blob: 314c0d3b3fa9aece3015a113b6d348b7bc2be4aa [file] [log] [blame]
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001/*
Mingming Cao617ba132006-10-11 01:20:53 -07002 * linux/fs/ext4/namei.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/namei.c
12 *
13 * Copyright (C) 1991, 1992 Linus Torvalds
14 *
15 * Big-endian to little-endian byte-swapping/bitmaps by
16 * David S. Miller (davem@caip.rutgers.edu), 1995
17 * Directory entry file type support and forward compatibility hooks
18 * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
19 * Hash Tree Directory indexing (c)
20 * Daniel Phillips, 2001
21 * Hash Tree Directory indexing porting
22 * Christopher Li, 2002
23 * Hash Tree Directory indexing cleanup
24 * Theodore Ts'o, 2002
25 */
26
27#include <linux/fs.h>
28#include <linux/pagemap.h>
Mingming Caodab291a2006-10-11 01:21:01 -070029#include <linux/jbd2.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070030#include <linux/time.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070031#include <linux/fcntl.h>
32#include <linux/stat.h>
33#include <linux/string.h>
34#include <linux/quotaops.h>
35#include <linux/buffer_head.h>
36#include <linux/bio.h>
Christoph Hellwig3dcf5452008-04-29 18:13:32 -040037#include "ext4.h"
38#include "ext4_jbd2.h"
Dave Kleikampac27a0e2006-10-11 01:20:50 -070039
Dave Kleikampac27a0e2006-10-11 01:20:50 -070040#include "xattr.h"
41#include "acl.h"
42
43/*
44 * define how far ahead to read directories while searching them.
45 */
46#define NAMEI_RA_CHUNKS 2
47#define NAMEI_RA_BLOCKS 4
Dave Kleikamp8c55e202007-05-24 13:04:54 -040048#define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070049#define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b))
50
Mingming Cao617ba132006-10-11 01:20:53 -070051static struct buffer_head *ext4_append(handle_t *handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -070052 struct inode *inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -050053 ext4_lblk_t *block, int *err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070054{
55 struct buffer_head *bh;
56
57 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
58
Akinobu Mitaa8716112008-04-17 10:38:59 -040059 bh = ext4_bread(handle, inode, *block, 1, err);
60 if (bh) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -070061 inode->i_size += inode->i_sb->s_blocksize;
Mingming Cao617ba132006-10-11 01:20:53 -070062 EXT4_I(inode)->i_disksize = inode->i_size;
Akinobu Mitaa8716112008-04-17 10:38:59 -040063 *err = ext4_journal_get_write_access(handle, bh);
64 if (*err) {
65 brelse(bh);
66 bh = NULL;
67 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -070068 }
69 return bh;
70}
71
72#ifndef assert
73#define assert(test) J_ASSERT(test)
74#endif
75
Dave Kleikampac27a0e2006-10-11 01:20:50 -070076#ifdef DX_DEBUG
77#define dxtrace(command) command
78#else
79#define dxtrace(command)
80#endif
81
82struct fake_dirent
83{
84 __le32 inode;
85 __le16 rec_len;
86 u8 name_len;
87 u8 file_type;
88};
89
90struct dx_countlimit
91{
92 __le16 limit;
93 __le16 count;
94};
95
96struct dx_entry
97{
98 __le32 hash;
99 __le32 block;
100};
101
102/*
103 * dx_root_info is laid out so that if it should somehow get overlaid by a
104 * dirent the two low bits of the hash version will be zero. Therefore, the
105 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
106 */
107
108struct dx_root
109{
110 struct fake_dirent dot;
111 char dot_name[4];
112 struct fake_dirent dotdot;
113 char dotdot_name[4];
114 struct dx_root_info
115 {
116 __le32 reserved_zero;
117 u8 hash_version;
118 u8 info_length; /* 8 */
119 u8 indirect_levels;
120 u8 unused_flags;
121 }
122 info;
123 struct dx_entry entries[0];
124};
125
126struct dx_node
127{
128 struct fake_dirent fake;
129 struct dx_entry entries[0];
130};
131
132
133struct dx_frame
134{
135 struct buffer_head *bh;
136 struct dx_entry *entries;
137 struct dx_entry *at;
138};
139
140struct dx_map_entry
141{
142 u32 hash;
Eric Sandeenef2b02d2007-09-18 22:46:42 -0700143 u16 offs;
144 u16 size;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700145};
146
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500147static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
148static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400149static inline unsigned dx_get_hash(struct dx_entry *entry);
150static void dx_set_hash(struct dx_entry *entry, unsigned value);
151static unsigned dx_get_count(struct dx_entry *entries);
152static unsigned dx_get_limit(struct dx_entry *entries);
153static void dx_set_count(struct dx_entry *entries, unsigned value);
154static void dx_set_limit(struct dx_entry *entries, unsigned value);
155static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
156static unsigned dx_node_limit(struct inode *dir);
Theodore Ts'of702ba02008-09-22 15:21:01 -0400157static struct dx_frame *dx_probe(const struct qstr *d_name,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700158 struct inode *dir,
159 struct dx_hash_info *hinfo,
160 struct dx_frame *frame,
161 int *err);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400162static void dx_release(struct dx_frame *frames);
Theodore Ts'o8bad4592009-02-14 21:46:54 -0500163static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400164 struct dx_hash_info *hinfo, struct dx_map_entry map[]);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700165static void dx_sort_map(struct dx_map_entry *map, unsigned count);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400166static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500167 struct dx_map_entry *offsets, int count, unsigned blocksize);
Theodore Ts'o8bad4592009-02-14 21:46:54 -0500168static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500169static void dx_insert_block(struct dx_frame *frame,
170 u32 hash, ext4_lblk_t block);
Mingming Cao617ba132006-10-11 01:20:53 -0700171static int ext4_htree_next_block(struct inode *dir, __u32 hash,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700172 struct dx_frame *frame,
173 struct dx_frame *frames,
174 __u32 *start_hash);
Theodore Ts'of702ba02008-09-22 15:21:01 -0400175static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
176 const struct qstr *d_name,
177 struct ext4_dir_entry_2 **res_dir,
178 int *err);
Mingming Cao617ba132006-10-11 01:20:53 -0700179static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700180 struct inode *inode);
181
182/*
Li Zefanf795e142008-07-11 19:27:31 -0400183 * p is at least 6 bytes before the end of page
184 */
185static inline struct ext4_dir_entry_2 *
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500186ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
Li Zefanf795e142008-07-11 19:27:31 -0400187{
188 return (struct ext4_dir_entry_2 *)((char *)p +
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500189 ext4_rec_len_from_disk(p->rec_len, blocksize));
Li Zefanf795e142008-07-11 19:27:31 -0400190}
191
192/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700193 * Future: use high four bits of block for coalesce-on-delete flags
194 * Mask them off for now.
195 */
196
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500197static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700198{
199 return le32_to_cpu(entry->block) & 0x00ffffff;
200}
201
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500202static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700203{
204 entry->block = cpu_to_le32(value);
205}
206
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400207static inline unsigned dx_get_hash(struct dx_entry *entry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700208{
209 return le32_to_cpu(entry->hash);
210}
211
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400212static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700213{
214 entry->hash = cpu_to_le32(value);
215}
216
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400217static inline unsigned dx_get_count(struct dx_entry *entries)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700218{
219 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
220}
221
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400222static inline unsigned dx_get_limit(struct dx_entry *entries)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700223{
224 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
225}
226
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400227static inline void dx_set_count(struct dx_entry *entries, unsigned value)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700228{
229 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
230}
231
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400232static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700233{
234 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
235}
236
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400237static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700238{
Mingming Cao617ba132006-10-11 01:20:53 -0700239 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
240 EXT4_DIR_REC_LEN(2) - infosize;
Li Zefand9c769b2008-07-11 19:27:31 -0400241 return entry_space / sizeof(struct dx_entry);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700242}
243
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400244static inline unsigned dx_node_limit(struct inode *dir)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700245{
Mingming Cao617ba132006-10-11 01:20:53 -0700246 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
Li Zefand9c769b2008-07-11 19:27:31 -0400247 return entry_space / sizeof(struct dx_entry);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700248}
249
250/*
251 * Debug
252 */
253#ifdef DX_DEBUG
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400254static void dx_show_index(char * label, struct dx_entry *entries)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700255{
Andrew Morton63f57932006-10-11 01:21:24 -0700256 int i, n = dx_get_count (entries);
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400257 printk(KERN_DEBUG "%s index ", label);
Andrew Morton63f57932006-10-11 01:21:24 -0700258 for (i = 0; i < n; i++) {
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400259 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500260 0, (unsigned long)dx_get_block(entries + i));
Andrew Morton63f57932006-10-11 01:21:24 -0700261 }
262 printk("\n");
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700263}
264
265struct stats
266{
267 unsigned names;
268 unsigned space;
269 unsigned bcount;
270};
271
Mingming Cao617ba132006-10-11 01:20:53 -0700272static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700273 int size, int show_names)
274{
275 unsigned names = 0, space = 0;
276 char *base = (char *) de;
277 struct dx_hash_info h = *hinfo;
278
279 printk("names: ");
280 while ((char *) de < base + size)
281 {
282 if (de->inode)
283 {
284 if (show_names)
285 {
286 int len = de->name_len;
287 char *name = de->name;
288 while (len--) printk("%c", *name++);
Mingming Cao617ba132006-10-11 01:20:53 -0700289 ext4fs_dirhash(de->name, de->name_len, &h);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700290 printk(":%x.%u ", h.hash,
291 ((char *) de - base));
292 }
Mingming Cao617ba132006-10-11 01:20:53 -0700293 space += EXT4_DIR_REC_LEN(de->name_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700294 names++;
295 }
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500296 de = ext4_next_entry(de, size);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700297 }
298 printk("(%i)\n", names);
299 return (struct stats) { names, space, 1 };
300}
301
302struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
303 struct dx_entry *entries, int levels)
304{
305 unsigned blocksize = dir->i_sb->s_blocksize;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400306 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700307 unsigned bcount = 0;
308 struct buffer_head *bh;
309 int err;
310 printk("%i indexed blocks...\n", count);
311 for (i = 0; i < count; i++, entries++)
312 {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500313 ext4_lblk_t block = dx_get_block(entries);
314 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700315 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
316 struct stats stats;
317 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
Mingming Cao617ba132006-10-11 01:20:53 -0700318 if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700319 stats = levels?
320 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
Mingming Cao617ba132006-10-11 01:20:53 -0700321 dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700322 names += stats.names;
323 space += stats.space;
324 bcount += stats.bcount;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400325 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700326 }
327 if (bcount)
Theodore Ts'o60e66792010-05-17 07:00:00 -0400328 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400329 levels ? "" : " ", names, space/bcount,
330 (space/bcount)*100/blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700331 return (struct stats) { names, space, bcount};
332}
333#endif /* DX_DEBUG */
334
335/*
336 * Probe for a directory leaf block to search.
337 *
338 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
339 * error in the directory index, and the caller should fall back to
340 * searching the directory normally. The callers of dx_probe **MUST**
341 * check for this error code, and make sure it never gets reflected
342 * back to userspace.
