blob: 8e07d2a5a13952836c820c32419f340b148a8ce2 [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"
Dave Kleikampac27a0e2006-10-11 01:20:50 -070030
Mingming Cao617ba132006-10-11 01:20:53 -070031static unsigned char ext4_filetype_table[] = {
Dave Kleikampac27a0e2006-10-11 01:20:50 -070032 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
33};
34
Theodore Ts'oaf5bc922008-09-08 22:25:24 -040035static int ext4_dx_readdir(struct file *filp,
36 void *dirent, filldir_t filldir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070037
38static unsigned char get_dtype(struct super_block *sb, int filetype)
39{
Mingming Cao617ba132006-10-11 01:20:53 -070040 if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE) ||
41 (filetype >= EXT4_FT_MAX))
Dave Kleikampac27a0e2006-10-11 01:20:50 -070042 return DT_UNKNOWN;
43
Mingming Cao617ba132006-10-11 01:20:53 -070044 return (ext4_filetype_table[filetype]);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070045}
46
Fan Yongd1f52732012-03-18 22:44:40 -040047/**
48 * Check if the given dir-inode refers to an htree-indexed directory
49 * (or a directory which chould potentially get coverted to use htree
50 * indexing).
51 *
52 * Return 1 if it is a dx dir, 0 if not
53 */
54static int is_dx_dir(struct inode *inode)
55{
56 struct super_block *sb = inode->i_sb;
57
58 if (EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
59 EXT4_FEATURE_COMPAT_DIR_INDEX) &&
60 ((ext4_test_inode_flag(inode, EXT4_INODE_INDEX)) ||
61 ((inode->i_size >> sb->s_blocksize_bits) == 1)))
62 return 1;
63
64 return 0;
65}
66
Theodore Ts'ocad3f002010-12-19 22:07:02 -050067/*
68 * Return 0 if the directory entry is OK, and 1 if there is a problem
69 *
70 * Note: this is the opposite of what ext2 and ext3 historically returned...
71 */
Theodore Ts'o60fd4da2010-07-27 11:54:40 -040072int __ext4_check_dir_entry(const char *function, unsigned int line,
Theodore Ts'of7c21172011-01-10 12:10:55 -050073 struct inode *dir, struct file *filp,
Theodore Ts'o60fd4da2010-07-27 11:54:40 -040074 struct ext4_dir_entry_2 *de,
75 struct buffer_head *bh,
76 unsigned int offset)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070077{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -040078 const char *error_msg = NULL;
Wei Yongjun3d0518f2009-02-14 23:01:36 -050079 const int rlen = ext4_rec_len_from_disk(de->rec_len,
80 dir->i_sb->s_blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070081
Theodore Ts'ocad3f002010-12-19 22:07:02 -050082 if (unlikely(rlen < EXT4_DIR_REC_LEN(1)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -070083 error_msg = "rec_len is smaller than minimal";
Theodore Ts'ocad3f002010-12-19 22:07:02 -050084 else if (unlikely(rlen % 4 != 0))
Dave Kleikampac27a0e2006-10-11 01:20:50 -070085 error_msg = "rec_len % 4 != 0";
Theodore Ts'ocad3f002010-12-19 22:07:02 -050086 else if (unlikely(rlen < EXT4_DIR_REC_LEN(de->name_len)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -070087 error_msg = "rec_len is too small for name_len";
Theodore Ts'ocad3f002010-12-19 22:07:02 -050088 else if (unlikely(((char *) de - bh->b_data) + rlen >
89 dir->i_sb->s_blocksize))
Dave Kleikampac27a0e2006-10-11 01:20:50 -070090 error_msg = "directory entry across blocks";
Theodore Ts'ocad3f002010-12-19 22:07:02 -050091 else if (unlikely(le32_to_cpu(de->inode) >
92 le32_to_cpu(EXT4_SB(dir->i_sb)->s_es->s_inodes_count)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -070093 error_msg = "inode out of bounds";
Theodore Ts'ocad3f002010-12-19 22:07:02 -050094 else
95 return 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -070096
Theodore Ts'of7c21172011-01-10 12:10:55 -050097 if (filp)
