blob: 680bb3388919a444b25e7dc0ee660b92ad475ce2 [file] [log] [blame]
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001/*
Mingming Cao617ba132006-10-11 01:20:53 -07002 * linux/fs/ext4/dir.c
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003 *
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
8 *
9 * from
10 *
11 * linux/fs/minix/dir.c
12 *
13 * Copyright (C) 1991, 1992 Linus Torvalds
14 *
Mingming Cao617ba132006-10-11 01:20:53 -070015 * ext4 directory handling functions
Dave Kleikampac27a0e2006-10-11 01:20:50 -070016 *
17 * Big-endian to little-endian byte-swapping/bitmaps by
18 * David S. Miller (davem@caip.rutgers.edu), 1995
19 *
20 * Hash Tree Directory indexing (c) 2001 Daniel Phillips
21 *
22 */
23
24#include <linux/fs.h>
Mingming Caodab291a2006-10-11 01:21:01 -070025#include <linux/jbd2.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070026#include <linux/buffer_head.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070027#include <linux/slab.h>
28#include <linux/rbtree.h>
Christoph Hellwig3dcf5452008-04-29 18:13:32 -040029#include "ext4.h"
Tao Ma65d165d2012-12-10 14:05:59 -050030#include "xattr.h"
Dave Kleikampac27a0e2006-10-11 01:20:50 -070031
Al Viro725bebb2013-05-17 16:08:53 -040032static int ext4_dx_readdir(struct file *, struct dir_context *);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070033
Fan Yongd1f52732012-03-18 22:44:40 -040034/**
35 * Check if the given dir-inode refers to an htree-indexed directory
Anatol Pomozov70261f52013-08-28 14:40:12 -040036 * (or a directory which could potentially get converted to use htree
Fan Yongd1f52732012-03-18 22:44:40 -040037 * indexing).
38 *
39 * Return 1 if it is a dx dir, 0 if not
40 */
41static int is_dx_dir(struct inode *inode)
42{
43 struct super_block *sb = inode->i_sb;
44
45 if (EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
46 EXT4_FEATURE_COMPAT_DIR_INDEX) &&
47 ((ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) ||
Tao Ma8af0f082013-04-19 17:53:09 -040048 ((inode->i_size >> sb->s_blocksize_bits) == 1) ||
49 ext4_has_inline_data(inode)))
Fan Yongd1f52732012-03-18 22:44:40 -040050 return 1;
51
52 return 0;
53}
54
Theodore Ts'ocad3f002010-12-19 22:07:02 -050055/*
56 * Return 0 if the directory entry is OK, and 1 if there is a problem
57 *
58 * Note: this is the opposite of what ext2 and ext3 historically returned...
Tao Ma65d165d2012-12-10 14:05:59 -050059 *
60 * bh passed here can be an inode block or a dir data block, depending
61 * on the inode inline data flag.
Theodore Ts'ocad3f002010-12-19 22:07:02 -050062 */
Theodore Ts'o60fd4da2010-07-27 11:54:40 -040063int __ext4_check_dir_entry(const char *function, unsigned int line,
Theodore Ts'of7c21172011-01-10 12:10:55 -050064 struct inode *dir, struct file *filp,
Theodore Ts'o60fd4da2010-07-27 11:54:40 -040065 struct ext4_dir_entry_2 *de,
Tao Ma226ba972012-12-10 14:05:58 -050066 struct buffer_head *bh, char *buf, int size,
Theodore Ts'o60fd4da2010-07-27 11:54:40 -040067 unsigned int offset)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070068{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -040069 const char *error_msg = NULL;
Wei Yongjun3d0518f2009-02-14 23:01:36 -050070 const int rlen = ext4_rec_len_from_disk(de->rec_len,
71 dir->i_sb->s_blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070072
