blob: 707d605bf76985dfe0b873b8fccda9c6459682b1 [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
Jiaying Zhang0562e0b2011-03-21 21:38:05 -040043#include <trace/events/ext4.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070044/*
45 * define how far ahead to read directories while searching them.
46 */
47#define NAMEI_RA_CHUNKS 2
48#define NAMEI_RA_BLOCKS 4
Dave Kleikamp8c55e202007-05-24 13:04:54 -040049#define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070050#define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b))
51
Mingming Cao617ba132006-10-11 01:20:53 -070052static struct buffer_head *ext4_append(handle_t *handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -070053 struct inode *inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -050054 ext4_lblk_t *block, int *err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070055{
56 struct buffer_head *bh;
57
58 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
59
Akinobu Mitaa8716112008-04-17 10:38:59 -040060 bh = ext4_bread(handle, inode, *block, 1, err);
61 if (bh) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -070062 inode->i_size += inode->i_sb->s_blocksize;
Mingming Cao617ba132006-10-11 01:20:53 -070063 EXT4_I(inode)->i_disksize = inode->i_size;
Akinobu Mitaa8716112008-04-17 10:38:59 -040064 *err = ext4_journal_get_write_access(handle, bh);
65 if (*err) {
66 brelse(bh);
67 bh = NULL;
68 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -070069 }
70 return bh;
71}
72
73#ifndef assert
74#define assert(test) J_ASSERT(test)
75#endif
76
Dave Kleikampac27a0e2006-10-11 01:20:50 -070077#ifdef DX_DEBUG
78#define dxtrace(command) command
79#else
80#define dxtrace(command)
81#endif
82
83struct fake_dirent
84{
85 __le32 inode;
86 __le16 rec_len;
87 u8 name_len;
88 u8 file_type;
89};
90
91struct dx_countlimit
92{
93 __le16 limit;
94 __le16 count;
95};
96
97struct dx_entry
98{
99 __le32 hash;
100 __le32 block;
101};
102
103/*
104 * dx_root_info is laid out so that if it should somehow get overlaid by a
105 * dirent the two low bits of the hash version will be zero. Therefore, the
106 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
107 */
108
109struct dx_root
110{
111 struct fake_dirent dot;
112 char dot_name[4];
113 struct fake_dirent dotdot;
114 char dotdot_name[4];
115 struct dx_root_info
116 {
117 __le32 reserved_zero;
118 u8 hash_version;
119 u8 info_length; /* 8 */
120 u8 indirect_levels;
121 u8 unused_flags;
122 }
123 info;
124 struct dx_entry entries[0];
125};
126
127struct dx_node
128{
129 struct fake_dirent fake;
130 struct dx_entry entries[0];
131};
132
133
134struct dx_frame
135{
136 struct buffer_head *bh;
137 struct dx_entry *entries;
138 struct dx_entry *at;
139};
140
141struct dx_map_entry
142{
143 u32 hash;
Eric Sandeenef2b02d2007-09-18 22:46:42 -0700144 u16 offs;
145 u16 size;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700146};
147
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500148static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
149static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400150static inline unsigned dx_get_hash(struct dx_entry *entry);
151static void dx_set_hash(struct dx_entry *entry, unsigned value);
152static unsigned dx_get_count(struct dx_entry *entries);
153static unsigned dx_get_limit(struct dx_entry *entries);
154static void dx_set_count(struct dx_entry *entries, unsigned value);
155static void dx_set_limit(struct dx_entry *entries, unsigned value);
156static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
157static unsigned dx_node_limit(struct inode *dir);
Theodore Ts'of702ba02008-09-22 15:21:01 -0400158static struct dx_frame *dx_probe(const struct qstr *d_name,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700159 struct inode *dir,
160 struct dx_hash_info *hinfo,
161 struct dx_frame *frame,
162 int *err);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400163static void dx_release(struct dx_frame *frames);
Theodore Ts'o8bad4592009-02-14 21:46:54 -0500164static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400165 struct dx_hash_info *hinfo, struct dx_map_entry map[]);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700166static void dx_sort_map(struct dx_map_entry *map, unsigned count);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400167static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500168 struct dx_map_entry *offsets, int count, unsigned blocksize);
Theodore Ts'o8bad4592009-02-14 21:46:54 -0500169static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500170static void dx_insert_block(struct dx_frame *frame,
171 u32 hash, ext4_lblk_t block);
Mingming Cao617ba132006-10-11 01:20:53 -0700172static int ext4_htree_next_block(struct inode *dir, __u32 hash,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700173 struct dx_frame *frame,
174 struct dx_frame *frames,
175 __u32 *start_hash);
Theodore Ts'of702ba02008-09-22 15:21:01 -0400176static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
177 const struct qstr *d_name,
178 struct ext4_dir_entry_2 **res_dir,
179 int *err);
Mingming Cao617ba132006-10-11 01:20:53 -0700180static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700181 struct inode *inode);
182
183/*
Li Zefanf795e142008-07-11 19:27:31 -0400184 * p is at least 6 bytes before the end of page
185 */
186static inline struct ext4_dir_entry_2 *
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500187ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
Li Zefanf795e142008-07-11 19:27:31 -0400188{
189 return (struct ext4_dir_entry_2 *)((char *)p +
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500190 ext4_rec_len_from_disk(p->rec_len, blocksize));
Li Zefanf795e142008-07-11 19:27:31 -0400191}
192
193/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700194 * Future: use high four bits of block for coalesce-on-delete flags
195 * Mask them off for now.
196 */
197
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500198static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700199{
200 return le32_to_cpu(entry->block) & 0x00ffffff;
201}
202
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500203static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700204{
205 entry->block = cpu_to_le32(value);
206}
207
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400208static inline unsigned dx_get_hash(struct dx_entry *entry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700209{
210 return le32_to_cpu(entry->hash);
211}
212
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400213static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700214{
215 entry->hash = cpu_to_le32(value);
216}
217
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400218static inline unsigned dx_get_count(struct dx_entry *entries)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700219{
220 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
221}
222
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400223static inline unsigned dx_get_limit(struct dx_entry *entries)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700224{
225 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
226}
227
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400228static inline void dx_set_count(struct dx_entry *entries, unsigned value)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700229{
230 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
231}
232
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400233static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700234{
235 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
236}
237
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400238static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700239{
Mingming Cao617ba132006-10-11 01:20:53 -0700240 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
241 EXT4_DIR_REC_LEN(2) - infosize;
Li Zefand9c769b2008-07-11 19:27:31 -0400242 return entry_space / sizeof(struct dx_entry);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700243}
244
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400245static inline unsigned dx_node_limit(struct inode *dir)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700246{
Mingming Cao617ba132006-10-11 01:20:53 -0700247 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
Li Zefand9c769b2008-07-11 19:27:31 -0400248 return entry_space / sizeof(struct dx_entry);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700249}
250
251/*
252 * Debug
253 */
254#ifdef DX_DEBUG
Theodore Ts'o47760042008-09-08 23:00:52 -0400255static void dx_show_index(char * label, struct dx_entry *entries)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700256{
Andrew Morton63f57932006-10-11 01:21:24 -0700257 int i, n = dx_get_count (entries);
Theodore Ts'o47760042008-09-08 23:00:52 -0400258 printk(KERN_DEBUG "%s index ", label);
Andrew Morton63f57932006-10-11 01:21:24 -0700259 for (i = 0; i < n; i++) {
Theodore Ts'o47760042008-09-08 23:00:52 -0400260 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500261 0, (unsigned long)dx_get_block(entries + i));
Andrew Morton63f57932006-10-11 01:21:24 -0700262 }
263 printk("\n");
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700264}
265
266struct stats
267{
268 unsigned names;
269 unsigned space;
270 unsigned bcount;
271};
272
Mingming Cao617ba132006-10-11 01:20:53 -0700273static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700274 int size, int show_names)
275{
276 unsigned names = 0, space = 0;
277 char *base = (char *) de;
278 struct dx_hash_info h = *hinfo;
279
280 printk("names: ");
281 while ((char *) de < base + size)
282 {
283 if (de->inode)
284 {
285 if (show_names)
286 {
287 int len = de->name_len;
288 char *name = de->name;
289 while (len--) printk("%c", *name++);
Mingming Cao617ba132006-10-11 01:20:53 -0700290 ext4fs_dirhash(de->name, de->name_len, &h);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700291 printk(":%x.%u ", h.hash,
292 ((char *) de - base));
293 }
Mingming Cao617ba132006-10-11 01:20:53 -0700294 space += EXT4_DIR_REC_LEN(de->name_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700295 names++;
296 }
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500297 de = ext4_next_entry(de, size);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700298 }
299 printk("(%i)\n", names);
300 return (struct stats) { names, space, 1 };
301}
302
303struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
304 struct dx_entry *entries, int levels)
305{
306 unsigned blocksize = dir->i_sb->s_blocksize;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400307 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700308 unsigned bcount = 0;
309 struct buffer_head *bh;
310 int err;
311 printk("%i indexed blocks...\n", count);
312 for (i = 0; i < count; i++, entries++)
313 {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500314 ext4_lblk_t block = dx_get_block(entries);
315 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700316 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
317 struct stats stats;
318 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
Mingming Cao617ba132006-10-11 01:20:53 -0700319 if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700320 stats = levels?
321 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
Mingming Cao617ba132006-10-11 01:20:53 -0700322 dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700323 names += stats.names;
324 space += stats.space;
325 bcount += stats.bcount;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400326 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700327 }
328 if (bcount)
Theodore Ts'o60e66792010-05-17 07:00:00 -0400329 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
Theodore Ts'o47760042008-09-08 23:00:52 -0400330 levels ? "" : " ", names, space/bcount,
331 (space/bcount)*100/blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700332 return (struct stats) { names, space, bcount};
333}
334#endif /* DX_DEBUG */
335
336/*
337 * Probe for a directory leaf block to search.
338 *
339 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
340 * error in the directory index, and the caller should fall back to
341 * searching the directory normally. The callers of dx_probe **MUST**
342 * check for this error code, and make sure it never gets reflected
343 * back to userspace.
