blob: a43e6617b35119e6bb46815e4f73aa2a518a4e30 [file] [log] [blame]
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001/*
Mingming Cao617ba132006-10-11 01:20:53 -07002 * linux/fs/ext4/namei.c
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003 *
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
8 *
9 * from
10 *
11 * linux/fs/minix/namei.c
12 *
13 * Copyright (C) 1991, 1992 Linus Torvalds
14 *
15 * Big-endian to little-endian byte-swapping/bitmaps by
16 * David S. Miller (davem@caip.rutgers.edu), 1995
17 * Directory entry file type support and forward compatibility hooks
18 * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
19 * Hash Tree Directory indexing (c)
20 * Daniel Phillips, 2001
21 * Hash Tree Directory indexing porting
22 * Christopher Li, 2002
23 * Hash Tree Directory indexing cleanup
24 * Theodore Ts'o, 2002
25 */
26
27#include <linux/fs.h>
28#include <linux/pagemap.h>
Mingming Caodab291a2006-10-11 01:21:01 -070029#include <linux/jbd2.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070030#include <linux/time.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070031#include <linux/fcntl.h>
32#include <linux/stat.h>
33#include <linux/string.h>
34#include <linux/quotaops.h>
35#include <linux/buffer_head.h>
36#include <linux/bio.h>
Christoph Hellwig3dcf5452008-04-29 18:13:32 -040037#include "ext4.h"
38#include "ext4_jbd2.h"
Dave Kleikampac27a0e2006-10-11 01:20:50 -070039
Dave Kleikampac27a0e2006-10-11 01:20:50 -070040#include "xattr.h"
41#include "acl.h"
42
43/*
44 * define how far ahead to read directories while searching them.
45 */
46#define NAMEI_RA_CHUNKS 2
47#define NAMEI_RA_BLOCKS 4
Dave Kleikamp8c55e202007-05-24 13:04:54 -040048#define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070049#define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b))
50
Mingming Cao617ba132006-10-11 01:20:53 -070051static struct buffer_head *ext4_append(handle_t *handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -070052 struct inode *inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -050053 ext4_lblk_t *block, int *err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070054{
55 struct buffer_head *bh;
56
57 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
58
Akinobu Mitaa8716112008-04-17 10:38:59 -040059 bh = ext4_bread(handle, inode, *block, 1, err);
60 if (bh) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -070061 inode->i_size += inode->i_sb->s_blocksize;
Mingming Cao617ba132006-10-11 01:20:53 -070062 EXT4_I(inode)->i_disksize = inode->i_size;
Akinobu Mitaa8716112008-04-17 10:38:59 -040063 *err = ext4_journal_get_write_access(handle, bh);
64 if (*err) {
65 brelse(bh);
66 bh = NULL;
67 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -070068 }
69 return bh;
70}
71
72#ifndef assert
73#define assert(test) J_ASSERT(test)
74#endif
75
Dave Kleikampac27a0e2006-10-11 01:20:50 -070076#ifdef DX_DEBUG
77#define dxtrace(command) command
78#else
79#define dxtrace(command)
80#endif
81
82struct fake_dirent
83{
84 __le32 inode;
85 __le16 rec_len;
86 u8 name_len;
87 u8 file_type;
88};
89
90struct dx_countlimit
91{
92 __le16 limit;
93 __le16 count;
94};
95
96struct dx_entry
97{
98 __le32 hash;
99 __le32 block;
100};
101
102/*
103 * dx_root_info is laid out so that if it should somehow get overlaid by a
104 * dirent the two low bits of the hash version will be zero. Therefore, the
105 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
106 */
107
108struct dx_root
109{
110 struct fake_dirent dot;
111 char dot_name[4];
112 struct fake_dirent dotdot;
113 char dotdot_name[4];
114 struct dx_root_info
115 {
116 __le32 reserved_zero;
117 u8 hash_version;
118 u8 info_length; /* 8 */
119 u8 indirect_levels;
120 u8 unused_flags;
121 }
122 info;
123 struct dx_entry entries[0];
124};
125
126struct dx_node
127{
128 struct fake_dirent fake;
129 struct dx_entry entries[0];
130};
131
132
133struct dx_frame
134{
135 struct buffer_head *bh;
136 struct dx_entry *entries;
137 struct dx_entry *at;
138};
139
140struct dx_map_entry
141{
142 u32 hash;
Eric Sandeenef2b02d2007-09-18 22:46:42 -0700143 u16 offs;
144 u16 size;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700145};
146
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500147static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
148static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400149static inline unsigned dx_get_hash(struct dx_entry *entry);
150static void dx_set_hash(struct dx_entry *entry, unsigned value);
151static unsigned dx_get_count(struct dx_entry *entries);
152static unsigned dx_get_limit(struct dx_entry *entries);
153static void dx_set_count(struct dx_entry *entries, unsigned value);
154static void dx_set_limit(struct dx_entry *entries, unsigned value);
155static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
156static unsigned dx_node_limit(struct inode *dir);
Theodore Ts'of702ba02008-09-22 15:21:01 -0400157static struct dx_frame *dx_probe(const struct qstr *d_name,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700158 struct inode *dir,
159 struct dx_hash_info *hinfo,
160 struct dx_frame *frame,
161 int *err);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400162static void dx_release(struct dx_frame *frames);
Theodore Ts'o8bad4592009-02-14 21:46:54 -0500163static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400164 struct dx_hash_info *hinfo, struct dx_map_entry map[]);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700165static void dx_sort_map(struct dx_map_entry *map, unsigned count);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400166static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500167 struct dx_map_entry *offsets, int count, unsigned blocksize);
Theodore Ts'o8bad4592009-02-14 21:46:54 -0500168static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500169static void dx_insert_block(struct dx_frame *frame,
170 u32 hash, ext4_lblk_t block);
Mingming Cao617ba132006-10-11 01:20:53 -0700171static int ext4_htree_next_block(struct inode *dir, __u32 hash,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700172 struct dx_frame *frame,
173 struct dx_frame *frames,
174 __u32 *start_hash);
Theodore Ts'of702ba02008-09-22 15:21:01 -0400175static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
176 const struct qstr *d_name,
177 struct ext4_dir_entry_2 **res_dir,
178 int *err);
Mingming Cao617ba132006-10-11 01:20:53 -0700179static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700180 struct inode *inode);
181
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500182unsigned int ext4_rec_len_from_disk(__le16 dlen, unsigned blocksize)
183{
184 unsigned len = le16_to_cpu(dlen);
185
186 if (len == EXT4_MAX_REC_LEN || len == 0)
187 return blocksize;
188 return (len & 65532) | ((len & 3) << 16);
189}
Theodore Ts'o60e66792010-05-17 07:00:00 -0400190
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500191__le16 ext4_rec_len_to_disk(unsigned len, unsigned blocksize)
192{
193 if ((len > blocksize) || (blocksize > (1 << 18)) || (len & 3))
194 BUG();
195 if (len < 65536)
196 return cpu_to_le16(len);
197 if (len == blocksize) {
198 if (blocksize == 65536)
199 return cpu_to_le16(EXT4_MAX_REC_LEN);
Theodore Ts'o60e66792010-05-17 07:00:00 -0400200 else
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500201 return cpu_to_le16(0);
202 }
203 return cpu_to_le16((len & 65532) | ((len >> 16) & 3));
204}
205
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700206/*
Li Zefanf795e142008-07-11 19:27:31 -0400207 * p is at least 6 bytes before the end of page
208 */
209static inline struct ext4_dir_entry_2 *
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500210ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
Li Zefanf795e142008-07-11 19:27:31 -0400211{
212 return (struct ext4_dir_entry_2 *)((char *)p +
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500213 ext4_rec_len_from_disk(p->rec_len, blocksize));
Li Zefanf795e142008-07-11 19:27:31 -0400214}
215
216/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700217 * Future: use high four bits of block for coalesce-on-delete flags
218 * Mask them off for now.
219 */
220
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500221static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700222{
223 return le32_to_cpu(entry->block) & 0x00ffffff;
224}
225
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500226static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700227{
228 entry->block = cpu_to_le32(value);
229}
230
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400231static inline unsigned dx_get_hash(struct dx_entry *entry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700232{
233 return le32_to_cpu(entry->hash);
234}
235
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400236static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700237{
238 entry->hash = cpu_to_le32(value);
239}
240
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400241static inline unsigned dx_get_count(struct dx_entry *entries)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700242{
243 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
244}
245
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400246static inline unsigned dx_get_limit(struct dx_entry *entries)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700247{
248 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
249}
250
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400251static inline void dx_set_count(struct dx_entry *entries, unsigned value)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700252{
253 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
254}
255
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400256static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700257{
258 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
259}
260
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400261static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700262{
Mingming Cao617ba132006-10-11 01:20:53 -0700263 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
264 EXT4_DIR_REC_LEN(2) - infosize;
Li Zefand9c769b2008-07-11 19:27:31 -0400265 return entry_space / sizeof(struct dx_entry);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700266}
267
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400268static inline unsigned dx_node_limit(struct inode *dir)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700269{
Mingming Cao617ba132006-10-11 01:20:53 -0700270 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
Li Zefand9c769b2008-07-11 19:27:31 -0400271 return entry_space / sizeof(struct dx_entry);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700272}
273
274/*
275 * Debug
276 */
277#ifdef DX_DEBUG
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400278static void dx_show_index(char * label, struct dx_entry *entries)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700279{
Andrew Morton63f57932006-10-11 01:21:24 -0700280 int i, n = dx_get_count (entries);
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400281 printk(KERN_DEBUG "%s index ", label);
Andrew Morton63f57932006-10-11 01:21:24 -0700282 for (i = 0; i < n; i++) {
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400283 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500284 0, (unsigned long)dx_get_block(entries + i));
Andrew Morton63f57932006-10-11 01:21:24 -0700285 }
286 printk("\n");
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700287}
288
289struct stats
290{
291 unsigned names;
292 unsigned space;
293 unsigned bcount;
294};
295
Mingming Cao617ba132006-10-11 01:20:53 -0700296static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700297 int size, int show_names)
298{
299 unsigned names = 0, space = 0;
300 char *base = (char *) de;
301 struct dx_hash_info h = *hinfo;
302
303 printk("names: ");
304 while ((char *) de < base + size)
305 {
306 if (de->inode)
307 {
308 if (show_names)
309 {
310 int len = de->name_len;
311 char *name = de->name;
312 while (len--) printk("%c", *name++);
Mingming Cao617ba132006-10-11 01:20:53 -0700313 ext4fs_dirhash(de->name, de->name_len, &h);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700314 printk(":%x.%u ", h.hash,
315 ((char *) de - base));
316 }
Mingming Cao617ba132006-10-11 01:20:53 -0700317 space += EXT4_DIR_REC_LEN(de->name_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700318 names++;
319 }
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500320 de = ext4_next_entry(de, size);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700321 }
322 printk("(%i)\n", names);
323 return (struct stats) { names, space, 1 };
324}
325
326struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
327 struct dx_entry *entries, int levels)
328{
329 unsigned blocksize = dir->i_sb->s_blocksize;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400330 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700331 unsigned bcount = 0;
332 struct buffer_head *bh;
333 int err;
334 printk("%i indexed blocks...\n", count);
335 for (i = 0; i < count; i++, entries++)
336 {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500337 ext4_lblk_t block = dx_get_block(entries);
338 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700339 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
340 struct stats stats;
341 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
Mingming Cao617ba132006-10-11 01:20:53 -0700342 if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700343 stats = levels?
344 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
Mingming Cao617ba132006-10-11 01:20:53 -0700345 dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700346 names += stats.names;
347 space += stats.space;
348 bcount += stats.bcount;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400349 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700350 }
351 if (bcount)
Theodore Ts'o60e66792010-05-17 07:00:00 -0400352 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400353 levels ? "" : " ", names, space/bcount,
354 (space/bcount)*100/blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700355 return (struct stats) { names, space, bcount};
356}
357#endif /* DX_DEBUG */
358
359/*
360 * Probe for a directory leaf block to search.
361 *
362 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
363 * error in the directory index, and the caller should fall back to
364 * searching the directory normally. The callers of dx_probe **MUST**
365 * check for this error code, and make sure it never gets reflected
366 * back to userspace.