Dan Carpenter4fda4002012-02-20 17:53:05 -050098 ext4_error_file(filp, function, line, bh->b_blocknr,
Theodore Ts'of7c21172011-01-10 12:10:55 -050099 "bad entry in directory: %s - offset=%u(%u), "
100 "inode=%u, rec_len=%d, name_len=%d",
Dan Carpenter4fda4002012-02-20 17:53:05 -0500101 error_msg, (unsigned) (offset % bh->b_size),
Theodore Ts'of7c21172011-01-10 12:10:55 -0500102 offset, le32_to_cpu(de->inode),
103 rlen, de->name_len);
104 else
Dan Carpenter4fda4002012-02-20 17:53:05 -0500105 ext4_error_inode(dir, function, line, bh->b_blocknr,
Theodore Ts'of7c21172011-01-10 12:10:55 -0500106 "bad entry in directory: %s - offset=%u(%u), "
107 "inode=%u, rec_len=%d, name_len=%d",
Dan Carpenter4fda4002012-02-20 17:53:05 -0500108 error_msg, (unsigned) (offset % bh->b_size),
Theodore Ts'of7c21172011-01-10 12:10:55 -0500109 offset, le32_to_cpu(de->inode),
110 rlen, de->name_len);
111
Theodore Ts'ocad3f002010-12-19 22:07:02 -0500112 return 1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700113}
114
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400115static int ext4_readdir(struct file *filp,
116 void *dirent, filldir_t filldir)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700117{
118 int error = 0;
Theodore Ts'o498e5f22008-11-05 00:14:04 -0500119 unsigned int offset;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700120 int i, stored;
Mingming Cao617ba132006-10-11 01:20:53 -0700121 struct ext4_dir_entry_2 *de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700122 int err;
Josef "Jeff" Sipek9d549892006-12-08 02:36:38 -0800123 struct inode *inode = filp->f_path.dentry->d_inode;
Fan Yongd1f52732012-03-18 22:44:40 -0400124 struct super_block *sb = inode->i_sb;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700125 int ret = 0;
Eric Sandeen9d9f1772008-10-09 11:15:52 -0400126 int dir_has_error = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700127
Fan Yongd1f52732012-03-18 22:44:40 -0400128 if (is_dx_dir(inode)) {
Mingming Cao617ba132006-10-11 01:20:53 -0700129 err = ext4_dx_readdir(filp, dirent, filldir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700130 if (err != ERR_BAD_DX_DIR) {
131 ret = err;
132 goto out;
133 }
134 /*
135 * We don't set the inode dirty flag since it's not
136 * critical that it get flushed back to the disk.
137 */
Theodore Ts'o07a03822010-06-14 09:54:48 -0400138 ext4_clear_inode_flag(filp->f_path.dentry->d_inode,
139 EXT4_INODE_INDEX);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700140 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700141 stored = 0;
142 offset = filp->f_pos & (sb->s_blocksize - 1);
143
144 while (!error && !stored && filp->f_pos < inode->i_size) {
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400145 struct ext4_map_blocks map;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700146 struct buffer_head *bh = NULL;
147
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400148 map.m_lblk = filp->f_pos >> EXT4_BLOCK_SIZE_BITS(sb);
149 map.m_len = 1;
150 err = ext4_map_blocks(NULL, inode, &map, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700151 if (err > 0) {
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400152 pgoff_t index = map.m_pblk >>
Fengguang Wudc7868f2007-07-19 01:48:04 -0700153 (PAGE_CACHE_SHIFT - inode->i_blkbits);
154 if (!ra_has_index(&filp->f_ra, index))
Rusty Russellcf914a72007-07-19 01:48:08 -0700155 page_cache_sync_readahead(
Fengguang Wudc7868f2007-07-19 01:48:04 -0700156 sb->s_bdev->bd_inode->i_mapping,
157 &filp->f_ra, filp,
Rusty Russellcf914a72007-07-19 01:48:08 -0700158 index, 1);
Fengguang Wuf4e6b492007-10-16 01:24:33 -0700159 filp->f_ra.prev_pos = (loff_t)index << PAGE_CACHE_SHIFT;
