blob: 8dde5ab239cc547f5caff3fd393d70ab17a6606d [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);
Tobias Klauser6ca7b132010-12-19 21:38:46 -05001040 if (IS_ERR(inode)) {
Bryan Donlane6f009b2009-02-22 21:20:25 -05001041 if (PTR_ERR(inode) == -ESTALE) {
Theodore Ts'o24676da2010-05-16 21:00:00 -04001042 EXT4_ERROR_INODE(dir,
1043 "deleted inode referenced: %u",
1044 ino);
Bryan Donlane6f009b2009-02-22 21:20:25 -05001045 return ERR_PTR(-EIO);
1046 } else {
1047 return ERR_CAST(inode);
1048 }
1049 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001050 }
1051 return d_splice_alias(inode, dentry);
1052}
1053
1054
Mingming Cao617ba132006-10-11 01:20:53 -07001055struct dentry *ext4_get_parent(struct dentry *child)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001056{
Theodore Ts'o498e5f22008-11-05 00:14:04 -05001057 __u32 ino;
Theodore Ts'of702ba02008-09-22 15:21:01 -04001058 static const struct qstr dotdot = {
1059 .name = "..",
1060 .len = 2,
1061 };
Mingming Cao617ba132006-10-11 01:20:53 -07001062 struct ext4_dir_entry_2 * de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001063 struct buffer_head *bh;
1064
Theodore Ts'of702ba02008-09-22 15:21:01 -04001065 bh = ext4_find_entry(child->d_inode, &dotdot, &de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001066 if (!bh)
1067 return ERR_PTR(-ENOENT);
1068 ino = le32_to_cpu(de->inode);
1069 brelse(bh);
1070
Mingming Cao617ba132006-10-11 01:20:53 -07001071 if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
Theodore Ts'o24676da2010-05-16 21:00:00 -04001072 EXT4_ERROR_INODE(child->d_inode,
1073 "bad parent inode number: %u", ino);
David Howells1d1fe1e2008-02-07 00:15:37 -08001074 return ERR_PTR(-EIO);
Vasily Averina6c15c22007-07-15 23:40:46 -07001075 }
1076
Christoph Hellwig44003722008-08-11 15:49:04 +02001077 return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001078}
1079
1080#define S_SHIFT 12
Mingming Cao617ba132006-10-11 01:20:53 -07001081static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1082 [S_IFREG >> S_SHIFT] = EXT4_FT_REG_FILE,
1083 [S_IFDIR >> S_SHIFT] = EXT4_FT_DIR,
1084 [S_IFCHR >> S_SHIFT] = EXT4_FT_CHRDEV,
1085 [S_IFBLK >> S_SHIFT] = EXT4_FT_BLKDEV,
1086 [S_IFIFO >> S_SHIFT] = EXT4_FT_FIFO,
1087 [S_IFSOCK >> S_SHIFT] = EXT4_FT_SOCK,
1088 [S_IFLNK >> S_SHIFT] = EXT4_FT_SYMLINK,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001089};
1090
Mingming Cao617ba132006-10-11 01:20:53 -07001091static inline void ext4_set_de_type(struct super_block *sb,
1092 struct ext4_dir_entry_2 *de,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001093 umode_t mode) {
Mingming Cao617ba132006-10-11 01:20:53 -07001094 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1095 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001096}
1097
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001098/*
1099 * Move count entries from end of map between two memory locations.
1100 * Returns pointer to last entry moved.
1101 */
Mingming Cao617ba132006-10-11 01:20:53 -07001102static struct ext4_dir_entry_2 *
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001103dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1104 unsigned blocksize)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001105{
1106 unsigned rec_len = 0;
1107
1108 while (count--) {
Theodore Ts'o60e66792010-05-17 07:00:00 -04001109 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
Toshiyuki Okajima9aee2282009-06-08 12:41:35 -04001110 (from + (map->offs<<2));
Mingming Cao617ba132006-10-11 01:20:53 -07001111 rec_len = EXT4_DIR_REC_LEN(de->name_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001112 memcpy (to, de, rec_len);
Mingming Cao617ba132006-10-11 01:20:53 -07001113 ((struct ext4_dir_entry_2 *) to)->rec_len =
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001114 ext4_rec_len_to_disk(rec_len, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001115 de->inode = 0;
1116 map++;
1117 to += rec_len;
1118 }
Mingming Cao617ba132006-10-11 01:20:53 -07001119 return (struct ext4_dir_entry_2 *) (to - rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001120}
1121
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001122/*
1123 * Compact each dir entry in the range to the minimal rec_len.
1124 * Returns pointer to last entry in range.
1125 */
Theodore Ts'o8bad4592009-02-14 21:46:54 -05001126static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001127{
Mingming Cao617ba132006-10-11 01:20:53 -07001128 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001129 unsigned rec_len = 0;
1130
1131 prev = to = de;
Theodore Ts'o8bad4592009-02-14 21:46:54 -05001132 while ((char*)de < base + blocksize) {
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001133 next = ext4_next_entry(de, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001134 if (de->inode && de->name_len) {
Mingming Cao617ba132006-10-11 01:20:53 -07001135 rec_len = EXT4_DIR_REC_LEN(de->name_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001136 if (de > to)
1137 memmove(to, de, rec_len);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001138 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001139 prev = to;
Mingming Cao617ba132006-10-11 01:20:53 -07001140 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001141 }
1142 de = next;
1143 }
1144 return prev;
1145}
1146
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001147/*
1148 * Split a full leaf block to make room for a new dir entry.
1149 * Allocate a new block, and move entries so that they are approx. equally full.
1150 * Returns pointer to de in block into which the new entry will be inserted.
1151 */
Mingming Cao617ba132006-10-11 01:20:53 -07001152static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001153 struct buffer_head **bh,struct dx_frame *frame,
1154 struct dx_hash_info *hinfo, int *error)
1155{
1156 unsigned blocksize = dir->i_sb->s_blocksize;
1157 unsigned count, continued;
1158 struct buffer_head *bh2;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001159 ext4_lblk_t newblock;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001160 u32 hash2;
1161 struct dx_map_entry *map;
1162 char *data1 = (*bh)->b_data, *data2;
Theodore Ts'o59e315b2008-12-06 16:58:39 -05001163 unsigned split, move, size;
Mingming Cao617ba132006-10-11 01:20:53 -07001164 struct ext4_dir_entry_2 *de = NULL, *de2;
Theodore Ts'o59e315b2008-12-06 16:58:39 -05001165 int err = 0, i;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001166
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001167 bh2 = ext4_append (handle, dir, &newblock, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001168 if (!(bh2)) {
1169 brelse(*bh);
1170 *bh = NULL;
1171 goto errout;
1172 }
1173
1174 BUFFER_TRACE(*bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001175 err = ext4_journal_get_write_access(handle, *bh);
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001176 if (err)
1177 goto journal_error;
1178
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001179 BUFFER_TRACE(frame->bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001180 err = ext4_journal_get_write_access(handle, frame->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001181 if (err)
1182 goto journal_error;
1183
1184 data2 = bh2->b_data;
1185
1186 /* create map in the end of data2 block */
1187 map = (struct dx_map_entry *) (data2 + blocksize);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001188 count = dx_make_map((struct ext4_dir_entry_2 *) data1,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001189 blocksize, hinfo, map);
1190 map -= count;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001191 dx_sort_map(map, count);
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001192 /* Split the existing block in the middle, size-wise */
1193 size = 0;
1194 move = 0;
1195 for (i = count-1; i >= 0; i--) {
1196 /* is more than half of this entry in 2nd half of the block? */
1197 if (size + map[i].size/2 > blocksize/2)
1198 break;
1199 size += map[i].size;
1200 move++;
1201 }
1202 /* map index at which we will split */
1203 split = count - move;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001204 hash2 = map[split].hash;
1205 continued = hash2 == map[split - 1].hash;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001206 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1207 (unsigned long)dx_get_block(frame->at),
1208 hash2, split, count-split));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001209
1210 /* Fancy dance to stay within two buffers */
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001211 de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001212 de = dx_pack_dirents(data1, blocksize);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001213 de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1214 blocksize);
1215 de2->rec_len = ext4_rec_len_to_disk(data2 + blocksize - (char *) de2,
1216 blocksize);
Mingming Cao617ba132006-10-11 01:20:53 -07001217 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1218 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001219
1220 /* Which block gets the new entry? */
1221 if (hinfo->hash >= hash2)
1222 {
1223 swap(*bh, bh2);
1224 de = de2;
1225 }
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001226 dx_insert_block(frame, hash2 + continued, newblock);
Frank Mayhar03901312009-01-07 00:06:22 -05001227 err = ext4_handle_dirty_metadata(handle, dir, bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001228 if (err)
1229 goto journal_error;
Frank Mayhar03901312009-01-07 00:06:22 -05001230 err = ext4_handle_dirty_metadata(handle, dir, frame->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001231 if (err)
1232 goto journal_error;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001233 brelse(bh2);
1234 dxtrace(dx_show_index("frame", frame->entries));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001235 return de;
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001236
1237journal_error:
1238 brelse(*bh);
1239 brelse(bh2);
1240 *bh = NULL;
1241 ext4_std_error(dir->i_sb, err);
1242errout:
1243 *error = err;
