blob: 17a17e10dd605198ddfe2dfed530d4cfff6812c1 [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}
190
191__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);
200 else
201 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'o4776004f2008-09-08 23:00:52 -0400352 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
353 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) {
Harvey Harrison46e665e2008-04-17 10:38:59 -0400386 ext4_warning(dir->i_sb, __func__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700387 "Unrecognised inode hash code %d",
388 root->info.hash_version);
389 brelse(bh);
390 *err = ERR_BAD_DX_DIR;
391 goto fail;
392 }
393 hinfo->hash_version = root->info.hash_version;
Theodore Ts'of99b2582008-10-28 13:21:44 -0400394 if (hinfo->hash_version <= DX_HASH_TEA)
395 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
Mingming Cao617ba132006-10-11 01:20:53 -0700396 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
Theodore Ts'of702ba02008-09-22 15:21:01 -0400397 if (d_name)
398 ext4fs_dirhash(d_name->name, d_name->len, hinfo);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700399 hash = hinfo->hash;
400
401 if (root->info.unused_flags & 1) {
Harvey Harrison46e665e2008-04-17 10:38:59 -0400402 ext4_warning(dir->i_sb, __func__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700403 "Unimplemented inode hash flags: %#06x",
404 root->info.unused_flags);
405 brelse(bh);
406 *err = ERR_BAD_DX_DIR;
407 goto fail;
408 }
409
410 if ((indirect = root->info.indirect_levels) > 1) {
Harvey Harrison46e665e2008-04-17 10:38:59 -0400411 ext4_warning(dir->i_sb, __func__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700412 "Unimplemented inode hash depth: %#06x",
413 root->info.indirect_levels);
414 brelse(bh);
415 *err = ERR_BAD_DX_DIR;
416 goto fail;
417 }
418
419 entries = (struct dx_entry *) (((char *)&root->info) +
420 root->info.info_length);
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700421
422 if (dx_get_limit(entries) != dx_root_limit(dir,
423 root->info.info_length)) {
Harvey Harrison46e665e2008-04-17 10:38:59 -0400424 ext4_warning(dir->i_sb, __func__,
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700425 "dx entry: limit != root limit");
426 brelse(bh);
427 *err = ERR_BAD_DX_DIR;
428 goto fail;
429 }
430
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400431 dxtrace(printk("Look up %x", hash));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700432 while (1)
433 {
434 count = dx_get_count(entries);
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700435 if (!count || count > dx_get_limit(entries)) {
Harvey Harrison46e665e2008-04-17 10:38:59 -0400436 ext4_warning(dir->i_sb, __func__,
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700437 "dx entry: no count or count > limit");
438 brelse(bh);
439 *err = ERR_BAD_DX_DIR;
440 goto fail2;
441 }
442
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700443 p = entries + 1;
444 q = entries + count - 1;
445 while (p <= q)
446 {
447 m = p + (q - p)/2;
448 dxtrace(printk("."));
449 if (dx_get_hash(m) > hash)
450 q = m - 1;
451 else
452 p = m + 1;
453 }
454
455 if (0) // linear search cross check
456 {
457 unsigned n = count - 1;
458 at = entries;
459 while (n--)
460 {
461 dxtrace(printk(","));
462 if (dx_get_hash(++at) > hash)
463 {
464 at--;
465 break;
466 }
467 }
468 assert (at == p - 1);
469 }
470
471 at = p - 1;
472 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
473 frame->bh = bh;
474 frame->entries = entries;
475 frame->at = at;
476 if (!indirect--) return frame;
Mingming Cao617ba132006-10-11 01:20:53 -0700477 if (!(bh = ext4_bread (NULL,dir, dx_get_block(at), 0, err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700478 goto fail2;
479 at = entries = ((struct dx_node *) bh->b_data)->entries;
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700480 if (dx_get_limit(entries) != dx_node_limit (dir)) {
Harvey Harrison46e665e2008-04-17 10:38:59 -0400481 ext4_warning(dir->i_sb, __func__,
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700482 "dx entry: limit != node limit");
483 brelse(bh);
484 *err = ERR_BAD_DX_DIR;
485 goto fail2;
486 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700487 frame++;
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700488 frame->bh = NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700489 }
490fail2:
491 while (frame >= frame_in) {
492 brelse(frame->bh);
493 frame--;
494 }
495fail:
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700496 if (*err == ERR_BAD_DX_DIR)
Harvey Harrison46e665e2008-04-17 10:38:59 -0400497 ext4_warning(dir->i_sb, __func__,
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700498 "Corrupt dir inode %ld, running e2fsck is "
499 "recommended.", dir->i_ino);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700500 return NULL;
501}
502
503static void dx_release (struct dx_frame *frames)
504{
505 if (frames[0].bh == NULL)
506 return;
507
508 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
509 brelse(frames[1].bh);
510 brelse(frames[0].bh);
511}
512
513/*
514 * This function increments the frame pointer to search the next leaf
515 * block, and reads in the necessary intervening nodes if the search
516 * should be necessary. Whether or not the search is necessary is
517 * controlled by the hash parameter. If the hash value is even, then
518 * the search is only continued if the next block starts with that
519 * hash value. This is used if we are searching for a specific file.