Theodore Ts'ocad3f002010-12-19 22:07:02 -050073 if (unlikely(rlen < EXT4_DIR_REC_LEN(1)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -070074 error_msg = "rec_len is smaller than minimal";
Theodore Ts'ocad3f002010-12-19 22:07:02 -050075 else if (unlikely(rlen % 4 != 0))
Dave Kleikampac27a0e2006-10-11 01:20:50 -070076 error_msg = "rec_len % 4 != 0";
Theodore Ts'ocad3f002010-12-19 22:07:02 -050077 else if (unlikely(rlen < EXT4_DIR_REC_LEN(de->name_len)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -070078 error_msg = "rec_len is too small for name_len";
Tao Ma226ba972012-12-10 14:05:58 -050079 else if (unlikely(((char *) de - buf) + rlen > size))
80 error_msg = "directory entry across range";
Theodore Ts'ocad3f002010-12-19 22:07:02 -050081 else if (unlikely(le32_to_cpu(de->inode) >
82 le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -070083 error_msg = "inode out of bounds";
Theodore Ts'ocad3f002010-12-19 22:07:02 -050084 else
85 return 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -070086
Theodore Ts'of7c21172011-01-10 12:10:55 -050087 if (filp)
Dan Carpenter4fda4002012-02-20 17:53:05 -050088 ext4_error_file(filp, function, line, bh->b_blocknr,
Theodore Ts'of7c21172011-01-10 12:10:55 -050089 "bad entry in directory: %s - offset=%u(%u), "
90 "inode=%u, rec_len=%d, name_len=%d",
Tao Ma226ba972012-12-10 14:05:58 -050091 error_msg, (unsigned) (offset % size),
Theodore Ts'of7c21172011-01-10 12:10:55 -050092 offset, le32_to_cpu(de->inode),
93 rlen, de->name_len);
94 else
Dan Carpenter4fda4002012-02-20 17:53:05 -050095 ext4_error_inode(dir, function, line, bh->b_blocknr,
Theodore Ts'of7c21172011-01-10 12:10:55 -050096 "bad entry in directory: %s - offset=%u(%u), "
97 "inode=%u, rec_len=%d, name_len=%d",
Tao Ma226ba972012-12-10 14:05:58 -050098 error_msg, (unsigned) (offset % size),
Theodore Ts'of7c21172011-01-10 12:10:55 -050099 offset, le32_to_cpu(de->inode),
100 rlen, de->name_len);
101
Theodore Ts'ocad3f002010-12-19 22:07:02 -0500102 return 1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700103}
104
Al Viro725bebb2013-05-17 16:08:53 -0400105static int ext4_readdir(struct file *file, struct dir_context *ctx)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700106{
Theodore Ts'o498e5f22008-11-05 00:14:04 -0500107 unsigned int offset;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700108 int i, stored;
Mingming Cao617ba132006-10-11 01:20:53 -0700109 struct ext4_dir_entry_2 *de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700110 int err;
Al Viro725bebb2013-05-17 16:08:53 -0400111 struct inode *inode = file_inode(file);
Fan Yongd1f52732012-03-18 22:44:40 -0400112 struct super_block *sb = inode->i_sb;
Eric Sandeen9d9f1772008-10-09 11:15:52 -0400113 int dir_has_error = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700114
Fan Yongd1f52732012-03-18 22:44:40 -0400115 if (is_dx_dir(inode)) {
Al Viro725bebb2013-05-17 16:08:53 -0400116 err = ext4_dx_readdir(file, ctx);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700117 if (err != ERR_BAD_DX_DIR) {
Al Viro725bebb2013-05-17 16:08:53 -0400118 return err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700119 }
120 /*
121 * We don't set the inode dirty flag since it's not
122 * critical that it get flushed back to the disk.
