1ac27a0ecSDave Kleikamp /* 2617ba13bSMingming Cao * linux/fs/ext4/dir.c 3ac27a0ecSDave Kleikamp * 4ac27a0ecSDave Kleikamp * Copyright (C) 1992, 1993, 1994, 1995 5ac27a0ecSDave Kleikamp * Remy Card (card@masi.ibp.fr) 6ac27a0ecSDave Kleikamp * Laboratoire MASI - Institut Blaise Pascal 7ac27a0ecSDave Kleikamp * Universite Pierre et Marie Curie (Paris VI) 8ac27a0ecSDave Kleikamp * 9ac27a0ecSDave Kleikamp * from 10ac27a0ecSDave Kleikamp * 11ac27a0ecSDave Kleikamp * linux/fs/minix/dir.c 12ac27a0ecSDave Kleikamp * 13ac27a0ecSDave Kleikamp * Copyright (C) 1991, 1992 Linus Torvalds 14ac27a0ecSDave Kleikamp * 15617ba13bSMingming Cao * ext4 directory handling functions 16ac27a0ecSDave Kleikamp * 17ac27a0ecSDave Kleikamp * Big-endian to little-endian byte-swapping/bitmaps by 18ac27a0ecSDave Kleikamp * David S. Miller (davem@caip.rutgers.edu), 1995 19ac27a0ecSDave Kleikamp * 20ac27a0ecSDave Kleikamp * Hash Tree Directory indexing (c) 2001 Daniel Phillips 21ac27a0ecSDave Kleikamp * 22ac27a0ecSDave Kleikamp */ 23ac27a0ecSDave Kleikamp 24ac27a0ecSDave Kleikamp #include <linux/fs.h> 25dab291afSMingming Cao #include <linux/jbd2.h> 26ac27a0ecSDave Kleikamp #include <linux/buffer_head.h> 27ac27a0ecSDave Kleikamp #include <linux/slab.h> 28ac27a0ecSDave Kleikamp #include <linux/rbtree.h> 293dcf5451SChristoph Hellwig #include "ext4.h" 30ac27a0ecSDave Kleikamp 31617ba13bSMingming Cao static unsigned char ext4_filetype_table[] = { 32ac27a0ecSDave Kleikamp DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK 33ac27a0ecSDave Kleikamp }; 34ac27a0ecSDave Kleikamp 35617ba13bSMingming Cao static int ext4_dx_readdir(struct file *filp, 36ac27a0ecSDave Kleikamp void *dirent, filldir_t filldir); 37ac27a0ecSDave Kleikamp 38ac27a0ecSDave Kleikamp static unsigned char get_dtype(struct super_block *sb, int filetype) 39ac27a0ecSDave Kleikamp { 40617ba13bSMingming Cao if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE) || 41617ba13bSMingming Cao (filetype >= EXT4_FT_MAX)) 42ac27a0ecSDave Kleikamp return DT_UNKNOWN; 43ac27a0ecSDave Kleikamp 44617ba13bSMingming Cao return (ext4_filetype_table[filetype]); 45ac27a0ecSDave Kleikamp } 46ac27a0ecSDave Kleikamp 47d1f5273eSFan Yong /** 48d1f5273eSFan Yong * Check if the given dir-inode refers to an htree-indexed directory 49d1f5273eSFan Yong * (or a directory which chould potentially get coverted to use htree 50d1f5273eSFan Yong * indexing). 51d1f5273eSFan Yong * 52d1f5273eSFan Yong * Return 1 if it is a dx dir, 0 if not 53d1f5273eSFan Yong */ 54d1f5273eSFan Yong static int is_dx_dir(struct inode *inode) 55d1f5273eSFan Yong { 56d1f5273eSFan Yong struct super_block *sb = inode->i_sb; 57d1f5273eSFan Yong 58d1f5273eSFan Yong if (EXT4_HAS_COMPAT_FEATURE(inode->i_sb, 59d1f5273eSFan Yong EXT4_FEATURE_COMPAT_DIR_INDEX) && 60d1f5273eSFan Yong ((ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) || 61d1f5273eSFan Yong ((inode->i_size >> sb->s_blocksize_bits) == 1))) 62d1f5273eSFan Yong return 1; 63d1f5273eSFan Yong 64d1f5273eSFan Yong return 0; 65d1f5273eSFan Yong } 66d1f5273eSFan Yong 67cad3f007STheodore Ts'o /* 68cad3f007STheodore Ts'o * Return 0 if the directory entry is OK, and 1 if there is a problem 69cad3f007STheodore Ts'o * 70cad3f007STheodore Ts'o * Note: this is the opposite of what ext2 and ext3 historically returned... 71cad3f007STheodore Ts'o */ 7260fd4da3STheodore Ts'o int __ext4_check_dir_entry(const char *function, unsigned int line, 73f7c21177STheodore Ts'o struct inode *dir, struct file *filp, 74617ba13bSMingming Cao struct ext4_dir_entry_2 *de, 75ac27a0ecSDave Kleikamp struct buffer_head *bh, 76498e5f24STheodore Ts'o unsigned int offset) 77ac27a0ecSDave Kleikamp { 78ac27a0ecSDave Kleikamp const char *error_msg = NULL; 793d0518f4SWei Yongjun const int rlen = ext4_rec_len_from_disk(de->rec_len, 803d0518f4SWei Yongjun dir->i_sb->s_blocksize); 81ac27a0ecSDave Kleikamp 82cad3f007STheodore Ts'o if (unlikely(rlen < EXT4_DIR_REC_LEN(1))) 83ac27a0ecSDave Kleikamp error_msg = "rec_len is smaller than minimal"; 84cad3f007STheodore Ts'o else if (unlikely(rlen % 4 != 0)) 85ac27a0ecSDave Kleikamp error_msg = "rec_len % 4 != 0"; 