367 */
368static struct dx_frame *
Theodore Ts'of702ba02008-09-22 15:21:01 -0400369dx_probe(const struct qstr *d_name, struct inode *dir,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700370 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
371{
372 unsigned count, indirect;
373 struct dx_entry *at, *entries, *p, *q, *m;
374 struct dx_root *root;
375 struct buffer_head *bh;
376 struct dx_frame *frame = frame_in;
377 u32 hash;
378
379 frame->bh = NULL;
Mingming Cao617ba132006-10-11 01:20:53 -0700380 if (!(bh = ext4_bread (NULL,dir, 0, 0, err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700381 goto fail;
382 root = (struct dx_root *) bh->b_data;
383 if (root->info.hash_version != DX_HASH_TEA &&
384 root->info.hash_version != DX_HASH_HALF_MD4 &&
385 root->info.hash_version != DX_HASH_LEGACY) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500386 ext4_warning(dir->i_sb, "Unrecognised inode hash code %d",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700387 root->info.hash_version);
388 brelse(bh);
389 *err = ERR_BAD_DX_DIR;
390 goto fail;
391 }
392 hinfo->hash_version = root->info.hash_version;
Theodore Ts'of99b2582008-10-28 13:21:44 -0400393 if (hinfo->hash_version <= DX_HASH_TEA)
394 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
Mingming Cao617ba132006-10-11 01:20:53 -0700395 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
Theodore Ts'of702ba02008-09-22 15:21:01 -0400396 if (d_name)
397 ext4fs_dirhash(d_name->name, d_name->len, hinfo);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700398 hash = hinfo->hash;
399
400 if (root->info.unused_flags & 1) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500401 ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700402 root->info.unused_flags);
403 brelse(bh);
404 *err = ERR_BAD_DX_DIR;
405 goto fail;
406 }
407
408 if ((indirect = root->info.indirect_levels) > 1) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500409 ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700410 root->info.indirect_levels);
411 brelse(bh);
412 *err = ERR_BAD_DX_DIR;
413 goto fail;
414 }
415
416 entries = (struct dx_entry *) (((char *)&root->info) +
417 root->info.info_length);
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700418
419 if (dx_get_limit(entries) != dx_root_limit(dir,
420 root->info.info_length)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500421 ext4_warning(dir->i_sb, "dx entry: limit != root limit");
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700422 brelse(bh);
423 *err = ERR_BAD_DX_DIR;
424 goto fail;
425 }
426
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400427 dxtrace(printk("Look up %x", hash));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700428 while (1)
429 {
430 count = dx_get_count(entries);
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700431 if (!count || count > dx_get_limit(entries)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500432 ext4_warning(dir->i_sb,
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700433 "dx entry: no count or count > limit");
434 brelse(bh);
435 *err = ERR_BAD_DX_DIR;
436 goto fail2;
437 }
438
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700439 p = entries + 1;
440 q = entries + count - 1;
441 while (p <= q)
442 {
443 m = p + (q - p)/2;
444 dxtrace(printk("."));
445 if (dx_get_hash(m) > hash)
446 q = m - 1;
447 else
448 p = m + 1;
449 }
450
451 if (0) // linear search cross check
452 {
453 unsigned n = count - 1;
454 at = entries;
455 while (n--)
456 {
457 dxtrace(printk(","));
458 if (dx_get_hash(++at) > hash)
459 {
460 at--;
461 break;
462 }
463 }
464 assert (at == p - 1);
465 }
466
467 at = p - 1;
468 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
469 frame->bh = bh;
470 frame->entries = entries;
471 frame->at = at;
472 if (!indirect--) return frame;
Mingming Cao617ba132006-10-11 01:20:53 -0700473 if (!(bh = ext4_bread (NULL,dir, dx_get_block(at), 0, err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700474 goto fail2;
475 at = entries = ((struct dx_node *) bh->b_data)->entries;
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700476 if (dx_get_limit(entries) != dx_node_limit (dir)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500477 ext4_warning(dir->i_sb,
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700478 "dx entry: limit != node limit");
479 brelse(bh);
480 *err = ERR_BAD_DX_DIR;
481 goto fail2;
482 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700483 frame++;
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700484 frame->bh = NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700485 }
486fail2:
487 while (frame >= frame_in) {
488 brelse(frame->bh);
489 frame--;
490 }
491fail:
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700492 if (*err == ERR_BAD_DX_DIR)
Eric Sandeen12062dd2010-02-15 14:19:27 -0500493 ext4_warning(dir->i_sb,
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700494 "Corrupt dir inode %ld, running e2fsck is "
495 "recommended.", dir->i_ino);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700496 return NULL;
497}
498
499static void dx_release (struct dx_frame *frames)
500{
501 if (frames[0].bh == NULL)
502 return;
503
504 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
505 brelse(frames[1].bh);
506 brelse(frames[0].bh);
507}
508
509/*
510 * This function increments the frame pointer to search the next leaf
511 * block, and reads in the necessary intervening nodes if the search
512 * should be necessary. Whether or not the search is necessary is
513 * controlled by the hash parameter. If the hash value is even, then
514 * the search is only continued if the next block starts with that
515 * hash value. This is used if we are searching for a specific file.
516 *
517 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
518 *
519 * This function returns 1 if the caller should continue to search,
520 * or 0 if it should not. If there is an error reading one of the
521 * index blocks, it will a negative error code.
522 *
523 * If start_hash is non-null, it will be filled in with the starting
524 * hash of the next page.
525 */
Mingming Cao617ba132006-10-11 01:20:53 -0700526static int ext4_htree_next_block(struct inode *dir, __u32 hash,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700527 struct dx_frame *frame,
528 struct dx_frame *frames,
529 __u32 *start_hash)
530{
531 struct dx_frame *p;
532 struct buffer_head *bh;
533 int err, num_frames = 0;
534 __u32 bhash;
535
536 p = frame;
537 /*
538 * Find the next leaf page by incrementing the frame pointer.
539 * If we run out of entries in the interior node, loop around and
540 * increment pointer in the parent node. When we break out of
541 * this loop, num_frames indicates the number of interior
542 * nodes need to be read.
543 */
544 while (1) {
545 if (++(p->at) < p->entries + dx_get_count(p->entries))
546 break;
547 if (p == frames)
548 return 0;
549 num_frames++;
550 p--;
551 }
552
553 /*
554 * If the hash is 1, then continue only if the next page has a
555 * continuation hash of any value. This is used for readdir
556 * handling. Otherwise, check to see if the hash matches the
557 * desired contiuation hash. If it doesn't, return since
558 * there's no point to read in the successive index pages.
559 */
560 bhash = dx_get_hash(p->at);
561 if (start_hash)
562 *start_hash = bhash;
563 if ((hash & 1) == 0) {
564 if ((bhash & ~1) != hash)
565 return 0;
566 }
567 /*
568 * If the hash is HASH_NB_ALWAYS, we always go to the next
569 * block so no check is necessary
570 */
571 while (num_frames--) {
Mingming Cao617ba132006-10-11 01:20:53 -0700572 if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at),
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700573 0, &err)))
574 return err; /* Failure */
575 p++;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400576 brelse(p->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700577 p->bh = bh;
578 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
579 }
580 return 1;
581}
582
583
584/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700585 * This function fills a red-black tree with information from a
586 * directory block. It returns the number directory entries loaded
587 * into the tree. If there is an error it is returned in err.
588 */
589static int htree_dirblock_to_tree(struct file *dir_file,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500590 struct inode *dir, ext4_lblk_t block,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700591 struct dx_hash_info *hinfo,
592 __u32 start_hash, __u32 start_minor_hash)
593{
594 struct buffer_head *bh;
Mingming Cao617ba132006-10-11 01:20:53 -0700595 struct ext4_dir_entry_2 *de, *top;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700596 int err, count = 0;
597
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500598 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
599 (unsigned long)block));
Mingming Cao617ba132006-10-11 01:20:53 -0700600 if (!(bh = ext4_bread (NULL, dir, block, 0, &err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700601 return err;
602
Mingming Cao617ba132006-10-11 01:20:53 -0700603 de = (struct ext4_dir_entry_2 *) bh->b_data;
604 top = (struct ext4_dir_entry_2 *) ((char *) de +
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700605 dir->i_sb->s_blocksize -
Mingming Cao617ba132006-10-11 01:20:53 -0700606 EXT4_DIR_REC_LEN(0));
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500607 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
Eric Sandeene6c40212006-12-06 20:36:28 -0800608 if (!ext4_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
609 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
610 +((char *)de - bh->b_data))) {
611 /* On error, skip the f_pos to the next block. */
612 dir_file->f_pos = (dir_file->f_pos |
613 (dir->i_sb->s_blocksize - 1)) + 1;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400614 brelse(bh);
Eric Sandeene6c40212006-12-06 20:36:28 -0800615 return count;
616 }
Mingming Cao617ba132006-10-11 01:20:53 -0700617 ext4fs_dirhash(de->name, de->name_len, hinfo);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700618 if ((hinfo->hash < start_hash) ||
619 ((hinfo->hash == start_hash) &&
620 (hinfo->minor_hash < start_minor_hash)))
621 continue;
622 if (de->inode == 0)
623 continue;
Mingming Cao617ba132006-10-11 01:20:53 -0700624 if ((err = ext4_htree_store_dirent(dir_file,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700625 hinfo->hash, hinfo->minor_hash, de)) != 0) {
626 brelse(bh);
627 return err;
628 }
629 count++;
630 }
631 brelse(bh);
632 return count;
633}
634
635
636/*
637 * This function fills a red-black tree with information from a
638 * directory. We start scanning the directory in hash order, starting
639 * at start_hash and start_minor_hash.
640 *
641 * This function returns the number of entries inserted into the tree,
642 * or a negative error code.
643 */
Mingming Cao617ba132006-10-11 01:20:53 -0700644int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700645 __u32 start_minor_hash, __u32 *next_hash)
646{
647 struct dx_hash_info hinfo;
Mingming Cao617ba132006-10-11 01:20:53 -0700648 struct ext4_dir_entry_2 *de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700649 struct dx_frame frames[2], *frame;
650 struct inode *dir;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500651 ext4_lblk_t block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700652 int count = 0;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500653 int ret, err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700654 __u32 hashval;
655
Theodore Ts'o60e66792010-05-17 07:00:00 -0400656 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400657 start_hash, start_minor_hash));
Josef "Jeff" Sipek9d549892006-12-08 02:36:38 -0800658 dir = dir_file->f_path.dentry->d_inode;
Dmitry Monakhov12e9b892010-05-16 22:00:00 -0400659 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
Mingming Cao617ba132006-10-11 01:20:53 -0700660 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
Theodore Ts'of99b2582008-10-28 13:21:44 -0400661 if (hinfo.hash_version <= DX_HASH_TEA)
662 hinfo.hash_version +=
663 EXT4_SB(dir->i_sb)->s_hash_unsigned;
Mingming Cao617ba132006-10-11 01:20:53 -0700664 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700665 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
666 start_hash, start_minor_hash);
667 *next_hash = ~0;
668 return count;
669 }
670 hinfo.hash = start_hash;
671 hinfo.minor_hash = 0;
Theodore Ts'of702ba02008-09-22 15:21:01 -0400672 frame = dx_probe(NULL, dir, &hinfo, frames, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700673 if (!frame)
674 return err;
675
676 /* Add '.' and '..' from the htree header */
677 if (!start_hash && !start_minor_hash) {
Mingming Cao617ba132006-10-11 01:20:53 -0700678 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
679 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700680 goto errout;
681 count++;
682 }
683 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
Mingming Cao617ba132006-10-11 01:20:53 -0700684 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500685 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
Mingming Cao617ba132006-10-11 01:20:53 -0700686 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700687 goto errout;
688 count++;
689 }
690
691 while (1) {
692 block = dx_get_block(frame->at);
693 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
694 start_hash, start_minor_hash);
695 if (ret < 0) {
696 err = ret;
697 goto errout;
698 }
699 count += ret;
700 hashval = ~0;
Mingming Cao617ba132006-10-11 01:20:53 -0700701 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700702 frame, frames, &hashval);
703 *next_hash = hashval;
704 if (ret < 0) {
705 err = ret;
706 goto errout;
707 }
708 /*
709 * Stop if: (a) there are no more entries, or
710 * (b) we have inserted at least one entry and the
711 * next hash value is not a continuation
712 */
713 if ((ret == 0) ||
714 (count && ((hashval & 1) == 0)))
715 break;
716 }
717 dx_release(frames);
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400718 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
719 "next hash: %x\n", count, *next_hash));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700720 return count;
721errout:
722 dx_release(frames);
723 return (err);
724}
725
726
727/*
728 * Directory block splitting, compacting
729 */
730
Eric Sandeenef2b02d2007-09-18 22:46:42 -0700731/*
732 * Create map of hash values, offsets, and sizes, stored at end of block.
733 * Returns number of entries mapped.