520 *
521 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
522 *
523 * This function returns 1 if the caller should continue to search,
524 * or 0 if it should not. If there is an error reading one of the
525 * index blocks, it will a negative error code.
526 *
527 * If start_hash is non-null, it will be filled in with the starting
528 * hash of the next page.
529 */
Mingming Cao617ba132006-10-11 01:20:53 -0700530static int ext4_htree_next_block(struct inode *dir, __u32 hash,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700531 struct dx_frame *frame,
532 struct dx_frame *frames,
533 __u32 *start_hash)
534{
535 struct dx_frame *p;
536 struct buffer_head *bh;
537 int err, num_frames = 0;
538 __u32 bhash;
539
540 p = frame;
541 /*
542 * Find the next leaf page by incrementing the frame pointer.
543 * If we run out of entries in the interior node, loop around and
544 * increment pointer in the parent node. When we break out of
545 * this loop, num_frames indicates the number of interior
546 * nodes need to be read.
547 */
548 while (1) {
549 if (++(p->at) < p->entries + dx_get_count(p->entries))
550 break;
551 if (p == frames)
552 return 0;
553 num_frames++;
554 p--;
555 }
556
557 /*
558 * If the hash is 1, then continue only if the next page has a
559 * continuation hash of any value. This is used for readdir
560 * handling. Otherwise, check to see if the hash matches the
561 * desired contiuation hash. If it doesn't, return since
562 * there's no point to read in the successive index pages.
563 */
564 bhash = dx_get_hash(p->at);
565 if (start_hash)
566 *start_hash = bhash;
567 if ((hash & 1) == 0) {
568 if ((bhash & ~1) != hash)
569 return 0;
570 }
571 /*
572 * If the hash is HASH_NB_ALWAYS, we always go to the next
573 * block so no check is necessary
574 */
575 while (num_frames--) {
Mingming Cao617ba132006-10-11 01:20:53 -0700576 if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at),
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700577 0, &err)))
578 return err; /* Failure */
579 p++;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400580 brelse(p->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700581 p->bh = bh;
582 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
583 }
584 return 1;
585}
586
587
588/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700589 * This function fills a red-black tree with information from a
590 * directory block. It returns the number directory entries loaded
591 * into the tree. If there is an error it is returned in err.