Theodore Ts'o2ed88682010-05-16 20:00:00 -0400160 bh = ext4_bread(NULL, inode, map.m_lblk, 0, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700161 }
162
163 /*
164 * We ignore I/O errors on directories so users have a chance
165 * of recovering data when there's a bad sector
166 */
167 if (!bh) {
Eric Sandeen9d9f1772008-10-09 11:15:52 -0400168 if (!dir_has_error) {
Theodore Ts'of7c21172011-01-10 12:10:55 -0500169 EXT4_ERROR_FILE(filp, 0,
170 "directory contains a "
171 "hole at offset %llu",
Eric Sandeen9d9f1772008-10-09 11:15:52 -0400172 (unsigned long long) filp->f_pos);
173 dir_has_error = 1;
174 }
Eric Sandeene6c40212006-12-06 20:36:28 -0800175 /* corrupt size? Maybe no more blocks to read */
176 if (filp->f_pos > inode->i_blocks << 9)
177 break;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700178 filp->f_pos += sb->s_blocksize - offset;
179 continue;
180 }
181
Darrick J. Wongb0336e82012-04-29 18:41:10 -0400182 /* Check the checksum */
183 if (!buffer_verified(bh) &&
184 !ext4_dirent_csum_verify(inode,
185 (struct ext4_dir_entry *)bh->b_data)) {
186 EXT4_ERROR_FILE(filp, 0, "directory fails checksum "
187 "at offset %llu",
188 (unsigned long long)filp->f_pos);
189 filp->f_pos += sb->s_blocksize - offset;
190 continue;
191 }
192 set_buffer_verified(bh);
193
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700194revalidate:
195 /* If the dir block has changed since the last call to
196 * readdir(2), then we might be pointing to an invalid
197 * dirent right now. Scan from the start of the block
198 * to make sure. */
199 if (filp->f_version != inode->i_version) {
200 for (i = 0; i < sb->s_blocksize && i < offset; ) {
Mingming Cao617ba132006-10-11 01:20:53 -0700201 de = (struct ext4_dir_entry_2 *)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700202 (bh->b_data + i);
203 /* It's too expensive to do a full
204 * dirent test each time round this
205 * loop, but we do have to test at
206 * least that it is non-zero. A
207 * failure will be detected in the
208 * dirent test below. */
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500209 if (ext4_rec_len_from_disk(de->rec_len,
210 sb->s_blocksize) < EXT4_DIR_REC_LEN(1))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700211 break;
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500212 i += ext4_rec_len_from_disk(de->rec_len,
213 sb->s_blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700214 }
215 offset = i;
216 filp->f_pos = (filp->f_pos & ~(sb->s_blocksize - 1))
217 | offset;
218 filp->f_version = inode->i_version;
219 }
220
221 while (!error && filp->f_pos < inode->i_size
222 && offset < sb->s_blocksize) {
Mingming Cao617ba132006-10-11 01:20:53 -0700223 de = (struct ext4_dir_entry_2 *) (bh->b_data + offset);
Theodore Ts'of7c21172011-01-10 12:10:55 -0500224 if (ext4_check_dir_entry(inode, filp, de,
Theodore Ts'ocad3f002010-12-19 22:07:02 -0500225 bh, offset)) {
Andrew Morton63f57932006-10-11 01:21:24 -0700226 /*
227 * On error, skip the f_pos to the next block
228 */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700229 filp->f_pos = (filp->f_pos |
230 (sb->s_blocksize - 1)) + 1;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400231 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700232 ret = stored;
233 goto out;
234 }
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500235 offset += ext4_rec_len_from_disk(de->rec_len,
236 sb->s_blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700237 if (le32_to_cpu(de->inode)) {
238 /* We might block in the next section
239 * if the data destination is
240 * currently swapped out. So, use a
241 * version stamp to detect whether or
242 * not the directory has been modified
243 * during the copy operation.