86cad3f007STheodore Ts'o else if (unlikely(rlen < EXT4_DIR_REC_LEN(de->name_len))) 87ac27a0ecSDave Kleikamp error_msg = "rec_len is too small for name_len"; 88cad3f007STheodore Ts'o else if (unlikely(((char *) de - bh->b_data) + rlen > 89cad3f007STheodore Ts'o dir->i_sb->s_blocksize)) 90ac27a0ecSDave Kleikamp error_msg = "directory entry across blocks"; 91cad3f007STheodore Ts'o else if (unlikely(le32_to_cpu(de->inode) > 92cad3f007STheodore Ts'o le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count))) 93ac27a0ecSDave Kleikamp error_msg = "inode out of bounds"; 94cad3f007STheodore Ts'o else 95cad3f007STheodore Ts'o return 0; 96ac27a0ecSDave Kleikamp 97f7c21177STheodore Ts'o if (filp) 984fda4003SDan Carpenter ext4_error_file(filp, function, line, bh->b_blocknr, 99f7c21177STheodore Ts'o "bad entry in directory: %s - offset=%u(%u), " 100f7c21177STheodore Ts'o "inode=%u, rec_len=%d, name_len=%d", 101f7c21177STheodore Ts'o error_msg, (unsigned) (offset % bh->b_size), 102f7c21177STheodore Ts'o offset, le32_to_cpu(de->inode), 103ac27a0ecSDave Kleikamp rlen, de->name_len); 104f7c21177STheodore Ts'o else 1054fda4003SDan Carpenter ext4_error_inode(dir, function, line, bh->b_blocknr, 106f7c21177STheodore Ts'o "bad entry in directory: %s - offset=%u(%u), " 107f7c21177STheodore Ts'o "inode=%u, rec_len=%d, name_len=%d", 108f7c21177STheodore Ts'o error_msg, (unsigned) (offset % bh->b_size), 109f7c21177STheodore Ts'o offset, le32_to_cpu(de->inode), 110f7c21177STheodore Ts'o rlen, de->name_len); 111f7c21177STheodore Ts'o 112cad3f007STheodore Ts'o return 1; 113ac27a0ecSDave Kleikamp } 114ac27a0ecSDave Kleikamp 115617ba13bSMingming Cao static int ext4_readdir(struct file *filp, 116ac27a0ecSDave Kleikamp void *dirent, filldir_t filldir) 117ac27a0ecSDave Kleikamp { 118ac27a0ecSDave Kleikamp int error = 0; 119498e5f24STheodore Ts'o unsigned int offset; 120ac27a0ecSDave Kleikamp int i, stored; 121617ba13bSMingming Cao struct ext4_dir_entry_2 *de; 122ac27a0ecSDave Kleikamp int err; 1239d549890SJosef "Jeff" Sipek struct inode *inode = filp->f_path.dentry->d_inode; 124d1f5273eSFan Yong struct super_block *sb = inode->i_sb; 125ac27a0ecSDave Kleikamp int ret = 0; 1269d9f1775SEric Sandeen int dir_has_error = 0; 127ac27a0ecSDave Kleikamp 128d1f5273eSFan Yong if (is_dx_dir(inode)) { 129617ba13bSMingming Cao err = ext4_dx_readdir(filp, dirent, filldir); 130ac27a0ecSDave Kleikamp if (err != ERR_BAD_DX_DIR) { 131ac27a0ecSDave Kleikamp ret = err; 132ac27a0ecSDave Kleikamp goto out; 133ac27a0ecSDave Kleikamp } 134ac27a0ecSDave Kleikamp /* 135ac27a0ecSDave Kleikamp * We don't set the inode dirty flag since it's not 136ac27a0ecSDave Kleikamp * critical that it get flushed back to the disk. 137ac27a0ecSDave Kleikamp */ 13807a03824STheodore Ts'o ext4_clear_inode_flag(filp->f_path.dentry->d_inode, 13907a03824STheodore Ts'o EXT4_INODE_INDEX); 140ac27a0ecSDave Kleikamp } 141ac27a0ecSDave Kleikamp stored = 0; 142ac27a0ecSDave Kleikamp offset = filp->f_pos & (sb->s_blocksize - 1); 143ac27a0ecSDave Kleikamp 144ac27a0ecSDave Kleikamp while (!error && !stored && filp->f_pos < inode->i_size) { 1452ed88685STheodore Ts'o struct ext4_map_blocks map; 146ac27a0ecSDave Kleikamp struct buffer_head *bh = NULL; 147ac27a0ecSDave Kleikamp 1482ed88685STheodore Ts'o map.m_lblk = filp->f_pos >> EXT4_BLOCK_SIZE_BITS(sb); 1492ed88685STheodore Ts'o map.m_len = 1; 1502ed88685STheodore Ts'o err = ext4_map_blocks(NULL, inode, &map, 0); 151ac27a0ecSDave Kleikamp if (err > 0) { 1522ed88685STheodore Ts'o pgoff_t index = map.m_pblk >> 153dc7868fcSFengguang Wu (PAGE_CACHE_SHIFT - inode->i_blkbits); 154dc7868fcSFengguang Wu if (!ra_has_index(&filp->f_ra, index)) 155cf914a7dSRusty Russell page_cache_sync_readahead( 156dc7868fcSFengguang Wu sb->s_bdev->bd_inode->i_mapping, 157dc7868fcSFengguang Wu &filp->f_ra, filp, 158cf914a7dSRusty Russell index, 1); 159f4e6b498SFengguang Wu filp->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT; 1602ed88685STheodore Ts'o bh = ext4_bread(NULL, inode, map.m_lblk, 0, &err); 161ac27a0ecSDave Kleikamp } 162ac27a0ecSDave Kleikamp 163ac27a0ecSDave Kleikamp /* 164ac27a0ecSDave Kleikamp * We ignore I/O errors on directories so users have a chance 165ac27a0ecSDave Kleikamp * of recovering data when there's a bad sector 166ac27a0ecSDave Kleikamp */ 167ac27a0ecSDave Kleikamp if (!bh) { 1689d9f1775SEric Sandeen if (!dir_has_error) { 169f7c21177STheodore Ts'o EXT4_ERROR_FILE(filp, 0, 170f7c21177STheodore Ts'o "directory contains a " 171f7c21177STheodore Ts'o "hole at offset %llu", 1729d9f1775SEric Sandeen (unsigned long long) filp->f_pos); 1739d9f1775SEric Sandeen dir_has_error = 1; 1749d9f1775SEric Sandeen } 175e6c40211SEric Sandeen /* corrupt size? Maybe no more blocks to read */ 176e6c40211SEric Sandeen if (filp->f_pos > inode->i_blocks << 9) 177e6c40211SEric Sandeen break; 178ac27a0ecSDave Kleikamp filp->f_pos += sb->s_blocksize - offset; 179ac27a0ecSDave Kleikamp continue; 180ac27a0ecSDave Kleikamp } 181ac27a0ecSDave Kleikamp 182b0336e8dSDarrick J. Wong /* Check the checksum */ 183b0336e8dSDarrick J. Wong if (!buffer_verified(bh) && 184b0336e8dSDarrick J. Wong !ext4_dirent_csum_verify(inode, 185b0336e8dSDarrick J. Wong (struct ext4_dir_entry *)bh->b_data)) { 186b0336e8dSDarrick J. Wong EXT4_ERROR_FILE(filp, 0, "directory fails checksum " 187b0336e8dSDarrick J. Wong "at offset %llu", 188b0336e8dSDarrick J. Wong (unsigned long long)filp->f_pos); 189b0336e8dSDarrick J. Wong filp->f_pos += sb->s_blocksize - offset; 190b0336e8dSDarrick J. Wong continue; 191b0336e8dSDarrick J. Wong } 192b0336e8dSDarrick J. Wong set_buffer_verified(bh); 193b0336e8dSDarrick J. Wong 194ac27a0ecSDave Kleikamp revalidate: 195ac27a0ecSDave Kleikamp /* If the dir block has changed since the last call to 196ac27a0ecSDave Kleikamp * readdir(2), then we might be pointing to an invalid 197ac27a0ecSDave Kleikamp * dirent right now. Scan from the start of the block 198ac27a0ecSDave Kleikamp * to make sure. */ 199ac27a0ecSDave Kleikamp if (filp->f_version != inode->i_version) { 200ac27a0ecSDave Kleikamp for (i = 0; i < sb->s_blocksize && i < offset; ) { 201617ba13bSMingming Cao de = (struct ext4_dir_entry_2 *) 202ac27a0ecSDave Kleikamp (bh->b_data + i); 203ac27a0ecSDave Kleikamp /* It's too expensive to do a full 204ac27a0ecSDave Kleikamp * dirent test each time round this 205ac27a0ecSDave Kleikamp * loop, but we do have to test at 206ac27a0ecSDave Kleikamp * least that it is non-zero. A 207ac27a0ecSDave Kleikamp * failure will be detected in the 208ac27a0ecSDave Kleikamp * dirent test below. */ 2093d0518f4SWei Yongjun if (ext4_rec_len_from_disk(de->rec_len, 2103d0518f4SWei Yongjun sb->s_blocksize) < EXT4_DIR_REC_LEN(1)) 211ac27a0ecSDave Kleikamp break; 2123d0518f4SWei Yongjun i += ext4_rec_len_from_disk(de->rec_len, 2133d0518f4SWei Yongjun sb->s_blocksize); 214ac27a0ecSDave Kleikamp } 215ac27a0ecSDave Kleikamp offset = i; 216ac27a0ecSDave Kleikamp filp->f_pos = (filp->f_pos & ~(sb->s_blocksize - 1)) 217ac27a0ecSDave Kleikamp | offset; 218ac27a0ecSDave Kleikamp filp->f_version = inode->i_version; 219ac27a0ecSDave Kleikamp } 220ac27a0ecSDave Kleikamp 221ac27a0ecSDave Kleikamp while (!error && filp->f_pos < inode->i_size 222ac27a0ecSDave Kleikamp && offset < sb->s_blocksize) { 223617ba13bSMingming Cao de = (struct ext4_dir_entry_2 *) (bh->b_data + offset); 224f7c21177STheodore Ts'o if (ext4_check_dir_entry(inode, filp, de, 225ac27a0ecSDave Kleikamp bh, offset)) { 22663f57933SAndrew Morton /* 22763f57933SAndrew Morton * On error, skip the f_pos to the next block 22863f57933SAndrew Morton */ 229ac27a0ecSDave Kleikamp filp->f_pos = (filp->f_pos | 230ac27a0ecSDave Kleikamp (sb->s_blocksize - 1)) + 1; 231ac27a0ecSDave Kleikamp brelse(bh); 232ac27a0ecSDave Kleikamp ret = stored; 233ac27a0ecSDave Kleikamp goto out; 234ac27a0ecSDave Kleikamp } 2353d0518f4SWei Yongjun offset += ext4_rec_len_from_disk(de->rec_len, 2363d0518f4SWei Yongjun sb->s_blocksize); 237ac27a0ecSDave Kleikamp if (le32_to_cpu(de->inode)) { 238ac27a0ecSDave Kleikamp /* We might block in the next section 239ac27a0ecSDave Kleikamp * if the data destination is 240ac27a0ecSDave Kleikamp * currently swapped out. So, use a 241ac27a0ecSDave Kleikamp * version stamp to detect whether or 242ac27a0ecSDave Kleikamp * not the directory has been modified 243ac27a0ecSDave Kleikamp * during the copy operation. 