123 */
Al Viro725bebb2013-05-17 16:08:53 -0400124 ext4_clear_inode_flag(file_inode(file),
Theodore Ts'o07a03822010-06-14 09:54:48 -0400125 EXT4_INODE_INDEX);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700126 }
Tao Ma8af0f082013-04-19 17:53:09 -0400127
128 if (ext4_has_inline_data(inode)) {
129 int has_inline_data = 1;
Al Viro725bebb2013-05-17 16:08:53 -0400130 int ret = ext4_read_inline_dir(file, ctx,
Tao Ma8af0f082013-04-19 17:53:09 -0400131 &has_inline_data);
132 if (has_inline_data)
133 return ret;
134 }
135
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700136 stored = 0;
Al Viro725bebb2013-05-17 16:08:53 -0400137 offset = ctx->pos & (sb->s_blocksize - 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700138
Al Viro725bebb2013-05-17 16:08:53 -0400139 while (ctx->pos < inode->i_size) {
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400140 struct ext4_map_blocks map;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700141 struct buffer_head *bh = NULL;
142
Al Viro725bebb2013-05-17 16:08:53 -0400143 map.m_lblk = ctx->pos >> EXT4_BLOCK_SIZE_BITS(sb);
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400144 map.m_len = 1;
145 err = ext4_map_blocks(NULL, inode, &map, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700146 if (err > 0) {
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400147 pgoff_t index = map.m_pblk >>
Fengguang Wudc7868f2007-07-19 01:48:04 -0700148 (PAGE_CACHE_SHIFT - inode->i_blkbits);
Al Viro725bebb2013-05-17 16:08:53 -0400149 if (!ra_has_index(&file->f_ra, index))
Rusty Russellcf914a72007-07-19 01:48:08 -0700150 page_cache_sync_readahead(
Fengguang Wudc7868f2007-07-19 01:48:04 -0700151 sb->s_bdev->bd_inode->i_mapping,
Al Viro725bebb2013-05-17 16:08:53 -0400152 &file->f_ra, file,
Rusty Russellcf914a72007-07-19 01:48:08 -0700153 index, 1);
Al Viro725bebb2013-05-17 16:08:53 -0400154 file->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400155 bh = ext4_bread(NULL, inode, map.m_lblk, 0, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700156 }
157
158 /*
159 * We ignore I/O errors on directories so users have a chance
160 * of recovering data when there's a bad sector
161 */
162 if (!bh) {
Eric Sandeen9d9f1772008-10-09 11:15:52 -0400163 if (!dir_has_error) {
Al Viro725bebb2013-05-17 16:08:53 -0400164 EXT4_ERROR_FILE(file, 0,
Theodore Ts'of7c21172011-01-10 12:10:55 -0500165 "directory contains a "
166 "hole at offset %llu",
Al Viro725bebb2013-05-17 16:08:53 -0400167 (unsigned long long) ctx->pos);
Eric Sandeen9d9f1772008-10-09 11:15:52 -0400168 dir_has_error = 1;
169 }
Eric Sandeene6c40212006-12-06 20:36:28 -0800170 /* corrupt size? Maybe no more blocks to read */
Al Viro725bebb2013-05-17 16:08:53 -0400171 if (ctx->pos > inode->i_blocks << 9)
Eric Sandeene6c40212006-12-06 20:36:28 -0800172 break;
Al Viro725bebb2013-05-17 16:08:53 -0400173 ctx->pos += sb->s_blocksize - offset;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700174 continue;
175 }
176
Darrick J. Wongb0336e82012-04-29 18:41:10 -0400177 /* Check the checksum */
178 if (!buffer_verified(bh) &&
179 !ext4_dirent_csum_verify(inode,
180 (struct ext4_dir_entry *)bh->b_data)) {
Al Viro725bebb2013-05-17 16:08:53 -0400181 EXT4_ERROR_FILE(file, 0, "directory fails checksum "
Darrick J. Wongb0336e82012-04-29 18:41:10 -0400182 "at offset %llu",
Al Viro725bebb2013-05-17 16:08:53 -0400183 (unsigned long long)ctx->pos);
184 ctx->pos += sb->s_blocksize - offset;
Guo Chaod5ac7772013-01-28 21:23:24 -0500185 brelse(bh);
Darrick J. Wongb0336e82012-04-29 18:41:10 -0400186 continue;
187 }
188 set_buffer_verified(bh);
189
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700190 /* If the dir block has changed since the last call to
191 * readdir(2), then we might be pointing to an invalid
192 * dirent right now. Scan from the start of the block
193 * to make sure. */
Al Viro725bebb2013-05-17 16:08:53 -0400194 if (file->f_version != inode->i_version) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700195 for (i = 0; i < sb->s_blocksize && i < offset; ) {
Mingming Cao617ba132006-10-11 01:20:53 -0700196 de = (struct ext4_dir_entry_2 *)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700197 (bh->b_data + i);
198 /* It's too expensive to do a full
199 * dirent test each time round this
200 * loop, but we do have to test at
201 * least that it is non-zero. A
202 * failure will be detected in the
203 * dirent test below. */
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500204 if (ext4_rec_len_from_disk(de->rec_len,
205 sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700206 break;
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500207 i += ext4_rec_len_from_disk(de->rec_len,