1244 return NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001245}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001246
1247/*
1248 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1249 * it points to a directory entry which is guaranteed to be large
1250 * enough for new directory entry. If de is NULL, then
1251 * add_dirent_to_buf will attempt search the directory block for
1252 * space. It will return -ENOSPC if no space is available, and -EIO
1253 * and -EEXIST if directory entry already exists.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001254 */
1255static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
Mingming Cao617ba132006-10-11 01:20:53 -07001256 struct inode *inode, struct ext4_dir_entry_2 *de,
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001257 struct buffer_head *bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001258{
1259 struct inode *dir = dentry->d_parent->d_inode;
1260 const char *name = dentry->d_name.name;
1261 int namelen = dentry->d_name.len;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05001262 unsigned int offset = 0;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001263 unsigned int blocksize = dir->i_sb->s_blocksize;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001264 unsigned short reclen;
1265 int nlen, rlen, err;
1266 char *top;
1267
Mingming Cao617ba132006-10-11 01:20:53 -07001268 reclen = EXT4_DIR_REC_LEN(namelen);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001269 if (!de) {
Mingming Cao617ba132006-10-11 01:20:53 -07001270 de = (struct ext4_dir_entry_2 *)bh->b_data;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001271 top = bh->b_data + blocksize - reclen;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001272 while ((char *) de <= top) {
Theodore Ts'of7c21172011-01-10 12:10:55 -05001273 if (ext4_check_dir_entry(dir, NULL, de, bh, offset))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001274 return -EIO;
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001275 if (ext4_match(namelen, name, de))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001276 return -EEXIST;
Mingming Cao617ba132006-10-11 01:20:53 -07001277 nlen = EXT4_DIR_REC_LEN(de->name_len);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001278 rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001279 if ((de->inode? rlen - nlen: rlen) >= reclen)
1280 break;
Mingming Cao617ba132006-10-11 01:20:53 -07001281 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001282 offset += rlen;
1283 }
1284 if ((char *) de > top)
1285 return -ENOSPC;
1286 }
1287 BUFFER_TRACE(bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001288 err = ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001289 if (err) {
Mingming Cao617ba132006-10-11 01:20:53 -07001290 ext4_std_error(dir->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001291 return err;
1292 }
1293
1294 /* By now the buffer is marked for journaling */
Mingming Cao617ba132006-10-11 01:20:53 -07001295 nlen = EXT4_DIR_REC_LEN(de->name_len);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001296 rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001297 if (de->inode) {
Mingming Cao617ba132006-10-11 01:20:53 -07001298 struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001299 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, blocksize);
1300 de->rec_len = ext4_rec_len_to_disk(nlen, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001301 de = de1;
1302 }
Mingming Cao617ba132006-10-11 01:20:53 -07001303 de->file_type = EXT4_FT_UNKNOWN;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001304 if (inode) {
1305 de->inode = cpu_to_le32(inode->i_ino);
Mingming Cao617ba132006-10-11 01:20:53 -07001306 ext4_set_de_type(dir->i_sb, de, inode->i_mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001307 } else
1308 de->inode = 0;
1309 de->name_len = namelen;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001310 memcpy(de->name, name, namelen);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001311 /*
1312 * XXX shouldn't update any times until successful
1313 * completion of syscall, but too many callers depend
1314 * on this.
1315 *
1316 * XXX similarly, too many callers depend on
Mingming Cao617ba132006-10-11 01:20:53 -07001317 * ext4_new_inode() setting the times, but error
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001318 * recovery deletes the inode, so the worst that can
1319 * happen is that the times are slightly out of date
1320 * and/or different from the directory change time.
1321 */
Kalpak Shahef7f3832007-07-18 09:15:20 -04001322 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
Mingming Cao617ba132006-10-11 01:20:53 -07001323 ext4_update_dx_flag(dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001324 dir->i_version++;
Mingming Cao617ba132006-10-11 01:20:53 -07001325 ext4_mark_inode_dirty(handle, dir);
Frank Mayhar03901312009-01-07 00:06:22 -05001326 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1327 err = ext4_handle_dirty_metadata(handle, dir, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001328 if (err)
Mingming Cao617ba132006-10-11 01:20:53 -07001329 ext4_std_error(dir->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001330 return 0;
1331}
1332
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001333/*
1334 * This converts a one block unindexed directory to a 3 block indexed
1335 * directory, and adds the dentry to the indexed directory.
1336 */
1337static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1338 struct inode *inode, struct buffer_head *bh)
1339{
1340 struct inode *dir = dentry->d_parent->d_inode;
1341 const char *name = dentry->d_name.name;
1342 int namelen = dentry->d_name.len;
1343 struct buffer_head *bh2;
1344 struct dx_root *root;
1345 struct dx_frame frames[2], *frame;
1346 struct dx_entry *entries;
Mingming Cao617ba132006-10-11 01:20:53 -07001347 struct ext4_dir_entry_2 *de, *de2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001348 char *data1, *top;
1349 unsigned len;
1350 int retval;
1351 unsigned blocksize;
1352 struct dx_hash_info hinfo;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001353 ext4_lblk_t block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001354 struct fake_dirent *fde;
1355
1356 blocksize = dir->i_sb->s_blocksize;
Theodore Ts'oe6b8bc02009-01-16 11:13:40 -05001357 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
Mingming Cao617ba132006-10-11 01:20:53 -07001358 retval = ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001359 if (retval) {
Mingming Cao617ba132006-10-11 01:20:53 -07001360 ext4_std_error(dir->i_sb, retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001361 brelse(bh);
1362 return retval;
1363 }
1364 root = (struct dx_root *) bh->b_data;
1365
Theodore Ts'oe6b8bc02009-01-16 11:13:40 -05001366 /* The 0th block becomes the root, move the dirents out */
1367 fde = &root->dotdot;
1368 de = (struct ext4_dir_entry_2 *)((char *)fde +
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001369 ext4_rec_len_from_disk(fde->rec_len, blocksize));
Theodore Ts'oe6b8bc02009-01-16 11:13:40 -05001370 if ((char *) de >= (((char *) root) + blocksize)) {
Theodore Ts'o24676da2010-05-16 21:00:00 -04001371 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
Theodore Ts'oe6b8bc02009-01-16 11:13:40 -05001372 brelse(bh);
1373 return -EIO;
1374 }
1375 len = ((char *) root) + blocksize - (char *) de;
1376
1377 /* Allocate new block for the 0th block's dirents */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001378 bh2 = ext4_append(handle, dir, &block, &retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001379 if (!(bh2)) {
1380 brelse(bh);
1381 return retval;
1382 }
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04001383 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001384 data1 = bh2->b_data;
1385
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001386 memcpy (data1, de, len);
Mingming Cao617ba132006-10-11 01:20:53 -07001387 de = (struct ext4_dir_entry_2 *) data1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001388 top = data1 + len;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001389 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001390 de = de2;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001391 de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1392 blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001393 /* Initialize the root; the dot dirents already exist */
Mingming Cao617ba132006-10-11 01:20:53 -07001394 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001395 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1396 blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001397 memset (&root->info, 0, sizeof(root->info));
1398 root->info.info_length = sizeof(root->info);
Mingming Cao617ba132006-10-11 01:20:53 -07001399 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001400 entries = root->entries;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001401 dx_set_block(entries, 1);
1402 dx_set_count(entries, 1);
1403 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001404
1405 /* Initialize as for dx_probe */
1406 hinfo.hash_version = root->info.hash_version;
Theodore Ts'of99b2582008-10-28 13:21:44 -04001407 if (hinfo.hash_version <= DX_HASH_TEA)
1408 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
Mingming Cao617ba132006-10-11 01:20:53 -07001409 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1410 ext4fs_dirhash(name, namelen, &hinfo);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001411 frame = frames;
1412 frame->entries = entries;
1413 frame->at = entries;
1414 frame->bh = bh;
1415 bh = bh2;
Allison Henderson6976a6f2011-05-15 00:19:41 -04001416
1417 ext4_handle_dirty_metadata(handle, dir, frame->bh);
1418 ext4_handle_dirty_metadata(handle, dir, bh);
1419
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001420 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
Jan Kara7ad8e4e2011-05-03 11:05:55 -04001421 if (!de) {
1422 /*
1423 * Even if the block split failed, we have to properly write
1424 * out all the changes we did so far. Otherwise we can end up
1425 * with corrupted filesystem.