343 */
344static struct dx_frame *
Theodore Ts'of702ba02008-09-22 15:21:01 -0400345dx_probe(const struct qstr *d_name, struct inode *dir,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700346 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
347{
348 unsigned count, indirect;
349 struct dx_entry *at, *entries, *p, *q, *m;
350 struct dx_root *root;
351 struct buffer_head *bh;
352 struct dx_frame *frame = frame_in;
353 u32 hash;
354
355 frame->bh = NULL;
Mingming Cao617ba132006-10-11 01:20:53 -0700356 if (!(bh = ext4_bread (NULL,dir, 0, 0, err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700357 goto fail;
358 root = (struct dx_root *) bh->b_data;
359 if (root->info.hash_version != DX_HASH_TEA &&
360 root->info.hash_version != DX_HASH_HALF_MD4 &&
361 root->info.hash_version != DX_HASH_LEGACY) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500362 ext4_warning(dir->i_sb, "Unrecognised inode hash code %d",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700363 root->info.hash_version);
364 brelse(bh);
365 *err = ERR_BAD_DX_DIR;
366 goto fail;
367 }
368 hinfo->hash_version = root->info.hash_version;
Theodore Ts'of99b2582008-10-28 13:21:44 -0400369 if (hinfo->hash_version <= DX_HASH_TEA)
370 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
Mingming Cao617ba132006-10-11 01:20:53 -0700371 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
Theodore Ts'of702ba02008-09-22 15:21:01 -0400372 if (d_name)
373 ext4fs_dirhash(d_name->name, d_name->len, hinfo);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700374 hash = hinfo->hash;
375
376 if (root->info.unused_flags & 1) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500377 ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700378 root->info.unused_flags);
379 brelse(bh);
380 *err = ERR_BAD_DX_DIR;
381 goto fail;
382 }
383
384 if ((indirect = root->info.indirect_levels) > 1) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500385 ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700386 root->info.indirect_levels);
387 brelse(bh);
388 *err = ERR_BAD_DX_DIR;
389 goto fail;
390 }
391
392 entries = (struct dx_entry *) (((char *)&root->info) +
393 root->info.info_length);
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700394
395 if (dx_get_limit(entries) != dx_root_limit(dir,
396 root->info.info_length)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500397 ext4_warning(dir->i_sb, "dx entry: limit != root limit");
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700398 brelse(bh);
399 *err = ERR_BAD_DX_DIR;
400 goto fail;
401 }
402
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400403 dxtrace(printk("Look up %x", hash));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700404 while (1)
405 {
406 count = dx_get_count(entries);
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700407 if (!count || count > dx_get_limit(entries)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500408 ext4_warning(dir->i_sb,
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700409 "dx entry: no count or count > limit");
410 brelse(bh);
411 *err = ERR_BAD_DX_DIR;
412 goto fail2;
413 }
414
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700415 p = entries + 1;
416 q = entries + count - 1;
417 while (p <= q)
418 {
419 m = p + (q - p)/2;
420 dxtrace(printk("."));
421 if (dx_get_hash(m) > hash)
422 q = m - 1;
423 else
424 p = m + 1;
425 }
426
427 if (0) // linear search cross check
428 {
429 unsigned n = count - 1;
430 at = entries;
431 while (n--)
432 {
433 dxtrace(printk(","));
434 if (dx_get_hash(++at) > hash)
435 {
436 at--;
437 break;
438 }
439 }
440 assert (at == p - 1);
441 }
442
443 at = p - 1;
444 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
445 frame->bh = bh;
446 frame->entries = entries;
447 frame->at = at;
448 if (!indirect--) return frame;
Mingming Cao617ba132006-10-11 01:20:53 -0700449 if (!(bh = ext4_bread (NULL,dir, dx_get_block(at), 0, err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700450 goto fail2;
451 at = entries = ((struct dx_node *) bh->b_data)->entries;
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700452 if (dx_get_limit(entries) != dx_node_limit (dir)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500453 ext4_warning(dir->i_sb,
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700454 "dx entry: limit != node limit");
455 brelse(bh);
456 *err = ERR_BAD_DX_DIR;
457 goto fail2;
458 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700459 frame++;
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700460 frame->bh = NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700461 }
462fail2:
463 while (frame >= frame_in) {
464 brelse(frame->bh);
465 frame--;
466 }
467fail:
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700468 if (*err == ERR_BAD_DX_DIR)
Eric Sandeen12062dd2010-02-15 14:19:27 -0500469 ext4_warning(dir->i_sb,
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700470 "Corrupt dir inode %ld, running e2fsck is "
471 "recommended.", dir->i_ino);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700472 return NULL;
473}
474
475static void dx_release (struct dx_frame *frames)
476{
477 if (frames[0].bh == NULL)
478 return;
479
480 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
481 brelse(frames[1].bh);
482 brelse(frames[0].bh);
483}
484
485/*
486 * This function increments the frame pointer to search the next leaf
487 * block, and reads in the necessary intervening nodes if the search
488 * should be necessary. Whether or not the search is necessary is
489 * controlled by the hash parameter. If the hash value is even, then
490 * the search is only continued if the next block starts with that
491 * hash value. This is used if we are searching for a specific file.
492 *
493 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
494 *
495 * This function returns 1 if the caller should continue to search,
496 * or 0 if it should not. If there is an error reading one of the
497 * index blocks, it will a negative error code.
498 *
499 * If start_hash is non-null, it will be filled in with the starting
500 * hash of the next page.
501 */
Mingming Cao617ba132006-10-11 01:20:53 -0700502static int ext4_htree_next_block(struct inode *dir, __u32 hash,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700503 struct dx_frame *frame,
504 struct dx_frame *frames,
505 __u32 *start_hash)
506{
507 struct dx_frame *p;
508 struct buffer_head *bh;
509 int err, num_frames = 0;
510 __u32 bhash;
511
512 p = frame;
513 /*
514 * Find the next leaf page by incrementing the frame pointer.
515 * If we run out of entries in the interior node, loop around and
516 * increment pointer in the parent node. When we break out of
517 * this loop, num_frames indicates the number of interior
518 * nodes need to be read.
519 */
520 while (1) {
521 if (++(p->at) < p->entries + dx_get_count(p->entries))
522 break;
523 if (p == frames)
524 return 0;
525 num_frames++;
526 p--;
527 }
528
529 /*
530 * If the hash is 1, then continue only if the next page has a
531 * continuation hash of any value. This is used for readdir
532 * handling. Otherwise, check to see if the hash matches the
533 * desired contiuation hash. If it doesn't, return since
534 * there's no point to read in the successive index pages.
535 */
536 bhash = dx_get_hash(p->at);
537 if (start_hash)
538 *start_hash = bhash;
539 if ((hash & 1) == 0) {
540 if ((bhash & ~1) != hash)
541 return 0;
542 }
543 /*
544 * If the hash is HASH_NB_ALWAYS, we always go to the next
545 * block so no check is necessary
546 */
547 while (num_frames--) {
Mingming Cao617ba132006-10-11 01:20:53 -0700548 if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at),
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700549 0, &err)))
550 return err; /* Failure */
551 p++;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400552 brelse(p->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700553 p->bh = bh;
554 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
555 }
556 return 1;
557}
558
559
560/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700561 * This function fills a red-black tree with information from a
562 * directory block. It returns the number directory entries loaded
563 * into the tree. If there is an error it is returned in err.
564 */
565static int htree_dirblock_to_tree(struct file *dir_file,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500566 struct inode *dir, ext4_lblk_t block,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700567 struct dx_hash_info *hinfo,
568 __u32 start_hash, __u32 start_minor_hash)
569{
570 struct buffer_head *bh;
Mingming Cao617ba132006-10-11 01:20:53 -0700571 struct ext4_dir_entry_2 *de, *top;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700572 int err, count = 0;
573
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500574 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
575 (unsigned long)block));
Mingming Cao617ba132006-10-11 01:20:53 -0700576 if (!(bh = ext4_bread (NULL, dir, block, 0, &err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700577 return err;
578
Mingming Cao617ba132006-10-11 01:20:53 -0700579 de = (struct ext4_dir_entry_2 *) bh->b_data;
580 top = (struct ext4_dir_entry_2 *) ((char *) de +
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700581 dir->i_sb->s_blocksize -
Mingming Cao617ba132006-10-11 01:20:53 -0700582 EXT4_DIR_REC_LEN(0));
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500583 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
Theodore Ts'o60fd4da2010-07-27 11:54:40 -0400584 if (!ext4_check_dir_entry(dir, de, bh,
Eric Sandeene6c40212006-12-06 20:36:28 -0800585 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
586 +((char *)de - bh->b_data))) {
587 /* On error, skip the f_pos to the next block. */
588 dir_file->f_pos = (dir_file->f_pos |
589 (dir->i_sb->s_blocksize - 1)) + 1;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400590 brelse(bh);
Eric Sandeene6c40212006-12-06 20:36:28 -0800591 return count;
592 }
Mingming Cao617ba132006-10-11 01:20:53 -0700593 ext4fs_dirhash(de->name, de->name_len, hinfo);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700594 if ((hinfo->hash < start_hash) ||
595 ((hinfo->hash == start_hash) &&
596 (hinfo->minor_hash < start_minor_hash)))
597 continue;
598 if (de->inode == 0)
599 continue;
Mingming Cao617ba132006-10-11 01:20:53 -0700600 if ((err = ext4_htree_store_dirent(dir_file,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700601 hinfo->hash, hinfo->minor_hash, de)) != 0) {
602 brelse(bh);
603 return err;
604 }
605 count++;
606 }
607 brelse(bh);
608 return count;
609}
610
611
612/*
613 * This function fills a red-black tree with information from a
614 * directory. We start scanning the directory in hash order, starting
615 * at start_hash and start_minor_hash.
616 *
617 * This function returns the number of entries inserted into the tree,
618 * or a negative error code.
619 */
Mingming Cao617ba132006-10-11 01:20:53 -0700620int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700621 __u32 start_minor_hash, __u32 *next_hash)
622{
623 struct dx_hash_info hinfo;
Mingming Cao617ba132006-10-11 01:20:53 -0700624 struct ext4_dir_entry_2 *de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700625 struct dx_frame frames[2], *frame;
626 struct inode *dir;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500627 ext4_lblk_t block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700628 int count = 0;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500629 int ret, err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700630 __u32 hashval;
631
Theodore Ts'o60e66792010-05-17 07:00:00 -0400632 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400633 start_hash, start_minor_hash));
Josef "Jeff" Sipek9d549892006-12-08 02:36:38 -0800634 dir = dir_file->f_path.dentry->d_inode;
Dmitry Monakhov12e9b892010-05-16 22:00:00 -0400635 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
Mingming Cao617ba132006-10-11 01:20:53 -0700636 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
Theodore Ts'of99b2582008-10-28 13:21:44 -0400637 if (hinfo.hash_version <= DX_HASH_TEA)
638 hinfo.hash_version +=
639 EXT4_SB(dir->i_sb)->s_hash_unsigned;
Mingming Cao617ba132006-10-11 01:20:53 -0700640 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700641 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
642 start_hash, start_minor_hash);
643 *next_hash = ~0;
644 return count;
645 }
646 hinfo.hash = start_hash;
647 hinfo.minor_hash = 0;
Theodore Ts'of702ba02008-09-22 15:21:01 -0400648 frame = dx_probe(NULL, dir, &hinfo, frames, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700649 if (!frame)
650 return err;
651
652 /* Add '.' and '..' from the htree header */
653 if (!start_hash && !start_minor_hash) {
Mingming Cao617ba132006-10-11 01:20:53 -0700654 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
655 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700656 goto errout;
657 count++;
658 }
659 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
Mingming Cao617ba132006-10-11 01:20:53 -0700660 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500661 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
Mingming Cao617ba132006-10-11 01:20:53 -0700662 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700663 goto errout;
664 count++;
665 }
666
667 while (1) {
668 block = dx_get_block(frame->at);
669 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
670 start_hash, start_minor_hash);
671 if (ret < 0) {
672 err = ret;
673 goto errout;
674 }
675 count += ret;
676 hashval = ~0;
Mingming Cao617ba132006-10-11 01:20:53 -0700677 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700678 frame, frames, &hashval);
679 *next_hash = hashval;
680 if (ret < 0) {
681 err = ret;
682 goto errout;
683 }
684 /*
685 * Stop if: (a) there are no more entries, or
686 * (b) we have inserted at least one entry and the
687 * next hash value is not a continuation
688 */
689 if ((ret == 0) ||
690 (count && ((hashval & 1) == 0)))
691 break;
692 }
693 dx_release(frames);
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400694 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
695 "next hash: %x\n", count, *next_hash));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700696 return count;
697errout:
698 dx_release(frames);
699 return (err);
700}
701
702
703/*
704 * Directory block splitting, compacting
705 */
706
Eric Sandeenef2b02d2007-09-18 22:46:42 -0700707/*
708 * Create map of hash values, offsets, and sizes, stored at end of block.
709 * Returns number of entries mapped.
710 */
Theodore Ts'o8bad4592009-02-14 21:46:54 -0500711static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
712 struct dx_hash_info *hinfo,
713 struct dx_map_entry *map_tail)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700714{
715 int count = 0;
716 char *base = (char *) de;
717 struct dx_hash_info h = *hinfo;
718
Theodore Ts'o8bad4592009-02-14 21:46:54 -0500719 while ((char *) de < base + blocksize) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700720 if (de->name_len && de->inode) {
Mingming Cao617ba132006-10-11 01:20:53 -0700721 ext4fs_dirhash(de->name, de->name_len, &h);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700722 map_tail--;
723 map_tail->hash = h.hash;
Toshiyuki Okajima9aee2282009-06-08 12:41:35 -0400724 map_tail->offs = ((char *) de - base)>>2;
Eric Sandeenef2b02d2007-09-18 22:46:42 -0700725 map_tail->size = le16_to_cpu(de->rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700726 count++;
727 cond_resched();
728 }
729 /* XXX: do we need to check rec_len == 0 case? -Chris */
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500730 de = ext4_next_entry(de, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700731 }
732 return count;
733}
734
Eric Sandeenef2b02d2007-09-18 22:46:42 -0700735/* Sort map by hash value */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700736static void dx_sort_map (struct dx_map_entry *map, unsigned count)
737{
Andrew Morton63f57932006-10-11 01:21:24 -0700738 struct dx_map_entry *p, *q, *top = map + count - 1;
739 int more;
740 /* Combsort until bubble sort doesn't suck */
741 while (count > 2) {
742 count = count*10/13;
743 if (count - 9 < 2) /* 9, 10 -> 11 */
744 count = 11;
745 for (p = top, q = p - count; q >= map; p--, q--)
746 if (p->hash < q->hash)
747 swap(*p, *q);
748 }
749 /* Garden variety bubble sort */
750 do {
751 more = 0;
752 q = top;
753 while (q-- > map) {
754 if (q[1].hash >= q[0].hash)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700755 continue;
Andrew Morton63f57932006-10-11 01:21:24 -0700756 swap(*(q+1), *q);
757 more = 1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700758 }
759 } while(more);
760}
761
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500762static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700763{
764 struct dx_entry *entries = frame->entries;
765 struct dx_entry *old = frame->at, *new = old + 1;
766 int count = dx_get_count(entries);
767
768 assert(count < dx_get_limit(entries));
769 assert(old < entries + count);
770 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
771 dx_set_hash(new, hash);
772 dx_set_block(new, block);
773 dx_set_count(entries, count + 1);
774}
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700775
Mingming Cao617ba132006-10-11 01:20:53 -0700776static void ext4_update_dx_flag(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700777{
Mingming Cao617ba132006-10-11 01:20:53 -0700778 if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
779 EXT4_FEATURE_COMPAT_DIR_INDEX))
Dmitry Monakhov12e9b892010-05-16 22:00:00 -0400780 ext4_clear_inode_flag(inode, EXT4_INODE_INDEX);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700781}
782
783/*
Mingming Cao617ba132006-10-11 01:20:53 -0700784 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700785 *
Mingming Cao617ba132006-10-11 01:20:53 -0700786 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700787 * `de != NULL' is guaranteed by caller.
