_can_truncate(inode)) || 1230c9eb13a9STheodore Ts'o is_bad_inode(inode)) 123191ef4cafSDuane Griffin goto bad_orphan; 123291ef4cafSDuane Griffin 12331d1fe1eeSDavid Howells if (NEXT_ORPHAN(inode) > max_ino) 12341d1fe1eeSDavid Howells goto bad_orphan; 12351d1fe1eeSDavid Howells brelse(bitmap_bh); 12361d1fe1eeSDavid Howells return inode; 12371d1fe1eeSDavid Howells 12381d1fe1eeSDavid Howells bad_orphan: 12397827a7f6STheodore Ts'o ext4_error(sb, "bad orphan inode %lu", ino); 12407827a7f6STheodore Ts'o if (bitmap_bh) 12417827a7f6STheodore Ts'o printk(KERN_ERR "ext4_test_bit(bit=%d, block=%llu) = %d\n", 1242ac27a0ecSDave Kleikamp bit, (unsigned long long)bitmap_bh->b_blocknr, 1243617ba13bSMingming Cao ext4_test_bit(bit, bitmap_bh->b_data)); 1244ac27a0ecSDave Kleikamp if (inode) { 12457827a7f6STheodore Ts'o printk(KERN_ERR "is_bad_inode(inode)=%d\n", 1246ac27a0ecSDave Kleikamp is_bad_inode(inode)); 12477827a7f6STheodore Ts'o printk(KERN_ERR "NEXT_ORPHAN(inode)=%u\n", 1248ac27a0ecSDave Kleikamp NEXT_ORPHAN(inode)); 12497827a7f6STheodore Ts'o printk(KERN_ERR "max_ino=%lu\n", max_ino); 12507827a7f6STheodore Ts'o printk(KERN_ERR "i_nlink=%u\n", inode->i_nlink); 1251ac27a0ecSDave Kleikamp /* Avoid freeing blocks if we got a bad deleted inode */ 12521d1fe1eeSDavid Howells if (inode->i_nlink == 0) 1253ac27a0ecSDave Kleikamp inode->i_blocks = 0; 1254ac27a0ecSDave Kleikamp iput(inode); 1255ac27a0ecSDave Kleikamp } 1256ac27a0ecSDave Kleikamp brelse(bitmap_bh); 12571d1fe1eeSDavid Howells return ERR_PTR(err); 1258ac27a0ecSDave Kleikamp } 1259ac27a0ecSDave Kleikamp 1260617ba13bSMingming Cao unsigned long ext4_count_free_inodes(struct super_block *sb) 1261ac27a0ecSDave Kleikamp { 1262ac27a0ecSDave Kleikamp unsigned long desc_count; 1263617ba13bSMingming Cao struct ext4_group_desc *gdp; 12648df9675fSTheodore Ts'o ext4_group_t i, ngroups = ext4_get_groups_count(sb); 1265617ba13bSMingming Cao #ifdef EXT4FS_DEBUG 1266617ba13bSMingming Cao struct ext4_super_block *es; 1267ac27a0ecSDave Kleikamp unsigned long bitmap_count, x; 1268ac27a0ecSDave Kleikamp struct buffer_head *bitmap_bh = NULL; 1269ac27a0ecSDave Kleikamp 1270617ba13bSMingming Cao es = EXT4_SB(sb)->s_es; 1271ac27a0ecSDave Kleikamp desc_count = 0; 1272ac27a0ecSDave Kleikamp bitmap_count = 0; 1273ac27a0ecSDave Kleikamp gdp = NULL; 12748df9675fSTheodore Ts'o for (i = 0; i < ngroups; i++) { 1275617ba13bSMingming Cao gdp = ext4_get_group_desc(sb, i, NULL); 1276ac27a0ecSDave Kleikamp if (!gdp) 1277ac27a0ecSDave Kleikamp continue; 1278560671a0SAneesh Kumar K.V desc_count += ext4_free_inodes_count(sb, gdp); 1279ac27a0ecSDave Kleikamp brelse(bitmap_bh); 1280e29d1cdeSEric Sandeen bitmap_bh = ext4_read_inode_bitmap(sb, i); 12819008a58eSDarrick J. Wong if (IS_ERR(bitmap_bh)) { 12829008a58eSDarrick J. Wong bitmap_bh = NULL; 1283ac27a0ecSDave Kleikamp continue; 12849008a58eSDarrick J. Wong } 1285ac27a0ecSDave Kleikamp 1286f6fb99caSTheodore Ts'o x = ext4_count_free(bitmap_bh->b_data, 1287f6fb99caSTheodore Ts'o EXT4_INODES_PER_GROUP(sb) / 8); 1288c549a95dSEric Sandeen printk(KERN_DEBUG "group %lu: stored = %d, counted = %lu\n", 1289785b4b3aSPeng Tao (unsigned long) i, ext4_free_inodes_count(sb, gdp), x); 1290ac27a0ecSDave Kleikamp bitmap_count += x; 1291ac27a0ecSDave Kleikamp } 1292ac27a0ecSDave Kleikamp brelse(bitmap_bh); 12934776004fSTheodore Ts'o printk(KERN_DEBUG "ext4_count_free_inodes: " 12944776004fSTheodore Ts'o "stored = %u, computed = %lu, %lu\n", 1295ac27a0ecSDave Kleikamp le32_to_cpu(es->s_free_inodes_count), desc_count, bitmap_count); 