734 */
Theodore Ts'o8bad4592009-02-14 21:46:54 -0500735static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
736 struct dx_hash_info *hinfo,
737 struct dx_map_entry *map_tail)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700738{
739 int count = 0;
740 char *base = (char *) de;
741 struct dx_hash_info h = *hinfo;
742
Theodore Ts'o8bad4592009-02-14 21:46:54 -0500743 while ((char *) de < base + blocksize) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700744 if (de->name_len && de->inode) {
Mingming Cao617ba132006-10-11 01:20:53 -0700745 ext4fs_dirhash(de->name, de->name_len, &h);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700746 map_tail--;
747 map_tail->hash = h.hash;
Toshiyuki Okajima9aee2282009-06-08 12:41:35 -0400748 map_tail->offs = ((char *) de - base)>>2;
Eric Sandeenef2b02d2007-09-18 22:46:42 -0700749 map_tail->size = le16_to_cpu(de->rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700750 count++;
751 cond_resched();
752 }
753 /* XXX: do we need to check rec_len == 0 case? -Chris */
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500754 de = ext4_next_entry(de, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700755 }
756 return count;
757}
758
Eric Sandeenef2b02d2007-09-18 22:46:42 -0700759/* Sort map by hash value */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700760static void dx_sort_map (struct dx_map_entry *map, unsigned count)
761{
Andrew Morton63f57932006-10-11 01:21:24 -0700762 struct dx_map_entry *p, *q, *top = map + count - 1;
763 int more;
764 /* Combsort until bubble sort doesn't suck */
765 while (count > 2) {
766 count = count*10/13;
767 if (count - 9 < 2) /* 9, 10 -> 11 */
768 count = 11;
769 for (p = top, q = p - count; q >= map; p--, q--)
770 if (p->hash < q->hash)
771 swap(*p, *q);
772 }
773 /* Garden variety bubble sort */
774 do {
775 more = 0;
776 q = top;
777 while (q-- > map) {
778 if (q[1].hash >= q[0].hash)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700779 continue;
Andrew Morton63f57932006-10-11 01:21:24 -0700780 swap(*(q+1), *q);
781 more = 1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700782 }
783 } while(more);
784}
785
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500786static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700787{
788 struct dx_entry *entries = frame->entries;
789 struct dx_entry *old = frame->at, *new = old + 1;
790 int count = dx_get_count(entries);
791
792 assert(count < dx_get_limit(entries));
793 assert(old < entries + count);
794 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
795 dx_set_hash(new, hash);
796 dx_set_block(new, block);
797 dx_set_count(entries, count + 1);
798}
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700799
Mingming Cao617ba132006-10-11 01:20:53 -0700800static void ext4_update_dx_flag(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700801{
Mingming Cao617ba132006-10-11 01:20:53 -0700802 if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
803 EXT4_FEATURE_COMPAT_DIR_INDEX))
Dmitry Monakhov12e9b892010-05-16 22:00:00 -0400804 ext4_clear_inode_flag(inode, EXT4_INODE_INDEX);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700805}
806
807/*
Mingming Cao617ba132006-10-11 01:20:53 -0700808 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700809 *
Mingming Cao617ba132006-10-11 01:20:53 -0700810 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700811 * `de != NULL' is guaranteed by caller.
812 */
Mingming Cao617ba132006-10-11 01:20:53 -0700813static inline int ext4_match (int len, const char * const name,
814 struct ext4_dir_entry_2 * de)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700815{
816 if (len != de->name_len)
817 return 0;
818 if (!de->inode)
819 return 0;
820 return !memcmp(name, de->name, len);
821}
822
823/*
824 * Returns 0 if not found, -1 on failure, and 1 on success
825 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400826static inline int search_dirblock(struct buffer_head *bh,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700827 struct inode *dir,
Theodore Ts'of702ba02008-09-22 15:21:01 -0400828 const struct qstr *d_name,
Theodore Ts'o498e5f22008-11-05 00:14:04 -0500829 unsigned int offset,
Mingming Cao617ba132006-10-11 01:20:53 -0700830 struct ext4_dir_entry_2 ** res_dir)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700831{
Mingming Cao617ba132006-10-11 01:20:53 -0700832 struct ext4_dir_entry_2 * de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700833 char * dlimit;
834 int de_len;
Theodore Ts'of702ba02008-09-22 15:21:01 -0400835 const char *name = d_name->name;
836 int namelen = d_name->len;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700837
Mingming Cao617ba132006-10-11 01:20:53 -0700838 de = (struct ext4_dir_entry_2 *) bh->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700839 dlimit = bh->b_data + dir->i_sb->s_blocksize;
840 while ((char *) de < dlimit) {
841 /* this code is executed quadratically often */
842 /* do minimal checking `by hand' */
843
844 if ((char *) de + namelen <= dlimit &&
Mingming Cao617ba132006-10-11 01:20:53 -0700845 ext4_match (namelen, name, de)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700846 /* found a match - just to be sure, do a full check */
Mingming Cao617ba132006-10-11 01:20:53 -0700847 if (!ext4_check_dir_entry("ext4_find_entry",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700848 dir, de, bh, offset))
849 return -1;
850 *res_dir = de;
851 return 1;
852 }
853 /* prevent looping on a bad block */
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500854 de_len = ext4_rec_len_from_disk(de->rec_len,
855 dir->i_sb->s_blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700856 if (de_len <= 0)
857 return -1;
858 offset += de_len;
Mingming Cao617ba132006-10-11 01:20:53 -0700859 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700860 }
861 return 0;
862}
863
864
865/*
Mingming Cao617ba132006-10-11 01:20:53 -0700866 * ext4_find_entry()
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700867 *
868 * finds an entry in the specified directory with the wanted name. It
869 * returns the cache buffer in which the entry was found, and the entry
870 * itself (as a parameter - res_dir). It does NOT read the inode of the
871 * entry - you'll have to do that yourself if you want to.
872 *
873 * The returned buffer_head has ->b_count elevated. The caller is expected
874 * to brelse() it when appropriate.
875 */
Theodore Ts'of702ba02008-09-22 15:21:01 -0400876static struct buffer_head * ext4_find_entry (struct inode *dir,
877 const struct qstr *d_name,
Mingming Cao617ba132006-10-11 01:20:53 -0700878 struct ext4_dir_entry_2 ** res_dir)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700879{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400880 struct super_block *sb;
881 struct buffer_head *bh_use[NAMEI_RA_SIZE];
882 struct buffer_head *bh, *ret = NULL;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500883 ext4_lblk_t start, block, b;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700884 int ra_max = 0; /* Number of bh's in the readahead
885 buffer, bh_use[] */
886 int ra_ptr = 0; /* Current index into readahead
887 buffer */
888 int num = 0;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500889 ext4_lblk_t nblocks;
890 int i, err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700891 int namelen;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700892
893 *res_dir = NULL;
894 sb = dir->i_sb;
Theodore Ts'of702ba02008-09-22 15:21:01 -0400895 namelen = d_name->len;
Mingming Cao617ba132006-10-11 01:20:53 -0700896 if (namelen > EXT4_NAME_LEN)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700897 return NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700898 if (is_dx(dir)) {
Theodore Ts'of702ba02008-09-22 15:21:01 -0400899 bh = ext4_dx_find_entry(dir, d_name, res_dir, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700900 /*
901 * On success, or if the error was file not found,
902 * return. Otherwise, fall back to doing a search the
903 * old fashioned way.
904 */
905 if (bh || (err != ERR_BAD_DX_DIR))
906 return bh;
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400907 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
908 "falling back\n"));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700909 }
Mingming Cao617ba132006-10-11 01:20:53 -0700910 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
911 start = EXT4_I(dir)->i_dir_start_lookup;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700912 if (start >= nblocks)
913 start = 0;
914 block = start;
915restart:
916 do {
917 /*
918 * We deal with the read-ahead logic here.
919 */
920 if (ra_ptr >= ra_max) {
921 /* Refill the readahead buffer */
922 ra_ptr = 0;
923 b = block;
924 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
925 /*
926 * Terminate if we reach the end of the
927 * directory and must wrap, or if our
928 * search has finished at this block.
929 */
930 if (b >= nblocks || (num && block == start)) {
931 bh_use[ra_max] = NULL;
932 break;
933 }
934 num++;
Mingming Cao617ba132006-10-11 01:20:53 -0700935 bh = ext4_getblk(NULL, dir, b++, 0, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700936 bh_use[ra_max] = bh;
937 if (bh)
938 ll_rw_block(READ_META, 1, &bh);
939 }
940 }
941 if ((bh = bh_use[ra_ptr++]) == NULL)
942 goto next;
943 wait_on_buffer(bh);
944 if (!buffer_uptodate(bh)) {
945 /* read error, skip block & hope for the best */
Theodore Ts'o24676da2010-05-16 21:00:00 -0400946 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
947 (unsigned long) block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700948 brelse(bh);
949 goto next;
950 }
Theodore Ts'of702ba02008-09-22 15:21:01 -0400951 i = search_dirblock(bh, dir, d_name,
Mingming Cao617ba132006-10-11 01:20:53 -0700952 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700953 if (i == 1) {
Mingming Cao617ba132006-10-11 01:20:53 -0700954 EXT4_I(dir)->i_dir_start_lookup = block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700955 ret = bh;
956 goto cleanup_and_exit;
957 } else {
958 brelse(bh);
959 if (i < 0)
960 goto cleanup_and_exit;
961 }
962 next:
963 if (++block >= nblocks)
964 block = 0;
965 } while (block != start);
966
967 /*
968 * If the directory has grown while we were searching, then
969 * search the last part of the directory before giving up.
970 */
971 block = nblocks;
Mingming Cao617ba132006-10-11 01:20:53 -0700972 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700973 if (block < nblocks) {
974 start = 0;
975 goto restart;
976 }
977
978cleanup_and_exit:
979 /* Clean up the read-ahead blocks */
980 for (; ra_ptr < ra_max; ra_ptr++)
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400981 brelse(bh_use[ra_ptr]);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700982 return ret;
983}
984
Theodore Ts'of702ba02008-09-22 15:21:01 -0400985static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
Mingming Cao617ba132006-10-11 01:20:53 -0700986 struct ext4_dir_entry_2 **res_dir, int *err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700987{
988 struct super_block * sb;
989 struct dx_hash_info hinfo;
990 u32 hash;
991 struct dx_frame frames[2], *frame;
Mingming Cao617ba132006-10-11 01:20:53 -0700992 struct ext4_dir_entry_2 *de, *top;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700993 struct buffer_head *bh;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500994 ext4_lblk_t block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700995 int retval;
Theodore Ts'of702ba02008-09-22 15:21:01 -0400996 int namelen = d_name->len;
997 const u8 *name = d_name->name;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700998
999 sb = dir->i_sb;
1000 /* NFS may look up ".." - look at dx_root directory block */
1001 if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
Theodore Ts'of702ba02008-09-22 15:21:01 -04001002 if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001003 return NULL;
1004 } else {
1005 frame = frames;
1006 frame->bh = NULL; /* for dx_release() */
1007 frame->at = (struct dx_entry *)frames; /* hack for zero entry*/
1008 dx_set_block(frame->at, 0); /* dx_root block is 0 */
1009 }
1010 hash = hinfo.hash;
1011 do {
1012 block = dx_get_block(frame->at);
Mingming Cao617ba132006-10-11 01:20:53 -07001013 if (!(bh = ext4_bread (NULL,dir, block, 0, err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001014 goto errout;
Mingming Cao617ba132006-10-11 01:20:53 -07001015 de = (struct ext4_dir_entry_2 *) bh->b_data;
1016 top = (struct ext4_dir_entry_2 *) ((char *) de + sb->s_blocksize -
1017 EXT4_DIR_REC_LEN(0));
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001018 for (; de < top; de = ext4_next_entry(de, sb->s_blocksize)) {
Duane Griffinf3b35f02008-07-11 19:27:31 -04001019 int off = (block << EXT4_BLOCK_SIZE_BITS(sb))
1020 + ((char *) de - bh->b_data);
1021
1022 if (!ext4_check_dir_entry(__func__, dir, de, bh, off)) {
1023 brelse(bh);
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001024 *err = ERR_BAD_DX_DIR;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001025 goto errout;
1026 }
Duane Griffinf3b35f02008-07-11 19:27:31 -04001027
1028 if (ext4_match(namelen, name, de)) {
1029 *res_dir = de;
1030 dx_release(frames);
1031 return bh;
1032 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001033 }
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001034 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001035 /* Check to see if we should continue to search */
Mingming Cao617ba132006-10-11 01:20:53 -07001036 retval = ext4_htree_next_block(dir, hash, frame,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001037 frames, NULL);
1038 if (retval < 0) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001039 ext4_warning(sb,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001040 "error reading index page in directory #%lu",
1041 dir->i_ino);
1042 *err = retval;
1043 goto errout;
1044 }
1045 } while (retval == 1);
1046
1047 *err = -ENOENT;
1048errout:
Theodore Ts'o4776004f2008-09-08 23:00:52 -04001049 dxtrace(printk(KERN_DEBUG "%s not found\n", name));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001050 dx_release (frames);
1051 return NULL;
1052}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001053
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001054static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001055{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001056 struct inode *inode;
1057 struct ext4_dir_entry_2 *de;
1058 struct buffer_head *bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001059
Mingming Cao617ba132006-10-11 01:20:53 -07001060 if (dentry->d_name.len > EXT4_NAME_LEN)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001061 return ERR_PTR(-ENAMETOOLONG);
1062
Theodore Ts'of702ba02008-09-22 15:21:01 -04001063 bh = ext4_find_entry(dir, &dentry->d_name, &de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001064 inode = NULL;
1065 if (bh) {
Theodore Ts'o498e5f22008-11-05 00:14:04 -05001066 __u32 ino = le32_to_cpu(de->inode);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001067 brelse(bh);
Mingming Cao617ba132006-10-11 01:20:53 -07001068 if (!ext4_valid_inum(dir->i_sb, ino)) {
Theodore Ts'o24676da2010-05-16 21:00:00 -04001069 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
David Howells1d1fe1e2008-02-07 00:15:37 -08001070 return ERR_PTR(-EIO);
Vasily Averina6c15c22007-07-15 23:40:46 -07001071 }
David Howells1d1fe1e2008-02-07 00:15:37 -08001072 inode = ext4_iget(dir->i_sb, ino);
Bryan Donlane6f009b2009-02-22 21:20:25 -05001073 if (unlikely(IS_ERR(inode))) {
1074 if (PTR_ERR(inode) == -ESTALE) {
Theodore Ts'o24676da2010-05-16 21:00:00 -04001075 EXT4_ERROR_INODE(dir,
1076 "deleted inode referenced: %u",
1077 ino);
Bryan Donlane6f009b2009-02-22 21:20:25 -05001078 return ERR_PTR(-EIO);
1079 } else {
1080 return ERR_CAST(inode);
1081 }
1082 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001083 }
1084 return d_splice_alias(inode, dentry);
1085}
1086
1087
Mingming Cao617ba132006-10-11 01:20:53 -07001088struct dentry *ext4_get_parent(struct dentry *child)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001089{
Theodore Ts'o498e5f22008-11-05 00:14:04 -05001090 __u32 ino;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001091 struct inode *inode;
Theodore Ts'of702ba02008-09-22 15:21:01 -04001092 static const struct qstr dotdot = {
1093 .name = "..",
1094 .len = 2,
1095 };
Mingming Cao617ba132006-10-11 01:20:53 -07001096 struct ext4_dir_entry_2 * de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001097 struct buffer_head *bh;
1098
Theodore Ts'of702ba02008-09-22 15:21:01 -04001099 bh = ext4_find_entry(child->d_inode, &dotdot, &de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001100 inode = NULL;
1101 if (!bh)
1102 return ERR_PTR(-ENOENT);
1103 ino = le32_to_cpu(de->inode);
1104 brelse(bh);
1105
Mingming Cao617ba132006-10-11 01:20:53 -07001106 if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
Theodore Ts'o24676da2010-05-16 21:00:00 -04001107 EXT4_ERROR_INODE(child->d_inode,
1108 "bad parent inode number: %u", ino);
David Howells1d1fe1e2008-02-07 00:15:37 -08001109 return ERR_PTR(-EIO);
Vasily Averina6c15c22007-07-15 23:40:46 -07001110 }
1111
Christoph Hellwig44003722008-08-11 15:49:04 +02001112 return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001113}
1114
1115#define S_SHIFT 12
Mingming Cao617ba132006-10-11 01:20:53 -07001116static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1117 [S_IFREG >> S_SHIFT] = EXT4_FT_REG_FILE,
1118 [S_IFDIR >> S_SHIFT] = EXT4_FT_DIR,
1119 [S_IFCHR >> S_SHIFT] = EXT4_FT_CHRDEV,
1120 [S_IFBLK >> S_SHIFT] = EXT4_FT_BLKDEV,
1121 [S_IFIFO >> S_SHIFT] = EXT4_FT_FIFO,
1122 [S_IFSOCK >> S_SHIFT] = EXT4_FT_SOCK,
1123 [S_IFLNK >> S_SHIFT] = EXT4_FT_SYMLINK,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001124};
1125
Mingming Cao617ba132006-10-11 01:20:53 -07001126static inline void ext4_set_de_type(struct super_block *sb,
1127 struct ext4_dir_entry_2 *de,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001128 umode_t mode) {
Mingming Cao617ba132006-10-11 01:20:53 -07001129 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1130 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001131}
1132
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001133/*
1134 * Move count entries from end of map between two memory locations.