788 */
Mingming Cao617ba132006-10-11 01:20:53 -0700789static inline int ext4_match (int len, const char * const name,
790 struct ext4_dir_entry_2 * de)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700791{
792 if (len != de->name_len)
793 return 0;
794 if (!de->inode)
795 return 0;
796 return !memcmp(name, de->name, len);
797}
798
799/*
800 * Returns 0 if not found, -1 on failure, and 1 on success
801 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400802static inline int search_dirblock(struct buffer_head *bh,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700803 struct inode *dir,
Theodore Ts'of702ba02008-09-22 15:21:01 -0400804 const struct qstr *d_name,
Theodore Ts'o498e5f22008-11-05 00:14:04 -0500805 unsigned int offset,
Mingming Cao617ba132006-10-11 01:20:53 -0700806 struct ext4_dir_entry_2 ** res_dir)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700807{
Mingming Cao617ba132006-10-11 01:20:53 -0700808 struct ext4_dir_entry_2 * de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700809 char * dlimit;
810 int de_len;
Theodore Ts'of702ba02008-09-22 15:21:01 -0400811 const char *name = d_name->name;
812 int namelen = d_name->len;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700813
Mingming Cao617ba132006-10-11 01:20:53 -0700814 de = (struct ext4_dir_entry_2 *) bh->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700815 dlimit = bh->b_data + dir->i_sb->s_blocksize;
816 while ((char *) de < dlimit) {
817 /* this code is executed quadratically often */
818 /* do minimal checking `by hand' */
819
820 if ((char *) de + namelen <= dlimit &&
Mingming Cao617ba132006-10-11 01:20:53 -0700821 ext4_match (namelen, name, de)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700822 /* found a match - just to be sure, do a full check */
Theodore Ts'o60fd4da2010-07-27 11:54:40 -0400823 if (!ext4_check_dir_entry(dir, de, bh, offset))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700824 return -1;
825 *res_dir = de;
826 return 1;
827 }
828 /* prevent looping on a bad block */
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500829 de_len = ext4_rec_len_from_disk(de->rec_len,
830 dir->i_sb->s_blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700831 if (de_len <= 0)
832 return -1;
833 offset += de_len;
Mingming Cao617ba132006-10-11 01:20:53 -0700834 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700835 }
836 return 0;
837}
838
839
840/*
Mingming Cao617ba132006-10-11 01:20:53 -0700841 * ext4_find_entry()
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700842 *
843 * finds an entry in the specified directory with the wanted name. It
844 * returns the cache buffer in which the entry was found, and the entry
845 * itself (as a parameter - res_dir). It does NOT read the inode of the
846 * entry - you'll have to do that yourself if you want to.
847 *
848 * The returned buffer_head has ->b_count elevated. The caller is expected
849 * to brelse() it when appropriate.
850 */
Theodore Ts'of702ba02008-09-22 15:21:01 -0400851static struct buffer_head * ext4_find_entry (struct inode *dir,
852 const struct qstr *d_name,
Mingming Cao617ba132006-10-11 01:20:53 -0700853 struct ext4_dir_entry_2 ** res_dir)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700854{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400855 struct super_block *sb;
856 struct buffer_head *bh_use[NAMEI_RA_SIZE];
857 struct buffer_head *bh, *ret = NULL;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500858 ext4_lblk_t start, block, b;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700859 int ra_max = 0; /* Number of bh's in the readahead
860 buffer, bh_use[] */
861 int ra_ptr = 0; /* Current index into readahead
862 buffer */
863 int num = 0;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500864 ext4_lblk_t nblocks;
865 int i, err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700866 int namelen;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700867
868 *res_dir = NULL;
869 sb = dir->i_sb;
Theodore Ts'of702ba02008-09-22 15:21:01 -0400870 namelen = d_name->len;
Mingming Cao617ba132006-10-11 01:20:53 -0700871 if (namelen > EXT4_NAME_LEN)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700872 return NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700873 if (is_dx(dir)) {
Theodore Ts'of702ba02008-09-22 15:21:01 -0400874 bh = ext4_dx_find_entry(dir, d_name, res_dir, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700875 /*
876 * On success, or if the error was file not found,
877 * return. Otherwise, fall back to doing a search the
878 * old fashioned way.
879 */
880 if (bh || (err != ERR_BAD_DX_DIR))
881 return bh;
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400882 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
883 "falling back\n"));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700884 }
Mingming Cao617ba132006-10-11 01:20:53 -0700885 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
886 start = EXT4_I(dir)->i_dir_start_lookup;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700887 if (start >= nblocks)
888 start = 0;
889 block = start;
890restart:
891 do {
892 /*
893 * We deal with the read-ahead logic here.
894 */
895 if (ra_ptr >= ra_max) {
896 /* Refill the readahead buffer */
897 ra_ptr = 0;
898 b = block;
899 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
900 /*
901 * Terminate if we reach the end of the
902 * directory and must wrap, or if our
903 * search has finished at this block.
904 */
905 if (b >= nblocks || (num && block == start)) {
906 bh_use[ra_max] = NULL;
907 break;
908 }
909 num++;
Mingming Cao617ba132006-10-11 01:20:53 -0700910 bh = ext4_getblk(NULL, dir, b++, 0, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700911 bh_use[ra_max] = bh;
912 if (bh)
913 ll_rw_block(READ_META, 1, &bh);
914 }
915 }
916 if ((bh = bh_use[ra_ptr++]) == NULL)
917 goto next;
918 wait_on_buffer(bh);
919 if (!buffer_uptodate(bh)) {
920 /* read error, skip block & hope for the best */
Theodore Ts'o24676da2010-05-16 21:00:00 -0400921 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
922 (unsigned long) block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700923 brelse(bh);
924 goto next;
925 }
Theodore Ts'of702ba02008-09-22 15:21:01 -0400926 i = search_dirblock(bh, dir, d_name,
Mingming Cao617ba132006-10-11 01:20:53 -0700927 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700928 if (i == 1) {
Mingming Cao617ba132006-10-11 01:20:53 -0700929 EXT4_I(dir)->i_dir_start_lookup = block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700930 ret = bh;
931 goto cleanup_and_exit;
932 } else {
933 brelse(bh);
934 if (i < 0)
935 goto cleanup_and_exit;
936 }
937 next:
938 if (++block >= nblocks)
939 block = 0;
940 } while (block != start);
941
942 /*
943 * If the directory has grown while we were searching, then
944 * search the last part of the directory before giving up.
945 */
946 block = nblocks;
Mingming Cao617ba132006-10-11 01:20:53 -0700947 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700948 if (block < nblocks) {
949 start = 0;
950 goto restart;
951 }
952
953cleanup_and_exit:
954 /* Clean up the read-ahead blocks */
955 for (; ra_ptr < ra_max; ra_ptr++)
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400956 brelse(bh_use[ra_ptr]);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700957 return ret;
958}
959
Theodore Ts'of702ba02008-09-22 15:21:01 -0400960static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
Mingming Cao617ba132006-10-11 01:20:53 -0700961 struct ext4_dir_entry_2 **res_dir, int *err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700962{
963 struct super_block * sb;
964 struct dx_hash_info hinfo;
965 u32 hash;
966 struct dx_frame frames[2], *frame;
Mingming Cao617ba132006-10-11 01:20:53 -0700967 struct ext4_dir_entry_2 *de, *top;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700968 struct buffer_head *bh;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500969 ext4_lblk_t block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700970 int retval;
Theodore Ts'of702ba02008-09-22 15:21:01 -0400971 int namelen = d_name->len;
972 const u8 *name = d_name->name;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700973
974 sb = dir->i_sb;
975 /* NFS may look up ".." - look at dx_root directory block */
976 if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
Theodore Ts'of702ba02008-09-22 15:21:01 -0400977 if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700978 return NULL;
979 } else {
980 frame = frames;
981 frame->bh = NULL; /* for dx_release() */
982 frame->at = (struct dx_entry *)frames; /* hack for zero entry*/
983 dx_set_block(frame->at, 0); /* dx_root block is 0 */
984 }
985 hash = hinfo.hash;
986 do {
987 block = dx_get_block(frame->at);
Mingming Cao617ba132006-10-11 01:20:53 -0700988 if (!(bh = ext4_bread (NULL,dir, block, 0, err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700989 goto errout;
Mingming Cao617ba132006-10-11 01:20:53 -0700990 de = (struct ext4_dir_entry_2 *) bh->b_data;
991 top = (struct ext4_dir_entry_2 *) ((char *) de + sb->s_blocksize -
992 EXT4_DIR_REC_LEN(0));
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500993 for (; de < top; de = ext4_next_entry(de, sb->s_blocksize)) {
Duane Griffinf3b35f02008-07-11 19:27:31 -0400994 int off = (block << EXT4_BLOCK_SIZE_BITS(sb))
995 + ((char *) de - bh->b_data);
996
Theodore Ts'o60fd4da2010-07-27 11:54:40 -0400997 if (!ext4_check_dir_entry(dir, de, bh, off)) {
Duane Griffinf3b35f02008-07-11 19:27:31 -0400998 brelse(bh);
Dmitriy Monakhovfedee542007-05-08 00:25:34 -0700999 *err = ERR_BAD_DX_DIR;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001000 goto errout;
1001 }
Duane Griffinf3b35f02008-07-11 19:27:31 -04001002
1003 if (ext4_match(namelen, name, de)) {
1004 *res_dir = de;
1005 dx_release(frames);
1006 return bh;
1007 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001008 }
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001009 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001010 /* Check to see if we should continue to search */
Mingming Cao617ba132006-10-11 01:20:53 -07001011 retval = ext4_htree_next_block(dir, hash, frame,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001012 frames, NULL);
1013 if (retval < 0) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001014 ext4_warning(sb,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001015 "error reading index page in directory #%lu",
1016 dir->i_ino);
1017 *err = retval;
1018 goto errout;
1019 }
1020 } while (retval == 1);
1021
1022 *err = -ENOENT;
1023errout:
Theodore Ts'o4776004f2008-09-08 23:00:52 -04001024 dxtrace(printk(KERN_DEBUG "%s not found\n", name));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001025 dx_release (frames);
1026 return NULL;
1027}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001028
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001029static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001030{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001031 struct inode *inode;
1032 struct ext4_dir_entry_2 *de;
1033 struct buffer_head *bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001034
Mingming Cao617ba132006-10-11 01:20:53 -07001035 if (dentry->d_name.len > EXT4_NAME_LEN)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001036 return ERR_PTR(-ENAMETOOLONG);
1037
Theodore Ts'of702ba02008-09-22 15:21:01 -04001038 bh = ext4_find_entry(dir, &dentry->d_name, &de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001039 inode = NULL;
1040 if (bh) {
Theodore Ts'o498e5f22008-11-05 00:14:04 -05001041 __u32 ino = le32_to_cpu(de->inode);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001042 brelse(bh);
Mingming Cao617ba132006-10-11 01:20:53 -07001043 if (!ext4_valid_inum(dir->i_sb, ino)) {
Theodore Ts'o24676da2010-05-16 21:00:00 -04001044 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
David Howells1d1fe1e2008-02-07 00:15:37 -08001045 return ERR_PTR(-EIO);
Vasily Averina6c15c22007-07-15 23:40:46 -07001046 }
David Howells1d1fe1e2008-02-07 00:15:37 -08001047 inode = ext4_iget(dir->i_sb, ino);
Bryan Donlane6f009b2009-02-22 21:20:25 -05001048 if (unlikely(IS_ERR(inode))) {
1049 if (PTR_ERR(inode) == -ESTALE) {
Theodore Ts'o24676da2010-05-16 21:00:00 -04001050 EXT4_ERROR_INODE(dir,
1051 "deleted inode referenced: %u",
1052 ino);
Bryan Donlane6f009b2009-02-22 21:20:25 -05001053 return ERR_PTR(-EIO);
1054 } else {
1055 return ERR_CAST(inode);
1056 }
1057 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001058 }
1059 return d_splice_alias(inode, dentry);
1060}
1061
1062
Mingming Cao617ba132006-10-11 01:20:53 -07001063struct dentry *ext4_get_parent(struct dentry *child)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001064{
Theodore Ts'o498e5f22008-11-05 00:14:04 -05001065 __u32 ino;
Theodore Ts'of702ba02008-09-22 15:21:01 -04001066 static const struct qstr dotdot = {
1067 .name = "..",
1068 .len = 2,
1069 };
Mingming Cao617ba132006-10-11 01:20:53 -07001070 struct ext4_dir_entry_2 * de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001071 struct buffer_head *bh;
1072
Theodore Ts'of702ba02008-09-22 15:21:01 -04001073 bh = ext4_find_entry(child->d_inode, &dotdot, &de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001074 if (!bh)
1075 return ERR_PTR(-ENOENT);
1076 ino = le32_to_cpu(de->inode);
1077 brelse(bh);
1078
Mingming Cao617ba132006-10-11 01:20:53 -07001079 if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
Theodore Ts'o24676da2010-05-16 21:00:00 -04001080 EXT4_ERROR_INODE(child->d_inode,
1081 "bad parent inode number: %u", ino);
David Howells1d1fe1e2008-02-07 00:15:37 -08001082 return ERR_PTR(-EIO);
Vasily Averina6c15c22007-07-15 23:40:46 -07001083 }
1084
Christoph Hellwig44003722008-08-11 15:49:04 +02001085 return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001086}
1087
1088#define S_SHIFT 12
Mingming Cao617ba132006-10-11 01:20:53 -07001089static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1090 [S_IFREG >> S_SHIFT] = EXT4_FT_REG_FILE,
1091 [S_IFDIR >> S_SHIFT] = EXT4_FT_DIR,
1092 [S_IFCHR >> S_SHIFT] = EXT4_FT_CHRDEV,
1093 [S_IFBLK >> S_SHIFT] = EXT4_FT_BLKDEV,
1094 [S_IFIFO >> S_SHIFT] = EXT4_FT_FIFO,
1095 [S_IFSOCK >> S_SHIFT] = EXT4_FT_SOCK,
1096 [S_IFLNK >> S_SHIFT] = EXT4_FT_SYMLINK,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001097};
1098
Mingming Cao617ba132006-10-11 01:20:53 -07001099static inline void ext4_set_de_type(struct super_block *sb,
1100 struct ext4_dir_entry_2 *de,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001101 umode_t mode) {
Mingming Cao617ba132006-10-11 01:20:53 -07001102 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1103 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001104}
1105
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001106/*
1107 * Move count entries from end of map between two memory locations.