1426 */
1427 ext4_mark_inode_dirty(handle, dir);
Jan Kara7ad8e4e2011-05-03 11:05:55 -04001428 dx_release(frames);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001429 return retval;
Jan Kara7ad8e4e2011-05-03 11:05:55 -04001430 }
1431 dx_release(frames);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001432
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001433 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1434 brelse(bh);
1435 return retval;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001436}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001437
1438/*
Mingming Cao617ba132006-10-11 01:20:53 -07001439 * ext4_add_entry()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001440 *
1441 * adds a file entry to the specified directory, using the same
Mingming Cao617ba132006-10-11 01:20:53 -07001442 * semantics as ext4_find_entry(). It returns NULL if it failed.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001443 *
1444 * NOTE!! The inode part of 'de' is left at 0 - which means you
1445 * may not sleep between calling this and putting something into
1446 * the entry, as someone else might have used it while you slept.
1447 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001448static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1449 struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001450{
1451 struct inode *dir = dentry->d_parent->d_inode;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001452 struct buffer_head *bh;
Mingming Cao617ba132006-10-11 01:20:53 -07001453 struct ext4_dir_entry_2 *de;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001454 struct super_block *sb;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001455 int retval;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001456 int dx_fallback=0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001457 unsigned blocksize;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001458 ext4_lblk_t block, blocks;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001459
1460 sb = dir->i_sb;
1461 blocksize = sb->s_blocksize;
1462 if (!dentry->d_name.len)
1463 return -EINVAL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001464 if (is_dx(dir)) {
Mingming Cao617ba132006-10-11 01:20:53 -07001465 retval = ext4_dx_add_entry(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001466 if (!retval || (retval != ERR_BAD_DX_DIR))
1467 return retval;
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04001468 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001469 dx_fallback++;
Mingming Cao617ba132006-10-11 01:20:53 -07001470 ext4_mark_inode_dirty(handle, dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001471 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001472 blocks = dir->i_size >> sb->s_blocksize_bits;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05001473 for (block = 0; block < blocks; block++) {
Mingming Cao617ba132006-10-11 01:20:53 -07001474 bh = ext4_bread(handle, dir, block, 0, &retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001475 if(!bh)
1476 return retval;
1477 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001478 if (retval != -ENOSPC) {
1479 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001480 return retval;
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001481 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001482
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001483 if (blocks == 1 && !dx_fallback &&
Theodore Ts'o1e424a32009-11-08 15:45:44 -05001484 EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1485 return make_indexed_dir(handle, dentry, inode, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001486 brelse(bh);
1487 }
Mingming Cao617ba132006-10-11 01:20:53 -07001488 bh = ext4_append(handle, dir, &block, &retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001489 if (!bh)
1490 return retval;
Mingming Cao617ba132006-10-11 01:20:53 -07001491 de = (struct ext4_dir_entry_2 *) bh->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001492 de->inode = 0;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001493 de->rec_len = ext4_rec_len_to_disk(blocksize, blocksize);
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001494 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1495 brelse(bh);
Frank Mayhar14ece102010-05-17 08:00:00 -04001496 if (retval == 0)
1497 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001498 return retval;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001499}
1500
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001501/*
1502 * Returns 0 for success, or a negative error value
1503 */
Mingming Cao617ba132006-10-11 01:20:53 -07001504static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001505 struct inode *inode)
1506{
1507 struct dx_frame frames[2], *frame;
1508 struct dx_entry *entries, *at;
1509 struct dx_hash_info hinfo;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001510 struct buffer_head *bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001511 struct inode *dir = dentry->d_parent->d_inode;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001512 struct super_block *sb = dir->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -07001513 struct ext4_dir_entry_2 *de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001514 int err;
1515
Theodore Ts'of702ba02008-09-22 15:21:01 -04001516 frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001517 if (!frame)
1518 return err;
1519 entries = frame->entries;
1520 at = frame->at;
1521
Mingming Cao617ba132006-10-11 01:20:53 -07001522 if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001523 goto cleanup;
1524
1525 BUFFER_TRACE(bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001526 err = ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001527 if (err)
1528 goto journal_error;
1529
1530 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001531 if (err != -ENOSPC)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001532 goto cleanup;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001533
1534 /* Block full, should compress but for now just split */
Theodore Ts'o47760042008-09-08 23:00:52 -04001535 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001536 dx_get_count(entries), dx_get_limit(entries)));
1537 /* Need to split index? */
1538 if (dx_get_count(entries) == dx_get_limit(entries)) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001539 ext4_lblk_t newblock;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001540 unsigned icount = dx_get_count(entries);
1541 int levels = frame - frames;
1542 struct dx_entry *entries2;
1543 struct dx_node *node2;
1544 struct buffer_head *bh2;
1545
1546 if (levels && (dx_get_count(frames->entries) ==
1547 dx_get_limit(frames->entries))) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001548 ext4_warning(sb, "Directory index full!");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001549 err = -ENOSPC;
1550 goto cleanup;
1551 }
Mingming Cao617ba132006-10-11 01:20:53 -07001552 bh2 = ext4_append (handle, dir, &newblock, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001553 if (!(bh2))
1554 goto cleanup;
1555 node2 = (struct dx_node *)(bh2->b_data);
1556 entries2 = node2->entries;
Andreas Schlick1f7bebb2009-09-10 23:16:07 -04001557 memset(&node2->fake, 0, sizeof(struct fake_dirent));
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001558 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
1559 sb->s_blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001560 BUFFER_TRACE(frame->bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001561 err = ext4_journal_get_write_access(handle, frame->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001562 if (err)
1563 goto journal_error;
1564 if (levels) {
1565 unsigned icount1 = icount/2, icount2 = icount - icount1;
1566 unsigned hash2 = dx_get_hash(entries + icount1);
Theodore Ts'o47760042008-09-08 23:00:52 -04001567 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
1568 icount1, icount2));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001569
1570 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
Mingming Cao617ba132006-10-11 01:20:53 -07001571 err = ext4_journal_get_write_access(handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001572 frames[0].bh);
1573 if (err)
1574 goto journal_error;
1575
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001576 memcpy((char *) entries2, (char *) (entries + icount1),
1577 icount2 * sizeof(struct dx_entry));
1578 dx_set_count(entries, icount1);
1579 dx_set_count(entries2, icount2);
1580 dx_set_limit(entries2, dx_node_limit(dir));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001581
1582 /* Which index block gets the new entry? */
1583 if (at - entries >= icount1) {
1584 frame->at = at = at - entries - icount1 + entries2;
1585 frame->entries = entries = entries2;
1586 swap(frame->bh, bh2);
1587 }
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001588 dx_insert_block(frames + 0, hash2, newblock);
1589 dxtrace(dx_show_index("node", frames[1].entries));
1590 dxtrace(dx_show_index("node",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001591 ((struct dx_node *) bh2->b_data)->entries));
Frank Mayhar03901312009-01-07 00:06:22 -05001592 err = ext4_handle_dirty_metadata(handle, inode, bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001593 if (err)
1594 goto journal_error;
1595 brelse (bh2);
1596 } else {
Theodore Ts'o47760042008-09-08 23:00:52 -04001597 dxtrace(printk(KERN_DEBUG
1598 "Creating second level index...\n"));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001599 memcpy((char *) entries2, (char *) entries,
1600 icount * sizeof(struct dx_entry));
1601 dx_set_limit(entries2, dx_node_limit(dir));
1602
1603 /* Set up root */
1604 dx_set_count(entries, 1);
1605 dx_set_block(entries + 0, newblock);
1606 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1607
1608 /* Add new access path frame */
1609 frame = frames + 1;
1610 frame->at = at = at - entries + entries2;
1611 frame->entries = entries = entries2;
1612 frame->bh = bh2;
Mingming Cao617ba132006-10-11 01:20:53 -07001613 err = ext4_journal_get_write_access(handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001614 frame->bh);
1615 if (err)
1616 goto journal_error;
1617 }
Theodore Ts'ob4097142011-01-10 12:46:59 -05001618 err = ext4_handle_dirty_metadata(handle, inode, frames[0].bh);
1619 if (err) {
1620 ext4_std_error(inode->i_sb, err);
1621 goto cleanup;
1622 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001623 }
1624 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1625 if (!de)
1626 goto cleanup;
1627 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001628 goto cleanup;
1629
1630journal_error:
Mingming Cao617ba132006-10-11 01:20:53 -07001631 ext4_std_error(dir->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001632cleanup:
1633 if (bh)
1634 brelse(bh);
1635 dx_release(frames);
1636 return err;
1637}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001638
1639/*
Mingming Cao617ba132006-10-11 01:20:53 -07001640 * ext4_delete_entry deletes a directory entry by merging it with the
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001641 * previous entry
1642 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001643static int ext4_delete_entry(handle_t *handle,
1644 struct inode *dir,
1645 struct ext4_dir_entry_2 *de_del,
1646 struct buffer_head *bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001647{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001648 struct ext4_dir_entry_2 *de, *pde;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001649 unsigned int blocksize = dir->i_sb->s_blocksize;
Theodore Ts'ob4097142011-01-10 12:46:59 -05001650 int i, err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001651
1652 i = 0;
1653 pde = NULL;
Mingming Cao617ba132006-10-11 01:20:53 -07001654 de = (struct ext4_dir_entry_2 *) bh->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001655 while (i < bh->b_size) {
Theodore Ts'of7c21172011-01-10 12:10:55 -05001656 if (ext4_check_dir_entry(dir, NULL, de, bh, i))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001657 return -EIO;
1658 if (de == de_del) {
1659 BUFFER_TRACE(bh, "get_write_access");
Theodore Ts'ob4097142011-01-10 12:46:59 -05001660 err = ext4_journal_get_write_access(handle, bh);
1661 if (unlikely(err)) {
1662 ext4_std_error(dir->i_sb, err);
1663 return err;
1664 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001665 if (pde)
Jan Karaa72d7f82008-01-28 23:58:27 -05001666 pde->rec_len = ext4_rec_len_to_disk(
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001667 ext4_rec_len_from_disk(pde->rec_len,
1668 blocksize) +
1669 ext4_rec_len_from_disk(de->rec_len,
1670 blocksize),
1671 blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001672 else
1673 de->inode = 0;
1674 dir->i_version++;
Frank Mayhar03901312009-01-07 00:06:22 -05001675 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
Theodore Ts'ob4097142011-01-10 12:46:59 -05001676 err = ext4_handle_dirty_metadata(handle, dir, bh);
1677 if (unlikely(err)) {
1678 ext4_std_error(dir->i_sb, err);
1679 return err;
1680 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001681 return 0;
1682 }
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001683 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001684 pde = de;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001685 de = ext4_next_entry(de, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001686 }
1687 return -ENOENT;
1688}
1689
Andreas Dilgerf8628a12007-07-18 08:38:01 -04001690/*
1691 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
1692 * since this indicates that nlinks count was previously 1.