244ac27a0ecSDave Kleikamp */ 2452b47c361SMathieu Desnoyers u64 version = filp->f_version; 246ac27a0ecSDave Kleikamp 247ac27a0ecSDave Kleikamp error = filldir(dirent, de->name, 248ac27a0ecSDave Kleikamp de->name_len, 249ac27a0ecSDave Kleikamp filp->f_pos, 250ac27a0ecSDave Kleikamp le32_to_cpu(de->inode), 251ac27a0ecSDave Kleikamp get_dtype(sb, de->file_type)); 252ac27a0ecSDave Kleikamp if (error) 253ac27a0ecSDave Kleikamp break; 254ac27a0ecSDave Kleikamp if (version != filp->f_version) 255ac27a0ecSDave Kleikamp goto revalidate; 256ac27a0ecSDave Kleikamp stored++; 257ac27a0ecSDave Kleikamp } 2583d0518f4SWei Yongjun filp->f_pos += ext4_rec_len_from_disk(de->rec_len, 2593d0518f4SWei Yongjun sb->s_blocksize); 260ac27a0ecSDave Kleikamp } 261ac27a0ecSDave Kleikamp offset = 0; 262ac27a0ecSDave Kleikamp brelse(bh); 263ac27a0ecSDave Kleikamp } 264ac27a0ecSDave Kleikamp out: 265ac27a0ecSDave Kleikamp return ret; 266ac27a0ecSDave Kleikamp } 267ac27a0ecSDave Kleikamp 268d1f5273eSFan Yong static inline int is_32bit_api(void) 269d1f5273eSFan Yong { 270d1f5273eSFan Yong #ifdef CONFIG_COMPAT 271d1f5273eSFan Yong return is_compat_task(); 272d1f5273eSFan Yong #else 273d1f5273eSFan Yong return (BITS_PER_LONG == 32); 274d1f5273eSFan Yong #endif 275d1f5273eSFan Yong } 276d1f5273eSFan Yong 277ac27a0ecSDave Kleikamp /* 278ac27a0ecSDave Kleikamp * These functions convert from the major/minor hash to an f_pos 279d1f5273eSFan Yong * value for dx directories 280ac27a0ecSDave Kleikamp * 281d1f5273eSFan Yong * Upper layer (for example NFS) should specify FMODE_32BITHASH or 282d1f5273eSFan Yong * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted 283d1f5273eSFan Yong * directly on both 32-bit and 64-bit nodes, under such case, neither 284d1f5273eSFan Yong * FMODE_32BITHASH nor FMODE_64BITHASH is specified. 285ac27a0ecSDave Kleikamp */ 286d1f5273eSFan Yong static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor) 287d1f5273eSFan Yong { 288d1f5273eSFan Yong if ((filp->f_mode & FMODE_32BITHASH) || 289d1f5273eSFan Yong (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 290d1f5273eSFan Yong return major >> 1; 291d1f5273eSFan Yong else 292d1f5273eSFan Yong return ((__u64)(major >> 1) << 32) | (__u64)minor; 293d1f5273eSFan Yong } 294d1f5273eSFan Yong 295d1f5273eSFan Yong static inline __u32 pos2maj_hash(struct file *filp, loff_t pos) 296d1f5273eSFan Yong { 297d1f5273eSFan Yong if ((filp->f_mode & FMODE_32BITHASH) || 298d1f5273eSFan Yong (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 299d1f5273eSFan Yong return (pos << 1) & 0xffffffff; 300d1f5273eSFan Yong else 301d1f5273eSFan Yong return ((pos >> 32) << 1) & 0xffffffff; 302d1f5273eSFan Yong } 303d1f5273eSFan Yong 304d1f5273eSFan Yong static inline __u32 pos2min_hash(struct file *filp, loff_t pos) 305d1f5273eSFan Yong { 306d1f5273eSFan Yong if ((filp->f_mode & FMODE_32BITHASH) || 307d1f5273eSFan Yong (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 308d1f5273eSFan Yong return 0; 309d1f5273eSFan Yong else 310d1f5273eSFan Yong return pos & 0xffffffff; 311d1f5273eSFan Yong } 312d1f5273eSFan Yong 313d1f5273eSFan Yong /* 314d1f5273eSFan Yong * Return 32- or 64-bit end-of-file for dx directories 315d1f5273eSFan Yong */ 316d1f5273eSFan Yong static inline loff_t ext4_get_htree_eof(struct file *filp) 317d1f5273eSFan Yong { 318d1f5273eSFan Yong if ((filp->f_mode & FMODE_32BITHASH) || 319d1f5273eSFan Yong (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api())) 320d1f5273eSFan Yong return EXT4_HTREE_EOF_32BIT; 321d1f5273eSFan Yong else 322d1f5273eSFan Yong return EXT4_HTREE_EOF_64BIT; 323d1f5273eSFan Yong } 324d1f5273eSFan Yong 325d1f5273eSFan Yong 326d1f5273eSFan Yong /* 327d1f5273eSFan Yong * ext4_dir_llseek() based on generic_file_llseek() to handle both 328d1f5273eSFan Yong * non-htree and htree directories, where the "offset" is in terms 329d1f5273eSFan Yong * of the filename hash value instead of the byte offset. 