344 */
345static struct dx_frame *
Theodore Ts'of702ba02008-09-22 15:21:01 -0400346dx_probe(const struct qstr *d_name, struct inode *dir,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700347 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
348{
349 unsigned count, indirect;
350 struct dx_entry *at, *entries, *p, *q, *m;
351 struct dx_root *root;
352 struct buffer_head *bh;
353 struct dx_frame *frame = frame_in;
354 u32 hash;
355
356 frame->bh = NULL;
Mingming Cao617ba132006-10-11 01:20:53 -0700357 if (!(bh = ext4_bread (NULL,dir, 0, 0, err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700358 goto fail;
359 root = (struct dx_root *) bh->b_data;
360 if (root->info.hash_version != DX_HASH_TEA &&
361 root->info.hash_version != DX_HASH_HALF_MD4 &&
362 root->info.hash_version != DX_HASH_LEGACY) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500363 ext4_warning(dir->i_sb, "Unrecognised inode hash code %d",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700364 root->info.hash_version);
365 brelse(bh);
366 *err = ERR_BAD_DX_DIR;
367 goto fail;
368 }
369 hinfo->hash_version = root->info.hash_version;
Theodore Ts'of99b2582008-10-28 13:21:44 -0400370 if (hinfo->hash_version <= DX_HASH_TEA)
371 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
Mingming Cao617ba132006-10-11 01:20:53 -0700372 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
Theodore Ts'of702ba02008-09-22 15:21:01 -0400373 if (d_name)
374 ext4fs_dirhash(d_name->name, d_name->len, hinfo);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700375 hash = hinfo->hash;
376
377 if (root->info.unused_flags & 1) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500378 ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700379 root->info.unused_flags);
380 brelse(bh);
381 *err = ERR_BAD_DX_DIR;
382 goto fail;
383 }
384
385 if ((indirect = root->info.indirect_levels) > 1) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500386 ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700387 root->info.indirect_levels);
388 brelse(bh);
389 *err = ERR_BAD_DX_DIR;
390 goto fail;
391 }
392
393 entries = (struct dx_entry *) (((char *)&root->info) +
394 root->info.info_length);
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700395
396 if (dx_get_limit(entries) != dx_root_limit(dir,
397 root->info.info_length)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500398 ext4_warning(dir->i_sb, "dx entry: limit != root limit");
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700399 brelse(bh);
400 *err = ERR_BAD_DX_DIR;
401 goto fail;
402 }
403
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400404 dxtrace(printk("Look up %x", hash));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700405 while (1)
406 {
407 count = dx_get_count(entries);
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700408 if (!count || count > dx_get_limit(entries)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500409 ext4_warning(dir->i_sb,
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700410 "dx entry: no count or count > limit");
411 brelse(bh);
412 *err = ERR_BAD_DX_DIR;
413 goto fail2;
414 }
415
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700416 p = entries + 1;
417 q = entries + count - 1;
418 while (p <= q)
419 {
420 m = p + (q - p)/2;
421 dxtrace(printk("."));
422 if (dx_get_hash(m) > hash)
423 q = m - 1;
424 else
425 p = m + 1;
426 }
427
428 if (0) // linear search cross check
429 {
430 unsigned n = count - 1;
431 at = entries;
432 while (n--)
433 {
434 dxtrace(printk(","));
435 if (dx_get_hash(++at) > hash)
436 {
437 at--;
438 break;
439 }
440 }
441 assert (at == p - 1);
442 }
443
444 at = p - 1;
445 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
446 frame->bh = bh;
447 frame->entries = entries;
448 frame->at = at;
449 if (!indirect--) return frame;
Mingming Cao617ba132006-10-11 01:20:53 -0700450 if (!(bh = ext4_bread (NULL,dir, dx_get_block(at), 0, err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700451 goto fail2;
452 at = entries = ((struct dx_node *) bh->b_data)->entries;
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700453 if (dx_get_limit(entries) != dx_node_limit (dir)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500454 ext4_warning(dir->i_sb,
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700455 "dx entry: limit != node limit");
456 brelse(bh);
457 *err = ERR_BAD_DX_DIR;
458 goto fail2;
459 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700460 frame++;
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700461 frame->bh = NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700462 }
463fail2:
464 while (frame >= frame_in) {
465 brelse(frame->bh);
466 frame--;
467 }
468fail:
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700469 if (*err == ERR_BAD_DX_DIR)
Eric Sandeen12062dd2010-02-15 14:19:27 -0500470 ext4_warning(dir->i_sb,
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700471 "Corrupt dir inode %ld, running e2fsck is "
472 "recommended.", dir->i_ino);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700473 return NULL;
474}
475
476static void dx_release (struct dx_frame *frames)
477{
478 if (frames[0].bh == NULL)
479 return;
480
481 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
482 brelse(frames[1].bh);
483 brelse(frames[0].bh);
484}
485
486/*
487 * This function increments the frame pointer to search the next leaf
488 * block, and reads in the necessary intervening nodes if the search
489 * should be necessary. Whether or not the search is necessary is
490 * controlled by the hash parameter. If the hash value is even, then
491 * the search is only continued if the next block starts with that
492 * hash value. This is used if we are searching for a specific file.
493 *
494 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
495 *
496 * This function returns 1 if the caller should continue to search,
497 * or 0 if it should not. If there is an error reading one of the
498 * index blocks, it will a negative error code.
499 *
500 * If start_hash is non-null, it will be filled in with the starting
501 * hash of the next page.
502 */
Mingming Cao617ba132006-10-11 01:20:53 -0700503static int ext4_htree_next_block(struct inode *dir, __u32 hash,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700504 struct dx_frame *frame,
505 struct dx_frame *frames,
506 __u32 *start_hash)
507{
508 struct dx_frame *p;
509 struct buffer_head *bh;
510 int err, num_frames = 0;
511 __u32 bhash;
512
513 p = frame;
514 /*
515 * Find the next leaf page by incrementing the frame pointer.
516 * If we run out of entries in the interior node, loop around and
517 * increment pointer in the parent node. When we break out of
518 * this loop, num_frames indicates the number of interior
519 * nodes need to be read.
520 */
521 while (1) {
522 if (++(p->at) < p->entries + dx_get_count(p->entries))
523 break;
524 if (p == frames)
525 return 0;
526 num_frames++;
527 p--;
528 }
529
530 /*
531 * If the hash is 1, then continue only if the next page has a
532 * continuation hash of any value. This is used for readdir
533 * handling. Otherwise, check to see if the hash matches the
534 * desired contiuation hash. If it doesn't, return since
535 * there's no point to read in the successive index pages.
536 */
537 bhash = dx_get_hash(p->at);
538 if (start_hash)
539 *start_hash = bhash;
540 if ((hash & 1) == 0) {
541 if ((bhash & ~1) != hash)
542 return 0;
543 }
544 /*
545 * If the hash is HASH_NB_ALWAYS, we always go to the next
546 * block so no check is necessary
547 */
548 while (num_frames--) {
Mingming Cao617ba132006-10-11 01:20:53 -0700549 if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at),
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700550 0, &err)))
551 return err; /* Failure */
552 p++;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400553 brelse(p->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700554 p->bh = bh;
555 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
556 }
557 return 1;
558}
559
560
561/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700562 * This function fills a red-black tree with information from a
563 * directory block. It returns the number directory entries loaded
564 * into the tree. If there is an error it is returned in err.
565 */
566static int htree_dirblock_to_tree(struct file *dir_file,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500567 struct inode *dir, ext4_lblk_t block,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700568 struct dx_hash_info *hinfo,
569 __u32 start_hash, __u32 start_minor_hash)
570{
571 struct buffer_head *bh;
Mingming Cao617ba132006-10-11 01:20:53 -0700572 struct ext4_dir_entry_2 *de, *top;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700573 int err, count = 0;
574
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500575 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
576 (unsigned long)block));
Mingming Cao617ba132006-10-11 01:20:53 -0700577 if (!(bh = ext4_bread (NULL, dir, block, 0, &err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700578 return err;
579
Mingming Cao617ba132006-10-11 01:20:53 -0700580 de = (struct ext4_dir_entry_2 *) bh->b_data;
581 top = (struct ext4_dir_entry_2 *) ((char *) de +
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700582 dir->i_sb->s_blocksize -
Mingming Cao617ba132006-10-11 01:20:53 -0700583 EXT4_DIR_REC_LEN(0));
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500584 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
Theodore Ts'of7c21172011-01-10 12:10:55 -0500585 if (ext4_check_dir_entry(dir, NULL, de, bh,
Theodore Ts'ocad3f002010-12-19 22:07:02 -0500586 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
587 + ((char *)de - bh->b_data))) {
Eric Sandeene6c40212006-12-06 20:36:28 -0800588 /* On error, skip the f_pos to the next block. */
589 dir_file->f_pos = (dir_file->f_pos |
590 (dir->i_sb->s_blocksize - 1)) + 1;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400591 brelse(bh);
Eric Sandeene6c40212006-12-06 20:36:28 -0800592 return count;
593 }
Mingming Cao617ba132006-10-11 01:20:53 -0700594 ext4fs_dirhash(de->name, de->name_len, hinfo);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700595 if ((hinfo->hash < start_hash) ||
596 ((hinfo->hash == start_hash) &&
597 (hinfo->minor_hash < start_minor_hash)))
598 continue;
599 if (de->inode == 0)
600 continue;
Mingming Cao617ba132006-10-11 01:20:53 -0700601 if ((err = ext4_htree_store_dirent(dir_file,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700602 hinfo->hash, hinfo->minor_hash, de)) != 0) {
603 brelse(bh);
604 return err;
605 }
606 count++;
607 }
608 brelse(bh);
609 return count;
610}
611
612
613/*
614 * This function fills a red-black tree with information from a
615 * directory. We start scanning the directory in hash order, starting
616 * at start_hash and start_minor_hash.
617 *
618 * This function returns the number of entries inserted into the tree,
619 * or a negative error code.
620 */
Mingming Cao617ba132006-10-11 01:20:53 -0700621int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700622 __u32 start_minor_hash, __u32 *next_hash)
623{
624 struct dx_hash_info hinfo;
Mingming Cao617ba132006-10-11 01:20:53 -0700625 struct ext4_dir_entry_2 *de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700626 struct dx_frame frames[2], *frame;
627 struct inode *dir;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500628 ext4_lblk_t block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700629 int count = 0;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500630 int ret, err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700631 __u32 hashval;
632
Theodore Ts'o60e66792010-05-17 07:00:00 -0400633 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
Theodore Ts'o47760042008-09-08 23:00:52 -0400634 start_hash, start_minor_hash));
Josef "Jeff" Sipek9d549892006-12-08 02:36:38 -0800635 dir = dir_file->f_path.dentry->d_inode;
Dmitry Monakhov12e9b892010-05-16 22:00:00 -0400636 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
Mingming Cao617ba132006-10-11 01:20:53 -0700637 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
Theodore Ts'of99b2582008-10-28 13:21:44 -0400638 if (hinfo.hash_version <= DX_HASH_TEA)
639 hinfo.hash_version +=
640 EXT4_SB(dir->i_sb)->s_hash_unsigned;
Mingming Cao617ba132006-10-11 01:20:53 -0700641 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700642 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
643 start_hash, start_minor_hash);
644 *next_hash = ~0;
645 return count;
646 }
647 hinfo.hash = start_hash;
648 hinfo.minor_hash = 0;
Theodore Ts'of702ba02008-09-22 15:21:01 -0400649 frame = dx_probe(NULL, dir, &hinfo, frames, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700650 if (!frame)
651 return err;
652
653 /* Add '.' and '..' from the htree header */
654 if (!start_hash && !start_minor_hash) {
Mingming Cao617ba132006-10-11 01:20:53 -0700655 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
656 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700657 goto errout;
658 count++;
659 }
660 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
Mingming Cao617ba132006-10-11 01:20:53 -0700661 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500662 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
Mingming Cao617ba132006-10-11 01:20:53 -0700663 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700664 goto errout;
665 count++;
666 }
667
668 while (1) {
669 block = dx_get_block(frame->at);
670 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
671 start_hash, start_minor_hash);
672 if (ret < 0) {
673 err = ret;
674 goto errout;
675 }
676 count += ret;
677 hashval = ~0;
Mingming Cao617ba132006-10-11 01:20:53 -0700678 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700679 frame, frames, &hashval);
680 *next_hash = hashval;
681 if (ret < 0) {
682 err = ret;
683 goto errout;
684 }
685 /*
686 * Stop if: (a) there are no more entries, or
687 * (b) we have inserted at least one entry and the
688 * next hash value is not a continuation
689 */
690 if ((ret == 0) ||
691 (count && ((hashval & 1) == 0)))
692 break;
693 }
694 dx_release(frames);
Theodore Ts'o47760042008-09-08 23:00:52 -0400695 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
696 "next hash: %x\n", count, *next_hash));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700697 return count;
698errout:
699 dx_release(frames);
700 return (err);
701}
702
703
704/*
705 * Directory block splitting, compacting
706 */
707
Eric Sandeenef2b02d2007-09-18 22:46:42 -0700708/*
709 * Create map of hash values, offsets, and sizes, stored at end of block.
710 * Returns number of entries mapped.
711 */
Theodore Ts'o8bad4592009-02-14 21:46:54 -0500712static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
713 struct dx_hash_info *hinfo,
714 struct dx_map_entry *map_tail)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700715{
716 int count = 0;
717 char *base = (char *) de;
718 struct dx_hash_info h = *hinfo;
719
Theodore Ts'o8bad4592009-02-14 21:46:54 -0500720 while ((char *) de < base + blocksize) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700721 if (de->name_len && de->inode) {
Mingming Cao617ba132006-10-11 01:20:53 -0700722 ext4fs_dirhash(de->name, de->name_len, &h);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700723 map_tail--;
724 map_tail->hash = h.hash;
Toshiyuki Okajima9aee2282009-06-08 12:41:35 -0400725 map_tail->offs = ((char *) de - base)>>2;
Eric Sandeenef2b02d2007-09-18 22:46:42 -0700726 map_tail->size = le16_to_cpu(de->rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700727 count++;
728 cond_resched();
729 }
730 /* XXX: do we need to check rec_len == 0 case? -Chris */
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500731 de = ext4_next_entry(de, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700732 }
733 return count;
734}
735
Eric Sandeenef2b02d2007-09-18 22:46:42 -0700736/* Sort map by hash value */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700737static void dx_sort_map (struct dx_map_entry *map, unsigned count)
738{
Andrew Morton63f57932006-10-11 01:21:24 -0700739 struct dx_map_entry *p, *q, *top = map + count - 1;
740 int more;
741 /* Combsort until bubble sort doesn't suck */
742 while (count > 2) {
743 count = count*10/13;
744 if (count - 9 < 2) /* 9, 10 -> 11 */
745 count = 11;
746 for (p = top, q = p - count; q >= map; p--, q--)
747 if (p->hash < q->hash)
748 swap(*p, *q);
749 }
750 /* Garden variety bubble sort */
751 do {
752 more = 0;
753 q = top;
754 while (q-- > map) {
755 if (q[1].hash >= q[0].hash)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700756 continue;
Andrew Morton63f57932006-10-11 01:21:24 -0700757 swap(*(q+1), *q);
758 more = 1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700759 }
760 } while(more);
761}
762
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500763static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700764{
765 struct dx_entry *entries = frame->entries;
766 struct dx_entry *old = frame->at, *new = old + 1;
767 int count = dx_get_count(entries);
768
769 assert(count < dx_get_limit(entries));
770 assert(old < entries + count);
771 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
772 dx_set_hash(new, hash);
773 dx_set_block(new, block);
774 dx_set_count(entries, count + 1);
775}
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700776
Mingming Cao617ba132006-10-11 01:20:53 -0700777static void ext4_update_dx_flag(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700778{
Mingming Cao617ba132006-10-11 01:20:53 -0700779 if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
780 EXT4_FEATURE_COMPAT_DIR_INDEX))
Dmitry Monakhov12e9b892010-05-16 22:00:00 -0400781 ext4_clear_inode_flag(inode, EXT4_INODE_INDEX);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700782}
783
784/*
Mingming Cao617ba132006-10-11 01:20:53 -0700785 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700786 *
Mingming Cao617ba132006-10-11 01:20:53 -0700787 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700788 * `de != NULL' is guaranteed by caller.