1693 */
1694static void ext4_inc_count(handle_t *handle, struct inode *inode)
1695{
1696 inc_nlink(inode);
1697 if (is_dx(inode) && inode->i_nlink > 1) {
1698 /* limit is 16-bit i_links_count */
1699 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
1700 inode->i_nlink = 1;
1701 EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
1702 EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
1703 }
1704 }
1705}
1706
1707/*
1708 * If a directory had nlink == 1, then we should let it be 1. This indicates
1709 * directory has >EXT4_LINK_MAX subdirs.
1710 */
1711static void ext4_dec_count(handle_t *handle, struct inode *inode)
1712{
1713 drop_nlink(inode);
1714 if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
1715 inc_nlink(inode);
1716}
1717
1718
Mingming Cao617ba132006-10-11 01:20:53 -07001719static int ext4_add_nondir(handle_t *handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001720 struct dentry *dentry, struct inode *inode)
1721{
Mingming Cao617ba132006-10-11 01:20:53 -07001722 int err = ext4_add_entry(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001723 if (!err) {
Mingming Cao617ba132006-10-11 01:20:53 -07001724 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001725 d_instantiate(dentry, inode);
Al Viro6b38e842008-12-30 02:03:31 -05001726 unlock_new_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001727 return 0;
1728 }
Eric Sandeen731b9a52007-02-10 01:46:16 -08001729 drop_nlink(inode);
Al Viro6b38e842008-12-30 02:03:31 -05001730 unlock_new_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001731 iput(inode);
1732 return err;
1733}
1734
1735/*
1736 * By the time this is called, we already have created
1737 * the directory cache entry for the new file, but it
1738 * is so far negative - it has no inode.
1739 *
1740 * If the create succeeds, we fill in the inode information
1741 * with d_instantiate().
1742 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001743static int ext4_create(struct inode *dir, struct dentry *dentry, int mode,
1744 struct nameidata *nd)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001745{
1746 handle_t *handle;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001747 struct inode *inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001748 int err, retries = 0;
1749
Christoph Hellwig871a2932010-03-03 09:05:07 -05001750 dquot_initialize(dir);
Christoph Hellwig907f4552010-03-03 09:05:06 -05001751
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001752retry:
Mingming Cao617ba132006-10-11 01:20:53 -07001753 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1754 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
Dmitry Monakhov5aca07e2009-12-08 22:42:15 -05001755 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001756 if (IS_ERR(handle))
1757 return PTR_ERR(handle);
1758
1759 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05001760 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001761
Andreas Dilger11013912009-06-13 11:45:35 -04001762 inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001763 err = PTR_ERR(inode);
1764 if (!IS_ERR(inode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07001765 inode->i_op = &ext4_file_inode_operations;
1766 inode->i_fop = &ext4_file_operations;
1767 ext4_set_aops(inode);
1768 err = ext4_add_nondir(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001769 }
Mingming Cao617ba132006-10-11 01:20:53 -07001770 ext4_journal_stop(handle);
1771 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001772 goto retry;
1773 return err;
1774}
1775
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001776static int ext4_mknod(struct inode *dir, struct dentry *dentry,
1777 int mode, dev_t rdev)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001778{
1779 handle_t *handle;
1780 struct inode *inode;
1781 int err, retries = 0;
1782
1783 if (!new_valid_dev(rdev))
1784 return -EINVAL;
1785
Christoph Hellwig871a2932010-03-03 09:05:07 -05001786 dquot_initialize(dir);
Christoph Hellwig907f4552010-03-03 09:05:06 -05001787
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001788retry:
Mingming Cao617ba132006-10-11 01:20:53 -07001789 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1790 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
Dmitry Monakhov5aca07e2009-12-08 22:42:15 -05001791 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001792 if (IS_ERR(handle))
1793 return PTR_ERR(handle);
1794
1795 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05001796 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001797
Andreas Dilger11013912009-06-13 11:45:35 -04001798 inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001799 err = PTR_ERR(inode);
1800 if (!IS_ERR(inode)) {
1801 init_special_inode(inode, inode->i_mode, rdev);
Theodore Ts'o03010a32008-10-10 20:02:48 -04001802#ifdef CONFIG_EXT4_FS_XATTR
Mingming Cao617ba132006-10-11 01:20:53 -07001803 inode->i_op = &ext4_special_inode_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001804#endif
Mingming Cao617ba132006-10-11 01:20:53 -07001805 err = ext4_add_nondir(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001806 }
Mingming Cao617ba132006-10-11 01:20:53 -07001807 ext4_journal_stop(handle);
1808 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001809 goto retry;
1810 return err;
1811}
1812
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001813static int ext4_mkdir(struct inode *dir, struct dentry *dentry, int mode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001814{
1815 handle_t *handle;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001816 struct inode *inode;
Namhyung Kimdabd9912011-01-10 12:11:16 -05001817 struct buffer_head *dir_block = NULL;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001818 struct ext4_dir_entry_2 *de;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001819 unsigned int blocksize = dir->i_sb->s_blocksize;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001820 int err, retries = 0;
1821
Andreas Dilgerf8628a12007-07-18 08:38:01 -04001822 if (EXT4_DIR_LINK_MAX(dir))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001823 return -EMLINK;
1824
Christoph Hellwig871a2932010-03-03 09:05:07 -05001825 dquot_initialize(dir);
Christoph Hellwig907f4552010-03-03 09:05:06 -05001826
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001827retry:
Mingming Cao617ba132006-10-11 01:20:53 -07001828 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1829 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
Dmitry Monakhov5aca07e2009-12-08 22:42:15 -05001830 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001831 if (IS_ERR(handle))
1832 return PTR_ERR(handle);
1833
1834 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05001835 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001836
Andreas Dilger11013912009-06-13 11:45:35 -04001837 inode = ext4_new_inode(handle, dir, S_IFDIR | mode,
1838 &dentry->d_name, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001839 err = PTR_ERR(inode);
1840 if (IS_ERR(inode))
1841 goto out_stop;
1842
Mingming Cao617ba132006-10-11 01:20:53 -07001843 inode->i_op = &ext4_dir_inode_operations;
1844 inode->i_fop = &ext4_dir_operations;
1845 inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001846 dir_block = ext4_bread(handle, inode, 0, 1, &err);
Aneesh Kumar K.V4cdeed82008-02-22 06:17:31 -05001847 if (!dir_block)
1848 goto out_clear_inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001849 BUFFER_TRACE(dir_block, "get_write_access");
Namhyung Kimdabd9912011-01-10 12:11:16 -05001850 err = ext4_journal_get_write_access(handle, dir_block);
1851 if (err)
1852 goto out_clear_inode;
Mingming Cao617ba132006-10-11 01:20:53 -07001853 de = (struct ext4_dir_entry_2 *) dir_block->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001854 de->inode = cpu_to_le32(inode->i_ino);
1855 de->name_len = 1;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001856 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
1857 blocksize);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001858 strcpy(de->name, ".");
Mingming Cao617ba132006-10-11 01:20:53 -07001859 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001860 de = ext4_next_entry(de, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001861 de->inode = cpu_to_le32(dir->i_ino);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001862 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(1),
1863 blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001864 de->name_len = 2;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001865 strcpy(de->name, "..");
Mingming Cao617ba132006-10-11 01:20:53 -07001866 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001867 inode->i_nlink = 2;
Frank Mayhar03901312009-01-07 00:06:22 -05001868 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
Namhyung Kimdabd9912011-01-10 12:11:16 -05001869 err = ext4_handle_dirty_metadata(handle, dir, dir_block);