330d1f5273eSFan Yong * 331d1f5273eSFan Yong * NOTE: offsets obtained *before* ext4_set_inode_flag(dir, EXT4_INODE_INDEX) 332d1f5273eSFan Yong * will be invalid once the directory was converted into a dx directory 333d1f5273eSFan Yong */ 334d1f5273eSFan Yong loff_t ext4_dir_llseek(struct file *file, loff_t offset, int origin) 335d1f5273eSFan Yong { 336d1f5273eSFan Yong struct inode *inode = file->f_mapping->host; 337d1f5273eSFan Yong loff_t ret = -EINVAL; 338d1f5273eSFan Yong int dx_dir = is_dx_dir(inode); 339d1f5273eSFan Yong 340d1f5273eSFan Yong mutex_lock(&inode->i_mutex); 341d1f5273eSFan Yong 342d1f5273eSFan Yong /* NOTE: relative offsets with dx directories might not work 343d1f5273eSFan Yong * as expected, as it is difficult to figure out the 344d1f5273eSFan Yong * correct offset between dx hashes */ 345d1f5273eSFan Yong 346d1f5273eSFan Yong switch (origin) { 347d1f5273eSFan Yong case SEEK_END: 348d1f5273eSFan Yong if (unlikely(offset > 0)) 349d1f5273eSFan Yong goto out_err; /* not supported for directories */ 350d1f5273eSFan Yong 351d1f5273eSFan Yong /* so only negative offsets are left, does that have a 352d1f5273eSFan Yong * meaning for directories at all? */ 353d1f5273eSFan Yong if (dx_dir) 354d1f5273eSFan Yong offset += ext4_get_htree_eof(file); 355d1f5273eSFan Yong else 356d1f5273eSFan Yong offset += inode->i_size; 357d1f5273eSFan Yong break; 358d1f5273eSFan Yong case SEEK_CUR: 359d1f5273eSFan Yong /* 360d1f5273eSFan Yong * Here we special-case the lseek(fd, 0, SEEK_CUR) 361d1f5273eSFan Yong * position-querying operation. Avoid rewriting the "same" 362d1f5273eSFan Yong * f_pos value back to the file because a concurrent read(), 363d1f5273eSFan Yong * write() or lseek() might have altered it 364d1f5273eSFan Yong */ 365d1f5273eSFan Yong if (offset == 0) { 366d1f5273eSFan Yong offset = file->f_pos; 367d1f5273eSFan Yong goto out_ok; 368d1f5273eSFan Yong } 369d1f5273eSFan Yong 370d1f5273eSFan Yong offset += file->f_pos; 371d1f5273eSFan Yong break; 372d1f5273eSFan Yong } 373d1f5273eSFan Yong 374d1f5273eSFan Yong if (unlikely(offset < 0)) 375d1f5273eSFan Yong goto out_err; 376d1f5273eSFan Yong 377d1f5273eSFan Yong if (!dx_dir) { 378d1f5273eSFan Yong if (offset > inode->i_sb->s_maxbytes) 379d1f5273eSFan Yong goto out_err; 380d1f5273eSFan Yong } else if (offset > ext4_get_htree_eof(file)) 381d1f5273eSFan Yong goto out_err; 382d1f5273eSFan Yong 383d1f5273eSFan Yong /* Special lock needed here? */ 384d1f5273eSFan Yong if (offset != file->f_pos) { 385d1f5273eSFan Yong file->f_pos = offset; 386d1f5273eSFan Yong file->f_version = 0; 387d1f5273eSFan Yong } 388d1f5273eSFan Yong 389d1f5273eSFan Yong out_ok: 390d1f5273eSFan Yong ret = offset; 391d1f5273eSFan Yong out_err: 392d1f5273eSFan Yong mutex_unlock(&inode->i_mutex); 393d1f5273eSFan Yong 394d1f5273eSFan Yong return ret; 395d1f5273eSFan Yong } 396ac27a0ecSDave Kleikamp 397ac27a0ecSDave Kleikamp /* 398ac27a0ecSDave Kleikamp * This structure holds the nodes of the red-black tree used to store 399ac27a0ecSDave Kleikamp * the directory entry in hash order. 400ac27a0ecSDave Kleikamp */ 401ac27a0ecSDave Kleikamp struct fname { 402ac27a0ecSDave Kleikamp __u32 hash; 403ac27a0ecSDave Kleikamp __u32 minor_hash; 404ac27a0ecSDave Kleikamp struct rb_node rb_hash; 405ac27a0ecSDave Kleikamp struct fname *next; 406ac27a0ecSDave Kleikamp __u32 inode; 407ac27a0ecSDave Kleikamp __u8 name_len; 408ac27a0ecSDave Kleikamp __u8 file_type; 409ac27a0ecSDave Kleikamp char name[0]; 410ac27a0ecSDave Kleikamp }; 411ac27a0ecSDave Kleikamp 412ac27a0ecSDave Kleikamp /* 413ac27a0ecSDave Kleikamp * This functoin implements a non-recursive way of freeing all of the 414ac27a0ecSDave Kleikamp * nodes in the red-black tree. 415ac27a0ecSDave Kleikamp */ 416ac27a0ecSDave Kleikamp static void free_rb_tree_fname(struct rb_root *root) 417ac27a0ecSDave Kleikamp { 418ac27a0ecSDave Kleikamp struct rb_node *n = root->rb_node; 419ac27a0ecSDave Kleikamp struct rb_node *parent; 420ac27a0ecSDave Kleikamp struct fname *fname; 421ac27a0ecSDave Kleikamp 422ac27a0ecSDave Kleikamp while (n) { 423ac27a0ecSDave Kleikamp /* Do the node's children first */ 42469baee06SShen Feng if (n->rb_left) { 425ac27a0ecSDave Kleikamp n = n->rb_left; 426ac27a0ecSDave