592 */
593static int htree_dirblock_to_tree(struct file *dir_file,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500594 struct inode *dir, ext4_lblk_t block,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700595 struct dx_hash_info *hinfo,
596 __u32 start_hash, __u32 start_minor_hash)
597{
598 struct buffer_head *bh;
Mingming Cao617ba132006-10-11 01:20:53 -0700599 struct ext4_dir_entry_2 *de, *top;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700600 int err, count = 0;
601
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500602 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
603 (unsigned long)block));
Mingming Cao617ba132006-10-11 01:20:53 -0700604 if (!(bh = ext4_bread (NULL, dir, block, 0, &err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700605 return err;
606
Mingming Cao617ba132006-10-11 01:20:53 -0700607 de = (struct ext4_dir_entry_2 *) bh->b_data;
608 top = (struct ext4_dir_entry_2 *) ((char *) de +
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700609 dir->i_sb->s_blocksize -
Mingming Cao617ba132006-10-11 01:20:53 -0700610 EXT4_DIR_REC_LEN(0));
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500611 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
Eric Sandeene6c40212006-12-06 20:36:28 -0800612 if (!ext4_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
613 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
614 +((char *)de - bh->b_data))) {
615 /* On error, skip the f_pos to the next block. */
616 dir_file->f_pos = (dir_file->f_pos |
617 (dir->i_sb->s_blocksize - 1)) + 1;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400618 brelse(bh);
Eric Sandeene6c40212006-12-06 20:36:28 -0800619 return count;
620 }
Mingming Cao617ba132006-10-11 01:20:53 -0700621 ext4fs_dirhash(de->name, de->name_len, hinfo);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700622 if ((hinfo->hash < start_hash) ||
623 ((hinfo->hash == start_hash) &&
624 (hinfo->minor_hash < start_minor_hash)))
625 continue;
626 if (de->inode == 0)
627 continue;
Mingming Cao617ba132006-10-11 01:20:53 -0700628 if ((err = ext4_htree_store_dirent(dir_file,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700629 hinfo->hash, hinfo->minor_hash, de)) != 0) {
630 brelse(bh);
631 return err;
632 }
633 count++;
634 }
635 brelse(bh);
636 return count;
637}
638
639
640/*
641 * This function fills a red-black tree with information from a
642 * directory. We start scanning the directory in hash order, starting
643 * at start_hash and start_minor_hash.
644 *
645 * This function returns the number of entries inserted into the tree,
646 * or a negative error code.
647 */
Mingming Cao617ba132006-10-11 01:20:53 -0700648int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700649 __u32 start_minor_hash, __u32 *next_hash)
650{
651 struct dx_hash_info hinfo;
Mingming Cao617ba132006-10-11 01:20:53 -0700652 struct ext4_dir_entry_2 *de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700653 struct dx_frame frames[2], *frame;
654 struct inode *dir;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500655 ext4_lblk_t block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700656 int count = 0;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500657 int ret, err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700658 __u32 hashval;
659
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400660 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
661 start_hash, start_minor_hash));
Josef "Jeff" Sipek9d549892006-12-08 02:36:38 -0800662 dir = dir_file->f_path.dentry->d_inode;
Mingming Cao617ba132006-10-11 01:20:53 -0700663 if (!(EXT4_I(dir)->i_flags & EXT4_INDEX_FL)) {
664 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
Theodore Ts'of99b2582008-10-28 13:21:44 -0400665 if (hinfo.hash_version <= DX_HASH_TEA)
666 hinfo.hash_version +=
667 EXT4_SB(dir->i_sb)->s_hash_unsigned;
Mingming Cao617ba132006-10-11 01:20:53 -0700668 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700669 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
670 start_hash, start_minor_hash);
671 *next_hash = ~0;
672 return count;
673 }
674 hinfo.hash = start_hash;
675 hinfo.minor_hash = 0;
Theodore Ts'of702ba02008-09-22 15:21:01 -0400676 frame = dx_probe(NULL, dir, &hinfo, frames, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700677 if (!frame)
678 return err;
679
680 /* Add '.' and '..' from the htree header */
681 if (!start_hash && !start_minor_hash) {
Mingming Cao617ba132006-10-11 01:20:53 -0700682 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
683 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700684 goto errout;
685 count++;
686 }
687 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
Mingming Cao617ba132006-10-11 01:20:53 -0700688 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500689 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
Mingming Cao617ba132006-10-11 01:20:53 -0700690 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700691 goto errout;
692 count++;
693 }
694
695 while (1) {
696 block = dx_get_block(frame->at);
697 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
698 start_hash, start_minor_hash);
699 if (ret < 0) {
700 err = ret;
701 goto errout;
702 }
703 count += ret;
704 hashval = ~0;
Mingming Cao617ba132006-10-11 01:20:53 -0700705 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700706 frame, frames, &hashval);
707 *next_hash = hashval;
708 if (ret < 0) {
709 err = ret;
710 goto errout;
711 }
712 /*
713 * Stop if: (a) there are no more entries, or
714 * (b) we have inserted at least one entry and the
715 * next hash value is not a continuation
716 */
717 if ((ret == 0) ||
718 (count && ((hashval & 1) == 0)))
719 break;
720 }
721 dx_release(frames);
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400722 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
723 "next hash: %x\n", count, *next_hash));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700724 return count;
725errout:
726 dx_release(frames);
727 return (err);
728}
729
730
731/*
732 * Directory block splitting, compacting
733 */
734
Eric Sandeenef2b02d2007-09-18 22:46:42 -0700735/*
736 * Create map of hash values, offsets, and sizes, stored at end of block.