789 */
Mingming Cao617ba132006-10-11 01:20:53 -0700790static inline int ext4_match (int len, const char * const name,
791 struct ext4_dir_entry_2 * de)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700792{
793 if (len != de->name_len)
794 return 0;
795 if (!de->inode)
796 return 0;
797 return !memcmp(name, de->name, len);
798}
799
800/*
801 * Returns 0 if not found, -1 on failure, and 1 on success
802 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400803static inline int search_dirblock(struct buffer_head *bh,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700804 struct inode *dir,
Theodore Ts'of702ba02008-09-22 15:21:01 -0400805 const struct qstr *d_name,
Theodore Ts'o498e5f22008-11-05 00:14:04 -0500806 unsigned int offset,
Mingming Cao617ba132006-10-11 01:20:53 -0700807 struct ext4_dir_entry_2 ** res_dir)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700808{
Mingming Cao617ba132006-10-11 01:20:53 -0700809 struct ext4_dir_entry_2 * de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700810 char * dlimit;
811 int de_len;
Theodore Ts'of702ba02008-09-22 15:21:01 -0400812 const char *name = d_name->name;
813 int namelen = d_name->len;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700814
Mingming Cao617ba132006-10-11 01:20:53 -0700815 de = (struct ext4_dir_entry_2 *) bh->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700816 dlimit = bh->b_data + dir->i_sb->s_blocksize;
817 while ((char *) de < dlimit) {
818 /* this code is executed quadratically often */
819 /* do minimal checking `by hand' */
820
821 if ((char *) de + namelen <= dlimit &&
Mingming Cao617ba132006-10-11 01:20:53 -0700822 ext4_match (namelen, name, de)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700823 /* found a match - just to be sure, do a full check */
Theodore Ts'of7c21172011-01-10 12:10:55 -0500824 if (ext4_check_dir_entry(dir, NULL, de, bh, offset))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700825 return -1;
826 *res_dir = de;
827 return 1;
828 }
829 /* prevent looping on a bad block */
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500830 de_len = ext4_rec_len_from_disk(de->rec_len,
831 dir->i_sb->s_blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700832 if (de_len <= 0)
833 return -1;
834 offset += de_len;
Mingming Cao617ba132006-10-11 01:20:53 -0700835 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700836 }
837 return 0;
838}
839
840
841/*
Mingming Cao617ba132006-10-11 01:20:53 -0700842 * ext4_find_entry()
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700843 *
844 * finds an entry in the specified directory with the wanted name. It
845 * returns the cache buffer in which the entry was found, and the entry
846 * itself (as a parameter - res_dir). It does NOT read the inode of the
847 * entry - you'll have to do that yourself if you want to.
848 *
849 * The returned buffer_head has ->b_count elevated. The caller is expected
850 * to brelse() it when appropriate.
851 */
Theodore Ts'of702ba02008-09-22 15:21:01 -0400852static struct buffer_head * ext4_find_entry (struct inode *dir,
853 const struct qstr *d_name,
Mingming Cao617ba132006-10-11 01:20:53 -0700854 struct ext4_dir_entry_2 ** res_dir)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700855{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400856 struct super_block *sb;
857 struct buffer_head *bh_use[NAMEI_RA_SIZE];
858 struct buffer_head *bh, *ret = NULL;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500859 ext4_lblk_t start, block, b;
Theodore Ts'o8941ec82010-10-27 21:30:08 -0400860 const u8 *name = d_name->name;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700861 int ra_max = 0; /* Number of bh's in the readahead
862 buffer, bh_use[] */
863 int ra_ptr = 0; /* Current index into readahead
864 buffer */
865 int num = 0;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500866 ext4_lblk_t nblocks;
867 int i, err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700868 int namelen;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700869
870 *res_dir = NULL;
871 sb = dir->i_sb;
Theodore Ts'of702ba02008-09-22 15:21:01 -0400872 namelen = d_name->len;
Mingming Cao617ba132006-10-11 01:20:53 -0700873 if (namelen > EXT4_NAME_LEN)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700874 return NULL;
Theodore Ts'o8941ec82010-10-27 21:30:08 -0400875 if ((namelen <= 2) && (name[0] == '.') &&
Aaro Koskinen6d5c3aa2010-12-14 21:45:31 -0500876 (name[1] == '.' || name[1] == '\0')) {
Theodore Ts'o8941ec82010-10-27 21:30:08 -0400877 /*
878 * "." or ".." will only be in the first block
879 * NFS may look up ".."; "." should be handled by the VFS
880 */
881 block = start = 0;
882 nblocks = 1;
883 goto restart;
884 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700885 if (is_dx(dir)) {
Theodore Ts'of702ba02008-09-22 15:21:01 -0400886 bh = ext4_dx_find_entry(dir, d_name, res_dir, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700887 /*
888 * On success, or if the error was file not found,
889 * return. Otherwise, fall back to doing a search the
890 * old fashioned way.
891 */
892 if (bh || (err != ERR_BAD_DX_DIR))
893 return bh;
Theodore Ts'o47760042008-09-08 23:00:52 -0400894 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
895 "falling back\n"));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700896 }
Mingming Cao617ba132006-10-11 01:20:53 -0700897 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
898 start = EXT4_I(dir)->i_dir_start_lookup;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700899 if (start >= nblocks)
900 start = 0;
901 block = start;
902restart:
903 do {
904 /*
905 * We deal with the read-ahead logic here.
906 */
907 if (ra_ptr >= ra_max) {
908 /* Refill the readahead buffer */
909 ra_ptr = 0;
910 b = block;
911 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
912 /*
913 * Terminate if we reach the end of the
914 * directory and must wrap, or if our
915 * search has finished at this block.
916 */
917 if (b >= nblocks || (num && block == start)) {
918 bh_use[ra_max] = NULL;
919 break;
920 }
921 num++;
Mingming Cao617ba132006-10-11 01:20:53 -0700922 bh = ext4_getblk(NULL, dir, b++, 0, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700923 bh_use[ra_max] = bh;
924 if (bh)
925 ll_rw_block(READ_META, 1, &bh);
926 }
927 }
928 if ((bh = bh_use[ra_ptr++]) == NULL)
929 goto next;
930 wait_on_buffer(bh);
931 if (!buffer_uptodate(bh)) {
932 /* read error, skip block & hope for the best */
Theodore Ts'o24676da2010-05-16 21:00:00 -0400933 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
934 (unsigned long) block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700935 brelse(bh);
936 goto next;
937 }
Theodore Ts'of702ba02008-09-22 15:21:01 -0400938 i = search_dirblock(bh, dir, d_name,
Mingming Cao617ba132006-10-11 01:20:53 -0700939 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700940 if (i == 1) {
Mingming Cao617ba132006-10-11 01:20:53 -0700941 EXT4_I(dir)->i_dir_start_lookup = block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700942 ret = bh;
943 goto cleanup_and_exit;
944 } else {
945 brelse(bh);
946 if (i < 0)
947 goto cleanup_and_exit;
948 }
949 next:
950 if (++block >= nblocks)
951 block = 0;
952 } while (block != start);
953
954 /*
955 * If the directory has grown while we were searching, then
956 * search the last part of the directory before giving up.
957 */
958 block = nblocks;
Mingming Cao617ba132006-10-11 01:20:53 -0700959 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700960 if (block < nblocks) {
961 start = 0;
962 goto restart;
963 }
964
965cleanup_and_exit:
966 /* Clean up the read-ahead blocks */
967 for (; ra_ptr < ra_max; ra_ptr++)
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400968 brelse(bh_use[ra_ptr]);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700969 return ret;
970}
971
Theodore Ts'of702ba02008-09-22 15:21:01 -0400972static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
Mingming Cao617ba132006-10-11 01:20:53 -0700973 struct ext4_dir_entry_2 **res_dir, int *err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700974{
Theodore Ts'o8941ec82010-10-27 21:30:08 -0400975 struct super_block * sb = dir->i_sb;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700976 struct dx_hash_info hinfo;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700977 struct dx_frame frames[2], *frame;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700978 struct buffer_head *bh;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500979 ext4_lblk_t block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700980 int retval;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700981
Theodore Ts'o8941ec82010-10-27 21:30:08 -0400982 if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err)))
983 return NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700984 do {
985 block = dx_get_block(frame->at);
Theodore Ts'o7845c042010-10-27 21:30:08 -0400986 if (!(bh = ext4_bread(NULL, dir, block, 0, err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700987 goto errout;
Duane Griffinf3b35f02008-07-11 19:27:31 -0400988
Theodore Ts'o7845c042010-10-27 21:30:08 -0400989 retval = search_dirblock(bh, dir, d_name,
990 block << EXT4_BLOCK_SIZE_BITS(sb),
991 res_dir);
992 if (retval == 1) { /* Success! */
993 dx_release(frames);
994 return bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700995 }
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400996 brelse(bh);
Theodore Ts'o7845c042010-10-27 21:30:08 -0400997 if (retval == -1) {
998 *err = ERR_BAD_DX_DIR;
999 goto errout;
1000 }
1001
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001002 /* Check to see if we should continue to search */
Theodore Ts'o8941ec82010-10-27 21:30:08 -04001003 retval = ext4_htree_next_block(dir, hinfo.hash, frame,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001004 frames, NULL);
1005 if (retval < 0) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001006 ext4_warning(sb,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001007 "error reading index page in directory #%lu",
1008 dir->i_ino);
1009 *err = retval;
1010 goto errout;
1011 }
1012 } while (retval == 1);
1013
1014 *err = -ENOENT;
1015errout:
Theodore Ts'o47760042008-09-08 23:00:52 -04001016 dxtrace(printk(KERN_DEBUG "%s not found\n", name));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001017 dx_release (frames);
1018 return NULL;
1019}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001020
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001021static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001022{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001023 struct inode *inode;
1024 struct ext4_dir_entry_2 *de;
1025 struct buffer_head *bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001026
Mingming Cao617ba132006-10-11 01:20:53 -07001027 if (dentry->d_name.len > EXT4_NAME_LEN)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001028 return ERR_PTR(-ENAMETOOLONG);
1029
Theodore Ts'of702ba02008-09-22 15:21:01 -04001030 bh = ext4_find_entry(dir, &dentry->d_name, &de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001031 inode = NULL;
1032 if (bh) {
Theodore Ts'o498e5f22008-11-05 00:14:04 -05001033 __u32 ino = le32_to_cpu(de->inode);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001034 brelse(bh);
Mingming Cao617ba132006-10-11 01:20:53 -07001035 if (!ext4_valid_inum(dir->i_sb, ino)) {
Theodore Ts'o24676da2010-05-16 21:00:00 -04001036 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
David Howells1d1fe1e2008-02-07 00:15:37 -08001037 return ERR_PTR(-EIO);
Vasily Averina6c15c22007-07-15 23:40:46 -07001038 }
David Howells1d1fe1e2008-02-07 00:15:37 -08001039 inode = ext4_iget(dir->i_sb, ino);
Al Viroa9049372011-07-08 21:20:11 -04001040 if (inode == ERR_PTR(-ESTALE)) {
1041 EXT4_ERROR_INODE(dir,
1042 "deleted inode referenced: %u",
1043 ino);
1044 return ERR_PTR(-EIO);
Bryan Donlane6f009b2009-02-22 21:20:25 -05001045 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001046 }
1047 return d_splice_alias(inode, dentry);
1048}
1049
1050
Mingming Cao617ba132006-10-11 01:20:53 -07001051struct dentry *ext4_get_parent(struct dentry *child)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001052{
Theodore Ts'o498e5f22008-11-05 00:14:04 -05001053 __u32 ino;
Theodore Ts'of702ba02008-09-22 15:21:01 -04001054 static const struct qstr dotdot = {
1055 .name = "..",
1056 .len = 2,
1057 };
Mingming Cao617ba132006-10-11 01:20:53 -07001058 struct ext4_dir_entry_2 * de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001059 struct buffer_head *bh;
1060
Theodore Ts'of702ba02008-09-22 15:21:01 -04001061 bh = ext4_find_entry(child->d_inode, &dotdot, &de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001062 if (!bh)
1063 return ERR_PTR(-ENOENT);
1064 ino = le32_to_cpu(de->inode);
1065 brelse(bh);
1066
Mingming Cao617ba132006-10-11 01:20:53 -07001067 if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
Theodore Ts'o24676da2010-05-16 21:00:00 -04001068 EXT4_ERROR_INODE(child->d_inode,
1069 "bad parent inode number: %u", ino);
David Howells1d1fe1e2008-02-07 00:15:37 -08001070 return ERR_PTR(-EIO);
Vasily Averina6c15c22007-07-15 23:40:46 -07001071 }
1072
Christoph Hellwig44003722008-08-11 15:49:04 +02001073 return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001074}
1075
1076#define S_SHIFT 12
Mingming Cao617ba132006-10-11 01:20:53 -07001077static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1078 [S_IFREG >> S_SHIFT] = EXT4_FT_REG_FILE,
1079 [S_IFDIR >> S_SHIFT] = EXT4_FT_DIR,
1080 [S_IFCHR >> S_SHIFT] = EXT4_FT_CHRDEV,
1081 [S_IFBLK >> S_SHIFT] = EXT4_FT_BLKDEV,
1082 [S_IFIFO >> S_SHIFT] = EXT4_FT_FIFO,
1083 [S_IFSOCK >> S_SHIFT] = EXT4_FT_SOCK,
1084 [S_IFLNK >> S_SHIFT] = EXT4_FT_SYMLINK,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001085};
1086
Mingming Cao617ba132006-10-11 01:20:53 -07001087static inline void ext4_set_de_type(struct super_block *sb,
1088 struct ext4_dir_entry_2 *de,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001089 umode_t mode) {
Mingming Cao617ba132006-10-11 01:20:53 -07001090 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1091 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001092}
1093
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001094/*
1095 * Move count entries from end of map between two memory locations.