208 sb->s_blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700209 }
210 offset = i;
Al Viro725bebb2013-05-17 16:08:53 -0400211 ctx->pos = (ctx->pos & ~(sb->s_blocksize - 1))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700212 | offset;
Al Viro725bebb2013-05-17 16:08:53 -0400213 file->f_version = inode->i_version;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700214 }
215
Al Viro725bebb2013-05-17 16:08:53 -0400216 while (ctx->pos < inode->i_size
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700217 && offset < sb->s_blocksize) {
Mingming Cao617ba132006-10-11 01:20:53 -0700218 de = (struct ext4_dir_entry_2 *) (bh->b_data + offset);
Al Viro725bebb2013-05-17 16:08:53 -0400219 if (ext4_check_dir_entry(inode, file, de, bh,
Tao Ma226ba972012-12-10 14:05:58 -0500220 bh->b_data, bh->b_size,
221 offset)) {
Andrew Morton63f57932006-10-11 01:21:24 -0700222 /*
Al Viro725bebb2013-05-17 16:08:53 -0400223 * On error, skip to the next block
Andrew Morton63f57932006-10-11 01:21:24 -0700224 */
Al Viro725bebb2013-05-17 16:08:53 -0400225 ctx->pos = (ctx->pos |
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700226 (sb->s_blocksize - 1)) + 1;
Al Viro725bebb2013-05-17 16:08:53 -0400227 break;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700228 }
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500229 offset += ext4_rec_len_from_disk(de->rec_len,
230 sb->s_blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700231 if (le32_to_cpu(de->inode)) {
Al Viro725bebb2013-05-17 16:08:53 -0400232 if (!dir_emit(ctx, de->name,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700233 de->name_len,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700234 le32_to_cpu(de->inode),
Al Viro725bebb2013-05-17 16:08:53 -0400235 get_dtype(sb, de->file_type))) {
236 brelse(bh);
237 return 0;
238 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700239 }
Al Viro725bebb2013-05-17 16:08:53 -0400240 ctx->pos += ext4_rec_len_from_disk(de->rec_len,
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500241 sb->s_blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700242 }
243 offset = 0;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400244 brelse(bh);
Al Viro725bebb2013-05-17 16:08:53 -0400245 if (ctx->pos < inode->i_size) {
246 if (!dir_relax(inode))
247 return 0;
248 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700249 }
Al Viro725bebb2013-05-17 16:08:53 -0400250 return 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700251}
252
Fan Yongd1f52732012-03-18 22:44:40 -0400253static inline int is_32bit_api(void)
254{
255#ifdef CONFIG_COMPAT
256 return is_compat_task();
257#else
258 return (BITS_PER_LONG == 32);
259#endif
260}
261
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700262/*
263 * These functions convert from the major/minor hash to an f_pos
Fan Yongd1f52732012-03-18 22:44:40 -0400264 * value for dx directories
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700265 *
Fan Yongd1f52732012-03-18 22:44:40 -0400266 * Upper layer (for example NFS) should specify FMODE_32BITHASH or
267 * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted
268 * directly on both 32-bit and 64-bit nodes, under such case, neither
269 * FMODE_32BITHASH nor FMODE_64BITHASH is specified.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700270 */
Fan Yongd1f52732012-03-18 22:44:40 -0400271static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor)
272{
273 if ((filp->f_mode & FMODE_32BITHASH) ||
274 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
275 return major >> 1;
276 else
277 return ((__u64)(major >> 1) << 32) | (__u64)minor;
278}
279
280static inline __u32 pos2maj_hash(struct file *filp, loff_t pos)
281{
282 if ((filp->f_mode & FMODE_32BITHASH) ||
283 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
284 return (pos << 1) & 0xffffffff;
285 else
286 return ((pos >> 32) << 1) & 0xffffffff;
287}
288
289static inline __u32 pos2min_hash(struct file *filp, loff_t pos)
290{
291 if ((filp->f_mode & FMODE_32BITHASH) ||
292 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
293 return 0;
294 else
295 return pos & 0xffffffff;
296}
297
298/*
299 * Return 32- or 64-bit end-of-file for dx directories
300 */
301static inline loff_t ext4_get_htree_eof(struct file *filp)
302{
303 if ((filp->f_mode & FMODE_32BITHASH) ||
304 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
305 return EXT4_HTREE_EOF_32BIT;
306 else
307 return EXT4_HTREE_EOF_64BIT;
308}
309
310
311/*
Eric Sandeenec7268c2012-04-30 13:14:03 -0500312 * ext4_dir_llseek() calls generic_file_llseek_size to handle htree
313 * directories, where the "offset" is in terms of the filename hash
314 * value instead of the byte offset.