1135 * Returns pointer to last entry moved.
1136 */
Mingming Cao617ba132006-10-11 01:20:53 -07001137static struct ext4_dir_entry_2 *
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001138dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1139 unsigned blocksize)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001140{
1141 unsigned rec_len = 0;
1142
1143 while (count--) {
Theodore Ts'o60e66792010-05-17 07:00:00 -04001144 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
Toshiyuki Okajima9aee2282009-06-08 12:41:35 -04001145 (from + (map->offs<<2));
Mingming Cao617ba132006-10-11 01:20:53 -07001146 rec_len = EXT4_DIR_REC_LEN(de->name_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001147 memcpy (to, de, rec_len);
Mingming Cao617ba132006-10-11 01:20:53 -07001148 ((struct ext4_dir_entry_2 *) to)->rec_len =
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001149 ext4_rec_len_to_disk(rec_len, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001150 de->inode = 0;
1151 map++;
1152 to += rec_len;
1153 }
Mingming Cao617ba132006-10-11 01:20:53 -07001154 return (struct ext4_dir_entry_2 *) (to - rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001155}
1156
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001157/*
1158 * Compact each dir entry in the range to the minimal rec_len.
1159 * Returns pointer to last entry in range.
1160 */
Theodore Ts'o8bad4592009-02-14 21:46:54 -05001161static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001162{
Mingming Cao617ba132006-10-11 01:20:53 -07001163 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001164 unsigned rec_len = 0;
1165
1166 prev = to = de;
Theodore Ts'o8bad4592009-02-14 21:46:54 -05001167 while ((char*)de < base + blocksize) {
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001168 next = ext4_next_entry(de, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001169 if (de->inode && de->name_len) {
Mingming Cao617ba132006-10-11 01:20:53 -07001170 rec_len = EXT4_DIR_REC_LEN(de->name_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001171 if (de > to)
1172 memmove(to, de, rec_len);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001173 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001174 prev = to;
Mingming Cao617ba132006-10-11 01:20:53 -07001175 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001176 }
1177 de = next;
1178 }
1179 return prev;
1180}
1181
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001182/*
1183 * Split a full leaf block to make room for a new dir entry.
1184 * Allocate a new block, and move entries so that they are approx. equally full.
1185 * Returns pointer to de in block into which the new entry will be inserted.
1186 */
Mingming Cao617ba132006-10-11 01:20:53 -07001187static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001188 struct buffer_head **bh,struct dx_frame *frame,
1189 struct dx_hash_info *hinfo, int *error)
1190{
1191 unsigned blocksize = dir->i_sb->s_blocksize;
1192 unsigned count, continued;
1193 struct buffer_head *bh2;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001194 ext4_lblk_t newblock;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001195 u32 hash2;
1196 struct dx_map_entry *map;
1197 char *data1 = (*bh)->b_data, *data2;
Theodore Ts'o59e315b2008-12-06 16:58:39 -05001198 unsigned split, move, size;
Mingming Cao617ba132006-10-11 01:20:53 -07001199 struct ext4_dir_entry_2 *de = NULL, *de2;
Theodore Ts'o59e315b2008-12-06 16:58:39 -05001200 int err = 0, i;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001201
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001202 bh2 = ext4_append (handle, dir, &newblock, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001203 if (!(bh2)) {
1204 brelse(*bh);
1205 *bh = NULL;
1206 goto errout;
1207 }
1208
1209 BUFFER_TRACE(*bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001210 err = ext4_journal_get_write_access(handle, *bh);
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001211 if (err)
1212 goto journal_error;
1213
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001214 BUFFER_TRACE(frame->bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001215 err = ext4_journal_get_write_access(handle, frame->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001216 if (err)
1217 goto journal_error;
1218
1219 data2 = bh2->b_data;
1220
1221 /* create map in the end of data2 block */
1222 map = (struct dx_map_entry *) (data2 + blocksize);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001223 count = dx_make_map((struct ext4_dir_entry_2 *) data1,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001224 blocksize, hinfo, map);
1225 map -= count;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001226 dx_sort_map(map, count);
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001227 /* Split the existing block in the middle, size-wise */
1228 size = 0;
1229 move = 0;
1230 for (i = count-1; i >= 0; i--) {
1231 /* is more than half of this entry in 2nd half of the block? */
1232 if (size + map[i].size/2 > blocksize/2)
1233 break;
1234 size += map[i].size;
1235 move++;
1236 }
1237 /* map index at which we will split */
1238 split = count - move;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001239 hash2 = map[split].hash;
1240 continued = hash2 == map[split - 1].hash;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001241 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1242 (unsigned long)dx_get_block(frame->at),
1243 hash2, split, count-split));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001244
1245 /* Fancy dance to stay within two buffers */
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001246 de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001247 de = dx_pack_dirents(data1, blocksize);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001248 de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1249 blocksize);
1250 de2->rec_len = ext4_rec_len_to_disk(data2 + blocksize - (char *) de2,
1251 blocksize);
Mingming Cao617ba132006-10-11 01:20:53 -07001252 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1253 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001254
1255 /* Which block gets the new entry? */
1256 if (hinfo->hash >= hash2)
1257 {
1258 swap(*bh, bh2);
1259 de = de2;
1260 }
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001261 dx_insert_block(frame, hash2 + continued, newblock);
Frank Mayhar03901312009-01-07 00:06:22 -05001262 err = ext4_handle_dirty_metadata(handle, dir, bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001263 if (err)
1264 goto journal_error;
Frank Mayhar03901312009-01-07 00:06:22 -05001265 err = ext4_handle_dirty_metadata(handle, dir, frame->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001266 if (err)
1267 goto journal_error;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001268 brelse(bh2);
1269 dxtrace(dx_show_index("frame", frame->entries));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001270 return de;
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001271
1272journal_error:
1273 brelse(*bh);
1274 brelse(bh2);
1275 *bh = NULL;
1276 ext4_std_error(dir->i_sb, err);
1277errout:
1278 *error = err;
1279 return NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001280}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001281
1282/*
1283 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1284 * it points to a directory entry which is guaranteed to be large
1285 * enough for new directory entry. If de is NULL, then
1286 * add_dirent_to_buf will attempt search the directory block for
1287 * space. It will return -ENOSPC if no space is available, and -EIO
1288 * and -EEXIST if directory entry already exists.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001289 */
1290static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
Mingming Cao617ba132006-10-11 01:20:53 -07001291 struct inode *inode, struct ext4_dir_entry_2 *de,
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001292 struct buffer_head *bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001293{
1294 struct inode *dir = dentry->d_parent->d_inode;
1295 const char *name = dentry->d_name.name;
1296 int namelen = dentry->d_name.len;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05001297 unsigned int offset = 0;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001298 unsigned int blocksize = dir->i_sb->s_blocksize;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001299 unsigned short reclen;
1300 int nlen, rlen, err;
1301 char *top;
1302
Mingming Cao617ba132006-10-11 01:20:53 -07001303 reclen = EXT4_DIR_REC_LEN(namelen);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001304 if (!de) {
Mingming Cao617ba132006-10-11 01:20:53 -07001305 de = (struct ext4_dir_entry_2 *)bh->b_data;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001306 top = bh->b_data + blocksize - reclen;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001307 while ((char *) de <= top) {
Mingming Cao617ba132006-10-11 01:20:53 -07001308 if (!ext4_check_dir_entry("ext4_add_entry", dir, de,
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001309 bh, offset))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001310 return -EIO;
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001311 if (ext4_match(namelen, name, de))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001312 return -EEXIST;
Mingming Cao617ba132006-10-11 01:20:53 -07001313 nlen = EXT4_DIR_REC_LEN(de->name_len);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001314 rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001315 if ((de->inode? rlen - nlen: rlen) >= reclen)
1316 break;
Mingming Cao617ba132006-10-11 01:20:53 -07001317 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001318 offset += rlen;
1319 }
1320 if ((char *) de > top)
1321 return -ENOSPC;
1322 }
1323 BUFFER_TRACE(bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001324 err = ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001325 if (err) {
Mingming Cao617ba132006-10-11 01:20:53 -07001326 ext4_std_error(dir->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001327 return err;
1328 }
1329
1330 /* By now the buffer is marked for journaling */
Mingming Cao617ba132006-10-11 01:20:53 -07001331 nlen = EXT4_DIR_REC_LEN(de->name_len);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001332 rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001333 if (de->inode) {
Mingming Cao617ba132006-10-11 01:20:53 -07001334 struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001335 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, blocksize);
1336 de->rec_len = ext4_rec_len_to_disk(nlen, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001337 de = de1;
1338 }
Mingming Cao617ba132006-10-11 01:20:53 -07001339 de->file_type = EXT4_FT_UNKNOWN;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001340 if (inode) {
1341 de->inode = cpu_to_le32(inode->i_ino);
Mingming Cao617ba132006-10-11 01:20:53 -07001342 ext4_set_de_type(dir->i_sb, de, inode->i_mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001343 } else
1344 de->inode = 0;
1345 de->name_len = namelen;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001346 memcpy(de->name, name, namelen);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001347 /*
1348 * XXX shouldn't update any times until successful
1349 * completion of syscall, but too many callers depend
1350 * on this.
1351 *
1352 * XXX similarly, too many callers depend on
Mingming Cao617ba132006-10-11 01:20:53 -07001353 * ext4_new_inode() setting the times, but error
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001354 * recovery deletes the inode, so the worst that can
1355 * happen is that the times are slightly out of date
1356 * and/or different from the directory change time.
1357 */
Kalpak Shahef7f3832007-07-18 09:15:20 -04001358 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
Mingming Cao617ba132006-10-11 01:20:53 -07001359 ext4_update_dx_flag(dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001360 dir->i_version++;
Mingming Cao617ba132006-10-11 01:20:53 -07001361 ext4_mark_inode_dirty(handle, dir);
Frank Mayhar03901312009-01-07 00:06:22 -05001362 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1363 err = ext4_handle_dirty_metadata(handle, dir, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001364 if (err)
Mingming Cao617ba132006-10-11 01:20:53 -07001365 ext4_std_error(dir->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001366 return 0;
1367}
1368
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001369/*
1370 * This converts a one block unindexed directory to a 3 block indexed
1371 * directory, and adds the dentry to the indexed directory.