1108 * Returns pointer to last entry moved.
1109 */
Mingming Cao617ba132006-10-11 01:20:53 -07001110static struct ext4_dir_entry_2 *
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001111dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1112 unsigned blocksize)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001113{
1114 unsigned rec_len = 0;
1115
1116 while (count--) {
Theodore Ts'o60e66792010-05-17 07:00:00 -04001117 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
Toshiyuki Okajima9aee2282009-06-08 12:41:35 -04001118 (from + (map->offs<<2));
Mingming Cao617ba132006-10-11 01:20:53 -07001119 rec_len = EXT4_DIR_REC_LEN(de->name_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001120 memcpy (to, de, rec_len);
Mingming Cao617ba132006-10-11 01:20:53 -07001121 ((struct ext4_dir_entry_2 *) to)->rec_len =
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001122 ext4_rec_len_to_disk(rec_len, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001123 de->inode = 0;
1124 map++;
1125 to += rec_len;
1126 }
Mingming Cao617ba132006-10-11 01:20:53 -07001127 return (struct ext4_dir_entry_2 *) (to - rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001128}
1129
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001130/*
1131 * Compact each dir entry in the range to the minimal rec_len.
1132 * Returns pointer to last entry in range.
1133 */
Theodore Ts'o8bad4592009-02-14 21:46:54 -05001134static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001135{
Mingming Cao617ba132006-10-11 01:20:53 -07001136 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001137 unsigned rec_len = 0;
1138
1139 prev = to = de;
Theodore Ts'o8bad4592009-02-14 21:46:54 -05001140 while ((char*)de < base + blocksize) {
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001141 next = ext4_next_entry(de, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001142 if (de->inode && de->name_len) {
Mingming Cao617ba132006-10-11 01:20:53 -07001143 rec_len = EXT4_DIR_REC_LEN(de->name_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001144 if (de > to)
1145 memmove(to, de, rec_len);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001146 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001147 prev = to;
Mingming Cao617ba132006-10-11 01:20:53 -07001148 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001149 }
1150 de = next;
1151 }
1152 return prev;
1153}
1154
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001155/*
1156 * Split a full leaf block to make room for a new dir entry.
1157 * Allocate a new block, and move entries so that they are approx. equally full.
1158 * Returns pointer to de in block into which the new entry will be inserted.
1159 */
Mingming Cao617ba132006-10-11 01:20:53 -07001160static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001161 struct buffer_head **bh,struct dx_frame *frame,
1162 struct dx_hash_info *hinfo, int *error)
1163{
1164 unsigned blocksize = dir->i_sb->s_blocksize;
1165 unsigned count, continued;
1166 struct buffer_head *bh2;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001167 ext4_lblk_t newblock;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001168 u32 hash2;
1169 struct dx_map_entry *map;
1170 char *data1 = (*bh)->b_data, *data2;
Theodore Ts'o59e315b2008-12-06 16:58:39 -05001171 unsigned split, move, size;
Mingming Cao617ba132006-10-11 01:20:53 -07001172 struct ext4_dir_entry_2 *de = NULL, *de2;
Theodore Ts'o59e315b2008-12-06 16:58:39 -05001173 int err = 0, i;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001174
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001175 bh2 = ext4_append (handle, dir, &newblock, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001176 if (!(bh2)) {
1177 brelse(*bh);
1178 *bh = NULL;
1179 goto errout;
1180 }
1181
1182 BUFFER_TRACE(*bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001183 err = ext4_journal_get_write_access(handle, *bh);
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001184 if (err)
1185 goto journal_error;
1186
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001187 BUFFER_TRACE(frame->bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001188 err = ext4_journal_get_write_access(handle, frame->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001189 if (err)
1190 goto journal_error;
1191
1192 data2 = bh2->b_data;
1193
1194 /* create map in the end of data2 block */
1195 map = (struct dx_map_entry *) (data2 + blocksize);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001196 count = dx_make_map((struct ext4_dir_entry_2 *) data1,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001197 blocksize, hinfo, map);
1198 map -= count;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001199 dx_sort_map(map, count);
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001200 /* Split the existing block in the middle, size-wise */
1201 size = 0;
1202 move = 0;
1203 for (i = count-1; i >= 0; i--) {
1204 /* is more than half of this entry in 2nd half of the block? */
1205 if (size + map[i].size/2 > blocksize/2)
1206 break;
1207 size += map[i].size;
1208 move++;
1209 }
1210 /* map index at which we will split */
1211 split = count - move;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001212 hash2 = map[split].hash;
1213 continued = hash2 == map[split - 1].hash;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001214 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1215 (unsigned long)dx_get_block(frame->at),
1216 hash2, split, count-split));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001217
1218 /* Fancy dance to stay within two buffers */
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001219 de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001220 de = dx_pack_dirents(data1, blocksize);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001221 de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1222 blocksize);
1223 de2->rec_len = ext4_rec_len_to_disk(data2 + blocksize - (char *) de2,
1224 blocksize);
Mingming Cao617ba132006-10-11 01:20:53 -07001225 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1226 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001227
1228 /* Which block gets the new entry? */
1229 if (hinfo->hash >= hash2)
1230 {
1231 swap(*bh, bh2);
1232 de = de2;
1233 }
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001234 dx_insert_block(frame, hash2 + continued, newblock);
Frank Mayhar03901312009-01-07 00:06:22 -05001235 err = ext4_handle_dirty_metadata(handle, dir, bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001236 if (err)
1237 goto journal_error;
Frank Mayhar03901312009-01-07 00:06:22 -05001238 err = ext4_handle_dirty_metadata(handle, dir, frame->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001239 if (err)
1240 goto journal_error;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001241 brelse(bh2);
1242 dxtrace(dx_show_index("frame", frame->entries));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001243 return de;
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001244
1245journal_error:
1246 brelse(*bh);
1247 brelse(bh2);
1248 *bh = NULL;
1249 ext4_std_error(dir->i_sb, err);
1250errout:
1251 *error = err;
1252 return NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001253}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001254
1255/*
1256 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1257 * it points to a directory entry which is guaranteed to be large
1258 * enough for new directory entry. If de is NULL, then
1259 * add_dirent_to_buf will attempt search the directory block for
1260 * space. It will return -ENOSPC if no space is available, and -EIO
1261 * and -EEXIST if directory entry already exists.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001262 */
1263static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
Mingming Cao617ba132006-10-11 01:20:53 -07001264 struct inode *inode, struct ext4_dir_entry_2 *de,
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001265 struct buffer_head *bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001266{
1267 struct inode *dir = dentry->d_parent->d_inode;
1268 const char *name = dentry->d_name.name;
1269 int namelen = dentry->d_name.len;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05001270 unsigned int offset = 0;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001271 unsigned int blocksize = dir->i_sb->s_blocksize;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001272 unsigned short reclen;
1273 int nlen, rlen, err;
1274 char *top;
1275
Mingming Cao617ba132006-10-11 01:20:53 -07001276 reclen = EXT4_DIR_REC_LEN(namelen);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001277 if (!de) {
Mingming Cao617ba132006-10-11 01:20:53 -07001278 de = (struct ext4_dir_entry_2 *)bh->b_data;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001279 top = bh->b_data + blocksize - reclen;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001280 while ((char *) de <= top) {
Theodore Ts'o60fd4da2010-07-27 11:54:40 -04001281 if (!ext4_check_dir_entry(dir, de, bh, offset))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001282 return -EIO;
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001283 if (ext4_match(namelen, name, de))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001284 return -EEXIST;
Mingming Cao617ba132006-10-11 01:20:53 -07001285 nlen = EXT4_DIR_REC_LEN(de->name_len);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001286 rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001287 if ((de->inode? rlen - nlen: rlen) >= reclen)
1288 break;
Mingming Cao617ba132006-10-11 01:20:53 -07001289 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001290 offset += rlen;
1291 }
1292 if ((char *) de > top)
1293 return -ENOSPC;
1294 }
1295 BUFFER_TRACE(bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001296 err = ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001297 if (err) {
Mingming Cao617ba132006-10-11 01:20:53 -07001298 ext4_std_error(dir->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001299 return err;
1300 }
1301
1302 /* By now the buffer is marked for journaling */
Mingming Cao617ba132006-10-11 01:20:53 -07001303 nlen = EXT4_DIR_REC_LEN(de->name_len);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001304 rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001305 if (de->inode) {
Mingming Cao617ba132006-10-11 01:20:53 -07001306 struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001307 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, blocksize);
1308 de->rec_len = ext4_rec_len_to_disk(nlen, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001309 de = de1;
1310 }
Mingming Cao617ba132006-10-11 01:20:53 -07001311 de->file_type = EXT4_FT_UNKNOWN;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001312 if (inode) {
1313 de->inode = cpu_to_le32(inode->i_ino);
Mingming Cao617ba132006-10-11 01:20:53 -07001314 ext4_set_de_type(dir->i_sb, de, inode->i_mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001315 } else
1316 de->inode = 0;
1317 de->name_len = namelen;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001318 memcpy(de->name, name, namelen);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001319 /*
1320 * XXX shouldn't update any times until successful
1321 * completion of syscall, but too many callers depend
1322 * on this.
1323 *
1324 * XXX similarly, too many callers depend on
Mingming Cao617ba132006-10-11 01:20:53 -07001325 * ext4_new_inode() setting the times, but error
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001326 * recovery deletes the inode, so the worst that can
1327 * happen is that the times are slightly out of date
1328 * and/or different from the directory change time.
1329 */
Kalpak Shahef7f3832007-07-18 09:15:20 -04001330 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
Mingming Cao617ba132006-10-11 01:20:53 -07001331 ext4_update_dx_flag(dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001332 dir->i_version++;
Mingming Cao617ba132006-10-11 01:20:53 -07001333 ext4_mark_inode_dirty(handle, dir);
Frank Mayhar03901312009-01-07 00:06:22 -05001334 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1335 err = ext4_handle_dirty_metadata(handle, dir, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001336 if (err)
Mingming Cao617ba132006-10-11 01:20:53 -07001337 ext4_std_error(dir->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001338 return 0;
1339}
1340
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001341/*
1342 * This converts a one block unindexed directory to a 3 block indexed
1343 * directory, and adds the dentry to the indexed directory.