244 */
Mathieu Desnoyers2b47c362007-10-16 23:27:21 -0700245 u64 version = filp->f_version;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700246
247 error = filldir(dirent, de->name,
248 de->name_len,
249 filp->f_pos,
250 le32_to_cpu(de->inode),
251 get_dtype(sb, de->file_type));
252 if (error)
253 break;
254 if (version != filp->f_version)
255 goto revalidate;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400256 stored++;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700257 }
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500258 filp->f_pos += ext4_rec_len_from_disk(de->rec_len,
259 sb->s_blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700260 }
261 offset = 0;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400262 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700263 }
264out:
265 return ret;
266}
267
Fan Yongd1f52732012-03-18 22:44:40 -0400268static inline int is_32bit_api(void)
269{
270#ifdef CONFIG_COMPAT
271 return is_compat_task();
272#else
273 return (BITS_PER_LONG == 32);
274#endif
275}
276
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700277/*
278 * These functions convert from the major/minor hash to an f_pos
Fan Yongd1f52732012-03-18 22:44:40 -0400279 * value for dx directories
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700280 *
Fan Yongd1f52732012-03-18 22:44:40 -0400281 * Upper layer (for example NFS) should specify FMODE_32BITHASH or
282 * FMODE_64BITHASH explicitly. On the other hand, we allow ext4 to be mounted
283 * directly on both 32-bit and 64-bit nodes, under such case, neither
284 * FMODE_32BITHASH nor FMODE_64BITHASH is specified.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700285 */
Fan Yongd1f52732012-03-18 22:44:40 -0400286static inline loff_t hash2pos(struct file *filp, __u32 major, __u32 minor)
287{
288 if ((filp->f_mode & FMODE_32BITHASH) ||
289 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
290 return major >> 1;
291 else
292 return ((__u64)(major >> 1) << 32) | (__u64)minor;
293}
294
295static inline __u32 pos2maj_hash(struct file *filp, loff_t pos)
296{
297 if ((filp->f_mode & FMODE_32BITHASH) ||
298 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
299 return (pos << 1) & 0xffffffff;
300 else
301 return ((pos >> 32) << 1) & 0xffffffff;
302}
303
304static inline __u32 pos2min_hash(struct file *filp, loff_t pos)
305{
306 if ((filp->f_mode & FMODE_32BITHASH) ||
307 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
308 return 0;
309 else
310 return pos & 0xffffffff;
311}
312
313/*
314 * Return 32- or 64-bit end-of-file for dx directories
315 */
316static inline loff_t ext4_get_htree_eof(struct file *filp)
317{
318 if ((filp->f_mode & FMODE_32BITHASH) ||
319 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
320 return EXT4_HTREE_EOF_32BIT;
321 else
322 return EXT4_HTREE_EOF_64BIT;
323}
324
325
326/*
Eric Sandeenec7268c2012-04-30 13:14:03 -0500327 * ext4_dir_llseek() calls generic_file_llseek_size to handle htree
328 * directories, where the "offset" is in terms of the filename hash
329 * value instead of the byte offset.
Fan Yongd1f52732012-03-18 22:44:40 -0400330 *
Eric Sandeenec7268c2012-04-30 13:14:03 -0500331 * Because we may return a 64-bit hash that is well beyond offset limits,
332 * we need to pass the max hash as the maximum allowable offset in
333 * the htree directory case.
334 *
335 * For non-htree, ext4_llseek already chooses the proper max offset.