Kleikamp continue; 427ac27a0ecSDave Kleikamp } 428ac27a0ecSDave Kleikamp if (n->rb_right) { 429ac27a0ecSDave Kleikamp n = n->rb_right; 430ac27a0ecSDave Kleikamp continue; 431ac27a0ecSDave Kleikamp } 432ac27a0ecSDave Kleikamp /* 433ac27a0ecSDave Kleikamp * The node has no children; free it, and then zero 434ac27a0ecSDave Kleikamp * out parent's link to it. Finally go to the 435ac27a0ecSDave Kleikamp * beginning of the loop and try to free the parent 436ac27a0ecSDave Kleikamp * node. 437ac27a0ecSDave Kleikamp */ 438ac27a0ecSDave Kleikamp parent = rb_parent(n); 439ac27a0ecSDave Kleikamp fname = rb_entry(n, struct fname, rb_hash); 440ac27a0ecSDave Kleikamp while (fname) { 441ac27a0ecSDave Kleikamp struct fname *old = fname; 442ac27a0ecSDave Kleikamp fname = fname->next; 443ac27a0ecSDave Kleikamp kfree(old); 444ac27a0ecSDave Kleikamp } 445ac27a0ecSDave Kleikamp if (!parent) 44664e290ecSVenkatesh Pallipadi *root = RB_ROOT; 447ac27a0ecSDave Kleikamp else if (parent->rb_left == n) 448ac27a0ecSDave Kleikamp parent->rb_left = NULL; 449ac27a0ecSDave Kleikamp else if (parent->rb_right == n) 450ac27a0ecSDave Kleikamp parent->rb_right = NULL; 451ac27a0ecSDave Kleikamp n = parent; 452ac27a0ecSDave Kleikamp } 453ac27a0ecSDave Kleikamp } 454ac27a0ecSDave Kleikamp 455ac27a0ecSDave Kleikamp 456d1f5273eSFan Yong static struct dir_private_info *ext4_htree_create_dir_info(struct file *filp, 457d1f5273eSFan Yong loff_t pos) 458ac27a0ecSDave Kleikamp { 459ac27a0ecSDave Kleikamp struct dir_private_info *p; 460ac27a0ecSDave Kleikamp 46169baee06SShen Feng p = kzalloc(sizeof(struct dir_private_info), GFP_KERNEL); 462ac27a0ecSDave Kleikamp if (!p) 463ac27a0ecSDave Kleikamp return NULL; 464d1f5273eSFan Yong p->curr_hash = pos2maj_hash(filp, pos); 465d1f5273eSFan Yong p->curr_minor_hash = pos2min_hash(filp, pos); 466ac27a0ecSDave Kleikamp return p; 467ac27a0ecSDave Kleikamp } 468ac27a0ecSDave Kleikamp 469617ba13bSMingming Cao void ext4_htree_free_dir_info(struct dir_private_info *p) 470ac27a0ecSDave Kleikamp { 471ac27a0ecSDave Kleikamp free_rb_tree_fname(&p->root); 472ac27a0ecSDave Kleikamp kfree(p); 473ac27a0ecSDave Kleikamp } 474ac27a0ecSDave Kleikamp 475ac27a0ecSDave Kleikamp /* 476ac27a0ecSDave Kleikamp * Given a directory entry, enter it into the fname rb tree. 477ac27a0ecSDave Kleikamp */ 478617ba13bSMingming Cao int ext4_htree_store_dirent(struct file *dir_file, __u32 hash, 479ac27a0ecSDave Kleikamp __u32 minor_hash, 480617ba13bSMingming Cao struct ext4_dir_entry_2 *dirent) 481ac27a0ecSDave Kleikamp { 482ac27a0ecSDave Kleikamp struct rb_node **p, *parent = NULL; 483ac27a0ecSDave Kleikamp struct fname *fname, *new_fn; 484ac27a0ecSDave Kleikamp struct dir_private_info *info; 485ac27a0ecSDave Kleikamp int len; 486ac27a0ecSDave Kleikamp 487a271fe85SJoe Perches info = dir_file->private_data; 488ac27a0ecSDave Kleikamp p = &info->root.rb_node; 489ac27a0ecSDave Kleikamp 490ac27a0ecSDave Kleikamp /* Create and allocate the fname structure */ 491ac27a0ecSDave Kleikamp len = sizeof(struct fname) + dirent->name_len + 1; 492ac27a0ecSDave Kleikamp new_fn = kzalloc(len, GFP_KERNEL); 493ac27a0ecSDave Kleikamp if (!new_fn) 494ac27a0ecSDave Kleikamp return -ENOMEM; 495ac27a0ecSDave Kleikamp new_fn->hash = hash; 496ac27a0ecSDave Kleikamp new_fn->minor_hash = minor_hash; 497ac27a0ecSDave Kleikamp new_fn->inode = le32_to_cpu(dirent->inode); 498ac27a0ecSDave Kleikamp new_fn->name_len = dirent->name_len; 499ac27a0ecSDave Kleikamp new_fn->file_type = dirent->file_type; 500ac27a0ecSDave Kleikamp memcpy(new_fn->name, dirent->name, dirent->name_len); 501ac27a0ecSDave Kleikamp new_fn->name[dirent->name_len] = 0; 502ac27a0ecSDave Kleikamp 503ac27a0ecSDave Kleikamp while (*p) { 504ac27a0ecSDave Kleikamp parent = *p; 505ac27a0ecSDave Kleikamp fname = rb_entry(parent, struct fname, rb_hash); 506ac27a0ecSDave Kleikamp 507ac27a0ecSDave Kleikamp /* 508ac27a0ecSDave Kleikamp * If the hash and minor hash match up, then we put 509ac27a0ecSDave Kleikamp * them on a linked list. This rarely happens... 