1344 */
1345static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1346 struct inode *inode, struct buffer_head *bh)
1347{
1348 struct inode *dir = dentry->d_parent->d_inode;
1349 const char *name = dentry->d_name.name;
1350 int namelen = dentry->d_name.len;
1351 struct buffer_head *bh2;
1352 struct dx_root *root;
1353 struct dx_frame frames[2], *frame;
1354 struct dx_entry *entries;
Mingming Cao617ba132006-10-11 01:20:53 -07001355 struct ext4_dir_entry_2 *de, *de2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001356 char *data1, *top;
1357 unsigned len;
1358 int retval;
1359 unsigned blocksize;
1360 struct dx_hash_info hinfo;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001361 ext4_lblk_t block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001362 struct fake_dirent *fde;
1363
1364 blocksize = dir->i_sb->s_blocksize;
Theodore Ts'oe6b8bc02009-01-16 11:13:40 -05001365 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
Mingming Cao617ba132006-10-11 01:20:53 -07001366 retval = ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001367 if (retval) {
Mingming Cao617ba132006-10-11 01:20:53 -07001368 ext4_std_error(dir->i_sb, retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001369 brelse(bh);
1370 return retval;
1371 }
1372 root = (struct dx_root *) bh->b_data;
1373
Theodore Ts'oe6b8bc02009-01-16 11:13:40 -05001374 /* The 0th block becomes the root, move the dirents out */
1375 fde = &root->dotdot;
1376 de = (struct ext4_dir_entry_2 *)((char *)fde +
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001377 ext4_rec_len_from_disk(fde->rec_len, blocksize));
Theodore Ts'oe6b8bc02009-01-16 11:13:40 -05001378 if ((char *) de >= (((char *) root) + blocksize)) {
Theodore Ts'o24676da2010-05-16 21:00:00 -04001379 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
Theodore Ts'oe6b8bc02009-01-16 11:13:40 -05001380 brelse(bh);
1381 return -EIO;
1382 }
1383 len = ((char *) root) + blocksize - (char *) de;
1384
1385 /* Allocate new block for the 0th block's dirents */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001386 bh2 = ext4_append(handle, dir, &block, &retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001387 if (!(bh2)) {
1388 brelse(bh);
1389 return retval;
1390 }
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04001391 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001392 data1 = bh2->b_data;
1393
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001394 memcpy (data1, de, len);
Mingming Cao617ba132006-10-11 01:20:53 -07001395 de = (struct ext4_dir_entry_2 *) data1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001396 top = data1 + len;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001397 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001398 de = de2;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001399 de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1400 blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001401 /* Initialize the root; the dot dirents already exist */
Mingming Cao617ba132006-10-11 01:20:53 -07001402 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001403 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1404 blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001405 memset (&root->info, 0, sizeof(root->info));
1406 root->info.info_length = sizeof(root->info);
Mingming Cao617ba132006-10-11 01:20:53 -07001407 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001408 entries = root->entries;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001409 dx_set_block(entries, 1);
1410 dx_set_count(entries, 1);
1411 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001412
1413 /* Initialize as for dx_probe */
1414 hinfo.hash_version = root->info.hash_version;
Theodore Ts'of99b2582008-10-28 13:21:44 -04001415 if (hinfo.hash_version <= DX_HASH_TEA)
1416 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
Mingming Cao617ba132006-10-11 01:20:53 -07001417 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1418 ext4fs_dirhash(name, namelen, &hinfo);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001419 frame = frames;
1420 frame->entries = entries;
1421 frame->at = entries;
1422 frame->bh = bh;
1423 bh = bh2;
1424 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1425 dx_release (frames);
1426 if (!(de))
1427 return retval;
1428
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001429 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1430 brelse(bh);
1431 return retval;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001432}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001433
1434/*
Mingming Cao617ba132006-10-11 01:20:53 -07001435 * ext4_add_entry()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001436 *
1437 * adds a file entry to the specified directory, using the same
Mingming Cao617ba132006-10-11 01:20:53 -07001438 * semantics as ext4_find_entry(). It returns NULL if it failed.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001439 *
1440 * NOTE!! The inode part of 'de' is left at 0 - which means you
1441 * may not sleep between calling this and putting something into
1442 * the entry, as someone else might have used it while you slept.
1443 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001444static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1445 struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001446{
1447 struct inode *dir = dentry->d_parent->d_inode;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001448 struct buffer_head *bh;
Mingming Cao617ba132006-10-11 01:20:53 -07001449 struct ext4_dir_entry_2 *de;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001450 struct super_block *sb;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001451 int retval;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001452 int dx_fallback=0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001453 unsigned blocksize;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001454 ext4_lblk_t block, blocks;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001455
1456 sb = dir->i_sb;
1457 blocksize = sb->s_blocksize;
1458 if (!dentry->d_name.len)
1459 return -EINVAL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001460 if (is_dx(dir)) {
Mingming Cao617ba132006-10-11 01:20:53 -07001461 retval = ext4_dx_add_entry(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001462 if (!retval || (retval != ERR_BAD_DX_DIR))
1463 return retval;
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04001464 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001465 dx_fallback++;
Mingming Cao617ba132006-10-11 01:20:53 -07001466 ext4_mark_inode_dirty(handle, dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001467 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001468 blocks = dir->i_size >> sb->s_blocksize_bits;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05001469 for (block = 0; block < blocks; block++) {
Mingming Cao617ba132006-10-11 01:20:53 -07001470 bh = ext4_bread(handle, dir, block, 0, &retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001471 if(!bh)
1472 return retval;
1473 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001474 if (retval != -ENOSPC) {
1475 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001476 return retval;
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001477 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001478
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001479 if (blocks == 1 && !dx_fallback &&
Theodore Ts'o1e424a32009-11-08 15:45:44 -05001480 EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1481 return make_indexed_dir(handle, dentry, inode, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001482 brelse(bh);
1483 }
Mingming Cao617ba132006-10-11 01:20:53 -07001484 bh = ext4_append(handle, dir, &block, &retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001485 if (!bh)
1486 return retval;
Mingming Cao617ba132006-10-11 01:20:53 -07001487 de = (struct ext4_dir_entry_2 *) bh->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001488 de->inode = 0;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001489 de->rec_len = ext4_rec_len_to_disk(blocksize, blocksize);
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001490 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1491 brelse(bh);
Frank Mayhar14ece102010-05-17 08:00:00 -04001492 if (retval == 0)
1493 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001494 return retval;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001495}
1496
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001497/*
1498 * Returns 0 for success, or a negative error value
1499 */
Mingming Cao617ba132006-10-11 01:20:53 -07001500static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001501 struct inode *inode)
1502{
1503 struct dx_frame frames[2], *frame;
1504 struct dx_entry *entries, *at;
1505 struct dx_hash_info hinfo;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001506 struct buffer_head *bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001507 struct inode *dir = dentry->d_parent->d_inode;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001508 struct super_block *sb = dir->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -07001509 struct ext4_dir_entry_2 *de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001510 int err;
1511
Theodore Ts'of702ba02008-09-22 15:21:01 -04001512 frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001513 if (!frame)
1514 return err;
1515 entries = frame->entries;
1516 at = frame->at;
1517
Mingming Cao617ba132006-10-11 01:20:53 -07001518 if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001519 goto cleanup;
1520
1521 BUFFER_TRACE(bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001522 err = ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001523 if (err)
1524 goto journal_error;
1525
1526 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001527 if (err != -ENOSPC)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001528 goto cleanup;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001529
1530 /* Block full, should compress but for now just split */
Theodore Ts'o4776004f2008-09-08 23:00:52 -04001531 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001532 dx_get_count(entries), dx_get_limit(entries)));
1533 /* Need to split index? */
1534 if (dx_get_count(entries) == dx_get_limit(entries)) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001535 ext4_lblk_t newblock;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001536 unsigned icount = dx_get_count(entries);
1537 int levels = frame - frames;
1538 struct dx_entry *entries2;
1539 struct dx_node *node2;
1540 struct buffer_head *bh2;
1541
1542 if (levels && (dx_get_count(frames->entries) ==
1543 dx_get_limit(frames->entries))) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001544 ext4_warning(sb, "Directory index full!");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001545 err = -ENOSPC;
1546 goto cleanup;
1547 }
Mingming Cao617ba132006-10-11 01:20:53 -07001548 bh2 = ext4_append (handle, dir, &newblock, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001549 if (!(bh2))
1550 goto cleanup;
1551 node2 = (struct dx_node *)(bh2->b_data);
1552 entries2 = node2->entries;
Andreas Schlick1f7bebb2009-09-10 23:16:07 -04001553 memset(&node2->fake, 0, sizeof(struct fake_dirent));
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001554 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
1555 sb->s_blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001556 BUFFER_TRACE(frame->bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001557 err = ext4_journal_get_write_access(handle, frame->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001558 if (err)
1559 goto journal_error;
1560 if (levels) {
1561 unsigned icount1 = icount/2, icount2 = icount - icount1;
1562 unsigned hash2 = dx_get_hash(entries + icount1);
Theodore Ts'o4776004f2008-09-08 23:00:52 -04001563 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
1564 icount1, icount2));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001565
1566 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
Mingming Cao617ba132006-10-11 01:20:53 -07001567 err = ext4_journal_get_write_access(handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001568 frames[0].bh);
1569 if (err)
1570 goto journal_error;
1571
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001572 memcpy((char *) entries2, (char *) (entries + icount1),
1573 icount2 * sizeof(struct dx_entry));
1574 dx_set_count(entries, icount1);
1575 dx_set_count(entries2, icount2);
1576 dx_set_limit(entries2, dx_node_limit(dir));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001577
1578 /* Which index block gets the new entry? */
1579 if (at - entries >= icount1) {
1580 frame->at = at = at - entries - icount1 + entries2;
1581 frame->entries = entries = entries2;
1582 swap(frame->bh, bh2);
1583 }
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001584 dx_insert_block(frames + 0, hash2, newblock);
1585 dxtrace(dx_show_index("node", frames[1].entries));
1586 dxtrace(dx_show_index("node",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001587 ((struct dx_node *) bh2->b_data)->entries));
Frank Mayhar03901312009-01-07 00:06:22 -05001588 err = ext4_handle_dirty_metadata(handle, inode, bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001589 if (err)
1590 goto journal_error;
1591 brelse (bh2);
1592 } else {
Theodore Ts'o4776004f2008-09-08 23:00:52 -04001593 dxtrace(printk(KERN_DEBUG
1594 "Creating second level index...\n"));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001595 memcpy((char *) entries2, (char *) entries,
1596 icount * sizeof(struct dx_entry));
1597 dx_set_limit(entries2, dx_node_limit(dir));
1598
1599 /* Set up root */
1600 dx_set_count(entries, 1);
1601 dx_set_block(entries + 0, newblock);
1602 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1603
1604 /* Add new access path frame */
1605 frame = frames + 1;
1606 frame->at = at = at - entries + entries2;
1607 frame->entries = entries = entries2;
1608 frame->bh = bh2;
Mingming Cao617ba132006-10-11 01:20:53 -07001609 err = ext4_journal_get_write_access(handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001610 frame->bh);
1611 if (err)
1612 goto journal_error;
1613 }
Frank Mayhar03901312009-01-07 00:06:22 -05001614 ext4_handle_dirty_metadata(handle, inode, frames[0].bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001615 }
1616 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1617 if (!de)
1618 goto cleanup;
1619 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001620 goto cleanup;
1621
1622journal_error:
Mingming Cao617ba132006-10-11 01:20:53 -07001623 ext4_std_error(dir->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001624cleanup:
1625 if (bh)
1626 brelse(bh);
1627 dx_release(frames);
1628 return err;
1629}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001630
1631/*
Mingming Cao617ba132006-10-11 01:20:53 -07001632 * ext4_delete_entry deletes a directory entry by merging it with the
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001633 * previous entry
1634 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001635static int ext4_delete_entry(handle_t *handle,
1636 struct inode *dir,
1637 struct ext4_dir_entry_2 *de_del,
1638 struct buffer_head *bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001639{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001640 struct ext4_dir_entry_2 *de, *pde;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001641 unsigned int blocksize = dir->i_sb->s_blocksize;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001642 int i;
1643
1644 i = 0;
1645 pde = NULL;
Mingming Cao617ba132006-10-11 01:20:53 -07001646 de = (struct ext4_dir_entry_2 *) bh->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001647 while (i < bh->b_size) {
Theodore Ts'o60fd4da2010-07-27 11:54:40 -04001648 if (!ext4_check_dir_entry(dir, de, bh, i))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001649 return -EIO;
1650 if (de == de_del) {
1651 BUFFER_TRACE(bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001652 ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001653 if (pde)
Jan Karaa72d7f82008-01-28 23:58:27 -05001654 pde->rec_len = ext4_rec_len_to_disk(
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001655 ext4_rec_len_from_disk(pde->rec_len,
1656 blocksize) +
1657 ext4_rec_len_from_disk(de->rec_len,
1658 blocksize),
1659 blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001660 else
1661 de->inode = 0;
1662 dir->i_version++;
Frank Mayhar03901312009-01-07 00:06:22 -05001663 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1664 ext4_handle_dirty_metadata(handle, dir, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001665 return 0;
1666 }
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001667 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001668 pde = de;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001669 de = ext4_next_entry(de, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001670 }
1671 return -ENOENT;
1672}
1673
Andreas Dilgerf8628a12007-07-18 08:38:01 -04001674/*
1675 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
1676 * since this indicates that nlinks count was previously 1.