737 * Returns number of entries mapped.
738 */
Theodore Ts'o8bad4592009-02-14 21:46:54 -0500739static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
740 struct dx_hash_info *hinfo,
741 struct dx_map_entry *map_tail)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700742{
743 int count = 0;
744 char *base = (char *) de;
745 struct dx_hash_info h = *hinfo;
746
Theodore Ts'o8bad4592009-02-14 21:46:54 -0500747 while ((char *) de < base + blocksize) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700748 if (de->name_len && de->inode) {
Mingming Cao617ba132006-10-11 01:20:53 -0700749 ext4fs_dirhash(de->name, de->name_len, &h);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700750 map_tail--;
751 map_tail->hash = h.hash;
Toshiyuki Okajima9aee2282009-06-08 12:41:35 -0400752 map_tail->offs = ((char *) de - base)>>2;
Eric Sandeenef2b02d2007-09-18 22:46:42 -0700753 map_tail->size = le16_to_cpu(de->rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700754 count++;
755 cond_resched();
756 }
757 /* XXX: do we need to check rec_len == 0 case? -Chris */
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500758 de = ext4_next_entry(de, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700759 }
760 return count;
761}
762
Eric Sandeenef2b02d2007-09-18 22:46:42 -0700763/* Sort map by hash value */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700764static void dx_sort_map (struct dx_map_entry *map, unsigned count)
765{
Andrew Morton63f57932006-10-11 01:21:24 -0700766 struct dx_map_entry *p, *q, *top = map + count - 1;
767 int more;
768 /* Combsort until bubble sort doesn't suck */
769 while (count > 2) {
770 count = count*10/13;
771 if (count - 9 < 2) /* 9, 10 -> 11 */
772 count = 11;
773 for (p = top, q = p - count; q >= map; p--, q--)
774 if (p->hash < q->hash)
775 swap(*p, *q);
776 }
777 /* Garden variety bubble sort */
778 do {
779 more = 0;
780 q = top;
781 while (q-- > map) {
782 if (q[1].hash >= q[0].hash)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700783 continue;
Andrew Morton63f57932006-10-11 01:21:24 -0700784 swap(*(q+1), *q);
785 more = 1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700786 }
787 } while(more);
788}
789
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500790static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700791{
792 struct dx_entry *entries = frame->entries;
793 struct dx_entry *old = frame->at, *new = old + 1;
794 int count = dx_get_count(entries);
795
796 assert(count < dx_get_limit(entries));
797 assert(old < entries + count);
798 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
799 dx_set_hash(new, hash);
800 dx_set_block(new, block);
801 dx_set_count(entries, count + 1);
802}
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700803
Mingming Cao617ba132006-10-11 01:20:53 -0700804static void ext4_update_dx_flag(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700805{
Mingming Cao617ba132006-10-11 01:20:53 -0700806 if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
807 EXT4_FEATURE_COMPAT_DIR_INDEX))
808 EXT4_I(inode)->i_flags &= ~EXT4_INDEX_FL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700809}
810
811/*
Mingming Cao617ba132006-10-11 01:20:53 -0700812 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700813 *
Mingming Cao617ba132006-10-11 01:20:53 -0700814 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700815 * `de != NULL' is guaranteed by caller.