1096 * Returns pointer to last entry moved.
1097 */
Mingming Cao617ba132006-10-11 01:20:53 -07001098static struct ext4_dir_entry_2 *
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001099dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1100 unsigned blocksize)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001101{
1102 unsigned rec_len = 0;
1103
1104 while (count--) {
Theodore Ts'o60e66792010-05-17 07:00:00 -04001105 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
Toshiyuki Okajima9aee2282009-06-08 12:41:35 -04001106 (from + (map->offs<<2));
Mingming Cao617ba132006-10-11 01:20:53 -07001107 rec_len = EXT4_DIR_REC_LEN(de->name_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001108 memcpy (to, de, rec_len);
Mingming Cao617ba132006-10-11 01:20:53 -07001109 ((struct ext4_dir_entry_2 *) to)->rec_len =
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001110 ext4_rec_len_to_disk(rec_len, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001111 de->inode = 0;
1112 map++;
1113 to += rec_len;
1114 }
Mingming Cao617ba132006-10-11 01:20:53 -07001115 return (struct ext4_dir_entry_2 *) (to - rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001116}
1117
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001118/*
1119 * Compact each dir entry in the range to the minimal rec_len.
1120 * Returns pointer to last entry in range.
1121 */
Theodore Ts'o8bad4592009-02-14 21:46:54 -05001122static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001123{
Mingming Cao617ba132006-10-11 01:20:53 -07001124 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001125 unsigned rec_len = 0;
1126
1127 prev = to = de;
Theodore Ts'o8bad4592009-02-14 21:46:54 -05001128 while ((char*)de < base + blocksize) {
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001129 next = ext4_next_entry(de, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001130 if (de->inode && de->name_len) {
Mingming Cao617ba132006-10-11 01:20:53 -07001131 rec_len = EXT4_DIR_REC_LEN(de->name_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001132 if (de > to)
1133 memmove(to, de, rec_len);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001134 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001135 prev = to;
Mingming Cao617ba132006-10-11 01:20:53 -07001136 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001137 }
1138 de = next;
1139 }
1140 return prev;
1141}
1142
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001143/*
1144 * Split a full leaf block to make room for a new dir entry.
1145 * Allocate a new block, and move entries so that they are approx. equally full.
1146 * Returns pointer to de in block into which the new entry will be inserted.
1147 */
Mingming Cao617ba132006-10-11 01:20:53 -07001148static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001149 struct buffer_head **bh,struct dx_frame *frame,
1150 struct dx_hash_info *hinfo, int *error)
1151{
1152 unsigned blocksize = dir->i_sb->s_blocksize;
1153 unsigned count, continued;
1154 struct buffer_head *bh2;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001155 ext4_lblk_t newblock;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001156 u32 hash2;
1157 struct dx_map_entry *map;
1158 char *data1 = (*bh)->b_data, *data2;
Theodore Ts'o59e315b2008-12-06 16:58:39 -05001159 unsigned split, move, size;
Mingming Cao617ba132006-10-11 01:20:53 -07001160 struct ext4_dir_entry_2 *de = NULL, *de2;
Theodore Ts'o59e315b2008-12-06 16:58:39 -05001161 int err = 0, i;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001162
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001163 bh2 = ext4_append (handle, dir, &newblock, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001164 if (!(bh2)) {
1165 brelse(*bh);
1166 *bh = NULL;
1167 goto errout;
1168 }
1169
1170 BUFFER_TRACE(*bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001171 err = ext4_journal_get_write_access(handle, *bh);
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001172 if (err)
1173 goto journal_error;
1174
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001175 BUFFER_TRACE(frame->bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001176 err = ext4_journal_get_write_access(handle, frame->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001177 if (err)
1178 goto journal_error;
1179
1180 data2 = bh2->b_data;
1181
1182 /* create map in the end of data2 block */
1183 map = (struct dx_map_entry *) (data2 + blocksize);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001184 count = dx_make_map((struct ext4_dir_entry_2 *) data1,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001185 blocksize, hinfo, map);
1186 map -= count;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001187 dx_sort_map(map, count);
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001188 /* Split the existing block in the middle, size-wise */
1189 size = 0;
1190 move = 0;
1191 for (i = count-1; i >= 0; i--) {
1192 /* is more than half of this entry in 2nd half of the block? */
1193 if (size + map[i].size/2 > blocksize/2)
1194 break;
1195 size += map[i].size;
1196 move++;
1197 }
1198 /* map index at which we will split */
1199 split = count - move;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001200 hash2 = map[split].hash;
1201 continued = hash2 == map[split - 1].hash;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001202 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1203 (unsigned long)dx_get_block(frame->at),
1204 hash2, split, count-split));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001205
1206 /* Fancy dance to stay within two buffers */
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001207 de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001208 de = dx_pack_dirents(data1, blocksize);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001209 de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1210 blocksize);
1211 de2->rec_len = ext4_rec_len_to_disk(data2 + blocksize - (char *) de2,
1212 blocksize);
Mingming Cao617ba132006-10-11 01:20:53 -07001213 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1214 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001215
1216 /* Which block gets the new entry? */
1217 if (hinfo->hash >= hash2)
1218 {
1219 swap(*bh, bh2);
1220 de = de2;
1221 }
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001222 dx_insert_block(frame, hash2 + continued, newblock);
Frank Mayhar03901312009-01-07 00:06:22 -05001223 err = ext4_handle_dirty_metadata(handle, dir, bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001224 if (err)
1225 goto journal_error;
Frank Mayhar03901312009-01-07 00:06:22 -05001226 err = ext4_handle_dirty_metadata(handle, dir, frame->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001227 if (err)
1228 goto journal_error;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001229 brelse(bh2);
1230 dxtrace(dx_show_index("frame", frame->entries));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001231 return de;
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001232
1233journal_error:
1234 brelse(*bh);
1235 brelse(bh2);
1236 *bh = NULL;
1237 ext4_std_error(dir->i_sb, err);
1238errout:
1239 *error = err;
1240 return NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001241}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001242
1243/*
1244 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1245 * it points to a directory entry which is guaranteed to be large
1246 * enough for new directory entry. If de is NULL, then
1247 * add_dirent_to_buf will attempt search the directory block for
1248 * space. It will return -ENOSPC if no space is available, and -EIO
1249 * and -EEXIST if directory entry already exists.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001250 */
1251static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
Mingming Cao617ba132006-10-11 01:20:53 -07001252 struct inode *inode, struct ext4_dir_entry_2 *de,
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001253 struct buffer_head *bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001254{
1255 struct inode *dir = dentry->d_parent->d_inode;
1256 const char *name = dentry->d_name.name;
1257 int namelen = dentry->d_name.len;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05001258 unsigned int offset = 0;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001259 unsigned int blocksize = dir->i_sb->s_blocksize;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001260 unsigned short reclen;
1261 int nlen, rlen, err;
1262 char *top;
1263
Mingming Cao617ba132006-10-11 01:20:53 -07001264 reclen = EXT4_DIR_REC_LEN(namelen);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001265 if (!de) {
Mingming Cao617ba132006-10-11 01:20:53 -07001266 de = (struct ext4_dir_entry_2 *)bh->b_data;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001267 top = bh->b_data + blocksize - reclen;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001268 while ((char *) de <= top) {
Theodore Ts'of7c21172011-01-10 12:10:55 -05001269 if (ext4_check_dir_entry(dir, NULL, de, bh, offset))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001270 return -EIO;
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001271 if (ext4_match(namelen, name, de))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001272 return -EEXIST;
Mingming Cao617ba132006-10-11 01:20:53 -07001273 nlen = EXT4_DIR_REC_LEN(de->name_len);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001274 rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001275 if ((de->inode? rlen - nlen: rlen) >= reclen)
1276 break;
Mingming Cao617ba132006-10-11 01:20:53 -07001277 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001278 offset += rlen;
1279 }
1280 if ((char *) de > top)
1281 return -ENOSPC;
1282 }
1283 BUFFER_TRACE(bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001284 err = ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001285 if (err) {
Mingming Cao617ba132006-10-11 01:20:53 -07001286 ext4_std_error(dir->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001287 return err;
1288 }
1289
1290 /* By now the buffer is marked for journaling */
Mingming Cao617ba132006-10-11 01:20:53 -07001291 nlen = EXT4_DIR_REC_LEN(de->name_len);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001292 rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001293 if (de->inode) {
Mingming Cao617ba132006-10-11 01:20:53 -07001294 struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001295 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, blocksize);
1296 de->rec_len = ext4_rec_len_to_disk(nlen, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001297 de = de1;
1298 }
Mingming Cao617ba132006-10-11 01:20:53 -07001299 de->file_type = EXT4_FT_UNKNOWN;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001300 if (inode) {
1301 de->inode = cpu_to_le32(inode->i_ino);
Mingming Cao617ba132006-10-11 01:20:53 -07001302 ext4_set_de_type(dir->i_sb, de, inode->i_mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001303 } else
1304 de->inode = 0;
1305 de->name_len = namelen;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001306 memcpy(de->name, name, namelen);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001307 /*
1308 * XXX shouldn't update any times until successful
1309 * completion of syscall, but too many callers depend
1310 * on this.
1311 *
1312 * XXX similarly, too many callers depend on
Mingming Cao617ba132006-10-11 01:20:53 -07001313 * ext4_new_inode() setting the times, but error
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001314 * recovery deletes the inode, so the worst that can
1315 * happen is that the times are slightly out of date
1316 * and/or different from the directory change time.