Fan Yongd1f52732012-03-18 22:44:40 -0400315 *
Eric Sandeenec7268c2012-04-30 13:14:03 -0500316 * Because we may return a 64-bit hash that is well beyond offset limits,
317 * we need to pass the max hash as the maximum allowable offset in
318 * the htree directory case.
319 *
320 * For non-htree, ext4_llseek already chooses the proper max offset.
Fan Yongd1f52732012-03-18 22:44:40 -0400321 */
Zheng Liud4e43952013-03-02 17:24:05 -0500322static loff_t ext4_dir_llseek(struct file *file, loff_t offset, int whence)
Fan Yongd1f52732012-03-18 22:44:40 -0400323{
324 struct inode *inode = file->f_mapping->host;
Fan Yongd1f52732012-03-18 22:44:40 -0400325 int dx_dir = is_dx_dir(inode);
Eric Sandeenec7268c2012-04-30 13:14:03 -0500326 loff_t htree_max = ext4_get_htree_eof(file);
Fan Yongd1f52732012-03-18 22:44:40 -0400327
Eric Sandeenec7268c2012-04-30 13:14:03 -0500328 if (likely(dx_dir))
Andrew Morton965c8e52012-12-17 15:59:39 -0800329 return generic_file_llseek_size(file, offset, whence,
Eric Sandeenec7268c2012-04-30 13:14:03 -0500330 htree_max, htree_max);
331 else
Andrew Morton965c8e52012-12-17 15:59:39 -0800332 return ext4_llseek(file, offset, whence);
Fan Yongd1f52732012-03-18 22:44:40 -0400333}
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700334
335/*
336 * This structure holds the nodes of the red-black tree used to store
337 * the directory entry in hash order.
338 */
339struct fname {
340 __u32 hash;
341 __u32 minor_hash;
342 struct rb_node rb_hash;
343 struct fname *next;
344 __u32 inode;
345 __u8 name_len;
346 __u8 file_type;
347 char name[0];
348};
349
350/*
351 * This functoin implements a non-recursive way of freeing all of the
352 * nodes in the red-black tree.
353 */
354static void free_rb_tree_fname(struct rb_root *root)
355{
356 struct rb_node *n = root->rb_node;
357 struct rb_node *parent;
358 struct fname *fname;
359
360 while (n) {
361 /* Do the node's children first */
Shen Feng69baee02008-07-11 19:27:31 -0400362 if (n->rb_left) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700363 n = n->rb_left;
364 continue;
365 }
366 if (n->rb_right) {
367 n = n->rb_right;
368 continue;
369 }
370 /*
371 * The node has no children; free it, and then zero
372 * out parent's link to it. Finally go to the
373 * beginning of the loop and try to free the parent
374 * node.
375 */
376 parent = rb_parent(n);
377 fname = rb_entry(n, struct fname, rb_hash);
378 while (fname) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400379 struct fname *old = fname;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700380 fname = fname->next;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400381 kfree(old);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700382 }
383 if (!parent)
Venkatesh Pallipadi64e290e2010-03-04 22:25:21 -0500384 *root = RB_ROOT;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700385 else if (parent->rb_left == n)
386 parent->rb_left = NULL;
387 else if (parent->rb_right == n)
388 parent->rb_right = NULL;
389 n = parent;
390 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700391}
392
393
Fan Yongd1f52732012-03-18 22:44:40 -0400394static struct dir_private_info *ext4_htree_create_dir_info(struct file *filp,
395 loff_t pos)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700396{
397 struct dir_private_info *p;
398
Shen Feng69baee02008-07-11 19:27:31 -0400399 p = kzalloc(sizeof(struct dir_private_info), GFP_KERNEL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700400 if (!p)
401 return NULL;
Fan Yongd1f52732012-03-18 22:44:40 -0400402 p->curr_hash = pos2maj_hash(filp, pos);
403 p->curr_minor_hash = pos2min_hash(filp, pos);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700404 return p;
405}
406
Mingming Cao617ba132006-10-11 01:20:53 -0700407void ext4_htree_free_dir_info(struct dir_private_info *p)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700408{
409 free_rb_tree_fname(&p->root);
410 kfree(p);
411}
412
413/*
414 * Given a directory entry, enter it into the fname rb tree.