1372 */
1373static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1374 struct inode *inode, struct buffer_head *bh)
1375{
1376 struct inode *dir = dentry->d_parent->d_inode;
1377 const char *name = dentry->d_name.name;
1378 int namelen = dentry->d_name.len;
1379 struct buffer_head *bh2;
1380 struct dx_root *root;
1381 struct dx_frame frames[2], *frame;
1382 struct dx_entry *entries;
Mingming Cao617ba132006-10-11 01:20:53 -07001383 struct ext4_dir_entry_2 *de, *de2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001384 char *data1, *top;
1385 unsigned len;
1386 int retval;
1387 unsigned blocksize;
1388 struct dx_hash_info hinfo;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001389 ext4_lblk_t block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001390 struct fake_dirent *fde;
1391
1392 blocksize = dir->i_sb->s_blocksize;
Theodore Ts'oe6b8bc02009-01-16 11:13:40 -05001393 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
Mingming Cao617ba132006-10-11 01:20:53 -07001394 retval = ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001395 if (retval) {
Mingming Cao617ba132006-10-11 01:20:53 -07001396 ext4_std_error(dir->i_sb, retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001397 brelse(bh);
1398 return retval;
1399 }
1400 root = (struct dx_root *) bh->b_data;
1401
Theodore Ts'oe6b8bc02009-01-16 11:13:40 -05001402 /* The 0th block becomes the root, move the dirents out */
1403 fde = &root->dotdot;
1404 de = (struct ext4_dir_entry_2 *)((char *)fde +
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001405 ext4_rec_len_from_disk(fde->rec_len, blocksize));
Theodore Ts'oe6b8bc02009-01-16 11:13:40 -05001406 if ((char *) de >= (((char *) root) + blocksize)) {
Theodore Ts'o24676da2010-05-16 21:00:00 -04001407 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
Theodore Ts'oe6b8bc02009-01-16 11:13:40 -05001408 brelse(bh);
1409 return -EIO;
1410 }
1411 len = ((char *) root) + blocksize - (char *) de;
1412
1413 /* Allocate new block for the 0th block's dirents */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001414 bh2 = ext4_append(handle, dir, &block, &retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001415 if (!(bh2)) {
1416 brelse(bh);
1417 return retval;
1418 }
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04001419 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001420 data1 = bh2->b_data;
1421
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001422 memcpy (data1, de, len);
Mingming Cao617ba132006-10-11 01:20:53 -07001423 de = (struct ext4_dir_entry_2 *) data1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001424 top = data1 + len;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001425 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001426 de = de2;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001427 de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1428 blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001429 /* Initialize the root; the dot dirents already exist */
Mingming Cao617ba132006-10-11 01:20:53 -07001430 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001431 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1432 blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001433 memset (&root->info, 0, sizeof(root->info));
1434 root->info.info_length = sizeof(root->info);
Mingming Cao617ba132006-10-11 01:20:53 -07001435 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001436 entries = root->entries;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001437 dx_set_block(entries, 1);
1438 dx_set_count(entries, 1);
1439 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001440
1441 /* Initialize as for dx_probe */
1442 hinfo.hash_version = root->info.hash_version;
Theodore Ts'of99b2582008-10-28 13:21:44 -04001443 if (hinfo.hash_version <= DX_HASH_TEA)
1444 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
Mingming Cao617ba132006-10-11 01:20:53 -07001445 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1446 ext4fs_dirhash(name, namelen, &hinfo);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001447 frame = frames;
1448 frame->entries = entries;
1449 frame->at = entries;
1450 frame->bh = bh;
1451 bh = bh2;
1452 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1453 dx_release (frames);
1454 if (!(de))
1455 return retval;
1456
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001457 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1458 brelse(bh);
1459 return retval;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001460}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001461
1462/*
Mingming Cao617ba132006-10-11 01:20:53 -07001463 * ext4_add_entry()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001464 *
1465 * adds a file entry to the specified directory, using the same
Mingming Cao617ba132006-10-11 01:20:53 -07001466 * semantics as ext4_find_entry(). It returns NULL if it failed.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001467 *
1468 * NOTE!! The inode part of 'de' is left at 0 - which means you
1469 * may not sleep between calling this and putting something into
1470 * the entry, as someone else might have used it while you slept.
1471 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001472static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1473 struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001474{
1475 struct inode *dir = dentry->d_parent->d_inode;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001476 struct buffer_head *bh;
Mingming Cao617ba132006-10-11 01:20:53 -07001477 struct ext4_dir_entry_2 *de;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001478 struct super_block *sb;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001479 int retval;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001480 int dx_fallback=0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001481 unsigned blocksize;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001482 ext4_lblk_t block, blocks;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001483
1484 sb = dir->i_sb;
1485 blocksize = sb->s_blocksize;
1486 if (!dentry->d_name.len)
1487 return -EINVAL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001488 if (is_dx(dir)) {
Mingming Cao617ba132006-10-11 01:20:53 -07001489 retval = ext4_dx_add_entry(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001490 if (!retval || (retval != ERR_BAD_DX_DIR))
1491 return retval;
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04001492 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001493 dx_fallback++;
Mingming Cao617ba132006-10-11 01:20:53 -07001494 ext4_mark_inode_dirty(handle, dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001495 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001496 blocks = dir->i_size >> sb->s_blocksize_bits;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05001497 for (block = 0; block < blocks; block++) {
Mingming Cao617ba132006-10-11 01:20:53 -07001498 bh = ext4_bread(handle, dir, block, 0, &retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001499 if(!bh)
1500 return retval;
1501 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001502 if (retval != -ENOSPC) {
1503 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001504 return retval;
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001505 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001506
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001507 if (blocks == 1 && !dx_fallback &&
Theodore Ts'o1e424a32009-11-08 15:45:44 -05001508 EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1509 return make_indexed_dir(handle, dentry, inode, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001510 brelse(bh);
1511 }
Mingming Cao617ba132006-10-11 01:20:53 -07001512 bh = ext4_append(handle, dir, &block, &retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001513 if (!bh)
1514 return retval;
Mingming Cao617ba132006-10-11 01:20:53 -07001515 de = (struct ext4_dir_entry_2 *) bh->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001516 de->inode = 0;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001517 de->rec_len = ext4_rec_len_to_disk(blocksize, blocksize);
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001518 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1519 brelse(bh);
Frank Mayhar14ece102010-05-17 08:00:00 -04001520 if (retval == 0)
1521 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001522 return retval;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001523}
1524
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001525/*
1526 * Returns 0 for success, or a negative error value
1527 */
Mingming Cao617ba132006-10-11 01:20:53 -07001528static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001529 struct inode *inode)
1530{
1531 struct dx_frame frames[2], *frame;
1532 struct dx_entry *entries, *at;
1533 struct dx_hash_info hinfo;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001534 struct buffer_head *bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001535 struct inode *dir = dentry->d_parent->d_inode;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001536 struct super_block *sb = dir->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -07001537 struct ext4_dir_entry_2 *de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001538 int err;
1539
Theodore Ts'of702ba02008-09-22 15:21:01 -04001540 frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001541 if (!frame)
1542 return err;
1543 entries = frame->entries;
1544 at = frame->at;
1545
Mingming Cao617ba132006-10-11 01:20:53 -07001546 if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001547 goto cleanup;
1548
1549 BUFFER_TRACE(bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001550 err = ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001551 if (err)
1552 goto journal_error;
1553
1554 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001555 if (err != -ENOSPC)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001556 goto cleanup;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001557
1558 /* Block full, should compress but for now just split */
Theodore Ts'o4776004f2008-09-08 23:00:52 -04001559 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001560 dx_get_count(entries), dx_get_limit(entries)));
1561 /* Need to split index? */
1562 if (dx_get_count(entries) == dx_get_limit(entries)) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001563 ext4_lblk_t newblock;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001564 unsigned icount = dx_get_count(entries);
1565 int levels = frame - frames;
1566 struct dx_entry *entries2;
1567 struct dx_node *node2;
1568 struct buffer_head *bh2;
1569
1570 if (levels && (dx_get_count(frames->entries) ==
1571 dx_get_limit(frames->entries))) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001572 ext4_warning(sb, "Directory index full!");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001573 err = -ENOSPC;
1574 goto cleanup;
1575 }
Mingming Cao617ba132006-10-11 01:20:53 -07001576 bh2 = ext4_append (handle, dir, &newblock, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001577 if (!(bh2))
1578 goto cleanup;
1579 node2 = (struct dx_node *)(bh2->b_data);
1580 entries2 = node2->entries;
Andreas Schlick1f7bebb2009-09-10 23:16:07 -04001581 memset(&node2->fake, 0, sizeof(struct fake_dirent));
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001582 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
1583 sb->s_blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001584 BUFFER_TRACE(frame->bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001585 err = ext4_journal_get_write_access(handle, frame->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001586 if (err)
1587 goto journal_error;
1588 if (levels) {
1589 unsigned icount1 = icount/2, icount2 = icount - icount1;
1590 unsigned hash2 = dx_get_hash(entries + icount1);
Theodore Ts'o4776004f2008-09-08 23:00:52 -04001591 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
1592 icount1, icount2));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001593
1594 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
Mingming Cao617ba132006-10-11 01:20:53 -07001595 err = ext4_journal_get_write_access(handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001596 frames[0].bh);
1597 if (err)
1598 goto journal_error;
1599
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001600 memcpy((char *) entries2, (char *) (entries + icount1),
1601 icount2 * sizeof(struct dx_entry));
1602 dx_set_count(entries, icount1);
1603 dx_set_count(entries2, icount2);
1604 dx_set_limit(entries2, dx_node_limit(dir));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001605
1606 /* Which index block gets the new entry? */
1607 if (at - entries >= icount1) {
1608 frame->at = at = at - entries - icount1 + entries2;
1609 frame->entries = entries = entries2;
1610 swap(frame->bh, bh2);
1611 }
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001612 dx_insert_block(frames + 0, hash2, newblock);
1613 dxtrace(dx_show_index("node", frames[1].entries));
1614 dxtrace(dx_show_index("node",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001615 ((struct dx_node *) bh2->b_data)->entries));
Frank Mayhar03901312009-01-07 00:06:22 -05001616 err = ext4_handle_dirty_metadata(handle, inode, bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001617 if (err)
1618 goto journal_error;
1619 brelse (bh2);
1620 } else {
Theodore Ts'o4776004f2008-09-08 23:00:52 -04001621 dxtrace(printk(KERN_DEBUG
1622 "Creating second level index...\n"));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001623 memcpy((char *) entries2, (char *) entries,
1624 icount * sizeof(struct dx_entry));
1625 dx_set_limit(entries2, dx_node_limit(dir));
1626
1627 /* Set up root */
1628 dx_set_count(entries, 1);
1629 dx_set_block(entries + 0, newblock);
1630 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1631
1632 /* Add new access path frame */
1633 frame = frames + 1;
1634 frame->at = at = at - entries + entries2;
1635 frame->entries = entries = entries2;
1636 frame->bh = bh2;
Mingming Cao617ba132006-10-11 01:20:53 -07001637 err = ext4_journal_get_write_access(handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001638 frame->bh);
1639 if (err)
1640 goto journal_error;
1641 }
Frank Mayhar03901312009-01-07 00:06:22 -05001642 ext4_handle_dirty_metadata(handle, inode, frames[0].bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001643 }
1644 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1645 if (!de)
1646 goto cleanup;
1647 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001648 goto cleanup;
1649
1650journal_error:
Mingming Cao617ba132006-10-11 01:20:53 -07001651 ext4_std_error(dir->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001652cleanup:
1653 if (bh)
1654 brelse(bh);
1655 dx_release(frames);
1656 return err;
1657}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001658
1659/*
Mingming Cao617ba132006-10-11 01:20:53 -07001660 * ext4_delete_entry deletes a directory entry by merging it with the
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001661 * previous entry
1662 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001663static int ext4_delete_entry(handle_t *handle,
1664 struct inode *dir,
1665 struct ext4_dir_entry_2 *de_del,
1666 struct buffer_head *bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001667{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001668 struct ext4_dir_entry_2 *de, *pde;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001669 unsigned int blocksize = dir->i_sb->s_blocksize;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001670 int i;
1671
1672 i = 0;
1673 pde = NULL;
Mingming Cao617ba132006-10-11 01:20:53 -07001674 de = (struct ext4_dir_entry_2 *) bh->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001675 while (i < bh->b_size) {
Mingming Cao617ba132006-10-11 01:20:53 -07001676 if (!ext4_check_dir_entry("ext4_delete_entry", dir, de, bh, i))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001677 return -EIO;
1678 if (de == de_del) {
1679 BUFFER_TRACE(bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001680 ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001681 if (pde)
Jan Karaa72d7f82008-01-28 23:58:27 -05001682 pde->rec_len = ext4_rec_len_to_disk(
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001683 ext4_rec_len_from_disk(pde->rec_len,
1684 blocksize) +
1685 ext4_rec_len_from_disk(de->rec_len,
1686 blocksize),
1687 blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001688 else
1689 de->inode = 0;
1690 dir->i_version++;
Frank Mayhar03901312009-01-07 00:06:22 -05001691 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1692 ext4_handle_dirty_metadata(handle, dir, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001693 return 0;
1694 }
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001695 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001696 pde = de;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001697 de = ext4_next_entry(de, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001698 }
1699 return -ENOENT;
1700}
1701
Andreas Dilgerf8628a12007-07-18 08:38:01 -04001702/*
1703 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
1704 * since this indicates that nlinks count was previously 1.