1296ac27a0ecSDave Kleikamp return desc_count; 1297ac27a0ecSDave Kleikamp #else 1298ac27a0ecSDave Kleikamp desc_count = 0; 12998df9675fSTheodore Ts'o for (i = 0; i < ngroups; i++) { 1300617ba13bSMingming Cao gdp = ext4_get_group_desc(sb, i, NULL); 1301ac27a0ecSDave Kleikamp if (!gdp) 1302ac27a0ecSDave Kleikamp continue; 1303560671a0SAneesh Kumar K.V desc_count += ext4_free_inodes_count(sb, gdp); 1304ac27a0ecSDave Kleikamp cond_resched(); 1305ac27a0ecSDave Kleikamp } 1306ac27a0ecSDave Kleikamp return desc_count; 1307ac27a0ecSDave Kleikamp #endif 1308ac27a0ecSDave Kleikamp } 1309ac27a0ecSDave Kleikamp 1310ac27a0ecSDave Kleikamp /* Called at mount-time, super-block is locked */ 1311617ba13bSMingming Cao unsigned long ext4_count_dirs(struct super_block * sb) 1312ac27a0ecSDave Kleikamp { 1313ac27a0ecSDave Kleikamp unsigned long count = 0; 13148df9675fSTheodore Ts'o ext4_group_t i, ngroups = ext4_get_groups_count(sb); 1315ac27a0ecSDave Kleikamp 13168df9675fSTheodore Ts'o for (i = 0; i < ngroups; i++) { 1317617ba13bSMingming Cao struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL); 1318ac27a0ecSDave Kleikamp if (!gdp) 1319ac27a0ecSDave Kleikamp continue; 1320560671a0SAneesh Kumar K.V count += ext4_used_dirs_count(sb, gdp); 1321ac27a0ecSDave Kleikamp } 1322ac27a0ecSDave Kleikamp return count; 1323ac27a0ecSDave Kleikamp } 1324bfff6873SLukas Czerner 1325bfff6873SLukas Czerner /* 1326bfff6873SLukas Czerner * Zeroes not yet zeroed inode table - just write zeroes through the whole 1327bfff6873SLukas Czerner * inode table. Must be called without any spinlock held. The only place 1328bfff6873SLukas Czerner * where it is called from on active part of filesystem is ext4lazyinit 1329bfff6873SLukas Czerner * thread, so we do not need any special locks, however we have to prevent 1330bfff6873SLukas Czerner * inode allocation from the current group, so we take alloc_sem lock, to 1331119c0d44STheodore Ts'o * block ext4_new_inode() until we are finished. 1332bfff6873SLukas Czerner */ 1333e0cbee3eSH Hartley Sweeten int ext4_init_inode_table(struct super_block *sb, ext4_group_t group, 1334bfff6873SLukas Czerner int barrier) 1335bfff6873SLukas Czerner { 1336bfff6873SLukas Czerner struct ext4_group_info *grp = ext4_get_group_info(sb, group); 1337bfff6873SLukas Czerner struct ext4_sb_info *sbi = EXT4_SB(sb); 1338bfff6873SLukas Czerner struct ext4_group_desc *gdp = NULL; 1339bfff6873SLukas Czerner struct buffer_head *group_desc_bh; 1340bfff6873SLukas Czerner handle_t *handle; 1341bfff6873SLukas Czerner ext4_fsblk_t blk; 1342bfff6873SLukas Czerner int num, ret = 0, used_blks = 0; 1343bfff6873SLukas Czerner 1344bfff6873SLukas Czerner /* This should not happen, but just to be sure check this */ 1345bc98a42cSDavid Howells if (sb_rdonly(sb)) { 1346bfff6873SLukas Czerner ret = 1; 1347bfff6873SLukas Czerner goto out; 1348bfff6873SLukas Czerner } 1349bfff6873SLukas Czerner 1350bfff6873SLukas Czerner gdp = ext4_get_group_desc(sb, group, &group_desc_bh); 1351bfff6873SLukas Czerner if (!gdp) 1352bfff6873SLukas Czerner goto out; 1353bfff6873SLukas Czerner 1354bfff6873SLukas Czerner /* 1355bfff6873SLukas Czerner * We do not need to lock this, because we are the only one 1356bfff6873SLukas Czerner * handling this flag. 1357bfff6873SLukas Czerner */ 1358bfff6873SLukas Czerner if (gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)) 1359bfff6873SLukas Czerner goto out; 1360bfff6873SLukas Czerner 13619924a92aSTheodore Ts'o handle = ext4_journal_start_sb(sb, EXT4_HT_MISC, 1); 1362bfff6873SLukas Czerner if (IS_ERR(handle)) { 1363bfff6873SLukas Czerner ret = PTR_ERR(handle); 1364bfff6873SLukas Czerner goto out; 1365bfff6873SLukas Czerner } 1366bfff6873SLukas Czerner 1367bfff6873SLukas Czerner down_write(&grp->alloc_sem); 1368bfff6873SLukas Czerner /* 1369bfff6873SLukas Czerner * If inode bitmap was already initialized there may be some 1370bfff6873SLukas Czerner * used inodes so we need to skip blocks with used inodes in 1371bfff6873SLukas Czerner * inode table. 