1317 */
Kalpak Shahef7f3832007-07-18 09:15:20 -04001318 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
Mingming Cao617ba132006-10-11 01:20:53 -07001319 ext4_update_dx_flag(dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001320 dir->i_version++;
Mingming Cao617ba132006-10-11 01:20:53 -07001321 ext4_mark_inode_dirty(handle, dir);
Frank Mayhar03901312009-01-07 00:06:22 -05001322 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1323 err = ext4_handle_dirty_metadata(handle, dir, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001324 if (err)
Mingming Cao617ba132006-10-11 01:20:53 -07001325 ext4_std_error(dir->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001326 return 0;
1327}
1328
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001329/*
1330 * This converts a one block unindexed directory to a 3 block indexed
1331 * directory, and adds the dentry to the indexed directory.
1332 */
1333static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1334 struct inode *inode, struct buffer_head *bh)
1335{
1336 struct inode *dir = dentry->d_parent->d_inode;
1337 const char *name = dentry->d_name.name;
1338 int namelen = dentry->d_name.len;
1339 struct buffer_head *bh2;
1340 struct dx_root *root;
1341 struct dx_frame frames[2], *frame;
1342 struct dx_entry *entries;
Mingming Cao617ba132006-10-11 01:20:53 -07001343 struct ext4_dir_entry_2 *de, *de2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001344 char *data1, *top;
1345 unsigned len;
1346 int retval;
1347 unsigned blocksize;
1348 struct dx_hash_info hinfo;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001349 ext4_lblk_t block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001350 struct fake_dirent *fde;
1351
1352 blocksize = dir->i_sb->s_blocksize;
Theodore Ts'oe6b8bc02009-01-16 11:13:40 -05001353 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
Mingming Cao617ba132006-10-11 01:20:53 -07001354 retval = ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001355 if (retval) {
Mingming Cao617ba132006-10-11 01:20:53 -07001356 ext4_std_error(dir->i_sb, retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001357 brelse(bh);
1358 return retval;
1359 }
1360 root = (struct dx_root *) bh->b_data;
1361
Theodore Ts'oe6b8bc02009-01-16 11:13:40 -05001362 /* The 0th block becomes the root, move the dirents out */
1363 fde = &root->dotdot;
1364 de = (struct ext4_dir_entry_2 *)((char *)fde +
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001365 ext4_rec_len_from_disk(fde->rec_len, blocksize));
Theodore Ts'oe6b8bc02009-01-16 11:13:40 -05001366 if ((char *) de >= (((char *) root) + blocksize)) {
Theodore Ts'o24676da2010-05-16 21:00:00 -04001367 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
Theodore Ts'oe6b8bc02009-01-16 11:13:40 -05001368 brelse(bh);
1369 return -EIO;
1370 }
1371 len = ((char *) root) + blocksize - (char *) de;
1372
1373 /* Allocate new block for the 0th block's dirents */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001374 bh2 = ext4_append(handle, dir, &block, &retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001375 if (!(bh2)) {
1376 brelse(bh);
1377 return retval;
1378 }
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04001379 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001380 data1 = bh2->b_data;
1381
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001382 memcpy (data1, de, len);
Mingming Cao617ba132006-10-11 01:20:53 -07001383 de = (struct ext4_dir_entry_2 *) data1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001384 top = data1 + len;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001385 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001386 de = de2;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001387 de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1388 blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001389 /* Initialize the root; the dot dirents already exist */
Mingming Cao617ba132006-10-11 01:20:53 -07001390 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001391 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1392 blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001393 memset (&root->info, 0, sizeof(root->info));
1394 root->info.info_length = sizeof(root->info);
Mingming Cao617ba132006-10-11 01:20:53 -07001395 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001396 entries = root->entries;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001397 dx_set_block(entries, 1);
1398 dx_set_count(entries, 1);
1399 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001400
1401 /* Initialize as for dx_probe */
1402 hinfo.hash_version = root->info.hash_version;
Theodore Ts'of99b2582008-10-28 13:21:44 -04001403 if (hinfo.hash_version <= DX_HASH_TEA)
1404 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
Mingming Cao617ba132006-10-11 01:20:53 -07001405 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1406 ext4fs_dirhash(name, namelen, &hinfo);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001407 frame = frames;
1408 frame->entries = entries;
1409 frame->at = entries;
1410 frame->bh = bh;
1411 bh = bh2;
Allison Henderson6976a6f2011-05-15 00:19:41 -04001412
1413 ext4_handle_dirty_metadata(handle, dir, frame->bh);
1414 ext4_handle_dirty_metadata(handle, dir, bh);
1415
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001416 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
Jan Kara7ad8e4e2011-05-03 11:05:55 -04001417 if (!de) {
1418 /*
1419 * Even if the block split failed, we have to properly write
1420 * out all the changes we did so far. Otherwise we can end up
1421 * with corrupted filesystem.
1422 */
1423 ext4_mark_inode_dirty(handle, dir);
Jan Kara7ad8e4e2011-05-03 11:05:55 -04001424 dx_release(frames);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001425 return retval;
Jan Kara7ad8e4e2011-05-03 11:05:55 -04001426 }
1427 dx_release(frames);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001428
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'o47760042008-09-08 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'o47760042008-09-08 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'o47760042008-09-08 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 }
Theodore Ts'ob4097142011-01-10 12:46:59 -05001614 err = ext4_handle_dirty_metadata(handle, inode, frames[0].bh);
1615 if (err) {
1616 ext4_std_error(inode->i_sb, err);
1617 goto cleanup;
1618 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001619 }
1620 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1621 if (!de)
1622 goto cleanup;
1623 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001624 goto cleanup;
1625
1626journal_error:
Mingming Cao617ba132006-10-11 01:20:53 -07001627 ext4_std_error(dir->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001628cleanup:
1629 if (bh)
1630 brelse(bh);
1631 dx_release(frames);
1632 return err;
1633}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001634
1635/*
Mingming Cao617ba132006-10-11 01:20:53 -07001636 * ext4_delete_entry deletes a directory entry by merging it with the
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001637 * previous entry
1638 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001639static int ext4_delete_entry(handle_t *handle,
1640 struct inode *dir,
1641 struct ext4_dir_entry_2 *de_del,
1642 struct buffer_head *bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001643{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001644 struct ext4_dir_entry_2 *de, *pde;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001645 unsigned int blocksize = dir->i_sb->s_blocksize;
Theodore Ts'ob4097142011-01-10 12:46:59 -05001646 int i, err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001647
1648 i = 0;
1649 pde = NULL;
Mingming Cao617ba132006-10-11 01:20:53 -07001650 de = (struct ext4_dir_entry_2 *) bh->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001651 while (i < bh->b_size) {
Theodore Ts'of7c21172011-01-10 12:10:55 -05001652 if (ext4_check_dir_entry(dir, NULL, de, bh, i))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001653 return -EIO;
1654 if (de == de_del) {
1655 BUFFER_TRACE(bh, "get_write_access");
Theodore Ts'ob4097142011-01-10 12:46:59 -05001656 err = ext4_journal_get_write_access(handle, bh);
1657 if (unlikely(err)) {
1658 ext4_std_error(dir->i_sb, err);
1659 return err;
1660 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001661 if (pde)
Jan Karaa72d7f82008-01-28 23:58:27 -05001662 pde->rec_len = ext4_rec_len_to_disk(
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001663 ext4_rec_len_from_disk(pde->rec_len,
1664 blocksize) +
1665 ext4_rec_len_from_disk(de->rec_len,
1666 blocksize),
1667 blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001668 else
1669 de->inode = 0;
1670 dir->i_version++;
Frank Mayhar03901312009-01-07 00:06:22 -05001671 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
Theodore Ts'ob4097142011-01-10 12:46:59 -05001672 err = ext4_handle_dirty_metadata(handle, dir, bh);
1673 if (unlikely(err)) {
1674 ext4_std_error(dir->i_sb, err);
1675 return err;
1676 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001677 return 0;
1678 }
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001679 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001680 pde = de;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001681 de = ext4_next_entry(de, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001682 }
1683 return -ENOENT;
1684}
1685
Andreas Dilgerf8628a12007-07-18 08:38:01 -04001686/*
1687 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
1688 * since this indicates that nlinks count was previously 1.
1689 */
1690static void ext4_inc_count(handle_t *handle, struct inode *inode)
1691{
1692 inc_nlink(inode);
1693 if (is_dx(inode) && inode->i_nlink > 1) {
1694 /* limit is 16-bit i_links_count */
1695 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
1696 inode->i_nlink = 1;
1697 EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
1698 EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
1699 }
1700 }
1701}
1702
1703/*
1704 * If a directory had nlink == 1, then we should let it be 1. This indicates
1705 * directory has >EXT4_LINK_MAX subdirs.
1706 */
1707static void ext4_dec_count(handle_t *handle, struct inode *inode)
1708{
1709 drop_nlink(inode);
1710 if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
1711 inc_nlink(inode);
1712}
1713
1714
Mingming Cao617ba132006-10-11 01:20:53 -07001715static int ext4_add_nondir(handle_t *handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001716 struct dentry *dentry, struct inode *inode)
1717{
Mingming Cao617ba132006-10-11 01:20:53 -07001718 int err = ext4_add_entry(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001719 if (!err) {
Mingming Cao617ba132006-10-11 01:20:53 -07001720 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001721 d_instantiate(dentry, inode);
Al Viro6b38e842008-12-30 02:03:31 -05001722 unlock_new_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001723 return 0;
1724 }
Eric Sandeen731b9a52007-02-10 01:46:16 -08001725 drop_nlink(inode);
Al Viro6b38e842008-12-30 02:03:31 -05001726 unlock_new_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001727 iput(inode);
1728 return err;
1729}
1730
1731/*
1732 * By the time this is called, we already have created
1733 * the directory cache entry for the new file, but it
1734 * is so far negative - it has no inode.
1735 *
1736 * If the create succeeds, we fill in the inode information
1737 * with d_instantiate().