1870 if (err)
1871 goto out_clear_inode;
1872 err = ext4_mark_inode_dirty(handle, inode);
1873 if (!err)
1874 err = ext4_add_entry(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001875 if (err) {
Aneesh Kumar K.V4cdeed82008-02-22 06:17:31 -05001876out_clear_inode:
1877 clear_nlink(inode);
Al Viro6b38e842008-12-30 02:03:31 -05001878 unlock_new_inode(inode);
Mingming Cao617ba132006-10-11 01:20:53 -07001879 ext4_mark_inode_dirty(handle, inode);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001880 iput(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001881 goto out_stop;
1882 }
Andreas Dilgerf8628a12007-07-18 08:38:01 -04001883 ext4_inc_count(handle, dir);
Mingming Cao617ba132006-10-11 01:20:53 -07001884 ext4_update_dx_flag(dir);
Namhyung Kimdabd9912011-01-10 12:11:16 -05001885 err = ext4_mark_inode_dirty(handle, dir);
1886 if (err)
1887 goto out_clear_inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001888 d_instantiate(dentry, inode);
Al Viro6b38e842008-12-30 02:03:31 -05001889 unlock_new_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001890out_stop:
Namhyung Kimdabd9912011-01-10 12:11:16 -05001891 brelse(dir_block);
Mingming Cao617ba132006-10-11 01:20:53 -07001892 ext4_journal_stop(handle);
1893 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001894 goto retry;
1895 return err;
1896}
1897
1898/*
1899 * routine to check that the specified directory is empty (for rmdir)
1900 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001901static int empty_dir(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001902{
Theodore Ts'o498e5f22008-11-05 00:14:04 -05001903 unsigned int offset;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001904 struct buffer_head *bh;
1905 struct ext4_dir_entry_2 *de, *de1;
1906 struct super_block *sb;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001907 int err = 0;
1908
1909 sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -07001910 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001911 !(bh = ext4_bread(NULL, inode, 0, 0, &err))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001912 if (err)
Theodore Ts'o24676da2010-05-16 21:00:00 -04001913 EXT4_ERROR_INODE(inode,
1914 "error %d reading directory lblock 0", err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001915 else
Eric Sandeen12062dd2010-02-15 14:19:27 -05001916 ext4_warning(inode->i_sb,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001917 "bad directory (dir #%lu) - no data block",
1918 inode->i_ino);
1919 return 1;
1920 }
Mingming Cao617ba132006-10-11 01:20:53 -07001921 de = (struct ext4_dir_entry_2 *) bh->b_data;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001922 de1 = ext4_next_entry(de, sb->s_blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001923 if (le32_to_cpu(de->inode) != inode->i_ino ||
1924 !le32_to_cpu(de1->inode) ||
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001925 strcmp(".", de->name) ||
1926 strcmp("..", de1->name)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001927 ext4_warning(inode->i_sb,
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001928 "bad directory (dir #%lu) - no `.' or `..'",
1929 inode->i_ino);
1930 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001931 return 1;
1932 }
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001933 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
1934 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
1935 de = ext4_next_entry(de1, sb->s_blocksize);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001936 while (offset < inode->i_size) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001937 if (!bh ||
Theodore Ts'o24676da2010-05-16 21:00:00 -04001938 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1939 unsigned int lblock;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001940 err = 0;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001941 brelse(bh);
Theodore Ts'o24676da2010-05-16 21:00:00 -04001942 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
1943 bh = ext4_bread(NULL, inode, lblock, 0, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001944 if (!bh) {
1945 if (err)
Theodore Ts'o24676da2010-05-16 21:00:00 -04001946 EXT4_ERROR_INODE(inode,
1947 "error %d reading directory "
1948 "lblock %u", err, lblock);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001949 offset += sb->s_blocksize;
1950 continue;
1951 }
Mingming Cao617ba132006-10-11 01:20:53 -07001952 de = (struct ext4_dir_entry_2 *) bh->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001953 }
Theodore Ts'of7c21172011-01-10 12:10:55 -05001954 if (ext4_check_dir_entry(inode, NULL, de, bh, offset)) {
Mingming Cao617ba132006-10-11 01:20:53 -07001955 de = (struct ext4_dir_entry_2 *)(bh->b_data +
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001956 sb->s_blocksize);
1957 offset = (offset | (sb->s_blocksize - 1)) + 1;
1958 continue;
1959 }
1960 if (le32_to_cpu(de->inode)) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001961 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001962 return 0;
1963 }
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001964 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
1965 de = ext4_next_entry(de, sb->s_blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001966 }
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001967 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001968 return 1;
1969}
1970
Mingming Cao617ba132006-10-11 01:20:53 -07001971/* ext4_orphan_add() links an unlinked or truncated inode into a list of
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001972 * such inodes, starting at the superblock, in case we crash before the
1973 * file is closed/deleted, or in case the inode truncate spans multiple
1974 * transactions and the last transaction is not recovered after a crash.
1975 *
1976 * At filesystem recovery time, we walk this list deleting unlinked
Mingming Cao617ba132006-10-11 01:20:53 -07001977 * inodes and truncating linked inodes in ext4_orphan_cleanup().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001978 */
Mingming Cao617ba132006-10-11 01:20:53 -07001979int ext4_orphan_add(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001980{
1981 struct super_block *sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -07001982 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001983 int err = 0, rc;
1984
Frank Mayhar03901312009-01-07 00:06:22 -05001985 if (!ext4_handle_valid(handle))
1986 return 0;
1987
Theodore Ts'o3b9d4ed2009-04-25 22:54:04 -04001988 mutex_lock(&EXT4_SB(sb)->s_orphan_lock);
Mingming Cao617ba132006-10-11 01:20:53 -07001989 if (!list_empty(&EXT4_I(inode)->i_orphan))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001990 goto out_unlock;
1991
Lukas Czernerafb86172011-07-11 18:47:04 -04001992 /*
1993 * Orphan handling is only valid for files with data blocks
1994 * being truncated, or files being unlinked. Note that we either
1995 * hold i_mutex, or the inode can not be referenced from outside,
1996 * so i_nlink should not be bumped due to race
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001997 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001998 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1999 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002000
Mingming Cao617ba132006-10-11 01:20:53 -07002001 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
2002 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002003 if (err)
2004 goto out_unlock;
2005
Mingming Cao617ba132006-10-11 01:20:53 -07002006 err = ext4_reserve_inode_write(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002007 if (err)
2008 goto out_unlock;
Dmitry Monakhov6e3617e2010-03-01 23:29:39 -05002009 /*
2010 * Due to previous errors inode may be already a part of on-disk
2011 * orphan list. If so skip on-disk list modification.
2012 */
2013 if (NEXT_ORPHAN(inode) && NEXT_ORPHAN(inode) <=
2014 (le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)))
2015 goto mem_insert;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002016
2017 /* Insert this inode at the head of the on-disk orphan list... */
Mingming Cao617ba132006-10-11 01:20:53 -07002018 NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
2019 EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
Curt Wohlgemuth73b50c12010-02-16 15:06:29 -05002020 err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
Mingming Cao617ba132006-10-11 01:20:53 -07002021 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002022 if (!err)
2023 err = rc;
2024
2025 /* Only add to the head of the in-memory list if all the
2026 * previous operations succeeded. If the orphan_add is going to
2027 * fail (possibly taking the journal offline), we can't risk
2028 * leaving the inode on the orphan list: stray orphan-list
2029 * entries can cause panics at unmount time.