816 */
Mingming Cao617ba132006-10-11 01:20:53 -0700817static inline int ext4_match (int len, const char * const name,
818 struct ext4_dir_entry_2 * de)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700819{
820 if (len != de->name_len)
821 return 0;
822 if (!de->inode)
823 return 0;
824 return !memcmp(name, de->name, len);
825}
826
827/*
828 * Returns 0 if not found, -1 on failure, and 1 on success
829 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400830static inline int search_dirblock(struct buffer_head *bh,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700831 struct inode *dir,
Theodore Ts'of702ba02008-09-22 15:21:01 -0400832 const struct qstr *d_name,
Theodore Ts'o498e5f22008-11-05 00:14:04 -0500833 unsigned int offset,
Mingming Cao617ba132006-10-11 01:20:53 -0700834 struct ext4_dir_entry_2 ** res_dir)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700835{
Mingming Cao617ba132006-10-11 01:20:53 -0700836 struct ext4_dir_entry_2 * de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700837 char * dlimit;
838 int de_len;
Theodore Ts'of702ba02008-09-22 15:21:01 -0400839 const char *name = d_name->name;
840 int namelen = d_name->len;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700841
Mingming Cao617ba132006-10-11 01:20:53 -0700842 de = (struct ext4_dir_entry_2 *) bh->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700843 dlimit = bh->b_data + dir->i_sb->s_blocksize;
844 while ((char *) de < dlimit) {
845 /* this code is executed quadratically often */
846 /* do minimal checking `by hand' */
847
848 if ((char *) de + namelen <= dlimit &&
Mingming Cao617ba132006-10-11 01:20:53 -0700849 ext4_match (namelen, name, de)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700850 /* found a match - just to be sure, do a full check */
Mingming Cao617ba132006-10-11 01:20:53 -0700851 if (!ext4_check_dir_entry("ext4_find_entry",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700852 dir, de, bh, offset))
853 return -1;
854 *res_dir = de;
855 return 1;
856 }
857 /* prevent looping on a bad block */
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500858 de_len = ext4_rec_len_from_disk(de->rec_len,
859 dir->i_sb->s_blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700860 if (de_len <= 0)
861 return -1;
862 offset += de_len;
Mingming Cao617ba132006-10-11 01:20:53 -0700863 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700864 }
865 return 0;
866}
867
868
869/*
Mingming Cao617ba132006-10-11 01:20:53 -0700870 * ext4_find_entry()
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700871 *
872 * finds an entry in the specified directory with the wanted name. It
873 * returns the cache buffer in which the entry was found, and the entry
874 * itself (as a parameter - res_dir). It does NOT read the inode of the
875 * entry - you'll have to do that yourself if you want to.
876 *
877 * The returned buffer_head has ->b_count elevated. The caller is expected
878 * to brelse() it when appropriate.
879 */
Theodore Ts'of702ba02008-09-22 15:21:01 -0400880static struct buffer_head * ext4_find_entry (struct inode *dir,
881 const struct qstr *d_name,
Mingming Cao617ba132006-10-11 01:20:53 -0700882 struct ext4_dir_entry_2 ** res_dir)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700883{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400884 struct super_block *sb;
885 struct buffer_head *bh_use[NAMEI_RA_SIZE];
886 struct buffer_head *bh, *ret = NULL;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500887 ext4_lblk_t start, block, b;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700888 int ra_max = 0; /* Number of bh's in the readahead
889 buffer, bh_use[] */
890 int ra_ptr = 0; /* Current index into readahead
891 buffer */
892 int num = 0;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500893 ext4_lblk_t nblocks;
894 int i, err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700895 int namelen;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700896
897 *res_dir = NULL;
898 sb = dir->i_sb;
Theodore Ts'of702ba02008-09-22 15:21:01 -0400899 namelen = d_name->len;
Mingming Cao617ba132006-10-11 01:20:53 -0700900 if (namelen > EXT4_NAME_LEN)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700901 return NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700902 if (is_dx(dir)) {
Theodore Ts'of702ba02008-09-22 15:21:01 -0400903 bh = ext4_dx_find_entry(dir, d_name, res_dir, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700904 /*
905 * On success, or if the error was file not found,
906 * return. Otherwise, fall back to doing a search the
907 * old fashioned way.