Fan Yongd1f52732012-03-18 22:44:40 -0400336 */
337loff_t ext4_dir_llseek(struct file *file, loff_t offset, int origin)
338{
339 struct inode *inode = file->f_mapping->host;
Fan Yongd1f52732012-03-18 22:44:40 -0400340 int dx_dir = is_dx_dir(inode);
Eric Sandeenec7268c2012-04-30 13:14:03 -0500341 loff_t htree_max = ext4_get_htree_eof(file);
Fan Yongd1f52732012-03-18 22:44:40 -0400342
Eric Sandeenec7268c2012-04-30 13:14:03 -0500343 if (likely(dx_dir))
344 return generic_file_llseek_size(file, offset, origin,
345 htree_max, htree_max);
346 else
347 return ext4_llseek(file, offset, origin);
Fan Yongd1f52732012-03-18 22:44:40 -0400348}
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700349
350/*
351 * This structure holds the nodes of the red-black tree used to store
352 * the directory entry in hash order.
353 */
354struct fname {
355 __u32 hash;
356 __u32 minor_hash;
357 struct rb_node rb_hash;
358 struct fname *next;
359 __u32 inode;
360 __u8 name_len;
361 __u8 file_type;
362 char name[0];
363};
364
365/*
366 * This functoin implements a non-recursive way of freeing all of the
367 * nodes in the red-black tree.
368 */
369static void free_rb_tree_fname(struct rb_root *root)
370{
371 struct rb_node *n = root->rb_node;
372 struct rb_node *parent;
373 struct fname *fname;
374
375 while (n) {
376 /* Do the node's children first */
Shen Feng69baee02008-07-11 19:27:31 -0400377 if (n->rb_left) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700378 n = n->rb_left;
379 continue;
380 }
381 if (n->rb_right) {
382 n = n->rb_right;
383 continue;
384 }
385 /*
386 * The node has no children; free it, and then zero
387 * out parent's link to it. Finally go to the
388 * beginning of the loop and try to free the parent
389 * node.
390 */
391 parent = rb_parent(n);
392 fname = rb_entry(n, struct fname, rb_hash);
393 while (fname) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400394 struct fname *old = fname;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700395 fname = fname->next;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400396 kfree(old);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700397 }
398 if (!parent)
Venkatesh Pallipadi64e290e2010-03-04 22:25:21 -0500399 *root = RB_ROOT;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700400 else if (parent->rb_left == n)
401 parent->rb_left = NULL;
402 else if (parent->rb_right == n)
403 parent->rb_right = NULL;
404 n = parent;
405 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700406}
407
408
Fan Yongd1f52732012-03-18 22:44:40 -0400409static struct dir_private_info *ext4_htree_create_dir_info(struct file *filp,
410 loff_t pos)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700411{
412 struct dir_private_info *p;
413
Shen Feng69baee02008-07-11 19:27:31 -0400414 p = kzalloc(sizeof(struct dir_private_info), GFP_KERNEL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700415 if (!p)
416 return NULL;
Fan Yongd1f52732012-03-18 22:44:40 -0400417 p->curr_hash = pos2maj_hash(filp, pos);
418 p->curr_minor_hash = pos2min_hash(filp, pos);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700419 return p;
420}
421
Mingming Cao617ba132006-10-11 01:20:53 -0700422void ext4_htree_free_dir_info(struct dir_private_info *p)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700423{
424 free_rb_tree_fname(&p->root);
425 kfree(p);
426}
427
428/*
429 * Given a directory entry, enter it into the fname rb tree.
430 */
Mingming Cao617ba132006-10-11 01:20:53 -0700431int ext4_htree_store_dirent(struct file *dir_file, __u32 hash,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700432 __u32 minor_hash,
Mingming Cao617ba132006-10-11 01:20:53 -0700433 struct ext4_dir_entry_2 *dirent)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700434{
435 struct rb_node **p, *parent = NULL;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400436 struct fname *fname, *new_fn;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700437 struct dir_private_info *info;
438 int len;
439
Joe Perchesa271fe82010-07-27 11:56:04 -0400440 info = dir_file->private_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700441 p = &info->root.rb_node;
442
443 /* Create and allocate the fname structure */
444 len = sizeof(struct fname) + dirent->name_len + 1;
445 new_fn = kzalloc(len, GFP_KERNEL);
446 if (!new_fn)
447 return -ENOMEM;
448 new_fn->hash = hash;
449 new_fn->minor_hash = minor_hash;
450 new_fn->inode = le32_to_cpu(dirent->inode);
451 new_fn->name_len = dirent->name_len;
452 new_fn->file_type = dirent->file_type;
453 memcpy(new_fn->name, dirent->name, dirent->name_len);
454 new_fn->name[dirent->name_len] = 0;
455
456 while (*p) {
457 parent = *p;
458 fname = rb_entry(parent, struct fname, rb_hash);
459
460 /*
461 * If the hash and minor hash match up, then we put
462 * them on a linked list. This rarely happens...