415 */
Mingming Cao617ba132006-10-11 01:20:53 -0700416int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700417 __u32 minor_hash,
Mingming Cao617ba132006-10-11 01:20:53 -0700418 struct ext4_dir_entry_2 *dirent)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700419{
420 struct rb_node **p, *parent = NULL;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400421 struct fname *fname, *new_fn;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700422 struct dir_private_info *info;
423 int len;
424
Joe Perchesa271fe82010-07-27 11:56:04 -0400425 info = dir_file->private_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700426 p = &info->root.rb_node;
427
428 /* Create and allocate the fname structure */
429 len = sizeof(struct fname) + dirent->name_len + 1;
430 new_fn = kzalloc(len, GFP_KERNEL);
431 if (!new_fn)
432 return -ENOMEM;
433 new_fn->hash = hash;
434 new_fn->minor_hash = minor_hash;
435 new_fn->inode = le32_to_cpu(dirent->inode);
436 new_fn->name_len = dirent->name_len;
437 new_fn->file_type = dirent->file_type;
438 memcpy(new_fn->name, dirent->name, dirent->name_len);
439 new_fn->name[dirent->name_len] = 0;
440
441 while (*p) {
442 parent = *p;
443 fname = rb_entry(parent, struct fname, rb_hash);
444
445 /*
446 * If the hash and minor hash match up, then we put
447 * them on a linked list. This rarely happens...
448 */
449 if ((new_fn->hash == fname->hash) &&
450 (new_fn->minor_hash == fname->minor_hash)) {
451 new_fn->next = fname->next;
452 fname->next = new_fn;
453 return 0;
454 }
455
456 if (new_fn->hash < fname->hash)
457 p = &(*p)->rb_left;
458 else if (new_fn->hash > fname->hash)
459 p = &(*p)->rb_right;
460 else if (new_fn->minor_hash < fname->minor_hash)
461 p = &(*p)->rb_left;
462 else /* if (new_fn->minor_hash > fname->minor_hash) */
463 p = &(*p)->rb_right;
464 }
465
466 rb_link_node(&new_fn->rb_hash, parent, p);
467 rb_insert_color(&new_fn->rb_hash, &info->root);
468 return 0;
469}
470
471
472
473/*
Mingming Cao617ba132006-10-11 01:20:53 -0700474 * This is a helper function for ext4_dx_readdir. It calls filldir
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700475 * for all entres on the fname linked list. (Normally there is only
476 * one entry on the linked list, unless there are 62 bit hash collisions.)
477 */
Al Viro725bebb2013-05-17 16:08:53 -0400478static int call_filldir(struct file *file, struct dir_context *ctx,
479 struct fname *fname)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700480{
Al Viro725bebb2013-05-17 16:08:53 -0400481 struct dir_private_info *info = file->private_data;
482 struct inode *inode = file_inode(file);
483 struct super_block *sb = inode->i_sb;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700484
485 if (!fname) {
Theodore Ts'o92b97812012-03-19 23:41:49 -0400486 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: "
487 "called with null fname?!?", __func__, __LINE__,
488 inode->i_ino, current->comm);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700489 return 0;
490 }
Al Viro725bebb2013-05-17 16:08:53 -0400491 ctx->pos = hash2pos(file, fname->hash, fname->minor_hash);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700492 while (fname) {
Al Viro725bebb2013-05-17 16:08:53 -0400493 if (!dir_emit(ctx, fname->name,
494 fname->name_len,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700495 fname->inode,
Al Viro725bebb2013-05-17 16:08:53 -0400496 get_dtype(sb, fname->file_type))) {
Theodore Ts'od0156412008-08-19 21:57:43 -0400497 info->extra_fname = fname;
Al Viro725bebb2013-05-17 16:08:53 -0400498 return 1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700499 }
500 fname = fname->next;
501 }
502 return 0;
503}
504
Al Viro725bebb2013-05-17 16:08:53 -0400505static int ext4_dx_readdir(struct file *file, struct dir_context *ctx)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700506{
Al Viro725bebb2013-05-17 16:08:53 -0400507 struct dir_private_info *info = file->private_data;
508 struct inode *inode = file_inode(file);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700509 struct fname *fname;
510 int ret;
511
512 if (!info) {
Al Viro725bebb2013-05-17 16:08:53 -0400513 info = ext4_htree_create_dir_info(file, ctx->pos);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700514 if (!info)
515 return -ENOMEM;
Al Viro725bebb2013-05-17 16:08:53 -0400516 file->private_data = info;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700517 }
518
Al Viro725bebb2013-05-17 16:08:53 -0400519 if (ctx->pos == ext4_get_htree_eof(file))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700520 return 0; /* EOF */
521
522 /* Some one has messed with f_pos; reset the world */
Al Viro725bebb2013-05-17 16:08:53 -0400523 if (info->last_pos != ctx->pos) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700524 free_rb_tree_fname(&info->root);
525 info->curr_node = NULL;
526 info->extra_fname = NULL;
Al Viro725bebb2013-05-17 16:08:53 -0400527 info->curr_hash = pos2maj_hash(file, ctx->pos);
528 info->curr_minor_hash = pos2min_hash(file, ctx->pos);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700529 }
530
531 /*
532 * If there are any leftover names on the hash collision
533 * chain, return them first.