1738 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001739static int ext4_create(struct inode *dir, struct dentry *dentry, int mode,
1740 struct nameidata *nd)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001741{
1742 handle_t *handle;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001743 struct inode *inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001744 int err, retries = 0;
1745
Christoph Hellwig871a2932010-03-03 09:05:07 -05001746 dquot_initialize(dir);
Christoph Hellwig907f4552010-03-03 09:05:06 -05001747
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001748retry:
Mingming Cao617ba132006-10-11 01:20:53 -07001749 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1750 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
Dmitry Monakhov5aca07e2009-12-08 22:42:15 -05001751 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001752 if (IS_ERR(handle))
1753 return PTR_ERR(handle);
1754
1755 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05001756 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001757
Andreas Dilger11013912009-06-13 11:45:35 -04001758 inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001759 err = PTR_ERR(inode);
1760 if (!IS_ERR(inode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07001761 inode->i_op = &ext4_file_inode_operations;
1762 inode->i_fop = &ext4_file_operations;
1763 ext4_set_aops(inode);
1764 err = ext4_add_nondir(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001765 }
Mingming Cao617ba132006-10-11 01:20:53 -07001766 ext4_journal_stop(handle);
1767 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001768 goto retry;
1769 return err;
1770}
1771
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001772static int ext4_mknod(struct inode *dir, struct dentry *dentry,
1773 int mode, dev_t rdev)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001774{
1775 handle_t *handle;
1776 struct inode *inode;
1777 int err, retries = 0;
1778
1779 if (!new_valid_dev(rdev))
1780 return -EINVAL;
1781
Christoph Hellwig871a2932010-03-03 09:05:07 -05001782 dquot_initialize(dir);
Christoph Hellwig907f4552010-03-03 09:05:06 -05001783
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001784retry:
Mingming Cao617ba132006-10-11 01:20:53 -07001785 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1786 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
Dmitry Monakhov5aca07e2009-12-08 22:42:15 -05001787 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001788 if (IS_ERR(handle))
1789 return PTR_ERR(handle);
1790
1791 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05001792 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001793
Andreas Dilger11013912009-06-13 11:45:35 -04001794 inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001795 err = PTR_ERR(inode);
1796 if (!IS_ERR(inode)) {
1797 init_special_inode(inode, inode->i_mode, rdev);
Theodore Ts'o03010a32008-10-10 20:02:48 -04001798#ifdef CONFIG_EXT4_FS_XATTR
Mingming Cao617ba132006-10-11 01:20:53 -07001799 inode->i_op = &ext4_special_inode_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001800#endif
Mingming Cao617ba132006-10-11 01:20:53 -07001801 err = ext4_add_nondir(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001802 }
Mingming Cao617ba132006-10-11 01:20:53 -07001803 ext4_journal_stop(handle);
1804 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001805 goto retry;
1806 return err;
1807}
1808
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001809static int ext4_mkdir(struct inode *dir, struct dentry *dentry, int mode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001810{
1811 handle_t *handle;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001812 struct inode *inode;
Namhyung Kimdabd9912011-01-10 12:11:16 -05001813 struct buffer_head *dir_block = NULL;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001814 struct ext4_dir_entry_2 *de;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001815 unsigned int blocksize = dir->i_sb->s_blocksize;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001816 int err, retries = 0;
1817
Andreas Dilgerf8628a12007-07-18 08:38:01 -04001818 if (EXT4_DIR_LINK_MAX(dir))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001819 return -EMLINK;
1820
Christoph Hellwig871a2932010-03-03 09:05:07 -05001821 dquot_initialize(dir);
Christoph Hellwig907f4552010-03-03 09:05:06 -05001822
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001823retry:
Mingming Cao617ba132006-10-11 01:20:53 -07001824 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1825 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
Dmitry Monakhov5aca07e2009-12-08 22:42:15 -05001826 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001827 if (IS_ERR(handle))
1828 return PTR_ERR(handle);
1829
1830 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05001831 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001832
Andreas Dilger11013912009-06-13 11:45:35 -04001833 inode = ext4_new_inode(handle, dir, S_IFDIR | mode,
1834 &dentry->d_name, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001835 err = PTR_ERR(inode);
1836 if (IS_ERR(inode))
1837 goto out_stop;
1838
Mingming Cao617ba132006-10-11 01:20:53 -07001839 inode->i_op = &ext4_dir_inode_operations;
1840 inode->i_fop = &ext4_dir_operations;
1841 inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001842 dir_block = ext4_bread(handle, inode, 0, 1, &err);
Aneesh Kumar K.V4cdeed82008-02-22 06:17:31 -05001843 if (!dir_block)
1844 goto out_clear_inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001845 BUFFER_TRACE(dir_block, "get_write_access");
Namhyung Kimdabd9912011-01-10 12:11:16 -05001846 err = ext4_journal_get_write_access(handle, dir_block);
1847 if (err)
1848 goto out_clear_inode;
Mingming Cao617ba132006-10-11 01:20:53 -07001849 de = (struct ext4_dir_entry_2 *) dir_block->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001850 de->inode = cpu_to_le32(inode->i_ino);
1851 de->name_len = 1;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001852 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
1853 blocksize);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001854 strcpy(de->name, ".");
Mingming Cao617ba132006-10-11 01:20:53 -07001855 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001856 de = ext4_next_entry(de, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001857 de->inode = cpu_to_le32(dir->i_ino);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001858 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(1),
1859 blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001860 de->name_len = 2;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001861 strcpy(de->name, "..");
Mingming Cao617ba132006-10-11 01:20:53 -07001862 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001863 inode->i_nlink = 2;
Frank Mayhar03901312009-01-07 00:06:22 -05001864 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
Namhyung Kimdabd9912011-01-10 12:11:16 -05001865 err = ext4_handle_dirty_metadata(handle, dir, dir_block);
1866 if (err)
1867 goto out_clear_inode;
1868 err = ext4_mark_inode_dirty(handle, inode);
1869 if (!err)
1870 err = ext4_add_entry(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001871 if (err) {
Aneesh Kumar K.V4cdeed82008-02-22 06:17:31 -05001872out_clear_inode:
1873 clear_nlink(inode);
Al Viro6b38e842008-12-30 02:03:31 -05001874 unlock_new_inode(inode);
Mingming Cao617ba132006-10-11 01:20:53 -07001875 ext4_mark_inode_dirty(handle, inode);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001876 iput(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001877 goto out_stop;
1878 }
Andreas Dilgerf8628a12007-07-18 08:38:01 -04001879 ext4_inc_count(handle, dir);
Mingming Cao617ba132006-10-11 01:20:53 -07001880 ext4_update_dx_flag(dir);
Namhyung Kimdabd9912011-01-10 12:11:16 -05001881 err = ext4_mark_inode_dirty(handle, dir);
1882 if (err)
1883 goto out_clear_inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001884 d_instantiate(dentry, inode);
Al Viro6b38e842008-12-30 02:03:31 -05001885 unlock_new_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001886out_stop:
Namhyung Kimdabd9912011-01-10 12:11:16 -05001887 brelse(dir_block);
Mingming Cao617ba132006-10-11 01:20:53 -07001888 ext4_journal_stop(handle);
1889 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001890 goto retry;
1891 return err;
1892}
1893
1894/*
1895 * routine to check that the specified directory is empty (for rmdir)
1896 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001897static int empty_dir(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001898{
Theodore Ts'o498e5f22008-11-05 00:14:04 -05001899 unsigned int offset;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001900 struct buffer_head *bh;
1901 struct ext4_dir_entry_2 *de, *de1;
1902 struct super_block *sb;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001903 int err = 0;
1904
1905 sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -07001906 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001907 !(bh = ext4_bread(NULL, inode, 0, 0, &err))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001908 if (err)
Theodore Ts'o24676da2010-05-16 21:00:00 -04001909 EXT4_ERROR_INODE(inode,
1910 "error %d reading directory lblock 0", err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001911 else
Eric Sandeen12062dd2010-02-15 14:19:27 -05001912 ext4_warning(inode->i_sb,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001913 "bad directory (dir #%lu) - no data block",
1914 inode->i_ino);
1915 return 1;
1916 }
Mingming Cao617ba132006-10-11 01:20:53 -07001917 de = (struct ext4_dir_entry_2 *) bh->b_data;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001918 de1 = ext4_next_entry(de, sb->s_blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001919 if (le32_to_cpu(de->inode) != inode->i_ino ||
1920 !le32_to_cpu(de1->inode) ||
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001921 strcmp(".", de->name) ||
1922 strcmp("..", de1->name)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001923 ext4_warning(inode->i_sb,
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001924 "bad directory (dir #%lu) - no `.' or `..'",
1925 inode->i_ino);
1926 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001927 return 1;
1928 }
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001929 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
1930 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
1931 de = ext4_next_entry(de1, sb->s_blocksize);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001932 while (offset < inode->i_size) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001933 if (!bh ||
Theodore Ts'o24676da2010-05-16 21:00:00 -04001934 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1935 unsigned int lblock;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001936 err = 0;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001937 brelse(bh);
Theodore Ts'o24676da2010-05-16 21:00:00 -04001938 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
1939 bh = ext4_bread(NULL, inode, lblock, 0, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001940 if (!bh) {
1941 if (err)
Theodore Ts'o24676da2010-05-16 21:00:00 -04001942 EXT4_ERROR_INODE(inode,
1943 "error %d reading directory "
1944 "lblock %u", err, lblock);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001945 offset += sb->s_blocksize;
1946 continue;
1947 }
Mingming Cao617ba132006-10-11 01:20:53 -07001948 de = (struct ext4_dir_entry_2 *) bh->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001949 }
Theodore Ts'of7c21172011-01-10 12:10:55 -05001950 if (ext4_check_dir_entry(inode, NULL, de, bh, offset)) {
Mingming Cao617ba132006-10-11 01:20:53 -07001951 de = (struct ext4_dir_entry_2 *)(bh->b_data +
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001952 sb->s_blocksize);
1953 offset = (offset | (sb->s_blocksize - 1)) + 1;
1954 continue;
1955 }
1956 if (le32_to_cpu(de->inode)) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001957 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001958 return 0;
1959 }
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001960 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
1961 de = ext4_next_entry(de, sb->s_blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001962 }
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001963 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001964 return 1;
1965}
1966
Mingming Cao617ba132006-10-11 01:20:53 -07001967/* ext4_orphan_add() links an unlinked or truncated inode into a list of
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001968 * such inodes, starting at the superblock, in case we crash before the
1969 * file is closed/deleted, or in case the inode truncate spans multiple
1970 * transactions and the last transaction is not recovered after a crash.
1971 *
1972 * At filesystem recovery time, we walk this list deleting unlinked
Mingming Cao617ba132006-10-11 01:20:53 -07001973 * inodes and truncating linked inodes in ext4_orphan_cleanup().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001974 */
Mingming Cao617ba132006-10-11 01:20:53 -07001975int ext4_orphan_add(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001976{
1977 struct super_block *sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -07001978 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001979 int err = 0, rc;
1980
Frank Mayhar03901312009-01-07 00:06:22 -05001981 if (!ext4_handle_valid(handle))
1982 return 0;
1983
Theodore Ts'o3b9d4ed2009-04-25 22:54:04 -04001984 mutex_lock(&EXT4_SB(sb)->s_orphan_lock);
Mingming Cao617ba132006-10-11 01:20:53 -07001985 if (!list_empty(&EXT4_I(inode)->i_orphan))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001986 goto out_unlock;
1987
1988 /* Orphan handling is only valid for files with data blocks
1989 * being truncated, or files being unlinked. */
1990
1991 /* @@@ FIXME: Observation from aviro:
Mingming Cao617ba132006-10-11 01:20:53 -07001992 * I think I can trigger J_ASSERT in ext4_orphan_add(). We block
Theodore Ts'o3b9d4ed2009-04-25 22:54:04 -04001993 * here (on s_orphan_lock), so race with ext4_link() which might bump
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001994 * ->i_nlink. For, say it, character device. Not a regular file,
1995 * not a directory, not a symlink and ->i_nlink > 0.
Theodore Ts'o3b9d4ed2009-04-25 22:54:04 -04001996 *
1997 * tytso, 4/25/2009: I'm not sure how that could happen;
1998 * shouldn't the fs core protect us from these sort of
1999 * unlink()/link() races?
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002000 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002001 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2002 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002003
Mingming Cao617ba132006-10-11 01:20:53 -07002004 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
2005 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002006 if (err)
2007 goto out_unlock;
2008
Mingming Cao617ba132006-10-11 01:20:53 -07002009 err = ext4_reserve_inode_write(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002010 if (err)
2011 goto out_unlock;
Dmitry Monakhov6e3617e2010-03-01 23:29:39 -05002012 /*
2013 * Due to previous errors inode may be already a part of on-disk
2014 * orphan list. If so skip on-disk list modification.
2015 */
2016 if (NEXT_ORPHAN(inode) && NEXT_ORPHAN(inode) <=
2017 (le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)))
2018 goto mem_insert;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002019
2020 /* Insert this inode at the head of the on-disk orphan list... */
Mingming Cao617ba132006-10-11 01:20:53 -07002021 NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
2022 EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
Curt Wohlgemuth73b50c12010-02-16 15:06:29 -05002023 err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
Mingming Cao617ba132006-10-11 01:20:53 -07002024 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002025 if (!err)
2026 err = rc;
2027
2028 /* Only add to the head of the in-memory list if all the
2029 * previous operations succeeded. If the orphan_add is going to
2030 * fail (possibly taking the journal offline), we can't risk
2031 * leaving the inode on the orphan list: stray orphan-list
2032 * entries can cause panics at unmount time.
2033 *
2034 * This is safe: on error we're going to ignore the orphan list
2035 * anyway on the next recovery. */
Dmitry Monakhov6e3617e2010-03-01 23:29:39 -05002036mem_insert:
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002037 if (!err)
Mingming Cao617ba132006-10-11 01:20:53 -07002038 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002039
2040 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2041 jbd_debug(4, "orphan inode %lu will point to %d\n",
2042 inode->i_ino, NEXT_ORPHAN(inode));
2043out_unlock:
Theodore Ts'o3b9d4ed2009-04-25 22:54:04 -04002044 mutex_unlock(&EXT4_SB(sb)->s_orphan_lock);
Mingming Cao617ba132006-10-11 01:20:53 -07002045 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002046 return err;
2047}
2048
2049/*
Mingming Cao617ba132006-10-11 01:20:53 -07002050 * ext4_orphan_del() removes an unlinked or truncated inode from the list
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002051 * of such inodes stored on disk, because it is finally being cleaned up.