1677 */
1678static void ext4_inc_count(handle_t *handle, struct inode *inode)
1679{
1680 inc_nlink(inode);
1681 if (is_dx(inode) && inode->i_nlink > 1) {
1682 /* limit is 16-bit i_links_count */
1683 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
1684 inode->i_nlink = 1;
1685 EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
1686 EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
1687 }
1688 }
1689}
1690
1691/*
1692 * If a directory had nlink == 1, then we should let it be 1. This indicates
1693 * directory has >EXT4_LINK_MAX subdirs.
1694 */
1695static void ext4_dec_count(handle_t *handle, struct inode *inode)
1696{
1697 drop_nlink(inode);
1698 if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
1699 inc_nlink(inode);
1700}
1701
1702
Mingming Cao617ba132006-10-11 01:20:53 -07001703static int ext4_add_nondir(handle_t *handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001704 struct dentry *dentry, struct inode *inode)
1705{
Mingming Cao617ba132006-10-11 01:20:53 -07001706 int err = ext4_add_entry(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001707 if (!err) {
Mingming Cao617ba132006-10-11 01:20:53 -07001708 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001709 d_instantiate(dentry, inode);
Al Viro6b38e842008-12-30 02:03:31 -05001710 unlock_new_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001711 return 0;
1712 }
Eric Sandeen731b9a52007-02-10 01:46:16 -08001713 drop_nlink(inode);
Al Viro6b38e842008-12-30 02:03:31 -05001714 unlock_new_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001715 iput(inode);
1716 return err;
1717}
1718
1719/*
1720 * By the time this is called, we already have created
1721 * the directory cache entry for the new file, but it
1722 * is so far negative - it has no inode.
1723 *
1724 * If the create succeeds, we fill in the inode information
1725 * with d_instantiate().
1726 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001727static int ext4_create(struct inode *dir, struct dentry *dentry, int mode,
1728 struct nameidata *nd)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001729{
1730 handle_t *handle;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001731 struct inode *inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001732 int err, retries = 0;
1733
Christoph Hellwig871a2932010-03-03 09:05:07 -05001734 dquot_initialize(dir);
Christoph Hellwig907f4552010-03-03 09:05:06 -05001735
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001736retry:
Mingming Cao617ba132006-10-11 01:20:53 -07001737 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1738 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
Dmitry Monakhov5aca07e2009-12-08 22:42:15 -05001739 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001740 if (IS_ERR(handle))
1741 return PTR_ERR(handle);
1742
1743 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05001744 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001745
Andreas Dilger11013912009-06-13 11:45:35 -04001746 inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001747 err = PTR_ERR(inode);
1748 if (!IS_ERR(inode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07001749 inode->i_op = &ext4_file_inode_operations;
1750 inode->i_fop = &ext4_file_operations;
1751 ext4_set_aops(inode);
1752 err = ext4_add_nondir(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001753 }
Mingming Cao617ba132006-10-11 01:20:53 -07001754 ext4_journal_stop(handle);
1755 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001756 goto retry;
1757 return err;
1758}
1759
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001760static int ext4_mknod(struct inode *dir, struct dentry *dentry,
1761 int mode, dev_t rdev)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001762{
1763 handle_t *handle;
1764 struct inode *inode;
1765 int err, retries = 0;
1766
1767 if (!new_valid_dev(rdev))
1768 return -EINVAL;
1769
Christoph Hellwig871a2932010-03-03 09:05:07 -05001770 dquot_initialize(dir);
Christoph Hellwig907f4552010-03-03 09:05:06 -05001771
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001772retry:
Mingming Cao617ba132006-10-11 01:20:53 -07001773 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1774 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
Dmitry Monakhov5aca07e2009-12-08 22:42:15 -05001775 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001776 if (IS_ERR(handle))
1777 return PTR_ERR(handle);
1778
1779 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05001780 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001781
Andreas Dilger11013912009-06-13 11:45:35 -04001782 inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001783 err = PTR_ERR(inode);
1784 if (!IS_ERR(inode)) {
1785 init_special_inode(inode, inode->i_mode, rdev);
Theodore Ts'o03010a32008-10-10 20:02:48 -04001786#ifdef CONFIG_EXT4_FS_XATTR
Mingming Cao617ba132006-10-11 01:20:53 -07001787 inode->i_op = &ext4_special_inode_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001788#endif
Mingming Cao617ba132006-10-11 01:20:53 -07001789 err = ext4_add_nondir(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001790 }
Mingming Cao617ba132006-10-11 01:20:53 -07001791 ext4_journal_stop(handle);
1792 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001793 goto retry;
1794 return err;
1795}
1796
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001797static int ext4_mkdir(struct inode *dir, struct dentry *dentry, int mode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001798{
1799 handle_t *handle;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001800 struct inode *inode;
1801 struct buffer_head *dir_block;
1802 struct ext4_dir_entry_2 *de;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001803 unsigned int blocksize = dir->i_sb->s_blocksize;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001804 int err, retries = 0;
1805
Andreas Dilgerf8628a12007-07-18 08:38:01 -04001806 if (EXT4_DIR_LINK_MAX(dir))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001807 return -EMLINK;
1808
Christoph Hellwig871a2932010-03-03 09:05:07 -05001809 dquot_initialize(dir);
Christoph Hellwig907f4552010-03-03 09:05:06 -05001810
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001811retry:
Mingming Cao617ba132006-10-11 01:20:53 -07001812 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1813 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
Dmitry Monakhov5aca07e2009-12-08 22:42:15 -05001814 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001815 if (IS_ERR(handle))
1816 return PTR_ERR(handle);
1817
1818 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05001819 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001820
Andreas Dilger11013912009-06-13 11:45:35 -04001821 inode = ext4_new_inode(handle, dir, S_IFDIR | mode,
1822 &dentry->d_name, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001823 err = PTR_ERR(inode);
1824 if (IS_ERR(inode))
1825 goto out_stop;
1826
Mingming Cao617ba132006-10-11 01:20:53 -07001827 inode->i_op = &ext4_dir_inode_operations;
1828 inode->i_fop = &ext4_dir_operations;
1829 inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001830 dir_block = ext4_bread(handle, inode, 0, 1, &err);
Aneesh Kumar K.V4cdeed82008-02-22 06:17:31 -05001831 if (!dir_block)
1832 goto out_clear_inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001833 BUFFER_TRACE(dir_block, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001834 ext4_journal_get_write_access(handle, dir_block);
1835 de = (struct ext4_dir_entry_2 *) dir_block->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001836 de->inode = cpu_to_le32(inode->i_ino);
1837 de->name_len = 1;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001838 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
1839 blocksize);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001840 strcpy(de->name, ".");
Mingming Cao617ba132006-10-11 01:20:53 -07001841 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001842 de = ext4_next_entry(de, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001843 de->inode = cpu_to_le32(dir->i_ino);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001844 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(1),
1845 blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001846 de->name_len = 2;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001847 strcpy(de->name, "..");
Mingming Cao617ba132006-10-11 01:20:53 -07001848 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001849 inode->i_nlink = 2;
Frank Mayhar03901312009-01-07 00:06:22 -05001850 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
1851 ext4_handle_dirty_metadata(handle, dir, dir_block);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001852 brelse(dir_block);
Mingming Cao617ba132006-10-11 01:20:53 -07001853 ext4_mark_inode_dirty(handle, inode);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001854 err = ext4_add_entry(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001855 if (err) {
Aneesh Kumar K.V4cdeed82008-02-22 06:17:31 -05001856out_clear_inode:
1857 clear_nlink(inode);
Al Viro6b38e842008-12-30 02:03:31 -05001858 unlock_new_inode(inode);
Mingming Cao617ba132006-10-11 01:20:53 -07001859 ext4_mark_inode_dirty(handle, inode);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001860 iput(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001861 goto out_stop;
1862 }
Andreas Dilgerf8628a12007-07-18 08:38:01 -04001863 ext4_inc_count(handle, dir);
Mingming Cao617ba132006-10-11 01:20:53 -07001864 ext4_update_dx_flag(dir);
1865 ext4_mark_inode_dirty(handle, dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001866 d_instantiate(dentry, inode);
Al Viro6b38e842008-12-30 02:03:31 -05001867 unlock_new_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001868out_stop:
Mingming Cao617ba132006-10-11 01:20:53 -07001869 ext4_journal_stop(handle);
1870 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001871 goto retry;
1872 return err;
1873}
1874
1875/*
1876 * routine to check that the specified directory is empty (for rmdir)
1877 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001878static int empty_dir(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001879{
Theodore Ts'o498e5f22008-11-05 00:14:04 -05001880 unsigned int offset;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001881 struct buffer_head *bh;
1882 struct ext4_dir_entry_2 *de, *de1;
1883 struct super_block *sb;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001884 int err = 0;
1885
1886 sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -07001887 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001888 !(bh = ext4_bread(NULL, inode, 0, 0, &err))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001889 if (err)
Theodore Ts'o24676da2010-05-16 21:00:00 -04001890 EXT4_ERROR_INODE(inode,
1891 "error %d reading directory lblock 0", err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001892 else
Eric Sandeen12062dd2010-02-15 14:19:27 -05001893 ext4_warning(inode->i_sb,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001894 "bad directory (dir #%lu) - no data block",
1895 inode->i_ino);
1896 return 1;
1897 }
Mingming Cao617ba132006-10-11 01:20:53 -07001898 de = (struct ext4_dir_entry_2 *) bh->b_data;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001899 de1 = ext4_next_entry(de, sb->s_blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001900 if (le32_to_cpu(de->inode) != inode->i_ino ||
1901 !le32_to_cpu(de1->inode) ||
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001902 strcmp(".", de->name) ||
1903 strcmp("..", de1->name)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001904 ext4_warning(inode->i_sb,
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001905 "bad directory (dir #%lu) - no `.' or `..'",
1906 inode->i_ino);
1907 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001908 return 1;
1909 }
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001910 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
1911 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
1912 de = ext4_next_entry(de1, sb->s_blocksize);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001913 while (offset < inode->i_size) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001914 if (!bh ||
Theodore Ts'o24676da2010-05-16 21:00:00 -04001915 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1916 unsigned int lblock;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001917 err = 0;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001918 brelse(bh);
Theodore Ts'o24676da2010-05-16 21:00:00 -04001919 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
1920 bh = ext4_bread(NULL, inode, lblock, 0, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001921 if (!bh) {
1922 if (err)
Theodore Ts'o24676da2010-05-16 21:00:00 -04001923 EXT4_ERROR_INODE(inode,
1924 "error %d reading directory "
1925 "lblock %u", err, lblock);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001926 offset += sb->s_blocksize;
1927 continue;
1928 }
Mingming Cao617ba132006-10-11 01:20:53 -07001929 de = (struct ext4_dir_entry_2 *) bh->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001930 }
Theodore Ts'o60fd4da2010-07-27 11:54:40 -04001931 if (!ext4_check_dir_entry(inode, de, bh, offset)) {
Mingming Cao617ba132006-10-11 01:20:53 -07001932 de = (struct ext4_dir_entry_2 *)(bh->b_data +
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001933 sb->s_blocksize);
1934 offset = (offset | (sb->s_blocksize - 1)) + 1;
1935 continue;
1936 }
1937 if (le32_to_cpu(de->inode)) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001938 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001939 return 0;
1940 }
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001941 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
1942 de = ext4_next_entry(de, sb->s_blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001943 }
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001944 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001945 return 1;
1946}
1947
Mingming Cao617ba132006-10-11 01:20:53 -07001948/* ext4_orphan_add() links an unlinked or truncated inode into a list of
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001949 * such inodes, starting at the superblock, in case we crash before the
1950 * file is closed/deleted, or in case the inode truncate spans multiple
1951 * transactions and the last transaction is not recovered after a crash.
1952 *
1953 * At filesystem recovery time, we walk this list deleting unlinked
Mingming Cao617ba132006-10-11 01:20:53 -07001954 * inodes and truncating linked inodes in ext4_orphan_cleanup().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001955 */
Mingming Cao617ba132006-10-11 01:20:53 -07001956int ext4_orphan_add(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001957{
1958 struct super_block *sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -07001959 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001960 int err = 0, rc;
1961
Frank Mayhar03901312009-01-07 00:06:22 -05001962 if (!ext4_handle_valid(handle))
1963 return 0;
1964
Theodore Ts'o3b9d4ed2009-04-25 22:54:04 -04001965 mutex_lock(&EXT4_SB(sb)->s_orphan_lock);
Mingming Cao617ba132006-10-11 01:20:53 -07001966 if (!list_empty(&EXT4_I(inode)->i_orphan))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001967 goto out_unlock;
1968
1969 /* Orphan handling is only valid for files with data blocks
1970 * being truncated, or files being unlinked. */
1971
1972 /* @@@ FIXME: Observation from aviro:
Mingming Cao617ba132006-10-11 01:20:53 -07001973 * I think I can trigger J_ASSERT in ext4_orphan_add(). We block
Theodore Ts'o3b9d4ed2009-04-25 22:54:04 -04001974 * here (on s_orphan_lock), so race with ext4_link() which might bump
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001975 * ->i_nlink. For, say it, character device. Not a regular file,
1976 * not a directory, not a symlink and ->i_nlink > 0.