2030 *
2031 * This is safe: on error we're going to ignore the orphan list
2032 * anyway on the next recovery. */
Dmitry Monakhov6e3617e2010-03-01 23:29:39 -05002033mem_insert:
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002034 if (!err)
Mingming Cao617ba132006-10-11 01:20:53 -07002035 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002036
2037 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2038 jbd_debug(4, "orphan inode %lu will point to %d\n",
2039 inode->i_ino, NEXT_ORPHAN(inode));
2040out_unlock:
Theodore Ts'o3b9d4ed2009-04-25 22:54:04 -04002041 mutex_unlock(&EXT4_SB(sb)->s_orphan_lock);
Mingming Cao617ba132006-10-11 01:20:53 -07002042 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002043 return err;
2044}
2045
2046/*
Mingming Cao617ba132006-10-11 01:20:53 -07002047 * ext4_orphan_del() removes an unlinked or truncated inode from the list
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002048 * of such inodes stored on disk, because it is finally being cleaned up.
2049 */
Mingming Cao617ba132006-10-11 01:20:53 -07002050int ext4_orphan_del(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002051{
2052 struct list_head *prev;
Mingming Cao617ba132006-10-11 01:20:53 -07002053 struct ext4_inode_info *ei = EXT4_I(inode);
2054 struct ext4_sb_info *sbi;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05002055 __u32 ino_next;
Mingming Cao617ba132006-10-11 01:20:53 -07002056 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002057 int err = 0;
2058
Curt Wohlgemuthd3d1faf2009-09-29 11:01:03 -04002059 /* ext4_handle_valid() assumes a valid handle_t pointer */
2060 if (handle && !ext4_handle_valid(handle))
Frank Mayhar03901312009-01-07 00:06:22 -05002061 return 0;
2062
Theodore Ts'o3b9d4ed2009-04-25 22:54:04 -04002063 mutex_lock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2064 if (list_empty(&ei->i_orphan))
2065 goto out;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002066
2067 ino_next = NEXT_ORPHAN(inode);
2068 prev = ei->i_orphan.prev;
Mingming Cao617ba132006-10-11 01:20:53 -07002069 sbi = EXT4_SB(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002070
2071 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2072
2073 list_del_init(&ei->i_orphan);
2074
2075 /* If we're on an error path, we may not have a valid
2076 * transaction handle with which to update the orphan list on
2077 * disk, but we still need to remove the inode from the linked
2078 * list in memory. */
Frank Mayhar03901312009-01-07 00:06:22 -05002079 if (sbi->s_journal && !handle)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002080 goto out;
2081
Mingming Cao617ba132006-10-11 01:20:53 -07002082 err = ext4_reserve_inode_write(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002083 if (err)
2084 goto out_err;
2085
2086 if (prev == &sbi->s_orphan) {
Theodore Ts'o498e5f22008-11-05 00:14:04 -05002087 jbd_debug(4, "superblock will point to %u\n", ino_next);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002088 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07002089 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002090 if (err)
2091 goto out_brelse;
2092 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
Curt Wohlgemuth73b50c12010-02-16 15:06:29 -05002093 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002094 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07002095 struct ext4_iloc iloc2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002096 struct inode *i_prev =
Mingming Cao617ba132006-10-11 01:20:53 -07002097 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002098
Theodore Ts'o498e5f22008-11-05 00:14:04 -05002099 jbd_debug(4, "orphan inode %lu will point to %u\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002100 i_prev->i_ino, ino_next);
Mingming Cao617ba132006-10-11 01:20:53 -07002101 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002102 if (err)
2103 goto out_brelse;
2104 NEXT_ORPHAN(i_prev) = ino_next;
Mingming Cao617ba132006-10-11 01:20:53 -07002105 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002106 }
2107 if (err)
2108 goto out_brelse;
2109 NEXT_ORPHAN(inode) = 0;
Mingming Cao617ba132006-10-11 01:20:53 -07002110 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002111
2112out_err:
Mingming Cao617ba132006-10-11 01:20:53 -07002113 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002114out:
Theodore Ts'o3b9d4ed2009-04-25 22:54:04 -04002115 mutex_unlock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002116 return err;
2117
2118out_brelse:
2119 brelse(iloc.bh);
2120 goto out_err;
2121}
2122
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002123static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002124{
2125 int retval;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002126 struct inode *inode;
2127 struct buffer_head *bh;
2128 struct ext4_dir_entry_2 *de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002129 handle_t *handle;
2130
2131 /* Initialize quotas before so that eventual writes go in
2132 * separate transaction */
Christoph Hellwig871a2932010-03-03 09:05:07 -05002133 dquot_initialize(dir);
2134 dquot_initialize(dentry->d_inode);
Christoph Hellwig907f4552010-03-03 09:05:06 -05002135
Mingming Cao617ba132006-10-11 01:20:53 -07002136 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002137 if (IS_ERR(handle))
2138 return PTR_ERR(handle);
2139
2140 retval = -ENOENT;
Theodore Ts'of702ba02008-09-22 15:21:01 -04002141 bh = ext4_find_entry(dir, &dentry->d_name, &de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002142 if (!bh)
2143 goto end_rmdir;
2144
2145 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05002146 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002147
2148 inode = dentry->d_inode;
2149
2150 retval = -EIO;
2151 if (le32_to_cpu(de->inode) != inode->i_ino)
2152 goto end_rmdir;
2153
2154 retval = -ENOTEMPTY;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002155 if (!empty_dir(inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002156 goto end_rmdir;
2157
Mingming Cao617ba132006-10-11 01:20:53 -07002158 retval = ext4_delete_entry(handle, dir, de, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002159 if (retval)
2160 goto end_rmdir;
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002161 if (!EXT4_DIR_LINK_EMPTY(inode))
Eric Sandeen12062dd2010-02-15 14:19:27 -05002162 ext4_warning(inode->i_sb,
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002163 "empty directory has too many links (%d)",
2164 inode->i_nlink);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002165 inode->i_version++;
2166 clear_nlink(inode);
2167 /* There's no need to set i_disksize: the fact that i_nlink is
2168 * zero will ensure that the right thing happens during any
2169 * recovery. */
2170 inode->i_size = 0;
Mingming Cao617ba132006-10-11 01:20:53 -07002171 ext4_orphan_add(handle, inode);
Kalpak Shahef7f3832007-07-18 09:15:20 -04002172 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002173 ext4_mark_inode_dirty(handle, inode);
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002174 ext4_dec_count(handle, dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002175 ext4_update_dx_flag(dir);
2176 ext4_mark_inode_dirty(handle, dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002177
2178end_rmdir:
Mingming Cao617ba132006-10-11 01:20:53 -07002179 ext4_journal_stop(handle);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002180 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002181 return retval;
2182}
2183
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002184static int ext4_unlink(struct inode *dir, struct dentry *dentry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002185{
2186 int retval;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002187 struct inode *inode;
2188 struct buffer_head *bh;
2189 struct ext4_dir_entry_2 *de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002190 handle_t *handle;
2191
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04002192 trace_ext4_unlink_enter(dir, dentry);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002193 /* Initialize quotas before so that eventual writes go
2194 * in separate transaction */
Christoph Hellwig871a2932010-03-03 09:05:07 -05002195 dquot_initialize(dir);
2196 dquot_initialize(dentry->d_inode);
Christoph Hellwig907f4552010-03-03 09:05:06 -05002197
Mingming Cao617ba132006-10-11 01:20:53 -07002198 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002199 if (IS_ERR(handle))
2200 return PTR_ERR(handle);
2201
2202 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05002203 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002204
2205 retval = -ENOENT;
Theodore Ts'of702ba02008-09-22 15:21:01 -04002206 bh = ext4_find_entry(dir, &dentry->d_name, &de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002207 if (!bh)
2208 goto end_unlink;
2209
2210 inode = dentry->d_inode;
2211
2212 retval = -EIO;
2213 if (le32_to_cpu(de->inode) != inode->i_ino)
2214 goto end_unlink;
2215
2216 if (!inode->i_nlink) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05002217 ext4_warning(inode->i_sb,
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002218 "Deleting nonexistent file (%lu), %d",
2219 inode->i_ino, inode->i_nlink);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002220 inode->i_nlink = 1;
2221 }
Mingming Cao617ba132006-10-11 01:20:53 -07002222 retval = ext4_delete_entry(handle, dir, de, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002223 if (retval)
2224 goto end_unlink;
Kalpak Shahef7f3832007-07-18 09:15:20 -04002225 dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002226 ext4_update_dx_flag(dir);
2227 ext4_mark_inode_dirty(handle, dir);
Theodore Ts'o825f1482008-02-15 15:00:38 -05002228 drop_nlink(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002229 if (!inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07002230 ext4_orphan_add(handle, inode);
Kalpak Shahef7f3832007-07-18 09:15:20 -04002231 inode->i_ctime = ext4_current_time(inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002232 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002233 retval = 0;
2234
2235end_unlink:
Mingming Cao617ba132006-10-11 01:20:53 -07002236 ext4_journal_stop(handle);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002237 brelse(bh);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04002238 trace_ext4_unlink_exit(dentry, retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002239 return retval;
2240}
2241
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002242static int ext4_symlink(struct inode *dir,
2243 struct dentry *dentry, const char *symname)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002244{
2245 handle_t *handle;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002246 struct inode *inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002247 int l, err, retries = 0;
Jan Karadf5e6222011-05-03 11:12:58 -04002248 int credits;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002249
2250 l = strlen(symname)+1;
2251 if (l > dir->i_sb->s_blocksize)
2252 return -ENAMETOOLONG;
2253
Christoph Hellwig871a2932010-03-03 09:05:07 -05002254 dquot_initialize(dir);
Christoph Hellwig907f4552010-03-03 09:05:06 -05002255
Jan Karadf5e6222011-05-03 11:12:58 -04002256 if (l > EXT4_N_BLOCKS * 4) {
2257 /*
2258 * For non-fast symlinks, we just allocate inode and put it on
2259 * orphan list in the first transaction => we need bitmap,
2260 * group descriptor, sb, inode block, quota blocks.