2052 */
Mingming Cao617ba132006-10-11 01:20:53 -07002053int ext4_orphan_del(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002054{
2055 struct list_head *prev;
Mingming Cao617ba132006-10-11 01:20:53 -07002056 struct ext4_inode_info *ei = EXT4_I(inode);
2057 struct ext4_sb_info *sbi;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05002058 __u32 ino_next;
Mingming Cao617ba132006-10-11 01:20:53 -07002059 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002060 int err = 0;
2061
Curt Wohlgemuthd3d1faf2009-09-29 11:01:03 -04002062 /* ext4_handle_valid() assumes a valid handle_t pointer */
2063 if (handle && !ext4_handle_valid(handle))
Frank Mayhar03901312009-01-07 00:06:22 -05002064 return 0;
2065
Theodore Ts'o3b9d4ed2009-04-25 22:54:04 -04002066 mutex_lock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2067 if (list_empty(&ei->i_orphan))
2068 goto out;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002069
2070 ino_next = NEXT_ORPHAN(inode);
2071 prev = ei->i_orphan.prev;
Mingming Cao617ba132006-10-11 01:20:53 -07002072 sbi = EXT4_SB(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002073
2074 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2075
2076 list_del_init(&ei->i_orphan);
2077
2078 /* If we're on an error path, we may not have a valid
2079 * transaction handle with which to update the orphan list on
2080 * disk, but we still need to remove the inode from the linked
2081 * list in memory. */
Frank Mayhar03901312009-01-07 00:06:22 -05002082 if (sbi->s_journal && !handle)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002083 goto out;
2084
Mingming Cao617ba132006-10-11 01:20:53 -07002085 err = ext4_reserve_inode_write(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002086 if (err)
2087 goto out_err;
2088
2089 if (prev == &sbi->s_orphan) {
Theodore Ts'o498e5f22008-11-05 00:14:04 -05002090 jbd_debug(4, "superblock will point to %u\n", ino_next);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002091 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07002092 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002093 if (err)
2094 goto out_brelse;
2095 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
Curt Wohlgemuth73b50c12010-02-16 15:06:29 -05002096 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002097 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07002098 struct ext4_iloc iloc2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002099 struct inode *i_prev =
Mingming Cao617ba132006-10-11 01:20:53 -07002100 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002101
Theodore Ts'o498e5f22008-11-05 00:14:04 -05002102 jbd_debug(4, "orphan inode %lu will point to %u\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002103 i_prev->i_ino, ino_next);
Mingming Cao617ba132006-10-11 01:20:53 -07002104 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002105 if (err)
2106 goto out_brelse;
2107 NEXT_ORPHAN(i_prev) = ino_next;
Mingming Cao617ba132006-10-11 01:20:53 -07002108 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002109 }
2110 if (err)
2111 goto out_brelse;
2112 NEXT_ORPHAN(inode) = 0;
Mingming Cao617ba132006-10-11 01:20:53 -07002113 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002114
2115out_err:
Mingming Cao617ba132006-10-11 01:20:53 -07002116 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002117out:
Theodore Ts'o3b9d4ed2009-04-25 22:54:04 -04002118 mutex_unlock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002119 return err;
2120
2121out_brelse:
2122 brelse(iloc.bh);
2123 goto out_err;
2124}
2125
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002126static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002127{
2128 int retval;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002129 struct inode *inode;
2130 struct buffer_head *bh;
2131 struct ext4_dir_entry_2 *de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002132 handle_t *handle;
2133
2134 /* Initialize quotas before so that eventual writes go in
2135 * separate transaction */
Christoph Hellwig871a2932010-03-03 09:05:07 -05002136 dquot_initialize(dir);
2137 dquot_initialize(dentry->d_inode);
Christoph Hellwig907f4552010-03-03 09:05:06 -05002138
Mingming Cao617ba132006-10-11 01:20:53 -07002139 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002140 if (IS_ERR(handle))
2141 return PTR_ERR(handle);
2142
2143 retval = -ENOENT;
Theodore Ts'of702ba02008-09-22 15:21:01 -04002144 bh = ext4_find_entry(dir, &dentry->d_name, &de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002145 if (!bh)
2146 goto end_rmdir;
2147
2148 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05002149 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002150
2151 inode = dentry->d_inode;
2152
2153 retval = -EIO;
2154 if (le32_to_cpu(de->inode) != inode->i_ino)
2155 goto end_rmdir;
2156
2157 retval = -ENOTEMPTY;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002158 if (!empty_dir(inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002159 goto end_rmdir;
2160
Mingming Cao617ba132006-10-11 01:20:53 -07002161 retval = ext4_delete_entry(handle, dir, de, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002162 if (retval)
2163 goto end_rmdir;
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002164 if (!EXT4_DIR_LINK_EMPTY(inode))
Eric Sandeen12062dd2010-02-15 14:19:27 -05002165 ext4_warning(inode->i_sb,
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002166 "empty directory has too many links (%d)",
2167 inode->i_nlink);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002168 inode->i_version++;
2169 clear_nlink(inode);
2170 /* There's no need to set i_disksize: the fact that i_nlink is
2171 * zero will ensure that the right thing happens during any
2172 * recovery. */
2173 inode->i_size = 0;
Mingming Cao617ba132006-10-11 01:20:53 -07002174 ext4_orphan_add(handle, inode);
Kalpak Shahef7f3832007-07-18 09:15:20 -04002175 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002176 ext4_mark_inode_dirty(handle, inode);
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002177 ext4_dec_count(handle, dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002178 ext4_update_dx_flag(dir);
2179 ext4_mark_inode_dirty(handle, dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002180
2181end_rmdir:
Mingming Cao617ba132006-10-11 01:20:53 -07002182 ext4_journal_stop(handle);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002183 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002184 return retval;
2185}
2186
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002187static int ext4_unlink(struct inode *dir, struct dentry *dentry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002188{
2189 int retval;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002190 struct inode *inode;
2191 struct buffer_head *bh;
2192 struct ext4_dir_entry_2 *de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002193 handle_t *handle;
2194
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04002195 trace_ext4_unlink_enter(dir, dentry);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002196 /* Initialize quotas before so that eventual writes go
2197 * in separate transaction */
Christoph Hellwig871a2932010-03-03 09:05:07 -05002198 dquot_initialize(dir);
2199 dquot_initialize(dentry->d_inode);
Christoph Hellwig907f4552010-03-03 09:05:06 -05002200
Mingming Cao617ba132006-10-11 01:20:53 -07002201 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002202 if (IS_ERR(handle))
2203 return PTR_ERR(handle);
2204
2205 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05002206 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002207
2208 retval = -ENOENT;
Theodore Ts'of702ba02008-09-22 15:21:01 -04002209 bh = ext4_find_entry(dir, &dentry->d_name, &de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002210 if (!bh)
2211 goto end_unlink;
2212
2213 inode = dentry->d_inode;
2214
2215 retval = -EIO;
2216 if (le32_to_cpu(de->inode) != inode->i_ino)
2217 goto end_unlink;
2218
2219 if (!inode->i_nlink) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05002220 ext4_warning(inode->i_sb,
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002221 "Deleting nonexistent file (%lu), %d",
2222 inode->i_ino, inode->i_nlink);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002223 inode->i_nlink = 1;
2224 }
Mingming Cao617ba132006-10-11 01:20:53 -07002225 retval = ext4_delete_entry(handle, dir, de, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002226 if (retval)
2227 goto end_unlink;
Kalpak Shahef7f3832007-07-18 09:15:20 -04002228 dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002229 ext4_update_dx_flag(dir);
2230 ext4_mark_inode_dirty(handle, dir);
Theodore Ts'o825f1482008-02-15 15:00:38 -05002231 drop_nlink(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002232 if (!inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07002233 ext4_orphan_add(handle, inode);
Kalpak Shahef7f3832007-07-18 09:15:20 -04002234 inode->i_ctime = ext4_current_time(inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002235 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002236 retval = 0;
2237
2238end_unlink:
Mingming Cao617ba132006-10-11 01:20:53 -07002239 ext4_journal_stop(handle);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002240 brelse(bh);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04002241 trace_ext4_unlink_exit(dentry, retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002242 return retval;
2243}
2244
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002245static int ext4_symlink(struct inode *dir,
2246 struct dentry *dentry, const char *symname)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002247{
2248 handle_t *handle;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002249 struct inode *inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002250 int l, err, retries = 0;
Jan Karadf5e6222011-05-03 11:12:58 -04002251 int credits;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002252
2253 l = strlen(symname)+1;
2254 if (l > dir->i_sb->s_blocksize)
2255 return -ENAMETOOLONG;
2256
Christoph Hellwig871a2932010-03-03 09:05:07 -05002257 dquot_initialize(dir);
Christoph Hellwig907f4552010-03-03 09:05:06 -05002258
Jan Karadf5e6222011-05-03 11:12:58 -04002259 if (l > EXT4_N_BLOCKS * 4) {
2260 /*
2261 * For non-fast symlinks, we just allocate inode and put it on
2262 * orphan list in the first transaction => we need bitmap,
2263 * group descriptor, sb, inode block, quota blocks.
2264 */
2265 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb);
2266 } else {
2267 /*
2268 * Fast symlink. We have to add entry to directory
2269 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
2270 * allocate new inode (bitmap, group descriptor, inode block,
2271 * quota blocks, sb is already counted in previous macros).
2272 */
2273 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2274 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2275 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb);
2276 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002277retry:
Jan Karadf5e6222011-05-03 11:12:58 -04002278 handle = ext4_journal_start(dir, credits);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002279 if (IS_ERR(handle))
2280 return PTR_ERR(handle);
2281
2282 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05002283 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002284
Andreas Dilger11013912009-06-13 11:45:35 -04002285 inode = ext4_new_inode(handle, dir, S_IFLNK|S_IRWXUGO,
2286 &dentry->d_name, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002287 err = PTR_ERR(inode);
2288 if (IS_ERR(inode))
2289 goto out_stop;
2290
Jan Karadf5e6222011-05-03 11:12:58 -04002291 if (l > EXT4_N_BLOCKS * 4) {
Mingming Cao617ba132006-10-11 01:20:53 -07002292 inode->i_op = &ext4_symlink_inode_operations;
2293 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002294 /*
Jan Karadf5e6222011-05-03 11:12:58 -04002295 * We cannot call page_symlink() with transaction started
2296 * because it calls into ext4_write_begin() which can wait
2297 * for transaction commit if we are running out of space
2298 * and thus we deadlock. So we have to stop transaction now
2299 * and restart it when symlink contents is written.
2300 *
2301 * To keep fs consistent in case of crash, we have to put inode
2302 * to orphan list in the mean time.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002303 */
Jan Karadf5e6222011-05-03 11:12:58 -04002304 drop_nlink(inode);
2305 err = ext4_orphan_add(handle, inode);
2306 ext4_journal_stop(handle);
2307 if (err)
2308 goto err_drop_inode;
Nick Piggin54566b22009-01-04 12:00:53 -08002309 err = __page_symlink(inode, symname, l, 1);
Jan Karadf5e6222011-05-03 11:12:58 -04002310 if (err)
2311 goto err_drop_inode;
2312 /*
2313 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
2314 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2315 */
2316 handle = ext4_journal_start(dir,
2317 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2318 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
2319 if (IS_ERR(handle)) {
2320 err = PTR_ERR(handle);
2321 goto err_drop_inode;
2322 }
2323 inc_nlink(inode);
2324 err = ext4_orphan_del(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002325 if (err) {
Jan Karadf5e6222011-05-03 11:12:58 -04002326 ext4_journal_stop(handle);
Theodore Ts'o825f1482008-02-15 15:00:38 -05002327 clear_nlink(inode);
Jan Karadf5e6222011-05-03 11:12:58 -04002328 goto err_drop_inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002329 }
2330 } else {
Aneesh Kumar K.Ve65187e2008-04-29 08:11:12 -04002331 /* clear the extent format for fast symlink */
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04002332 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
Mingming Cao617ba132006-10-11 01:20:53 -07002333 inode->i_op = &ext4_fast_symlink_inode_operations;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002334 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002335 inode->i_size = l-1;
2336 }
Mingming Cao617ba132006-10-11 01:20:53 -07002337 EXT4_I(inode)->i_disksize = inode->i_size;
2338 err = ext4_add_nondir(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002339out_stop:
Mingming Cao617ba132006-10-11 01:20:53 -07002340 ext4_journal_stop(handle);
2341 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002342 goto retry;
2343 return err;
Jan Karadf5e6222011-05-03 11:12:58 -04002344err_drop_inode:
2345 unlock_new_inode(inode);
2346 iput(inode);
2347 return err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002348}
2349
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002350static int ext4_link(struct dentry *old_dentry,
2351 struct inode *dir, struct dentry *dentry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002352{
2353 handle_t *handle;
2354 struct inode *inode = old_dentry->d_inode;
2355 int err, retries = 0;
2356
Theodore Ts'ob05ab1d2009-08-29 21:08:08 -04002357 if (inode->i_nlink >= EXT4_LINK_MAX)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002358 return -EMLINK;
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002359
Christoph Hellwig871a2932010-03-03 09:05:07 -05002360 dquot_initialize(dir);
Christoph Hellwig907f4552010-03-03 09:05:06 -05002361
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002362retry:
Mingming Cao617ba132006-10-11 01:20:53 -07002363 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2364 EXT4_INDEX_EXTRA_TRANS_BLOCKS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002365 if (IS_ERR(handle))
2366 return PTR_ERR(handle);
2367
2368 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05002369 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002370
Kalpak Shahef7f3832007-07-18 09:15:20 -04002371 inode->i_ctime = ext4_current_time(inode);
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002372 ext4_inc_count(handle, inode);
Al Viro7de9c6e2010-10-23 11:11:40 -04002373 ihold(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002374
Al Viro6b38e842008-12-30 02:03:31 -05002375 err = ext4_add_entry(handle, dentry, inode);
2376 if (!err) {
2377 ext4_mark_inode_dirty(handle, inode);
2378 d_instantiate(dentry, inode);
2379 } else {
2380 drop_nlink(inode);
2381 iput(inode);
2382 }
Mingming Cao617ba132006-10-11 01:20:53 -07002383 ext4_journal_stop(handle);
2384 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002385 goto retry;
2386 return err;
2387}
2388
Wei Yongjun3d0518f2009-02-14 23:01:36 -05002389#define PARENT_INO(buffer, size) \
2390 (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer), size)->inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002391
2392/*
2393 * Anybody can rename anything with this: the permission checks are left to the
2394 * higher-level routines.