Theodore Ts'o3b9d4ed2009-04-25 22:54:04 -04001977 *
1978 * tytso, 4/25/2009: I'm not sure how that could happen;
1979 * shouldn't the fs core protect us from these sort of
1980 * unlink()/link() races?
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001981 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001982 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1983 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001984
Mingming Cao617ba132006-10-11 01:20:53 -07001985 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1986 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001987 if (err)
1988 goto out_unlock;
1989
Mingming Cao617ba132006-10-11 01:20:53 -07001990 err = ext4_reserve_inode_write(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001991 if (err)
1992 goto out_unlock;
Dmitry Monakhov6e3617e2010-03-01 23:29:39 -05001993 /*
1994 * Due to previous errors inode may be already a part of on-disk
1995 * orphan list. If so skip on-disk list modification.
1996 */
1997 if (NEXT_ORPHAN(inode) && NEXT_ORPHAN(inode) <=
1998 (le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)))
1999 goto mem_insert;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002000
2001 /* Insert this inode at the head of the on-disk orphan list... */
Mingming Cao617ba132006-10-11 01:20:53 -07002002 NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
2003 EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
Curt Wohlgemuth73b50c12010-02-16 15:06:29 -05002004 err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
Mingming Cao617ba132006-10-11 01:20:53 -07002005 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002006 if (!err)
2007 err = rc;
2008
2009 /* Only add to the head of the in-memory list if all the
2010 * previous operations succeeded. If the orphan_add is going to
2011 * fail (possibly taking the journal offline), we can't risk
2012 * leaving the inode on the orphan list: stray orphan-list
2013 * entries can cause panics at unmount time.
2014 *
2015 * This is safe: on error we're going to ignore the orphan list
2016 * anyway on the next recovery. */
Dmitry Monakhov6e3617e2010-03-01 23:29:39 -05002017mem_insert:
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002018 if (!err)
Mingming Cao617ba132006-10-11 01:20:53 -07002019 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002020
2021 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2022 jbd_debug(4, "orphan inode %lu will point to %d\n",
2023 inode->i_ino, NEXT_ORPHAN(inode));
2024out_unlock:
Theodore Ts'o3b9d4ed2009-04-25 22:54:04 -04002025 mutex_unlock(&EXT4_SB(sb)->s_orphan_lock);
Mingming Cao617ba132006-10-11 01:20:53 -07002026 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002027 return err;
2028}
2029
2030/*
Mingming Cao617ba132006-10-11 01:20:53 -07002031 * ext4_orphan_del() removes an unlinked or truncated inode from the list
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002032 * of such inodes stored on disk, because it is finally being cleaned up.
2033 */
Mingming Cao617ba132006-10-11 01:20:53 -07002034int ext4_orphan_del(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002035{
2036 struct list_head *prev;
Mingming Cao617ba132006-10-11 01:20:53 -07002037 struct ext4_inode_info *ei = EXT4_I(inode);
2038 struct ext4_sb_info *sbi;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05002039 __u32 ino_next;
Mingming Cao617ba132006-10-11 01:20:53 -07002040 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002041 int err = 0;
2042
Curt Wohlgemuthd3d1faf2009-09-29 11:01:03 -04002043 /* ext4_handle_valid() assumes a valid handle_t pointer */
2044 if (handle && !ext4_handle_valid(handle))
Frank Mayhar03901312009-01-07 00:06:22 -05002045 return 0;
2046
Theodore Ts'o3b9d4ed2009-04-25 22:54:04 -04002047 mutex_lock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2048 if (list_empty(&ei->i_orphan))
2049 goto out;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002050
2051 ino_next = NEXT_ORPHAN(inode);
2052 prev = ei->i_orphan.prev;
Mingming Cao617ba132006-10-11 01:20:53 -07002053 sbi = EXT4_SB(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002054
2055 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2056
2057 list_del_init(&ei->i_orphan);
2058
2059 /* If we're on an error path, we may not have a valid
2060 * transaction handle with which to update the orphan list on
2061 * disk, but we still need to remove the inode from the linked
2062 * list in memory. */
Frank Mayhar03901312009-01-07 00:06:22 -05002063 if (sbi->s_journal && !handle)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002064 goto out;
2065
Mingming Cao617ba132006-10-11 01:20:53 -07002066 err = ext4_reserve_inode_write(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002067 if (err)
2068 goto out_err;
2069
2070 if (prev == &sbi->s_orphan) {
Theodore Ts'o498e5f22008-11-05 00:14:04 -05002071 jbd_debug(4, "superblock will point to %u\n", ino_next);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002072 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07002073 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002074 if (err)
2075 goto out_brelse;
2076 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
Curt Wohlgemuth73b50c12010-02-16 15:06:29 -05002077 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002078 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07002079 struct ext4_iloc iloc2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002080 struct inode *i_prev =
Mingming Cao617ba132006-10-11 01:20:53 -07002081 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002082
Theodore Ts'o498e5f22008-11-05 00:14:04 -05002083 jbd_debug(4, "orphan inode %lu will point to %u\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002084 i_prev->i_ino, ino_next);
Mingming Cao617ba132006-10-11 01:20:53 -07002085 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002086 if (err)
2087 goto out_brelse;
2088 NEXT_ORPHAN(i_prev) = ino_next;
Mingming Cao617ba132006-10-11 01:20:53 -07002089 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002090 }
2091 if (err)
2092 goto out_brelse;
2093 NEXT_ORPHAN(inode) = 0;
Mingming Cao617ba132006-10-11 01:20:53 -07002094 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002095
2096out_err:
Mingming Cao617ba132006-10-11 01:20:53 -07002097 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002098out:
Theodore Ts'o3b9d4ed2009-04-25 22:54:04 -04002099 mutex_unlock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002100 return err;
2101
2102out_brelse:
2103 brelse(iloc.bh);
2104 goto out_err;
2105}
2106
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002107static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002108{
2109 int retval;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002110 struct inode *inode;
2111 struct buffer_head *bh;
2112 struct ext4_dir_entry_2 *de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002113 handle_t *handle;
2114
2115 /* Initialize quotas before so that eventual writes go in
2116 * separate transaction */
Christoph Hellwig871a2932010-03-03 09:05:07 -05002117 dquot_initialize(dir);
2118 dquot_initialize(dentry->d_inode);
Christoph Hellwig907f4552010-03-03 09:05:06 -05002119
Mingming Cao617ba132006-10-11 01:20:53 -07002120 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002121 if (IS_ERR(handle))
2122 return PTR_ERR(handle);
2123
2124 retval = -ENOENT;
Theodore Ts'of702ba02008-09-22 15:21:01 -04002125 bh = ext4_find_entry(dir, &dentry->d_name, &de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002126 if (!bh)
2127 goto end_rmdir;
2128
2129 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05002130 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002131
2132 inode = dentry->d_inode;
2133
2134 retval = -EIO;
2135 if (le32_to_cpu(de->inode) != inode->i_ino)
2136 goto end_rmdir;
2137
2138 retval = -ENOTEMPTY;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002139 if (!empty_dir(inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002140 goto end_rmdir;
2141
Mingming Cao617ba132006-10-11 01:20:53 -07002142 retval = ext4_delete_entry(handle, dir, de, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002143 if (retval)
2144 goto end_rmdir;
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002145 if (!EXT4_DIR_LINK_EMPTY(inode))
Eric Sandeen12062dd2010-02-15 14:19:27 -05002146 ext4_warning(inode->i_sb,
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002147 "empty directory has too many links (%d)",
2148 inode->i_nlink);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002149 inode->i_version++;
2150 clear_nlink(inode);
2151 /* There's no need to set i_disksize: the fact that i_nlink is
2152 * zero will ensure that the right thing happens during any
2153 * recovery. */
2154 inode->i_size = 0;
Mingming Cao617ba132006-10-11 01:20:53 -07002155 ext4_orphan_add(handle, inode);
Kalpak Shahef7f3832007-07-18 09:15:20 -04002156 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002157 ext4_mark_inode_dirty(handle, inode);
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002158 ext4_dec_count(handle, dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002159 ext4_update_dx_flag(dir);
2160 ext4_mark_inode_dirty(handle, dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002161
2162end_rmdir:
Mingming Cao617ba132006-10-11 01:20:53 -07002163 ext4_journal_stop(handle);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002164 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002165 return retval;
2166}
2167
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002168static int ext4_unlink(struct inode *dir, struct dentry *dentry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002169{
2170 int retval;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002171 struct inode *inode;
2172 struct buffer_head *bh;
2173 struct ext4_dir_entry_2 *de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002174 handle_t *handle;
2175
2176 /* Initialize quotas before so that eventual writes go
2177 * in separate transaction */
Christoph Hellwig871a2932010-03-03 09:05:07 -05002178 dquot_initialize(dir);
2179 dquot_initialize(dentry->d_inode);
Christoph Hellwig907f4552010-03-03 09:05:06 -05002180
Mingming Cao617ba132006-10-11 01:20:53 -07002181 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002182 if (IS_ERR(handle))
2183 return PTR_ERR(handle);
2184
2185 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05002186 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002187
2188 retval = -ENOENT;
Theodore Ts'of702ba02008-09-22 15:21:01 -04002189 bh = ext4_find_entry(dir, &dentry->d_name, &de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002190 if (!bh)
2191 goto end_unlink;
2192
2193 inode = dentry->d_inode;
2194
2195 retval = -EIO;
2196 if (le32_to_cpu(de->inode) != inode->i_ino)
2197 goto end_unlink;
2198
2199 if (!inode->i_nlink) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05002200 ext4_warning(inode->i_sb,
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002201 "Deleting nonexistent file (%lu), %d",
2202 inode->i_ino, inode->i_nlink);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002203 inode->i_nlink = 1;
2204 }
Mingming Cao617ba132006-10-11 01:20:53 -07002205 retval = ext4_delete_entry(handle, dir, de, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002206 if (retval)
2207 goto end_unlink;
Kalpak Shahef7f3832007-07-18 09:15:20 -04002208 dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002209 ext4_update_dx_flag(dir);
2210 ext4_mark_inode_dirty(handle, dir);
Theodore Ts'o825f1482008-02-15 15:00:38 -05002211 drop_nlink(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002212 if (!inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07002213 ext4_orphan_add(handle, inode);
Kalpak Shahef7f3832007-07-18 09:15:20 -04002214 inode->i_ctime = ext4_current_time(inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002215 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002216 retval = 0;
2217
2218end_unlink:
Mingming Cao617ba132006-10-11 01:20:53 -07002219 ext4_journal_stop(handle);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002220 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002221 return retval;
2222}
2223
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002224static int ext4_symlink(struct inode *dir,
2225 struct dentry *dentry, const char *symname)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002226{
2227 handle_t *handle;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002228 struct inode *inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002229 int l, err, retries = 0;
2230
2231 l = strlen(symname)+1;
2232 if (l > dir->i_sb->s_blocksize)
2233 return -ENAMETOOLONG;
2234
Christoph Hellwig871a2932010-03-03 09:05:07 -05002235 dquot_initialize(dir);
Christoph Hellwig907f4552010-03-03 09:05:06 -05002236
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002237retry:
Mingming Cao617ba132006-10-11 01:20:53 -07002238 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2239 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 5 +
Dmitry Monakhov5aca07e2009-12-08 22:42:15 -05002240 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002241 if (IS_ERR(handle))
2242 return PTR_ERR(handle);
2243
2244 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05002245 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002246
Andreas Dilger11013912009-06-13 11:45:35 -04002247 inode = ext4_new_inode(handle, dir, S_IFLNK|S_IRWXUGO,
2248 &dentry->d_name, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002249 err = PTR_ERR(inode);
2250 if (IS_ERR(inode))
2251 goto out_stop;
2252
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002253 if (l > sizeof(EXT4_I(inode)->i_data)) {
Mingming Cao617ba132006-10-11 01:20:53 -07002254 inode->i_op = &ext4_symlink_inode_operations;
2255 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002256 /*
Mingming Cao617ba132006-10-11 01:20:53 -07002257 * page_symlink() calls into ext4_prepare/commit_write.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002258 * We have a transaction open. All is sweetness. It also sets
2259 * i_size in generic_commit_write().