510ac27a0ecSDave Kleikamp */ 511ac27a0ecSDave Kleikamp if ((new_fn->hash == fname->hash) && 512ac27a0ecSDave Kleikamp (new_fn->minor_hash == fname->minor_hash)) { 513ac27a0ecSDave Kleikamp new_fn->next = fname->next; 514ac27a0ecSDave Kleikamp fname->next = new_fn; 515ac27a0ecSDave Kleikamp return 0; 516ac27a0ecSDave Kleikamp } 517ac27a0ecSDave Kleikamp 518ac27a0ecSDave Kleikamp if (new_fn->hash < fname->hash) 519ac27a0ecSDave Kleikamp p = &(*p)->rb_left; 520ac27a0ecSDave Kleikamp else if (new_fn->hash > fname->hash) 521ac27a0ecSDave Kleikamp p = &(*p)->rb_right; 522ac27a0ecSDave Kleikamp else if (new_fn->minor_hash < fname->minor_hash) 523ac27a0ecSDave Kleikamp p = &(*p)->rb_left; 524ac27a0ecSDave Kleikamp else /* if (new_fn->minor_hash > fname->minor_hash) */ 525ac27a0ecSDave Kleikamp p = &(*p)->rb_right; 526ac27a0ecSDave Kleikamp } 527ac27a0ecSDave Kleikamp 528ac27a0ecSDave Kleikamp rb_link_node(&new_fn->rb_hash, parent, p); 529ac27a0ecSDave Kleikamp rb_insert_color(&new_fn->rb_hash, &info->root); 530ac27a0ecSDave Kleikamp return 0; 531ac27a0ecSDave Kleikamp } 532ac27a0ecSDave Kleikamp 533ac27a0ecSDave Kleikamp 534ac27a0ecSDave Kleikamp 535ac27a0ecSDave Kleikamp /* 536617ba13bSMingming Cao * This is a helper function for ext4_dx_readdir. It calls filldir 537ac27a0ecSDave Kleikamp * for all entres on the fname linked list. (Normally there is only 538ac27a0ecSDave Kleikamp * one entry on the linked list, unless there are 62 bit hash collisions.) 539ac27a0ecSDave Kleikamp */ 540ac27a0ecSDave Kleikamp static int call_filldir(struct file *filp, void *dirent, 541ac27a0ecSDave Kleikamp filldir_t filldir, struct fname *fname) 542ac27a0ecSDave Kleikamp { 543ac27a0ecSDave Kleikamp struct dir_private_info *info = filp->private_data; 544ac27a0ecSDave Kleikamp loff_t curr_pos; 5459d549890SJosef "Jeff" Sipek struct inode *inode = filp->f_path.dentry->d_inode; 546ac27a0ecSDave Kleikamp struct super_block *sb; 547ac27a0ecSDave Kleikamp int error; 548ac27a0ecSDave Kleikamp 549ac27a0ecSDave Kleikamp sb = inode->i_sb; 550ac27a0ecSDave Kleikamp 551ac27a0ecSDave Kleikamp if (!fname) { 55292b97816STheodore Ts'o ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: " 55392b97816STheodore Ts'o "called with null fname?!?", __func__, __LINE__, 55492b97816STheodore Ts'o inode->i_ino, current->comm); 555ac27a0ecSDave Kleikamp return 0; 556ac27a0ecSDave Kleikamp } 557d1f5273eSFan Yong curr_pos = hash2pos(filp, fname->hash, fname->minor_hash); 558ac27a0ecSDave Kleikamp while (fname) { 559ac27a0ecSDave Kleikamp error = filldir(dirent, fname->name, 560ac27a0ecSDave Kleikamp fname->name_len, curr_pos, 561ac27a0ecSDave Kleikamp fname->inode, 562ac27a0ecSDave Kleikamp get_dtype(sb, fname->file_type)); 563ac27a0ecSDave Kleikamp if (error) { 564ac27a0ecSDave Kleikamp filp->f_pos = curr_pos; 565d0156417STheodore Ts'o info->extra_fname = fname; 566ac27a0ecSDave Kleikamp return error; 567ac27a0ecSDave Kleikamp } 568ac27a0ecSDave Kleikamp fname = fname->next; 569ac27a0ecSDave Kleikamp } 570ac27a0ecSDave Kleikamp return 0; 571ac27a0ecSDave Kleikamp } 572ac27a0ecSDave Kleikamp 573617ba13bSMingming Cao static int ext4_dx_readdir(struct file *filp, 574ac27a0ecSDave Kleikamp void *dirent, filldir_t filldir) 575ac27a0ecSDave Kleikamp { 576ac27a0ecSDave Kleikamp struct dir_private_info *info = filp->private_data; 5779d549890SJosef "Jeff" Sipek struct inode *inode = filp->f_path.dentry->d_inode; 578ac27a0ecSDave Kleikamp struct fname *fname; 579ac27a0ecSDave Kleikamp int ret; 580ac27a0ecSDave Kleikamp 581ac27a0ecSDave Kleikamp if (!info) { 582d1f5273eSFan Yong info = ext4_htree_create_dir_info(filp, filp->f_pos); 583ac27a0ecSDave Kleikamp if (!info) 584ac27a0ecSDave Kleikamp return -ENOMEM; 585ac27a0ecSDave Kleikamp filp->private_data = info; 586ac27a0ecSDave Kleikamp } 587ac27a0ecSDave Kleikamp 588d1f5273eSFan Yong if (filp->f_pos == ext4_get_htree_eof(filp)) 589ac27a0ecSDave Kleikamp return 0; /* EOF */ 590ac27a0ecSDave Kleikamp 591ac27a0ecSDave Kleikamp /* Some one has messed with f_pos; reset the world */ 592ac27a0ecSDave Kleikamp if (info->last_pos != filp->f_pos) { 593ac27a0ecSDave Kleikamp free_rb_tree_fname(&info->root); 594ac27a0ecSDave Kleikamp info->curr_node = NULL; 595ac27a0ecSDave Kleikamp info->extra_fname = NULL; 596d1f5273eSFan Yong info->curr_hash = pos2maj_hash(filp, filp->f_pos); 597d1f5273eSFan Yong info->curr_minor_hash = pos2min_hash(filp, filp->f_pos); 598ac27a0ecSDave Kleikamp } 599ac27a0ecSDave Kleikamp 600ac27a0ecSDave Kleikamp /* 601ac27a0ecSDave Kleikamp * If there are any leftover names on the hash collision 602ac27a0ecSDave Kleikamp * chain, return them first. 