2395 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002396static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
2397 struct inode *new_dir, struct dentry *new_dentry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002398{
2399 handle_t *handle;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002400 struct inode *old_inode, *new_inode;
2401 struct buffer_head *old_bh, *new_bh, *dir_bh;
2402 struct ext4_dir_entry_2 *old_de, *new_de;
Theodore Ts'o8750c6d2009-02-23 23:05:27 -05002403 int retval, force_da_alloc = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002404
Christoph Hellwig871a2932010-03-03 09:05:07 -05002405 dquot_initialize(old_dir);
2406 dquot_initialize(new_dir);
Christoph Hellwig907f4552010-03-03 09:05:06 -05002407
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002408 old_bh = new_bh = dir_bh = NULL;
2409
2410 /* Initialize quotas before so that eventual writes go
2411 * in separate transaction */
2412 if (new_dentry->d_inode)
Christoph Hellwig871a2932010-03-03 09:05:07 -05002413 dquot_initialize(new_dentry->d_inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002414 handle = ext4_journal_start(old_dir, 2 *
2415 EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2416 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002417 if (IS_ERR(handle))
2418 return PTR_ERR(handle);
2419
2420 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
Frank Mayhar03901312009-01-07 00:06:22 -05002421 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002422
Theodore Ts'of702ba02008-09-22 15:21:01 -04002423 old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002424 /*
2425 * Check for inode number is _not_ due to possible IO errors.
2426 * We might rmdir the source, keep it as pwd of some process
2427 * and merrily kill the link to whatever was created under the
2428 * same name. Goodbye sticky bit ;-<
2429 */
2430 old_inode = old_dentry->d_inode;
2431 retval = -ENOENT;
2432 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2433 goto end_rename;
2434
2435 new_inode = new_dentry->d_inode;
Theodore Ts'of702ba02008-09-22 15:21:01 -04002436 new_bh = ext4_find_entry(new_dir, &new_dentry->d_name, &new_de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002437 if (new_bh) {
2438 if (!new_inode) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002439 brelse(new_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002440 new_bh = NULL;
2441 }
2442 }
2443 if (S_ISDIR(old_inode->i_mode)) {
2444 if (new_inode) {
2445 retval = -ENOTEMPTY;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002446 if (!empty_dir(new_inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002447 goto end_rename;
2448 }
2449 retval = -EIO;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002450 dir_bh = ext4_bread(handle, old_inode, 0, 0, &retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002451 if (!dir_bh)
2452 goto end_rename;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05002453 if (le32_to_cpu(PARENT_INO(dir_bh->b_data,
2454 old_dir->i_sb->s_blocksize)) != old_dir->i_ino)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002455 goto end_rename;
2456 retval = -EMLINK;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002457 if (!new_inode && new_dir != old_dir &&
Aneesh Kumar K.V2c94eb82009-08-28 21:43:15 -04002458 EXT4_DIR_LINK_MAX(new_dir))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002459 goto end_rename;
Amir Goldsteinef607892011-03-20 21:18:44 -04002460 BUFFER_TRACE(dir_bh, "get_write_access");
2461 retval = ext4_journal_get_write_access(handle, dir_bh);
2462 if (retval)
2463 goto end_rename;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002464 }
2465 if (!new_bh) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002466 retval = ext4_add_entry(handle, new_dentry, old_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002467 if (retval)
2468 goto end_rename;
2469 } else {
2470 BUFFER_TRACE(new_bh, "get write access");
Amir Goldsteinef607892011-03-20 21:18:44 -04002471 retval = ext4_journal_get_write_access(handle, new_bh);
2472 if (retval)
2473 goto end_rename;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002474 new_de->inode = cpu_to_le32(old_inode->i_ino);
Mingming Cao617ba132006-10-11 01:20:53 -07002475 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2476 EXT4_FEATURE_INCOMPAT_FILETYPE))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002477 new_de->file_type = old_de->file_type;
2478 new_dir->i_version++;
Jan Kara53b7e9f2008-04-29 22:02:11 -04002479 new_dir->i_ctime = new_dir->i_mtime =
2480 ext4_current_time(new_dir);
2481 ext4_mark_inode_dirty(handle, new_dir);
Frank Mayhar03901312009-01-07 00:06:22 -05002482 BUFFER_TRACE(new_bh, "call ext4_handle_dirty_metadata");
Theodore Ts'ob4097142011-01-10 12:46:59 -05002483 retval = ext4_handle_dirty_metadata(handle, new_dir, new_bh);
2484 if (unlikely(retval)) {
2485 ext4_std_error(new_dir->i_sb, retval);
2486 goto end_rename;
2487 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002488 brelse(new_bh);
2489 new_bh = NULL;
2490 }
2491
2492 /*
2493 * Like most other Unix systems, set the ctime for inodes on a
2494 * rename.
2495 */
Kalpak Shahef7f3832007-07-18 09:15:20 -04002496 old_inode->i_ctime = ext4_current_time(old_inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002497 ext4_mark_inode_dirty(handle, old_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002498
2499 /*
2500 * ok, that's it
2501 */
2502 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2503 old_de->name_len != old_dentry->d_name.len ||
2504 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
Mingming Cao617ba132006-10-11 01:20:53 -07002505 (retval = ext4_delete_entry(handle, old_dir,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002506 old_de, old_bh)) == -ENOENT) {
2507 /* old_de could have moved from under us during htree split, so
2508 * make sure that we are deleting the right entry. We might
2509 * also be pointing to a stale entry in the unused part of
2510 * old_bh so just checking inum and the name isn't enough. */
2511 struct buffer_head *old_bh2;
Mingming Cao617ba132006-10-11 01:20:53 -07002512 struct ext4_dir_entry_2 *old_de2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002513
Theodore Ts'of702ba02008-09-22 15:21:01 -04002514 old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002515 if (old_bh2) {
Mingming Cao617ba132006-10-11 01:20:53 -07002516 retval = ext4_delete_entry(handle, old_dir,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002517 old_de2, old_bh2);
2518 brelse(old_bh2);
2519 }
2520 }
2521 if (retval) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05002522 ext4_warning(old_dir->i_sb,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002523 "Deleting old file (%lu), %d, error=%d",
2524 old_dir->i_ino, old_dir->i_nlink, retval);
2525 }
2526
2527 if (new_inode) {
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002528 ext4_dec_count(handle, new_inode);
Kalpak Shahef7f3832007-07-18 09:15:20 -04002529 new_inode->i_ctime = ext4_current_time(new_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002530 }
Kalpak Shahef7f3832007-07-18 09:15:20 -04002531 old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002532 ext4_update_dx_flag(old_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002533 if (dir_bh) {
Wei Yongjun3d0518f2009-02-14 23:01:36 -05002534 PARENT_INO(dir_bh->b_data, new_dir->i_sb->s_blocksize) =
2535 cpu_to_le32(new_dir->i_ino);
Frank Mayhar03901312009-01-07 00:06:22 -05002536 BUFFER_TRACE(dir_bh, "call ext4_handle_dirty_metadata");
Theodore Ts'ob4097142011-01-10 12:46:59 -05002537 retval = ext4_handle_dirty_metadata(handle, old_dir, dir_bh);
2538 if (retval) {
2539 ext4_std_error(old_dir->i_sb, retval);
2540 goto end_rename;
2541 }
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002542 ext4_dec_count(handle, old_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002543 if (new_inode) {
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002544 /* checked empty_dir above, can't have another parent,
Theodore Ts'o825f1482008-02-15 15:00:38 -05002545 * ext4_dec_count() won't work for many-linked dirs */
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002546 new_inode->i_nlink = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002547 } else {
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002548 ext4_inc_count(handle, new_dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002549 ext4_update_dx_flag(new_dir);
2550 ext4_mark_inode_dirty(handle, new_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002551 }
2552 }
Mingming Cao617ba132006-10-11 01:20:53 -07002553 ext4_mark_inode_dirty(handle, old_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002554 if (new_inode) {
Mingming Cao617ba132006-10-11 01:20:53 -07002555 ext4_mark_inode_dirty(handle, new_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002556 if (!new_inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07002557 ext4_orphan_add(handle, new_inode);
Theodore Ts'oafd46722009-03-16 23:12:23 -04002558 if (!test_opt(new_dir->i_sb, NO_AUTO_DA_ALLOC))
2559 force_da_alloc = 1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002560 }
2561 retval = 0;
2562
2563end_rename:
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002564 brelse(dir_bh);
2565 brelse(old_bh);
2566 brelse(new_bh);
Mingming Cao617ba132006-10-11 01:20:53 -07002567 ext4_journal_stop(handle);
Theodore Ts'o8750c6d2009-02-23 23:05:27 -05002568 if (retval == 0 && force_da_alloc)
2569 ext4_alloc_da_blocks(old_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002570 return retval;
2571}
2572
2573/*
2574 * directories can handle most operations...
2575 */
Arjan van de Ven754661f2007-02-12 00:55:38 -08002576const struct inode_operations ext4_dir_inode_operations = {
Mingming Cao617ba132006-10-11 01:20:53 -07002577 .create = ext4_create,
2578 .lookup = ext4_lookup,
2579 .link = ext4_link,
2580 .unlink = ext4_unlink,
2581 .symlink = ext4_symlink,
2582 .mkdir = ext4_mkdir,
2583 .rmdir = ext4_rmdir,
2584 .mknod = ext4_mknod,
2585 .rename = ext4_rename,
2586 .setattr = ext4_setattr,
Theodore Ts'o03010a32008-10-10 20:02:48 -04002587#ifdef CONFIG_EXT4_FS_XATTR
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002588 .setxattr = generic_setxattr,
2589 .getxattr = generic_getxattr,
Mingming Cao617ba132006-10-11 01:20:53 -07002590 .listxattr = ext4_listxattr,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002591 .removexattr = generic_removexattr,
2592#endif
Linus Torvalds1d5ccd12009-08-28 12:12:24 -07002593 .check_acl = ext4_check_acl,
Aneesh Kumar K.Vabc87462009-05-02 22:54:32 -04002594 .fiemap = ext4_fiemap,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002595};
2596
Arjan van de Ven754661f2007-02-12 00:55:38 -08002597const struct inode_operations ext4_special_inode_operations = {
Mingming Cao617ba132006-10-11 01:20:53 -07002598 .setattr = ext4_setattr,
Theodore Ts'o03010a32008-10-10 20:02:48 -04002599#ifdef CONFIG_EXT4_FS_XATTR
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002600 .setxattr = generic_setxattr,
2601 .getxattr = generic_getxattr,
Mingming Cao617ba132006-10-11 01:20:53 -07002602 .listxattr = ext4_listxattr,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002603 .removexattr = generic_removexattr,
2604#endif
Linus Torvalds1d5ccd12009-08-28 12:12:24 -07002605 .check_acl = ext4_check_acl,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002606};