2260 */
Nick Piggin54566b22009-01-04 12:00:53 -08002261 err = __page_symlink(inode, symname, l, 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002262 if (err) {
Theodore Ts'o825f1482008-02-15 15:00:38 -05002263 clear_nlink(inode);
Al Viro6b38e842008-12-30 02:03:31 -05002264 unlock_new_inode(inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002265 ext4_mark_inode_dirty(handle, inode);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002266 iput(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002267 goto out_stop;
2268 }
2269 } else {
Aneesh Kumar K.Ve65187e2008-04-29 08:11:12 -04002270 /* clear the extent format for fast symlink */
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04002271 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
Mingming Cao617ba132006-10-11 01:20:53 -07002272 inode->i_op = &ext4_fast_symlink_inode_operations;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002273 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002274 inode->i_size = l-1;
2275 }
Mingming Cao617ba132006-10-11 01:20:53 -07002276 EXT4_I(inode)->i_disksize = inode->i_size;
2277 err = ext4_add_nondir(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002278out_stop:
Mingming Cao617ba132006-10-11 01:20:53 -07002279 ext4_journal_stop(handle);
2280 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002281 goto retry;
2282 return err;
2283}
2284
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002285static int ext4_link(struct dentry *old_dentry,
2286 struct inode *dir, struct dentry *dentry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002287{
2288 handle_t *handle;
2289 struct inode *inode = old_dentry->d_inode;
2290 int err, retries = 0;
2291
Theodore Ts'ob05ab1d2009-08-29 21:08:08 -04002292 if (inode->i_nlink >= EXT4_LINK_MAX)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002293 return -EMLINK;
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002294
Christoph Hellwig871a2932010-03-03 09:05:07 -05002295 dquot_initialize(dir);
Christoph Hellwig907f4552010-03-03 09:05:06 -05002296
Eric Sandeen2988a772007-02-10 01:46:16 -08002297 /*
2298 * Return -ENOENT if we've raced with unlink and i_nlink is 0. Doing
2299 * otherwise has the potential to corrupt the orphan inode list.
2300 */
2301 if (inode->i_nlink == 0)
2302 return -ENOENT;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002303
2304retry:
Mingming Cao617ba132006-10-11 01:20:53 -07002305 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2306 EXT4_INDEX_EXTRA_TRANS_BLOCKS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002307 if (IS_ERR(handle))
2308 return PTR_ERR(handle);
2309
2310 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05002311 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002312
Kalpak Shahef7f3832007-07-18 09:15:20 -04002313 inode->i_ctime = ext4_current_time(inode);
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002314 ext4_inc_count(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002315 atomic_inc(&inode->i_count);
2316
Al Viro6b38e842008-12-30 02:03:31 -05002317 err = ext4_add_entry(handle, dentry, inode);
2318 if (!err) {
2319 ext4_mark_inode_dirty(handle, inode);
2320 d_instantiate(dentry, inode);
2321 } else {
2322 drop_nlink(inode);
2323 iput(inode);
2324 }
Mingming Cao617ba132006-10-11 01:20:53 -07002325 ext4_journal_stop(handle);
2326 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002327 goto retry;
2328 return err;
2329}
2330
Wei Yongjun3d0518f2009-02-14 23:01:36 -05002331#define PARENT_INO(buffer, size) \
2332 (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer), size)->inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002333
2334/*
2335 * Anybody can rename anything with this: the permission checks are left to the
2336 * higher-level routines.
2337 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002338static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
2339 struct inode *new_dir, struct dentry *new_dentry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002340{
2341 handle_t *handle;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002342 struct inode *old_inode, *new_inode;
2343 struct buffer_head *old_bh, *new_bh, *dir_bh;
2344 struct ext4_dir_entry_2 *old_de, *new_de;
Theodore Ts'o8750c6d2009-02-23 23:05:27 -05002345 int retval, force_da_alloc = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002346
Christoph Hellwig871a2932010-03-03 09:05:07 -05002347 dquot_initialize(old_dir);
2348 dquot_initialize(new_dir);
Christoph Hellwig907f4552010-03-03 09:05:06 -05002349
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002350 old_bh = new_bh = dir_bh = NULL;
2351
2352 /* Initialize quotas before so that eventual writes go
2353 * in separate transaction */
2354 if (new_dentry->d_inode)
Christoph Hellwig871a2932010-03-03 09:05:07 -05002355 dquot_initialize(new_dentry->d_inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002356 handle = ext4_journal_start(old_dir, 2 *
2357 EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2358 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002359 if (IS_ERR(handle))
2360 return PTR_ERR(handle);
2361
2362 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
Frank Mayhar03901312009-01-07 00:06:22 -05002363 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002364
Theodore Ts'of702ba02008-09-22 15:21:01 -04002365 old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002366 /*
2367 * Check for inode number is _not_ due to possible IO errors.
2368 * We might rmdir the source, keep it as pwd of some process
2369 * and merrily kill the link to whatever was created under the
2370 * same name. Goodbye sticky bit ;-<
2371 */
2372 old_inode = old_dentry->d_inode;
2373 retval = -ENOENT;
2374 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2375 goto end_rename;
2376
2377 new_inode = new_dentry->d_inode;
Theodore Ts'of702ba02008-09-22 15:21:01 -04002378 new_bh = ext4_find_entry(new_dir, &new_dentry->d_name, &new_de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002379 if (new_bh) {
2380 if (!new_inode) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002381 brelse(new_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002382 new_bh = NULL;
2383 }
2384 }
2385 if (S_ISDIR(old_inode->i_mode)) {
2386 if (new_inode) {
2387 retval = -ENOTEMPTY;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002388 if (!empty_dir(new_inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002389 goto end_rename;
2390 }
2391 retval = -EIO;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002392 dir_bh = ext4_bread(handle, old_inode, 0, 0, &retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002393 if (!dir_bh)
2394 goto end_rename;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05002395 if (le32_to_cpu(PARENT_INO(dir_bh->b_data,
2396 old_dir->i_sb->s_blocksize)) != old_dir->i_ino)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002397 goto end_rename;
2398 retval = -EMLINK;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002399 if (!new_inode && new_dir != old_dir &&
Aneesh Kumar K.V2c94eb82009-08-28 21:43:15 -04002400 EXT4_DIR_LINK_MAX(new_dir))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002401 goto end_rename;
2402 }
2403 if (!new_bh) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002404 retval = ext4_add_entry(handle, new_dentry, old_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002405 if (retval)
2406 goto end_rename;
2407 } else {
2408 BUFFER_TRACE(new_bh, "get write access");
Mingming Cao617ba132006-10-11 01:20:53 -07002409 ext4_journal_get_write_access(handle, new_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002410 new_de->inode = cpu_to_le32(old_inode->i_ino);
Mingming Cao617ba132006-10-11 01:20:53 -07002411 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2412 EXT4_FEATURE_INCOMPAT_FILETYPE))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002413 new_de->file_type = old_de->file_type;
2414 new_dir->i_version++;
Jan Kara53b7e9f2008-04-29 22:02:11 -04002415 new_dir->i_ctime = new_dir->i_mtime =
2416 ext4_current_time(new_dir);
2417 ext4_mark_inode_dirty(handle, new_dir);
Frank Mayhar03901312009-01-07 00:06:22 -05002418 BUFFER_TRACE(new_bh, "call ext4_handle_dirty_metadata");
2419 ext4_handle_dirty_metadata(handle, new_dir, new_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002420 brelse(new_bh);
2421 new_bh = NULL;
2422 }
2423
2424 /*
2425 * Like most other Unix systems, set the ctime for inodes on a
2426 * rename.
2427 */
Kalpak Shahef7f3832007-07-18 09:15:20 -04002428 old_inode->i_ctime = ext4_current_time(old_inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002429 ext4_mark_inode_dirty(handle, old_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002430
2431 /*
2432 * ok, that's it
2433 */
2434 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2435 old_de->name_len != old_dentry->d_name.len ||
2436 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
Mingming Cao617ba132006-10-11 01:20:53 -07002437 (retval = ext4_delete_entry(handle, old_dir,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002438 old_de, old_bh)) == -ENOENT) {
2439 /* old_de could have moved from under us during htree split, so
2440 * make sure that we are deleting the right entry. We might
2441 * also be pointing to a stale entry in the unused part of
2442 * old_bh so just checking inum and the name isn't enough. */
2443 struct buffer_head *old_bh2;
Mingming Cao617ba132006-10-11 01:20:53 -07002444 struct ext4_dir_entry_2 *old_de2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002445
Theodore Ts'of702ba02008-09-22 15:21:01 -04002446 old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002447 if (old_bh2) {
Mingming Cao617ba132006-10-11 01:20:53 -07002448 retval = ext4_delete_entry(handle, old_dir,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002449 old_de2, old_bh2);
2450 brelse(old_bh2);
2451 }
2452 }
2453 if (retval) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05002454 ext4_warning(old_dir->i_sb,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002455 "Deleting old file (%lu), %d, error=%d",
2456 old_dir->i_ino, old_dir->i_nlink, retval);
2457 }
2458
2459 if (new_inode) {
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002460 ext4_dec_count(handle, new_inode);
Kalpak Shahef7f3832007-07-18 09:15:20 -04002461 new_inode->i_ctime = ext4_current_time(new_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002462 }
Kalpak Shahef7f3832007-07-18 09:15:20 -04002463 old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002464 ext4_update_dx_flag(old_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002465 if (dir_bh) {
2466 BUFFER_TRACE(dir_bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07002467 ext4_journal_get_write_access(handle, dir_bh);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05002468 PARENT_INO(dir_bh->b_data, new_dir->i_sb->s_blocksize) =
2469 cpu_to_le32(new_dir->i_ino);
Frank Mayhar03901312009-01-07 00:06:22 -05002470 BUFFER_TRACE(dir_bh, "call ext4_handle_dirty_metadata");
2471 ext4_handle_dirty_metadata(handle, old_dir, dir_bh);
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002472 ext4_dec_count(handle, old_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002473 if (new_inode) {
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002474 /* checked empty_dir above, can't have another parent,
Theodore Ts'o825f1482008-02-15 15:00:38 -05002475 * ext4_dec_count() won't work for many-linked dirs */
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002476 new_inode->i_nlink = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002477 } else {
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002478 ext4_inc_count(handle, new_dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002479 ext4_update_dx_flag(new_dir);
2480 ext4_mark_inode_dirty(handle, new_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002481 }
2482 }
Mingming Cao617ba132006-10-11 01:20:53 -07002483 ext4_mark_inode_dirty(handle, old_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002484 if (new_inode) {
Mingming Cao617ba132006-10-11 01:20:53 -07002485 ext4_mark_inode_dirty(handle, new_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002486 if (!new_inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07002487 ext4_orphan_add(handle, new_inode);
Theodore Ts'oafd46722009-03-16 23:12:23 -04002488 if (!test_opt(new_dir->i_sb, NO_AUTO_DA_ALLOC))
2489 force_da_alloc = 1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002490 }
2491 retval = 0;
2492
2493end_rename:
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002494 brelse(dir_bh);
2495 brelse(old_bh);
2496 brelse(new_bh);
Mingming Cao617ba132006-10-11 01:20:53 -07002497 ext4_journal_stop(handle);
Theodore Ts'o8750c6d2009-02-23 23:05:27 -05002498 if (retval == 0 && force_da_alloc)
2499 ext4_alloc_da_blocks(old_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002500 return retval;
2501}
2502
2503/*
2504 * directories can handle most operations...
2505 */
Arjan van de Ven754661f2007-02-12 00:55:38 -08002506const struct inode_operations ext4_dir_inode_operations = {
Mingming Cao617ba132006-10-11 01:20:53 -07002507 .create = ext4_create,
2508 .lookup = ext4_lookup,
2509 .link = ext4_link,
2510 .unlink = ext4_unlink,
2511 .symlink = ext4_symlink,
2512 .mkdir = ext4_mkdir,
2513 .rmdir = ext4_rmdir,
2514 .mknod = ext4_mknod,
2515 .rename = ext4_rename,
2516 .setattr = ext4_setattr,
Theodore Ts'o03010a32008-10-10 20:02:48 -04002517#ifdef CONFIG_EXT4_FS_XATTR
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002518 .setxattr = generic_setxattr,
2519 .getxattr = generic_getxattr,
Mingming Cao617ba132006-10-11 01:20:53 -07002520 .listxattr = ext4_listxattr,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002521 .removexattr = generic_removexattr,
2522#endif
Linus Torvalds1d5ccd12009-08-28 12:12:24 -07002523 .check_acl = ext4_check_acl,
Aneesh Kumar K.Vabc87462009-05-02 22:54:32 -04002524 .fiemap = ext4_fiemap,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002525};
2526
Arjan van de Ven754661f2007-02-12 00:55:38 -08002527const struct inode_operations ext4_special_inode_operations = {
Mingming Cao617ba132006-10-11 01:20:53 -07002528 .setattr = ext4_setattr,
Theodore Ts'o03010a32008-10-10 20:02:48 -04002529#ifdef CONFIG_EXT4_FS_XATTR
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002530 .setxattr = generic_setxattr,
2531 .getxattr = generic_getxattr,
Mingming Cao617ba132006-10-11 01:20:53 -07002532 .listxattr = ext4_listxattr,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002533 .removexattr = generic_removexattr,
2534#endif
Linus Torvalds1d5ccd12009-08-28 12:12:24 -07002535 .check_acl = ext4_check_acl,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002536};