1372bfff6873SLukas Czerner */ 1373bfff6873SLukas Czerner if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_UNINIT))) 1374bfff6873SLukas Czerner used_blks = DIV_ROUND_UP((EXT4_INODES_PER_GROUP(sb) - 1375bfff6873SLukas Czerner ext4_itable_unused_count(sb, gdp)), 1376bfff6873SLukas Czerner sbi->s_inodes_per_block); 1377bfff6873SLukas Czerner 1378857ac889SLukas Czerner if ((used_blks < 0) || (used_blks > sbi->s_itb_per_group)) { 13791084f252STheodore Ts'o ext4_error(sb, "Something is wrong with group %u: " 13801084f252STheodore Ts'o "used itable blocks: %d; " 13811084f252STheodore Ts'o "itable unused count: %u", 1382857ac889SLukas Czerner group, used_blks, 1383857ac889SLukas Czerner ext4_itable_unused_count(sb, gdp)); 1384857ac889SLukas Czerner ret = 1; 138533853a0dSYongqiang Yang goto err_out; 1386857ac889SLukas Czerner } 1387857ac889SLukas Czerner 1388bfff6873SLukas Czerner blk = ext4_inode_table(sb, gdp) + used_blks; 1389bfff6873SLukas Czerner num = sbi->s_itb_per_group - used_blks; 1390bfff6873SLukas Czerner 1391bfff6873SLukas Czerner BUFFER_TRACE(group_desc_bh, "get_write_access"); 1392bfff6873SLukas Czerner ret = ext4_journal_get_write_access(handle, 1393bfff6873SLukas Czerner group_desc_bh); 1394bfff6873SLukas Czerner if (ret) 1395bfff6873SLukas Czerner goto err_out; 1396bfff6873SLukas Czerner 1397bfff6873SLukas Czerner /* 1398bfff6873SLukas Czerner * Skip zeroout if the inode table is full. But we set the ZEROED 1399bfff6873SLukas Czerner * flag anyway, because obviously, when it is full it does not need 1400bfff6873SLukas Czerner * further zeroing. 1401bfff6873SLukas Czerner */ 1402bfff6873SLukas Czerner if (unlikely(num == 0)) 1403bfff6873SLukas Czerner goto skip_zeroout; 1404bfff6873SLukas Czerner 1405bfff6873SLukas Czerner ext4_debug("going to zero out inode table in group %d\n", 1406bfff6873SLukas Czerner group); 1407a107e5a3STheodore Ts'o ret = sb_issue_zeroout(sb, blk, num, GFP_NOFS); 1408bfff6873SLukas Czerner if (ret < 0) 1409bfff6873SLukas Czerner goto err_out; 1410a107e5a3STheodore Ts'o if (barrier) 1411a107e5a3STheodore Ts'o blkdev_issue_flush(sb->s_bdev, GFP_NOFS, NULL); 1412bfff6873SLukas Czerner 1413bfff6873SLukas Czerner skip_zeroout: 1414bfff6873SLukas Czerner ext4_lock_group(sb, group); 1415bfff6873SLukas Czerner gdp->bg_flags |= cpu_to_le16(EXT4_BG_INODE_ZEROED); 1416feb0ab32SDarrick J. Wong ext4_group_desc_csum_set(sb, group, gdp); 1417bfff6873SLukas Czerner ext4_unlock_group(sb, group); 1418bfff6873SLukas Czerner 1419bfff6873SLukas Czerner BUFFER_TRACE(group_desc_bh, 1420bfff6873SLukas Czerner "call ext4_handle_dirty_metadata"); 1421bfff6873SLukas Czerner ret = ext4_handle_dirty_metadata(handle, NULL, 1422bfff6873SLukas Czerner group_desc_bh); 1423bfff6873SLukas Czerner 1424bfff6873SLukas Czerner err_out: 1425bfff6873SLukas Czerner up_write(&grp->alloc_sem); 1426bfff6873SLukas Czerner ext4_journal_stop(handle); 1427bfff6873SLukas Czerner out: 1428bfff6873SLukas Czerner return ret; 1429bfff6873SLukas Czerner } 1430