534 */
Theodore Ts'od0156412008-08-19 21:57:43 -0400535 if (info->extra_fname) {
Al Viro725bebb2013-05-17 16:08:53 -0400536 if (call_filldir(file, ctx, info->extra_fname))
Theodore Ts'od0156412008-08-19 21:57:43 -0400537 goto finished;
Theodore Ts'od0156412008-08-19 21:57:43 -0400538 info->extra_fname = NULL;
Theodore Ts'o3c37fc82008-10-25 11:39:08 -0400539 goto next_node;
Theodore Ts'od0156412008-08-19 21:57:43 -0400540 } else if (!info->curr_node)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700541 info->curr_node = rb_first(&info->root);
542
543 while (1) {
544 /*
545 * Fill the rbtree if we have no more entries,
546 * or the inode has changed since we last read in the
547 * cached entries.
548 */
549 if ((!info->curr_node) ||
Al Viro725bebb2013-05-17 16:08:53 -0400550 (file->f_version != inode->i_version)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700551 info->curr_node = NULL;
552 free_rb_tree_fname(&info->root);
Al Viro725bebb2013-05-17 16:08:53 -0400553 file->f_version = inode->i_version;
554 ret = ext4_htree_fill_tree(file, info->curr_hash,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700555 info->curr_minor_hash,
556 &info->next_hash);
557 if (ret < 0)
558 return ret;
559 if (ret == 0) {
Al Viro725bebb2013-05-17 16:08:53 -0400560 ctx->pos = ext4_get_htree_eof(file);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700561 break;
562 }
563 info->curr_node = rb_first(&info->root);
564 }
565
566 fname = rb_entry(info->curr_node, struct fname, rb_hash);
567 info->curr_hash = fname->hash;
568 info->curr_minor_hash = fname->minor_hash;
Al Viro725bebb2013-05-17 16:08:53 -0400569 if (call_filldir(file, ctx, fname))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700570 break;
Theodore Ts'o3c37fc82008-10-25 11:39:08 -0400571 next_node:
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700572 info->curr_node = rb_next(info->curr_node);
Theodore Ts'o3c37fc82008-10-25 11:39:08 -0400573 if (info->curr_node) {
574 fname = rb_entry(info->curr_node, struct fname,
575 rb_hash);
576 info->curr_hash = fname->hash;
577 info->curr_minor_hash = fname->minor_hash;
578 } else {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700579 if (info->next_hash == ~0) {
Al Viro725bebb2013-05-17 16:08:53 -0400580 ctx->pos = ext4_get_htree_eof(file);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700581 break;
582 }
583 info->curr_hash = info->next_hash;
584 info->curr_minor_hash = 0;
585 }
586 }
587finished:
Al Viro725bebb2013-05-17 16:08:53 -0400588 info->last_pos = ctx->pos;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700589 return 0;
590}
591
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400592static int ext4_release_dir(struct inode *inode, struct file *filp)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700593{
Andrew Morton63f57932006-10-11 01:21:24 -0700594 if (filp->private_data)
Mingming Cao617ba132006-10-11 01:20:53 -0700595 ext4_htree_free_dir_info(filp->private_data);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700596
597 return 0;
598}
Fan Yongd1f52732012-03-18 22:44:40 -0400599
600const struct file_operations ext4_dir_operations = {
601 .llseek = ext4_dir_llseek,
602 .read = generic_read_dir,
Al Viro725bebb2013-05-17 16:08:53 -0400603 .iterate = ext4_readdir,
Fan Yongd1f52732012-03-18 22:44:40 -0400604 .unlocked_ioctl = ext4_ioctl,
605#ifdef CONFIG_COMPAT
606 .compat_ioctl = ext4_compat_ioctl,
607#endif
608 .fsync = ext4_sync_file,
609 .release = ext4_release_dir,
610};