1705 */
1706static void ext4_inc_count(handle_t *handle, struct inode *inode)
1707{
1708 inc_nlink(inode);
1709 if (is_dx(inode) && inode->i_nlink > 1) {
1710 /* limit is 16-bit i_links_count */
1711 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
1712 inode->i_nlink = 1;
1713 EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
1714 EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
1715 }
1716 }
1717}
1718
1719/*
1720 * If a directory had nlink == 1, then we should let it be 1. This indicates
1721 * directory has >EXT4_LINK_MAX subdirs.
1722 */
1723static void ext4_dec_count(handle_t *handle, struct inode *inode)
1724{
1725 drop_nlink(inode);
1726 if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
1727 inc_nlink(inode);
1728}
1729
1730
Mingming Cao617ba132006-10-11 01:20:53 -07001731static int ext4_add_nondir(handle_t *handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001732 struct dentry *dentry, struct inode *inode)
1733{
Mingming Cao617ba132006-10-11 01:20:53 -07001734 int err = ext4_add_entry(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001735 if (!err) {
Mingming Cao617ba132006-10-11 01:20:53 -07001736 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001737 d_instantiate(dentry, inode);
Al Viro6b38e842008-12-30 02:03:31 -05001738 unlock_new_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001739 return 0;
1740 }
Eric Sandeen731b9a52007-02-10 01:46:16 -08001741 drop_nlink(inode);
Al Viro6b38e842008-12-30 02:03:31 -05001742 unlock_new_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001743 iput(inode);
1744 return err;
1745}
1746
1747/*
1748 * By the time this is called, we already have created
1749 * the directory cache entry for the new file, but it
1750 * is so far negative - it has no inode.
1751 *
1752 * If the create succeeds, we fill in the inode information
1753 * with d_instantiate().
1754 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001755static int ext4_create(struct inode *dir, struct dentry *dentry, int mode,
1756 struct nameidata *nd)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001757{
1758 handle_t *handle;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001759 struct inode *inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001760 int err, retries = 0;
1761
Christoph Hellwig871a2932010-03-03 09:05:07 -05001762 dquot_initialize(dir);
Christoph Hellwig907f4552010-03-03 09:05:06 -05001763
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001764retry:
Mingming Cao617ba132006-10-11 01:20:53 -07001765 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1766 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
Dmitry Monakhov5aca07e2009-12-08 22:42:15 -05001767 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001768 if (IS_ERR(handle))
1769 return PTR_ERR(handle);
1770
1771 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05001772 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001773
Andreas Dilger11013912009-06-13 11:45:35 -04001774 inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001775 err = PTR_ERR(inode);
1776 if (!IS_ERR(inode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07001777 inode->i_op = &ext4_file_inode_operations;
1778 inode->i_fop = &ext4_file_operations;
1779 ext4_set_aops(inode);
1780 err = ext4_add_nondir(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001781 }
Mingming Cao617ba132006-10-11 01:20:53 -07001782 ext4_journal_stop(handle);
1783 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001784 goto retry;
1785 return err;
1786}
1787
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001788static int ext4_mknod(struct inode *dir, struct dentry *dentry,
1789 int mode, dev_t rdev)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001790{
1791 handle_t *handle;
1792 struct inode *inode;
1793 int err, retries = 0;
1794
1795 if (!new_valid_dev(rdev))
1796 return -EINVAL;
1797
Christoph Hellwig871a2932010-03-03 09:05:07 -05001798 dquot_initialize(dir);
Christoph Hellwig907f4552010-03-03 09:05:06 -05001799
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001800retry:
Mingming Cao617ba132006-10-11 01:20:53 -07001801 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1802 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
Dmitry Monakhov5aca07e2009-12-08 22:42:15 -05001803 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001804 if (IS_ERR(handle))
1805 return PTR_ERR(handle);
1806
1807 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05001808 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001809
Andreas Dilger11013912009-06-13 11:45:35 -04001810 inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001811 err = PTR_ERR(inode);
1812 if (!IS_ERR(inode)) {
1813 init_special_inode(inode, inode->i_mode, rdev);
Theodore Ts'o03010a32008-10-10 20:02:48 -04001814#ifdef CONFIG_EXT4_FS_XATTR
Mingming Cao617ba132006-10-11 01:20:53 -07001815 inode->i_op = &ext4_special_inode_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001816#endif
Mingming Cao617ba132006-10-11 01:20:53 -07001817 err = ext4_add_nondir(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001818 }
Mingming Cao617ba132006-10-11 01:20:53 -07001819 ext4_journal_stop(handle);
1820 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001821 goto retry;
1822 return err;
1823}
1824
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001825static int ext4_mkdir(struct inode *dir, struct dentry *dentry, int mode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001826{
1827 handle_t *handle;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001828 struct inode *inode;
1829 struct buffer_head *dir_block;
1830 struct ext4_dir_entry_2 *de;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001831 unsigned int blocksize = dir->i_sb->s_blocksize;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001832 int err, retries = 0;
1833
Andreas Dilgerf8628a12007-07-18 08:38:01 -04001834 if (EXT4_DIR_LINK_MAX(dir))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001835 return -EMLINK;
1836
Christoph Hellwig871a2932010-03-03 09:05:07 -05001837 dquot_initialize(dir);
Christoph Hellwig907f4552010-03-03 09:05:06 -05001838
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001839retry:
Mingming Cao617ba132006-10-11 01:20:53 -07001840 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1841 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
Dmitry Monakhov5aca07e2009-12-08 22:42:15 -05001842 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001843 if (IS_ERR(handle))
1844 return PTR_ERR(handle);
1845
1846 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05001847 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001848
Andreas Dilger11013912009-06-13 11:45:35 -04001849 inode = ext4_new_inode(handle, dir, S_IFDIR | mode,
1850 &dentry->d_name, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001851 err = PTR_ERR(inode);
1852 if (IS_ERR(inode))
1853 goto out_stop;
1854
Mingming Cao617ba132006-10-11 01:20:53 -07001855 inode->i_op = &ext4_dir_inode_operations;
1856 inode->i_fop = &ext4_dir_operations;
1857 inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001858 dir_block = ext4_bread(handle, inode, 0, 1, &err);
Aneesh Kumar K.V4cdeed82008-02-22 06:17:31 -05001859 if (!dir_block)
1860 goto out_clear_inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001861 BUFFER_TRACE(dir_block, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001862 ext4_journal_get_write_access(handle, dir_block);
1863 de = (struct ext4_dir_entry_2 *) dir_block->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001864 de->inode = cpu_to_le32(inode->i_ino);
1865 de->name_len = 1;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001866 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
1867 blocksize);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001868 strcpy(de->name, ".");
Mingming Cao617ba132006-10-11 01:20:53 -07001869 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001870 de = ext4_next_entry(de, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001871 de->inode = cpu_to_le32(dir->i_ino);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001872 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(1),
1873 blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001874 de->name_len = 2;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001875 strcpy(de->name, "..");
Mingming Cao617ba132006-10-11 01:20:53 -07001876 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001877 inode->i_nlink = 2;
Frank Mayhar03901312009-01-07 00:06:22 -05001878 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
1879 ext4_handle_dirty_metadata(handle, dir, dir_block);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001880 brelse(dir_block);
Mingming Cao617ba132006-10-11 01:20:53 -07001881 ext4_mark_inode_dirty(handle, inode);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001882 err = ext4_add_entry(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001883 if (err) {
Aneesh Kumar K.V4cdeed82008-02-22 06:17:31 -05001884out_clear_inode:
1885 clear_nlink(inode);
Al Viro6b38e842008-12-30 02:03:31 -05001886 unlock_new_inode(inode);
Mingming Cao617ba132006-10-11 01:20:53 -07001887 ext4_mark_inode_dirty(handle, inode);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001888 iput(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001889 goto out_stop;
1890 }
Andreas Dilgerf8628a12007-07-18 08:38:01 -04001891 ext4_inc_count(handle, dir);
Mingming Cao617ba132006-10-11 01:20:53 -07001892 ext4_update_dx_flag(dir);
1893 ext4_mark_inode_dirty(handle, dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001894 d_instantiate(dentry, inode);
Al Viro6b38e842008-12-30 02:03:31 -05001895 unlock_new_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001896out_stop:
Mingming Cao617ba132006-10-11 01:20:53 -07001897 ext4_journal_stop(handle);
1898 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001899 goto retry;
1900 return err;
1901}
1902
1903/*
1904 * routine to check that the specified directory is empty (for rmdir)
1905 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001906static int empty_dir(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001907{
Theodore Ts'o498e5f22008-11-05 00:14:04 -05001908 unsigned int offset;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001909 struct buffer_head *bh;
1910 struct ext4_dir_entry_2 *de, *de1;
1911 struct super_block *sb;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001912 int err = 0;
1913
1914 sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -07001915 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001916 !(bh = ext4_bread(NULL, inode, 0, 0, &err))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001917 if (err)
Theodore Ts'o24676da2010-05-16 21:00:00 -04001918 EXT4_ERROR_INODE(inode,
1919 "error %d reading directory lblock 0", err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001920 else
Eric Sandeen12062dd2010-02-15 14:19:27 -05001921 ext4_warning(inode->i_sb,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001922 "bad directory (dir #%lu) - no data block",
1923 inode->i_ino);
1924 return 1;
1925 }
Mingming Cao617ba132006-10-11 01:20:53 -07001926 de = (struct ext4_dir_entry_2 *) bh->b_data;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001927 de1 = ext4_next_entry(de, sb->s_blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001928 if (le32_to_cpu(de->inode) != inode->i_ino ||
1929 !le32_to_cpu(de1->inode) ||
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001930 strcmp(".", de->name) ||
1931 strcmp("..", de1->name)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001932 ext4_warning(inode->i_sb,
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001933 "bad directory (dir #%lu) - no `.' or `..'",
1934 inode->i_ino);
1935 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001936 return 1;
1937 }
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001938 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
1939 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
1940 de = ext4_next_entry(de1, sb->s_blocksize);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001941 while (offset < inode->i_size) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001942 if (!bh ||
Theodore Ts'o24676da2010-05-16 21:00:00 -04001943 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1944 unsigned int lblock;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001945 err = 0;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001946 brelse(bh);
Theodore Ts'o24676da2010-05-16 21:00:00 -04001947 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
1948 bh = ext4_bread(NULL, inode, lblock, 0, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001949 if (!bh) {
1950 if (err)
Theodore Ts'o24676da2010-05-16 21:00:00 -04001951 EXT4_ERROR_INODE(inode,
1952 "error %d reading directory "
1953 "lblock %u", err, lblock);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001954 offset += sb->s_blocksize;
1955 continue;
1956 }
Mingming Cao617ba132006-10-11 01:20:53 -07001957 de = (struct ext4_dir_entry_2 *) bh->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001958 }
Mingming Cao617ba132006-10-11 01:20:53 -07001959 if (!ext4_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1960 de = (struct ext4_dir_entry_2 *)(bh->b_data +
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001961 sb->s_blocksize);
1962 offset = (offset | (sb->s_blocksize - 1)) + 1;
1963 continue;
1964 }
1965 if (le32_to_cpu(de->inode)) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001966 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001967 return 0;
1968 }
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001969 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
1970 de = ext4_next_entry(de, sb->s_blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001971 }
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001972 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001973 return 1;
1974}
1975
Mingming Cao617ba132006-10-11 01:20:53 -07001976/* ext4_orphan_add() links an unlinked or truncated inode into a list of
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001977 * such inodes, starting at the superblock, in case we crash before the
1978 * file is closed/deleted, or in case the inode truncate spans multiple
1979 * transactions and the last transaction is not recovered after a crash.
1980 *
1981 * At filesystem recovery time, we walk this list deleting unlinked
Mingming Cao617ba132006-10-11 01:20:53 -07001982 * inodes and truncating linked inodes in ext4_orphan_cleanup().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001983 */
Mingming Cao617ba132006-10-11 01:20:53 -07001984int ext4_orphan_add(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001985{
1986 struct super_block *sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -07001987 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001988 int err = 0, rc;
1989
Frank Mayhar03901312009-01-07 00:06:22 -05001990 if (!ext4_handle_valid(handle))
1991 return 0;
1992
Theodore Ts'o3b9d4ed2009-04-25 22:54:04 -04001993 mutex_lock(&EXT4_SB(sb)->s_orphan_lock);
Mingming Cao617ba132006-10-11 01:20:53 -07001994 if (!list_empty(&EXT4_I(inode)->i_orphan))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001995 goto out_unlock;
1996
1997 /* Orphan handling is only valid for files with data blocks
1998 * being truncated, or files being unlinked. */
1999
2000 /* @@@ FIXME: Observation from aviro:
Mingming Cao617ba132006-10-11 01:20:53 -07002001 * I think I can trigger J_ASSERT in ext4_orphan_add(). We block
Theodore Ts'o3b9d4ed2009-04-25 22:54:04 -04002002 * here (on s_orphan_lock), so race with ext4_link() which might bump
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002003 * ->i_nlink. For, say it, character device. Not a regular file,
2004 * not a directory, not a symlink and ->i_nlink > 0.