603ac27a0ecSDave Kleikamp */ 604d0156417STheodore Ts'o if (info->extra_fname) { 605d0156417STheodore Ts'o if (call_filldir(filp, dirent, filldir, info->extra_fname)) 606ac27a0ecSDave Kleikamp goto finished; 607d0156417STheodore Ts'o info->extra_fname = NULL; 6083c37fc86STheodore Ts'o goto next_node; 609d0156417STheodore Ts'o } else if (!info->curr_node) 610ac27a0ecSDave Kleikamp info->curr_node = rb_first(&info->root); 611ac27a0ecSDave Kleikamp 612ac27a0ecSDave Kleikamp while (1) { 613ac27a0ecSDave Kleikamp /* 614ac27a0ecSDave Kleikamp * Fill the rbtree if we have no more entries, 615ac27a0ecSDave Kleikamp * or the inode has changed since we last read in the 616ac27a0ecSDave Kleikamp * cached entries. 617ac27a0ecSDave Kleikamp */ 618ac27a0ecSDave Kleikamp if ((!info->curr_node) || 619ac27a0ecSDave Kleikamp (filp->f_version != inode->i_version)) { 620ac27a0ecSDave Kleikamp info->curr_node = NULL; 621ac27a0ecSDave Kleikamp free_rb_tree_fname(&info->root); 622ac27a0ecSDave Kleikamp filp->f_version = inode->i_version; 623617ba13bSMingming Cao ret = ext4_htree_fill_tree(filp, info->curr_hash, 624ac27a0ecSDave Kleikamp info->curr_minor_hash, 625ac27a0ecSDave Kleikamp &info->next_hash); 626ac27a0ecSDave Kleikamp if (ret < 0) 627ac27a0ecSDave Kleikamp return ret; 628ac27a0ecSDave Kleikamp if (ret == 0) { 629d1f5273eSFan Yong filp->f_pos = ext4_get_htree_eof(filp); 630ac27a0ecSDave Kleikamp break; 631ac27a0ecSDave Kleikamp } 632ac27a0ecSDave Kleikamp info->curr_node = rb_first(&info->root); 633ac27a0ecSDave Kleikamp } 634ac27a0ecSDave Kleikamp 635ac27a0ecSDave Kleikamp fname = rb_entry(info->curr_node, struct fname, rb_hash); 636ac27a0ecSDave Kleikamp info->curr_hash = fname->hash; 637ac27a0ecSDave Kleikamp info->curr_minor_hash = fname->minor_hash; 638ac27a0ecSDave Kleikamp if (call_filldir(filp, dirent, filldir, fname)) 639ac27a0ecSDave Kleikamp break; 6403c37fc86STheodore Ts'o next_node: 641ac27a0ecSDave Kleikamp info->curr_node = rb_next(info->curr_node); 6423c37fc86STheodore Ts'o if (info->curr_node) { 6433c37fc86STheodore Ts'o fname = rb_entry(info->curr_node, struct fname, 6443c37fc86STheodore Ts'o rb_hash); 6453c37fc86STheodore Ts'o info->curr_hash = fname->hash; 6463c37fc86STheodore Ts'o info->curr_minor_hash = fname->minor_hash; 6473c37fc86STheodore Ts'o } else { 648ac27a0ecSDave Kleikamp if (info->next_hash == ~0) { 649d1f5273eSFan Yong filp->f_pos = ext4_get_htree_eof(filp); 650ac27a0ecSDave Kleikamp break; 651ac27a0ecSDave Kleikamp } 652ac27a0ecSDave Kleikamp info->curr_hash = info->next_hash; 653ac27a0ecSDave Kleikamp info->curr_minor_hash = 0; 654ac27a0ecSDave Kleikamp } 655ac27a0ecSDave Kleikamp } 656ac27a0ecSDave Kleikamp finished: 657ac27a0ecSDave Kleikamp info->last_pos = filp->f_pos; 658ac27a0ecSDave Kleikamp return 0; 659ac27a0ecSDave Kleikamp } 660ac27a0ecSDave Kleikamp 661617ba13bSMingming Cao static int ext4_release_dir(struct inode *inode, struct file *filp) 662ac27a0ecSDave Kleikamp { 663ac27a0ecSDave Kleikamp if (filp->private_data) 664617ba13bSMingming Cao ext4_htree_free_dir_info(filp->private_data); 665ac27a0ecSDave Kleikamp 666ac27a0ecSDave Kleikamp return 0; 667ac27a0ecSDave Kleikamp } 668d1f5273eSFan Yong 669d1f5273eSFan Yong const struct file_operations ext4_dir_operations = { 670d1f5273eSFan Yong .llseek = ext4_dir_llseek, 671d1f5273eSFan Yong .read = generic_read_dir, 672d1f5273eSFan Yong .readdir = ext4_readdir, 673d1f5273eSFan Yong .unlocked_ioctl = ext4_ioctl, 674d1f5273eSFan Yong #ifdef CONFIG_COMPAT 675d1f5273eSFan Yong .compat_ioctl = ext4_compat_ioctl, 676d1f5273eSFan Yong #endif 677d1f5273eSFan Yong .fsync = ext4_sync_file, 678d1f5273eSFan Yong .release = ext4_release_dir, 679d1f5273eSFan Yong }; 680