2261 */
2262 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb);
2263 } else {
2264 /*
2265 * Fast symlink. We have to add entry to directory
2266 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
2267 * allocate new inode (bitmap, group descriptor, inode block,
2268 * quota blocks, sb is already counted in previous macros).
2269 */
2270 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2271 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2272 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb);
2273 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002274retry:
Jan Karadf5e6222011-05-03 11:12:58 -04002275 handle = ext4_journal_start(dir, credits);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002276 if (IS_ERR(handle))
2277 return PTR_ERR(handle);
2278
2279 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05002280 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002281
Andreas Dilger11013912009-06-13 11:45:35 -04002282 inode = ext4_new_inode(handle, dir, S_IFLNK|S_IRWXUGO,
2283 &dentry->d_name, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002284 err = PTR_ERR(inode);
2285 if (IS_ERR(inode))
2286 goto out_stop;
2287
Jan Karadf5e6222011-05-03 11:12:58 -04002288 if (l > EXT4_N_BLOCKS * 4) {
Mingming Cao617ba132006-10-11 01:20:53 -07002289 inode->i_op = &ext4_symlink_inode_operations;
2290 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002291 /*
Jan Karadf5e6222011-05-03 11:12:58 -04002292 * We cannot call page_symlink() with transaction started
2293 * because it calls into ext4_write_begin() which can wait
2294 * for transaction commit if we are running out of space
2295 * and thus we deadlock. So we have to stop transaction now
2296 * and restart it when symlink contents is written.
2297 *
2298 * To keep fs consistent in case of crash, we have to put inode
2299 * to orphan list in the mean time.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002300 */
Jan Karadf5e6222011-05-03 11:12:58 -04002301 drop_nlink(inode);
2302 err = ext4_orphan_add(handle, inode);
2303 ext4_journal_stop(handle);
2304 if (err)
2305 goto err_drop_inode;
Nick Piggin54566b22009-01-04 12:00:53 -08002306 err = __page_symlink(inode, symname, l, 1);
Jan Karadf5e6222011-05-03 11:12:58 -04002307 if (err)
2308 goto err_drop_inode;
2309 /*
2310 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
2311 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2312 */
2313 handle = ext4_journal_start(dir,
2314 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2315 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
2316 if (IS_ERR(handle)) {
2317 err = PTR_ERR(handle);
2318 goto err_drop_inode;
2319 }
2320 inc_nlink(inode);
2321 err = ext4_orphan_del(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002322 if (err) {
Jan Karadf5e6222011-05-03 11:12:58 -04002323 ext4_journal_stop(handle);
Theodore Ts'o825f1482008-02-15 15:00:38 -05002324 clear_nlink(inode);
Jan Karadf5e6222011-05-03 11:12:58 -04002325 goto err_drop_inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002326 }
2327 } else {
Aneesh Kumar K.Ve65187e2008-04-29 08:11:12 -04002328 /* clear the extent format for fast symlink */
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04002329 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
Mingming Cao617ba132006-10-11 01:20:53 -07002330 inode->i_op = &ext4_fast_symlink_inode_operations;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002331 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002332 inode->i_size = l-1;
2333 }
Mingming Cao617ba132006-10-11 01:20:53 -07002334 EXT4_I(inode)->i_disksize = inode->i_size;
2335 err = ext4_add_nondir(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002336out_stop:
Mingming Cao617ba132006-10-11 01:20:53 -07002337 ext4_journal_stop(handle);
2338 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002339 goto retry;
2340 return err;
Jan Karadf5e6222011-05-03 11:12:58 -04002341err_drop_inode:
2342 unlock_new_inode(inode);
2343 iput(inode);
2344 return err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002345}
2346
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002347static int ext4_link(struct dentry *old_dentry,
2348 struct inode *dir, struct dentry *dentry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002349{
2350 handle_t *handle;
2351 struct inode *inode = old_dentry->d_inode;
2352 int err, retries = 0;
2353
Theodore Ts'ob05ab1d2009-08-29 21:08:08 -04002354 if (inode->i_nlink >= EXT4_LINK_MAX)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002355 return -EMLINK;
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002356
Christoph Hellwig871a2932010-03-03 09:05:07 -05002357 dquot_initialize(dir);
Christoph Hellwig907f4552010-03-03 09:05:06 -05002358
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002359retry:
Mingming Cao617ba132006-10-11 01:20:53 -07002360 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2361 EXT4_INDEX_EXTRA_TRANS_BLOCKS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002362 if (IS_ERR(handle))
2363 return PTR_ERR(handle);
2364
2365 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05002366 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002367
Kalpak Shahef7f3832007-07-18 09:15:20 -04002368 inode->i_ctime = ext4_current_time(inode);
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002369 ext4_inc_count(handle, inode);
Al Viro7de9c6e2010-10-23 11:11:40 -04002370 ihold(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002371
Al Viro6b38e842008-12-30 02:03:31 -05002372 err = ext4_add_entry(handle, dentry, inode);
2373 if (!err) {
2374 ext4_mark_inode_dirty(handle, inode);
2375 d_instantiate(dentry, inode);
2376 } else {
2377 drop_nlink(inode);
2378 iput(inode);
2379 }
Mingming Cao617ba132006-10-11 01:20:53 -07002380 ext4_journal_stop(handle);
2381 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002382 goto retry;
2383 return err;
2384}
2385
Wei Yongjun3d0518f2009-02-14 23:01:36 -05002386#define PARENT_INO(buffer, size) \
2387 (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer), size)->inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002388
2389/*
2390 * Anybody can rename anything with this: the permission checks are left to the
2391 * higher-level routines.
2392 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002393static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
2394 struct inode *new_dir, struct dentry *new_dentry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002395{
2396 handle_t *handle;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002397 struct inode *old_inode, *new_inode;
2398 struct buffer_head *old_bh, *new_bh, *dir_bh;
2399 struct ext4_dir_entry_2 *old_de, *new_de;
Theodore Ts'o8750c6d2009-02-23 23:05:27 -05002400 int retval, force_da_alloc = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002401
Christoph Hellwig871a2932010-03-03 09:05:07 -05002402 dquot_initialize(old_dir);
2403 dquot_initialize(new_dir);
Christoph Hellwig907f4552010-03-03 09:05:06 -05002404
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002405 old_bh = new_bh = dir_bh = NULL;
2406
2407 /* Initialize quotas before so that eventual writes go
2408 * in separate transaction */
2409 if (new_dentry->d_inode)
Christoph Hellwig871a2932010-03-03 09:05:07 -05002410 dquot_initialize(new_dentry->d_inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002411 handle = ext4_journal_start(old_dir, 2 *
2412 EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2413 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002414 if (IS_ERR(handle))
2415 return PTR_ERR(handle);
2416
2417 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
Frank Mayhar03901312009-01-07 00:06:22 -05002418 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002419
Theodore Ts'of702ba02008-09-22 15:21:01 -04002420 old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002421 /*
2422 * Check for inode number is _not_ due to possible IO errors.