Theodore Ts'o3b9d4ed2009-04-25 22:54:04 -04002005 *
2006 * tytso, 4/25/2009: I'm not sure how that could happen;
2007 * shouldn't the fs core protect us from these sort of
2008 * unlink()/link() races?
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002009 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002010 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2011 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002012
Mingming Cao617ba132006-10-11 01:20:53 -07002013 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
2014 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002015 if (err)
2016 goto out_unlock;
2017
Mingming Cao617ba132006-10-11 01:20:53 -07002018 err = ext4_reserve_inode_write(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002019 if (err)
2020 goto out_unlock;
Dmitry Monakhov6e3617e2010-03-01 23:29:39 -05002021 /*
2022 * Due to previous errors inode may be already a part of on-disk
2023 * orphan list. If so skip on-disk list modification.
2024 */
2025 if (NEXT_ORPHAN(inode) && NEXT_ORPHAN(inode) <=
2026 (le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)))
2027 goto mem_insert;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002028
2029 /* Insert this inode at the head of the on-disk orphan list... */
Mingming Cao617ba132006-10-11 01:20:53 -07002030 NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
2031 EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
Curt Wohlgemuth73b50c12010-02-16 15:06:29 -05002032 err = ext4_handle_dirty_metadata(handle, NULL, EXT4_SB(sb)->s_sbh);
Mingming Cao617ba132006-10-11 01:20:53 -07002033 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002034 if (!err)
2035 err = rc;
2036
2037 /* Only add to the head of the in-memory list if all the
2038 * previous operations succeeded. If the orphan_add is going to
2039 * fail (possibly taking the journal offline), we can't risk
2040 * leaving the inode on the orphan list: stray orphan-list
2041 * entries can cause panics at unmount time.
2042 *
2043 * This is safe: on error we're going to ignore the orphan list
2044 * anyway on the next recovery. */
Dmitry Monakhov6e3617e2010-03-01 23:29:39 -05002045mem_insert:
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002046 if (!err)
Mingming Cao617ba132006-10-11 01:20:53 -07002047 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002048
2049 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2050 jbd_debug(4, "orphan inode %lu will point to %d\n",
2051 inode->i_ino, NEXT_ORPHAN(inode));
2052out_unlock:
Theodore Ts'o3b9d4ed2009-04-25 22:54:04 -04002053 mutex_unlock(&EXT4_SB(sb)->s_orphan_lock);
Mingming Cao617ba132006-10-11 01:20:53 -07002054 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002055 return err;
2056}
2057
2058/*
Mingming Cao617ba132006-10-11 01:20:53 -07002059 * ext4_orphan_del() removes an unlinked or truncated inode from the list
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002060 * of such inodes stored on disk, because it is finally being cleaned up.
2061 */
Mingming Cao617ba132006-10-11 01:20:53 -07002062int ext4_orphan_del(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002063{
2064 struct list_head *prev;
Mingming Cao617ba132006-10-11 01:20:53 -07002065 struct ext4_inode_info *ei = EXT4_I(inode);
2066 struct ext4_sb_info *sbi;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05002067 __u32 ino_next;
Mingming Cao617ba132006-10-11 01:20:53 -07002068 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002069 int err = 0;
2070
Curt Wohlgemuthd3d1faf2009-09-29 11:01:03 -04002071 /* ext4_handle_valid() assumes a valid handle_t pointer */
2072 if (handle && !ext4_handle_valid(handle))
Frank Mayhar03901312009-01-07 00:06:22 -05002073 return 0;
2074
Theodore Ts'o3b9d4ed2009-04-25 22:54:04 -04002075 mutex_lock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2076 if (list_empty(&ei->i_orphan))
2077 goto out;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002078
2079 ino_next = NEXT_ORPHAN(inode);
2080 prev = ei->i_orphan.prev;
Mingming Cao617ba132006-10-11 01:20:53 -07002081 sbi = EXT4_SB(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002082
2083 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2084
2085 list_del_init(&ei->i_orphan);
2086
2087 /* If we're on an error path, we may not have a valid
2088 * transaction handle with which to update the orphan list on
2089 * disk, but we still need to remove the inode from the linked
2090 * list in memory. */
Frank Mayhar03901312009-01-07 00:06:22 -05002091 if (sbi->s_journal && !handle)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002092 goto out;
2093
Mingming Cao617ba132006-10-11 01:20:53 -07002094 err = ext4_reserve_inode_write(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002095 if (err)
2096 goto out_err;
2097
2098 if (prev == &sbi->s_orphan) {
Theodore Ts'o498e5f22008-11-05 00:14:04 -05002099 jbd_debug(4, "superblock will point to %u\n", ino_next);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002100 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07002101 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002102 if (err)
2103 goto out_brelse;
2104 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
Curt Wohlgemuth73b50c12010-02-16 15:06:29 -05002105 err = ext4_handle_dirty_metadata(handle, NULL, sbi->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002106 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07002107 struct ext4_iloc iloc2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002108 struct inode *i_prev =
Mingming Cao617ba132006-10-11 01:20:53 -07002109 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002110
Theodore Ts'o498e5f22008-11-05 00:14:04 -05002111 jbd_debug(4, "orphan inode %lu will point to %u\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002112 i_prev->i_ino, ino_next);
Mingming Cao617ba132006-10-11 01:20:53 -07002113 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002114 if (err)
2115 goto out_brelse;
2116 NEXT_ORPHAN(i_prev) = ino_next;
Mingming Cao617ba132006-10-11 01:20:53 -07002117 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002118 }
2119 if (err)
2120 goto out_brelse;
2121 NEXT_ORPHAN(inode) = 0;
Mingming Cao617ba132006-10-11 01:20:53 -07002122 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002123
2124out_err:
Mingming Cao617ba132006-10-11 01:20:53 -07002125 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002126out:
Theodore Ts'o3b9d4ed2009-04-25 22:54:04 -04002127 mutex_unlock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002128 return err;
2129
2130out_brelse:
2131 brelse(iloc.bh);
2132 goto out_err;
2133}
2134
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002135static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002136{
2137 int retval;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002138 struct inode *inode;
2139 struct buffer_head *bh;
2140 struct ext4_dir_entry_2 *de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002141 handle_t *handle;
2142
2143 /* Initialize quotas before so that eventual writes go in
2144 * separate transaction */
Christoph Hellwig871a2932010-03-03 09:05:07 -05002145 dquot_initialize(dir);
2146 dquot_initialize(dentry->d_inode);
Christoph Hellwig907f4552010-03-03 09:05:06 -05002147
Mingming Cao617ba132006-10-11 01:20:53 -07002148 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002149 if (IS_ERR(handle))
2150 return PTR_ERR(handle);
2151
2152 retval = -ENOENT;
Theodore Ts'of702ba02008-09-22 15:21:01 -04002153 bh = ext4_find_entry(dir, &dentry->d_name, &de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002154 if (!bh)
2155 goto end_rmdir;
2156
2157 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05002158 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002159
2160 inode = dentry->d_inode;
2161
2162 retval = -EIO;
2163 if (le32_to_cpu(de->inode) != inode->i_ino)
2164 goto end_rmdir;
2165
2166 retval = -ENOTEMPTY;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002167 if (!empty_dir(inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002168 goto end_rmdir;
2169
Mingming Cao617ba132006-10-11 01:20:53 -07002170 retval = ext4_delete_entry(handle, dir, de, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002171 if (retval)
2172 goto end_rmdir;
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002173 if (!EXT4_DIR_LINK_EMPTY(inode))
Eric Sandeen12062dd2010-02-15 14:19:27 -05002174 ext4_warning(inode->i_sb,
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002175 "empty directory has too many links (%d)",
2176 inode->i_nlink);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002177 inode->i_version++;
2178 clear_nlink(inode);
2179 /* There's no need to set i_disksize: the fact that i_nlink is
2180 * zero will ensure that the right thing happens during any
2181 * recovery. */
2182 inode->i_size = 0;
Mingming Cao617ba132006-10-11 01:20:53 -07002183 ext4_orphan_add(handle, inode);
Kalpak Shahef7f3832007-07-18 09:15:20 -04002184 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002185 ext4_mark_inode_dirty(handle, inode);
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002186 ext4_dec_count(handle, dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002187 ext4_update_dx_flag(dir);
2188 ext4_mark_inode_dirty(handle, dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002189
2190end_rmdir:
Mingming Cao617ba132006-10-11 01:20:53 -07002191 ext4_journal_stop(handle);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002192 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002193 return retval;
2194}
2195
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002196static int ext4_unlink(struct inode *dir, struct dentry *dentry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002197{
2198 int retval;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002199 struct inode *inode;
2200 struct buffer_head *bh;
2201 struct ext4_dir_entry_2 *de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002202 handle_t *handle;
2203
2204 /* Initialize quotas before so that eventual writes go
2205 * in separate transaction */
Christoph Hellwig871a2932010-03-03 09:05:07 -05002206 dquot_initialize(dir);
2207 dquot_initialize(dentry->d_inode);
Christoph Hellwig907f4552010-03-03 09:05:06 -05002208
Mingming Cao617ba132006-10-11 01:20:53 -07002209 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002210 if (IS_ERR(handle))
2211 return PTR_ERR(handle);
2212
2213 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05002214 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002215
2216 retval = -ENOENT;
Theodore Ts'of702ba02008-09-22 15:21:01 -04002217 bh = ext4_find_entry(dir, &dentry->d_name, &de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002218 if (!bh)
2219 goto end_unlink;
2220
2221 inode = dentry->d_inode;
2222
2223 retval = -EIO;
2224 if (le32_to_cpu(de->inode) != inode->i_ino)
2225 goto end_unlink;
2226
2227 if (!inode->i_nlink) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05002228 ext4_warning(inode->i_sb,
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002229 "Deleting nonexistent file (%lu), %d",
2230 inode->i_ino, inode->i_nlink);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002231 inode->i_nlink = 1;
2232 }
Mingming Cao617ba132006-10-11 01:20:53 -07002233 retval = ext4_delete_entry(handle, dir, de, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002234 if (retval)
2235 goto end_unlink;
Kalpak Shahef7f3832007-07-18 09:15:20 -04002236 dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002237 ext4_update_dx_flag(dir);
2238 ext4_mark_inode_dirty(handle, dir);
Theodore Ts'o825f1482008-02-15 15:00:38 -05002239 drop_nlink(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002240 if (!inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07002241 ext4_orphan_add(handle, inode);
Kalpak Shahef7f3832007-07-18 09:15:20 -04002242 inode->i_ctime = ext4_current_time(inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002243 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002244 retval = 0;
2245
2246end_unlink:
Mingming Cao617ba132006-10-11 01:20:53 -07002247 ext4_journal_stop(handle);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002248 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002249 return retval;
2250}
2251
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002252static int ext4_symlink(struct inode *dir,
2253 struct dentry *dentry, const char *symname)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002254{
2255 handle_t *handle;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002256 struct inode *inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002257 int l, err, retries = 0;
2258
2259 l = strlen(symname)+1;
2260 if (l > dir->i_sb->s_blocksize)
2261 return -ENAMETOOLONG;
2262
Christoph Hellwig871a2932010-03-03 09:05:07 -05002263 dquot_initialize(dir);
Christoph Hellwig907f4552010-03-03 09:05:06 -05002264
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002265retry:
Mingming Cao617ba132006-10-11 01:20:53 -07002266 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2267 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 5 +
Dmitry Monakhov5aca07e2009-12-08 22:42:15 -05002268 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002269 if (IS_ERR(handle))
2270 return PTR_ERR(handle);
2271
2272 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05002273 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002274
Andreas Dilger11013912009-06-13 11:45:35 -04002275 inode = ext4_new_inode(handle, dir, S_IFLNK|S_IRWXUGO,
2276 &dentry->d_name, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002277 err = PTR_ERR(inode);
2278 if (IS_ERR(inode))
2279 goto out_stop;
2280
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002281 if (l > sizeof(EXT4_I(inode)->i_data)) {
Mingming Cao617ba132006-10-11 01:20:53 -07002282 inode->i_op = &ext4_symlink_inode_operations;
2283 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002284 /*
Mingming Cao617ba132006-10-11 01:20:53 -07002285 * page_symlink() calls into ext4_prepare/commit_write.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002286 * We have a transaction open. All is sweetness. It also sets
2287 * i_size in generic_commit_write().