463 */
464 if ((new_fn->hash == fname->hash) &&
465 (new_fn->minor_hash == fname->minor_hash)) {
466 new_fn->next = fname->next;
467 fname->next = new_fn;
468 return 0;
469 }
470
471 if (new_fn->hash < fname->hash)
472 p = &(*p)->rb_left;
473 else if (new_fn->hash > fname->hash)
474 p = &(*p)->rb_right;
475 else if (new_fn->minor_hash < fname->minor_hash)
476 p = &(*p)->rb_left;
477 else /* if (new_fn->minor_hash > fname->minor_hash) */
478 p = &(*p)->rb_right;
479 }
480
481 rb_link_node(&new_fn->rb_hash, parent, p);
482 rb_insert_color(&new_fn->rb_hash, &info->root);
483 return 0;
484}
485
486
487
488/*
Mingming Cao617ba132006-10-11 01:20:53 -0700489 * This is a helper function for ext4_dx_readdir. It calls filldir
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700490 * for all entres on the fname linked list. (Normally there is only
491 * one entry on the linked list, unless there are 62 bit hash collisions.)
492 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400493static int call_filldir(struct file *filp, void *dirent,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700494 filldir_t filldir, struct fname *fname)
495{
496 struct dir_private_info *info = filp->private_data;
497 loff_t curr_pos;
Josef "Jeff" Sipek9d549892006-12-08 02:36:38 -0800498 struct inode *inode = filp->f_path.dentry->d_inode;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400499 struct super_block *sb;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700500 int error;
501
502 sb = inode->i_sb;
503
504 if (!fname) {
Theodore Ts'o92b97812012-03-19 23:41:49 -0400505 ext4_msg(sb, KERN_ERR, "%s:%d: inode #%lu: comm %s: "
506 "called with null fname?!?", __func__, __LINE__,
507 inode->i_ino, current->comm);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700508 return 0;
509 }
Fan Yongd1f52732012-03-18 22:44:40 -0400510 curr_pos = hash2pos(filp, fname->hash, fname->minor_hash);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700511 while (fname) {
512 error = filldir(dirent, fname->name,
513 fname->name_len, curr_pos,
514 fname->inode,
515 get_dtype(sb, fname->file_type));
516 if (error) {
517 filp->f_pos = curr_pos;
Theodore Ts'od0156412008-08-19 21:57:43 -0400518 info->extra_fname = fname;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700519 return error;
520 }
521 fname = fname->next;
522 }
523 return 0;
524}
525
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400526static int ext4_dx_readdir(struct file *filp,
527 void *dirent, filldir_t filldir)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700528{
529 struct dir_private_info *info = filp->private_data;
Josef "Jeff" Sipek9d549892006-12-08 02:36:38 -0800530 struct inode *inode = filp->f_path.dentry->d_inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700531 struct fname *fname;
532 int ret;
533
534 if (!info) {
Fan Yongd1f52732012-03-18 22:44:40 -0400535 info = ext4_htree_create_dir_info(filp, filp->f_pos);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700536 if (!info)
537 return -ENOMEM;
538 filp->private_data = info;
539 }
540
Fan Yongd1f52732012-03-18 22:44:40 -0400541 if (filp->f_pos == ext4_get_htree_eof(filp))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700542 return 0; /* EOF */
543
544 /* Some one has messed with f_pos; reset the world */
545 if (info->last_pos != filp->f_pos) {
546 free_rb_tree_fname(&info->root);
547 info->curr_node = NULL;
548 info->extra_fname = NULL;
Fan Yongd1f52732012-03-18 22:44:40 -0400549 info->curr_hash = pos2maj_hash(filp, filp->f_pos);
550 info->curr_minor_hash = pos2min_hash(filp, filp->f_pos);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700551 }
552
553 /*
554 * If there are any leftover names on the hash collision
555 * chain, return them first.