2423 * We might rmdir the source, keep it as pwd of some process
2424 * and merrily kill the link to whatever was created under the
2425 * same name. Goodbye sticky bit ;-<
2426 */
2427 old_inode = old_dentry->d_inode;
2428 retval = -ENOENT;
2429 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2430 goto end_rename;
2431
2432 new_inode = new_dentry->d_inode;
Theodore Ts'of702ba02008-09-22 15:21:01 -04002433 new_bh = ext4_find_entry(new_dir, &new_dentry->d_name, &new_de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002434 if (new_bh) {
2435 if (!new_inode) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002436 brelse(new_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002437 new_bh = NULL;
2438 }
2439 }
2440 if (S_ISDIR(old_inode->i_mode)) {
2441 if (new_inode) {
2442 retval = -ENOTEMPTY;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002443 if (!empty_dir(new_inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002444 goto end_rename;
2445 }
2446 retval = -EIO;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002447 dir_bh = ext4_bread(handle, old_inode, 0, 0, &retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002448 if (!dir_bh)
2449 goto end_rename;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05002450 if (le32_to_cpu(PARENT_INO(dir_bh->b_data,
2451 old_dir->i_sb->s_blocksize)) != old_dir->i_ino)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002452 goto end_rename;
2453 retval = -EMLINK;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002454 if (!new_inode && new_dir != old_dir &&
Aneesh Kumar K.V2c94eb82009-08-28 21:43:15 -04002455 EXT4_DIR_LINK_MAX(new_dir))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002456 goto end_rename;
Amir Goldsteinef607892011-03-20 21:18:44 -04002457 BUFFER_TRACE(dir_bh, "get_write_access");
2458 retval = ext4_journal_get_write_access(handle, dir_bh);
2459 if (retval)
2460 goto end_rename;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002461 }
2462 if (!new_bh) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002463 retval = ext4_add_entry(handle, new_dentry, old_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002464 if (retval)
2465 goto end_rename;
2466 } else {
2467 BUFFER_TRACE(new_bh, "get write access");
Amir Goldsteinef607892011-03-20 21:18:44 -04002468 retval = ext4_journal_get_write_access(handle, new_bh);
2469 if (retval)
2470 goto end_rename;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002471 new_de->inode = cpu_to_le32(old_inode->i_ino);
Mingming Cao617ba132006-10-11 01:20:53 -07002472 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2473 EXT4_FEATURE_INCOMPAT_FILETYPE))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002474 new_de->file_type = old_de->file_type;
2475 new_dir->i_version++;
Jan Kara53b7e9f2008-04-29 22:02:11 -04002476 new_dir->i_ctime = new_dir->i_mtime =
2477 ext4_current_time(new_dir);
2478 ext4_mark_inode_dirty(handle, new_dir);
Frank Mayhar03901312009-01-07 00:06:22 -05002479 BUFFER_TRACE(new_bh, "call ext4_handle_dirty_metadata");
Theodore Ts'ob4097142011-01-10 12:46:59 -05002480 retval = ext4_handle_dirty_metadata(handle, new_dir, new_bh);
2481 if (unlikely(retval)) {
2482 ext4_std_error(new_dir->i_sb, retval);
2483 goto end_rename;
2484 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002485 brelse(new_bh);
2486 new_bh = NULL;
2487 }
2488
2489 /*
2490 * Like most other Unix systems, set the ctime for inodes on a
2491 * rename.
2492 */
Kalpak Shahef7f3832007-07-18 09:15:20 -04002493 old_inode->i_ctime = ext4_current_time(old_inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002494 ext4_mark_inode_dirty(handle, old_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002495
2496 /*
2497 * ok, that's it
2498 */
2499 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2500 old_de->name_len != old_dentry->d_name.len ||
2501 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
Mingming Cao617ba132006-10-11 01:20:53 -07002502 (retval = ext4_delete_entry(handle, old_dir,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002503 old_de, old_bh)) == -ENOENT) {
2504 /* old_de could have moved from under us during htree split, so
2505 * make sure that we are deleting the right entry. We might
2506 * also be pointing to a stale entry in the unused part of
2507 * old_bh so just checking inum and the name isn't enough. */
2508 struct buffer_head *old_bh2;
Mingming Cao617ba132006-10-11 01:20:53 -07002509 struct ext4_dir_entry_2 *old_de2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002510
Theodore Ts'of702ba02008-09-22 15:21:01 -04002511 old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002512 if (old_bh2) {
Mingming Cao617ba132006-10-11 01:20:53 -07002513 retval = ext4_delete_entry(handle, old_dir,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002514 old_de2, old_bh2);
2515 brelse(old_bh2);
2516 }
2517 }
2518 if (retval) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05002519 ext4_warning(old_dir->i_sb,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002520 "Deleting old file (%lu), %d, error=%d",
2521 old_dir->i_ino, old_dir->i_nlink, retval);
2522 }
2523
2524 if (new_inode) {
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002525 ext4_dec_count(handle, new_inode);
Kalpak Shahef7f3832007-07-18 09:15:20 -04002526 new_inode->i_ctime = ext4_current_time(new_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002527 }
Kalpak Shahef7f3832007-07-18 09:15:20 -04002528 old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002529 ext4_update_dx_flag(old_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002530 if (dir_bh) {
Wei Yongjun3d0518f2009-02-14 23:01:36 -05002531 PARENT_INO(dir_bh->b_data, new_dir->i_sb->s_blocksize) =
2532 cpu_to_le32(new_dir->i_ino);
Frank Mayhar03901312009-01-07 00:06:22 -05002533 BUFFER_TRACE(dir_bh, "call ext4_handle_dirty_metadata");
Theodore Ts'ob4097142011-01-10 12:46:59 -05002534 retval = ext4_handle_dirty_metadata(handle, old_dir, dir_bh);
2535 if (retval) {
2536 ext4_std_error(old_dir->i_sb, retval);
2537 goto end_rename;
2538 }
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002539 ext4_dec_count(handle, old_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002540 if (new_inode) {
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002541 /* checked empty_dir above, can't have another parent,
Theodore Ts'o825f1482008-02-15 15:00:38 -05002542 * ext4_dec_count() won't work for many-linked dirs */
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002543 new_inode->i_nlink = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002544 } else {
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002545 ext4_inc_count(handle, new_dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002546 ext4_update_dx_flag(new_dir);
2547 ext4_mark_inode_dirty(handle, new_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002548 }
2549 }
Mingming Cao617ba132006-10-11 01:20:53 -07002550 ext4_mark_inode_dirty(handle, old_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002551 if (new_inode) {
Mingming Cao617ba132006-10-11 01:20:53 -07002552 ext4_mark_inode_dirty(handle, new_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002553 if (!new_inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07002554 ext4_orphan_add(handle, new_inode);
Theodore Ts'oafd46722009-03-16 23:12:23 -04002555 if (!test_opt(new_dir->i_sb, NO_AUTO_DA_ALLOC))
2556 force_da_alloc = 1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002557 }
2558 retval = 0;
2559
2560end_rename:
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002561 brelse(dir_bh);
2562 brelse(old_bh);
2563 brelse(new_bh);
Mingming Cao617ba132006-10-11 01:20:53 -07002564 ext4_journal_stop(handle);
Theodore Ts'o8750c6d2009-02-23 23:05:27 -05002565 if (retval == 0 && force_da_alloc)
2566 ext4_alloc_da_blocks(old_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002567 return retval;
2568}
2569
2570/*
2571 * directories can handle most operations...
2572 */
Arjan van de Ven754661f2007-02-12 00:55:38 -08002573const struct inode_operations ext4_dir_inode_operations = {
Mingming Cao617ba132006-10-11 01:20:53 -07002574 .create = ext4_create,
2575 .lookup = ext4_lookup,
2576 .link = ext4_link,
2577 .unlink = ext4_unlink,
2578 .symlink = ext4_symlink,
2579 .mkdir = ext4_mkdir,
2580 .rmdir = ext4_rmdir,
2581 .mknod = ext4_mknod,
2582 .rename = ext4_rename,
2583 .setattr = ext4_setattr,
Theodore Ts'o03010a32008-10-10 20:02:48 -04002584#ifdef CONFIG_EXT4_FS_XATTR
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002585 .setxattr = generic_setxattr,
2586 .getxattr = generic_getxattr,
Mingming Cao617ba132006-10-11 01:20:53 -07002587 .listxattr = ext4_listxattr,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002588 .removexattr = generic_removexattr,
2589#endif
Linus Torvalds1d5ccd12009-08-28 12:12:24 -07002590 .check_acl = ext4_check_acl,
Aneesh Kumar K.Vabc87462009-05-02 22:54:32 -04002591 .fiemap = ext4_fiemap,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002592};
2593
Arjan van de Ven754661f2007-02-12 00:55:38 -08002594const struct inode_operations ext4_special_inode_operations = {
Mingming Cao617ba132006-10-11 01:20:53 -07002595 .setattr = ext4_setattr,
Theodore Ts'o03010a32008-10-10 20:02:48 -04002596#ifdef CONFIG_EXT4_FS_XATTR
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002597 .setxattr = generic_setxattr,
2598 .getxattr = generic_getxattr,
Mingming Cao617ba132006-10-11 01:20:53 -07002599 .listxattr = ext4_listxattr,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002600 .removexattr = generic_removexattr,
2601#endif
Linus Torvalds1d5ccd12009-08-28 12:12:24 -07002602 .check_acl = ext4_check_acl,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002603};