2288 */
Nick Piggin54566b22009-01-04 12:00:53 -08002289 err = __page_symlink(inode, symname, l, 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002290 if (err) {
Theodore Ts'o825f1482008-02-15 15:00:38 -05002291 clear_nlink(inode);
Al Viro6b38e842008-12-30 02:03:31 -05002292 unlock_new_inode(inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002293 ext4_mark_inode_dirty(handle, inode);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002294 iput(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002295 goto out_stop;
2296 }
2297 } else {
Aneesh Kumar K.Ve65187e2008-04-29 08:11:12 -04002298 /* clear the extent format for fast symlink */
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04002299 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
Mingming Cao617ba132006-10-11 01:20:53 -07002300 inode->i_op = &ext4_fast_symlink_inode_operations;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002301 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002302 inode->i_size = l-1;
2303 }
Mingming Cao617ba132006-10-11 01:20:53 -07002304 EXT4_I(inode)->i_disksize = inode->i_size;
2305 err = ext4_add_nondir(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002306out_stop:
Mingming Cao617ba132006-10-11 01:20:53 -07002307 ext4_journal_stop(handle);
2308 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002309 goto retry;
2310 return err;
2311}
2312
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002313static int ext4_link(struct dentry *old_dentry,
2314 struct inode *dir, struct dentry *dentry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002315{
2316 handle_t *handle;
2317 struct inode *inode = old_dentry->d_inode;
2318 int err, retries = 0;
2319
Theodore Ts'ob05ab1d2009-08-29 21:08:08 -04002320 if (inode->i_nlink >= EXT4_LINK_MAX)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002321 return -EMLINK;
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002322
Christoph Hellwig871a2932010-03-03 09:05:07 -05002323 dquot_initialize(dir);
Christoph Hellwig907f4552010-03-03 09:05:06 -05002324
Eric Sandeen2988a772007-02-10 01:46:16 -08002325 /*
2326 * Return -ENOENT if we've raced with unlink and i_nlink is 0. Doing
2327 * otherwise has the potential to corrupt the orphan inode list.
2328 */
2329 if (inode->i_nlink == 0)
2330 return -ENOENT;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002331
2332retry:
Mingming Cao617ba132006-10-11 01:20:53 -07002333 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2334 EXT4_INDEX_EXTRA_TRANS_BLOCKS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002335 if (IS_ERR(handle))
2336 return PTR_ERR(handle);
2337
2338 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05002339 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002340
Kalpak Shahef7f3832007-07-18 09:15:20 -04002341 inode->i_ctime = ext4_current_time(inode);
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002342 ext4_inc_count(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002343 atomic_inc(&inode->i_count);
2344
Al Viro6b38e842008-12-30 02:03:31 -05002345 err = ext4_add_entry(handle, dentry, inode);
2346 if (!err) {
2347 ext4_mark_inode_dirty(handle, inode);
2348 d_instantiate(dentry, inode);
2349 } else {
2350 drop_nlink(inode);
2351 iput(inode);
2352 }
Mingming Cao617ba132006-10-11 01:20:53 -07002353 ext4_journal_stop(handle);
2354 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002355 goto retry;
2356 return err;
2357}
2358
Wei Yongjun3d0518f2009-02-14 23:01:36 -05002359#define PARENT_INO(buffer, size) \
2360 (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer), size)->inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002361
2362/*
2363 * Anybody can rename anything with this: the permission checks are left to the
2364 * higher-level routines.
2365 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002366static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
2367 struct inode *new_dir, struct dentry *new_dentry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002368{
2369 handle_t *handle;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002370 struct inode *old_inode, *new_inode;
2371 struct buffer_head *old_bh, *new_bh, *dir_bh;
2372 struct ext4_dir_entry_2 *old_de, *new_de;
Theodore Ts'o8750c6d2009-02-23 23:05:27 -05002373 int retval, force_da_alloc = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002374
Christoph Hellwig871a2932010-03-03 09:05:07 -05002375 dquot_initialize(old_dir);
2376 dquot_initialize(new_dir);
Christoph Hellwig907f4552010-03-03 09:05:06 -05002377
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002378 old_bh = new_bh = dir_bh = NULL;
2379
2380 /* Initialize quotas before so that eventual writes go
2381 * in separate transaction */
2382 if (new_dentry->d_inode)
Christoph Hellwig871a2932010-03-03 09:05:07 -05002383 dquot_initialize(new_dentry->d_inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002384 handle = ext4_journal_start(old_dir, 2 *
2385 EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2386 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002387 if (IS_ERR(handle))
2388 return PTR_ERR(handle);
2389
2390 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
Frank Mayhar03901312009-01-07 00:06:22 -05002391 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002392
Theodore Ts'of702ba02008-09-22 15:21:01 -04002393 old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002394 /*
2395 * Check for inode number is _not_ due to possible IO errors.
2396 * We might rmdir the source, keep it as pwd of some process
2397 * and merrily kill the link to whatever was created under the
2398 * same name. Goodbye sticky bit ;-<
2399 */
2400 old_inode = old_dentry->d_inode;
2401 retval = -ENOENT;
2402 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2403 goto end_rename;
2404
2405 new_inode = new_dentry->d_inode;
Theodore Ts'of702ba02008-09-22 15:21:01 -04002406 new_bh = ext4_find_entry(new_dir, &new_dentry->d_name, &new_de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002407 if (new_bh) {
2408 if (!new_inode) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002409 brelse(new_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002410 new_bh = NULL;
2411 }
2412 }
2413 if (S_ISDIR(old_inode->i_mode)) {
2414 if (new_inode) {
2415 retval = -ENOTEMPTY;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002416 if (!empty_dir(new_inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002417 goto end_rename;
2418 }
2419 retval = -EIO;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002420 dir_bh = ext4_bread(handle, old_inode, 0, 0, &retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002421 if (!dir_bh)
2422 goto end_rename;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05002423 if (le32_to_cpu(PARENT_INO(dir_bh->b_data,
2424 old_dir->i_sb->s_blocksize)) != old_dir->i_ino)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002425 goto end_rename;
2426 retval = -EMLINK;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002427 if (!new_inode && new_dir != old_dir &&
Aneesh Kumar K.V2c94eb82009-08-28 21:43:15 -04002428 EXT4_DIR_LINK_MAX(new_dir))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002429 goto end_rename;
2430 }
2431 if (!new_bh) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002432 retval = ext4_add_entry(handle, new_dentry, old_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002433 if (retval)
2434 goto end_rename;
2435 } else {
2436 BUFFER_TRACE(new_bh, "get write access");
Mingming Cao617ba132006-10-11 01:20:53 -07002437 ext4_journal_get_write_access(handle, new_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002438 new_de->inode = cpu_to_le32(old_inode->i_ino);
Mingming Cao617ba132006-10-11 01:20:53 -07002439 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2440 EXT4_FEATURE_INCOMPAT_FILETYPE))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002441 new_de->file_type = old_de->file_type;
2442 new_dir->i_version++;
Jan Kara53b7e9f2008-04-29 22:02:11 -04002443 new_dir->i_ctime = new_dir->i_mtime =
2444 ext4_current_time(new_dir);
2445 ext4_mark_inode_dirty(handle, new_dir);
Frank Mayhar03901312009-01-07 00:06:22 -05002446 BUFFER_TRACE(new_bh, "call ext4_handle_dirty_metadata");
2447 ext4_handle_dirty_metadata(handle, new_dir, new_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002448 brelse(new_bh);
2449 new_bh = NULL;
2450 }
2451
2452 /*
2453 * Like most other Unix systems, set the ctime for inodes on a
2454 * rename.
2455 */
Kalpak Shahef7f3832007-07-18 09:15:20 -04002456 old_inode->i_ctime = ext4_current_time(old_inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002457 ext4_mark_inode_dirty(handle, old_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002458
2459 /*
2460 * ok, that's it
2461 */
2462 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2463 old_de->name_len != old_dentry->d_name.len ||
2464 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
Mingming Cao617ba132006-10-11 01:20:53 -07002465 (retval = ext4_delete_entry(handle, old_dir,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002466 old_de, old_bh)) == -ENOENT) {
2467 /* old_de could have moved from under us during htree split, so
2468 * make sure that we are deleting the right entry. We might
2469 * also be pointing to a stale entry in the unused part of
2470 * old_bh so just checking inum and the name isn't enough. */
2471 struct buffer_head *old_bh2;
Mingming Cao617ba132006-10-11 01:20:53 -07002472 struct ext4_dir_entry_2 *old_de2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002473
Theodore Ts'of702ba02008-09-22 15:21:01 -04002474 old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002475 if (old_bh2) {
Mingming Cao617ba132006-10-11 01:20:53 -07002476 retval = ext4_delete_entry(handle, old_dir,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002477 old_de2, old_bh2);
2478 brelse(old_bh2);
2479 }
2480 }
2481 if (retval) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05002482 ext4_warning(old_dir->i_sb,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002483 "Deleting old file (%lu), %d, error=%d",
2484 old_dir->i_ino, old_dir->i_nlink, retval);
2485 }
2486
2487 if (new_inode) {
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002488 ext4_dec_count(handle, new_inode);
Kalpak Shahef7f3832007-07-18 09:15:20 -04002489 new_inode->i_ctime = ext4_current_time(new_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002490 }
Kalpak Shahef7f3832007-07-18 09:15:20 -04002491 old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002492 ext4_update_dx_flag(old_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002493 if (dir_bh) {
2494 BUFFER_TRACE(dir_bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07002495 ext4_journal_get_write_access(handle, dir_bh);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05002496 PARENT_INO(dir_bh->b_data, new_dir->i_sb->s_blocksize) =
2497 cpu_to_le32(new_dir->i_ino);
Frank Mayhar03901312009-01-07 00:06:22 -05002498 BUFFER_TRACE(dir_bh, "call ext4_handle_dirty_metadata");
2499 ext4_handle_dirty_metadata(handle, old_dir, dir_bh);
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002500 ext4_dec_count(handle, old_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002501 if (new_inode) {
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002502 /* checked empty_dir above, can't have another parent,
Theodore Ts'o825f1482008-02-15 15:00:38 -05002503 * ext4_dec_count() won't work for many-linked dirs */
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002504 new_inode->i_nlink = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002505 } else {
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002506 ext4_inc_count(handle, new_dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002507 ext4_update_dx_flag(new_dir);
2508 ext4_mark_inode_dirty(handle, new_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002509 }
2510 }
Mingming Cao617ba132006-10-11 01:20:53 -07002511 ext4_mark_inode_dirty(handle, old_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002512 if (new_inode) {
Mingming Cao617ba132006-10-11 01:20:53 -07002513 ext4_mark_inode_dirty(handle, new_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002514 if (!new_inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07002515 ext4_orphan_add(handle, new_inode);
Theodore Ts'oafd46722009-03-16 23:12:23 -04002516 if (!test_opt(new_dir->i_sb, NO_AUTO_DA_ALLOC))
2517 force_da_alloc = 1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002518 }
2519 retval = 0;
2520
2521end_rename:
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002522 brelse(dir_bh);
2523 brelse(old_bh);
2524 brelse(new_bh);
Mingming Cao617ba132006-10-11 01:20:53 -07002525 ext4_journal_stop(handle);
Theodore Ts'o8750c6d2009-02-23 23:05:27 -05002526 if (retval == 0 && force_da_alloc)
2527 ext4_alloc_da_blocks(old_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002528 return retval;
2529}
2530
2531/*
2532 * directories can handle most operations...
2533 */
Arjan van de Ven754661f2007-02-12 00:55:38 -08002534const struct inode_operations ext4_dir_inode_operations = {
Mingming Cao617ba132006-10-11 01:20:53 -07002535 .create = ext4_create,
2536 .lookup = ext4_lookup,
2537 .link = ext4_link,
2538 .unlink = ext4_unlink,
2539 .symlink = ext4_symlink,
2540 .mkdir = ext4_mkdir,
2541 .rmdir = ext4_rmdir,
2542 .mknod = ext4_mknod,
2543 .rename = ext4_rename,
2544 .setattr = ext4_setattr,
Theodore Ts'o03010a32008-10-10 20:02:48 -04002545#ifdef CONFIG_EXT4_FS_XATTR
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002546 .setxattr = generic_setxattr,
2547 .getxattr = generic_getxattr,
Mingming Cao617ba132006-10-11 01:20:53 -07002548 .listxattr = ext4_listxattr,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002549 .removexattr = generic_removexattr,
2550#endif
Linus Torvalds1d5ccd12009-08-28 12:12:24 -07002551 .check_acl = ext4_check_acl,
Aneesh Kumar K.Vabc87462009-05-02 22:54:32 -04002552 .fiemap = ext4_fiemap,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002553};
2554
Arjan van de Ven754661f2007-02-12 00:55:38 -08002555const struct inode_operations ext4_special_inode_operations = {
Mingming Cao617ba132006-10-11 01:20:53 -07002556 .setattr = ext4_setattr,
Theodore Ts'o03010a32008-10-10 20:02:48 -04002557#ifdef CONFIG_EXT4_FS_XATTR
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002558 .setxattr = generic_setxattr,
2559 .getxattr = generic_getxattr,
Mingming Cao617ba132006-10-11 01:20:53 -07002560 .listxattr = ext4_listxattr,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002561 .removexattr = generic_removexattr,
2562#endif
Linus Torvalds1d5ccd12009-08-28 12:12:24 -07002563 .check_acl = ext4_check_acl,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002564};