556 */
Theodore Ts'od0156412008-08-19 21:57:43 -0400557 if (info->extra_fname) {
558 if (call_filldir(filp, dirent, filldir, info->extra_fname))
559 goto finished;
Theodore Ts'od0156412008-08-19 21:57:43 -0400560 info->extra_fname = NULL;
Theodore Ts'o3c37fc82008-10-25 11:39:08 -0400561 goto next_node;
Theodore Ts'od0156412008-08-19 21:57:43 -0400562 } else if (!info->curr_node)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700563 info->curr_node = rb_first(&info->root);
564
565 while (1) {
566 /*
567 * Fill the rbtree if we have no more entries,
568 * or the inode has changed since we last read in the
569 * cached entries.
570 */
571 if ((!info->curr_node) ||
572 (filp->f_version != inode->i_version)) {
573 info->curr_node = NULL;
574 free_rb_tree_fname(&info->root);
575 filp->f_version = inode->i_version;
Mingming Cao617ba132006-10-11 01:20:53 -0700576 ret = ext4_htree_fill_tree(filp, info->curr_hash,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700577 info->curr_minor_hash,
578 &info->next_hash);
579 if (ret < 0)
580 return ret;
581 if (ret == 0) {
Fan Yongd1f52732012-03-18 22:44:40 -0400582 filp->f_pos = ext4_get_htree_eof(filp);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700583 break;
584 }
585 info->curr_node = rb_first(&info->root);
586 }
587
588 fname = rb_entry(info->curr_node, struct fname, rb_hash);
589 info->curr_hash = fname->hash;
590 info->curr_minor_hash = fname->minor_hash;
591 if (call_filldir(filp, dirent, filldir, fname))
592 break;
Theodore Ts'o3c37fc82008-10-25 11:39:08 -0400593 next_node:
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700594 info->curr_node = rb_next(info->curr_node);
Theodore Ts'o3c37fc82008-10-25 11:39:08 -0400595 if (info->curr_node) {
596 fname = rb_entry(info->curr_node, struct fname,
597 rb_hash);
598 info->curr_hash = fname->hash;
599 info->curr_minor_hash = fname->minor_hash;
600 } else {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700601 if (info->next_hash == ~0) {
Fan Yongd1f52732012-03-18 22:44:40 -0400602 filp->f_pos = ext4_get_htree_eof(filp);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700603 break;
604 }
605 info->curr_hash = info->next_hash;
606 info->curr_minor_hash = 0;
607 }
608 }
609finished:
610 info->last_pos = filp->f_pos;
611 return 0;
612}
613
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400614static int ext4_release_dir(struct inode *inode, struct file *filp)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700615{
Andrew Morton63f57932006-10-11 01:21:24 -0700616 if (filp->private_data)
Mingming Cao617ba132006-10-11 01:20:53 -0700617 ext4_htree_free_dir_info(filp->private_data);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700618
619 return 0;
620}
Fan Yongd1f52732012-03-18 22:44:40 -0400621
622const struct file_operations ext4_dir_operations = {
623 .llseek = ext4_dir_llseek,
624 .read = generic_read_dir,
625 .readdir = ext4_readdir,
626 .unlocked_ioctl = ext4_ioctl,
627#ifdef CONFIG_COMPAT
628 .compat_ioctl = ext4_compat_ioctl,
629#endif
630 .fsync = ext4_sync_file,
631 .release = ext4_release_dir,
632};