908 */
909 if (bh || (err != ERR_BAD_DX_DIR))
910 return bh;
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400911 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
912 "falling back\n"));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700913 }
Mingming Cao617ba132006-10-11 01:20:53 -0700914 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
915 start = EXT4_I(dir)->i_dir_start_lookup;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700916 if (start >= nblocks)
917 start = 0;
918 block = start;
919restart:
920 do {
921 /*
922 * We deal with the read-ahead logic here.
923 */
924 if (ra_ptr >= ra_max) {
925 /* Refill the readahead buffer */
926 ra_ptr = 0;
927 b = block;
928 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
929 /*
930 * Terminate if we reach the end of the
931 * directory and must wrap, or if our
932 * search has finished at this block.
933 */
934 if (b >= nblocks || (num && block == start)) {
935 bh_use[ra_max] = NULL;
936 break;
937 }
938 num++;
Mingming Cao617ba132006-10-11 01:20:53 -0700939 bh = ext4_getblk(NULL, dir, b++, 0, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700940 bh_use[ra_max] = bh;
941 if (bh)
942 ll_rw_block(READ_META, 1, &bh);
943 }
944 }
945 if ((bh = bh_use[ra_ptr++]) == NULL)
946 goto next;
947 wait_on_buffer(bh);
948 if (!buffer_uptodate(bh)) {
949 /* read error, skip block & hope for the best */
Harvey Harrison46e665e2008-04-17 10:38:59 -0400950 ext4_error(sb, __func__, "reading directory #%lu "
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500951 "offset %lu", dir->i_ino,
952 (unsigned long)block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700953 brelse(bh);
954 goto next;
955 }
Theodore Ts'of702ba02008-09-22 15:21:01 -0400956 i = search_dirblock(bh, dir, d_name,
Mingming Cao617ba132006-10-11 01:20:53 -0700957 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700958 if (i == 1) {
Mingming Cao617ba132006-10-11 01:20:53 -0700959 EXT4_I(dir)->i_dir_start_lookup = block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700960 ret = bh;
961 goto cleanup_and_exit;
962 } else {
963 brelse(bh);
964 if (i < 0)
965 goto cleanup_and_exit;
966 }
967 next:
968 if (++block >= nblocks)
969 block = 0;
970 } while (block != start);
971
972 /*
973 * If the directory has grown while we were searching, then
974 * search the last part of the directory before giving up.
975 */
976 block = nblocks;
Mingming Cao617ba132006-10-11 01:20:53 -0700977 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700978 if (block < nblocks) {
979 start = 0;
980 goto restart;
981 }
982
983cleanup_and_exit:
984 /* Clean up the read-ahead blocks */
985 for (; ra_ptr < ra_max; ra_ptr++)
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400986 brelse(bh_use[ra_ptr]);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700987 return ret;
988}
989
Theodore Ts'of702ba02008-09-22 15:21:01 -0400990static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
Mingming Cao617ba132006-10-11 01:20:53 -0700991 struct ext4_dir_entry_2 **res_dir, int *err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700992{
993 struct super_block * sb;
994 struct dx_hash_info hinfo;
995 u32 hash;
996 struct dx_frame frames[2], *frame;
Mingming Cao617ba132006-10-11 01:20:53 -0700997 struct ext4_dir_entry_2 *de, *top;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700998 struct buffer_head *bh;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500999 ext4_lblk_t block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001000 int retval;
Theodore Ts'of702ba02008-09-22 15:21:01 -04001001 int namelen = d_name->len;
1002 const u8 *name = d_name->name;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001003
1004 sb = dir->i_sb;
1005 /* NFS may look up ".." - look at dx_root directory block */
1006 if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
Theodore Ts'of702ba02008-09-22 15:21:01 -04001007 if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001008 return NULL;
1009 } else {
1010 frame = frames;
1011 frame->bh = NULL; /* for dx_release() */
1012 frame->at = (struct dx_entry *)frames; /* hack for zero entry*/
1013 dx_set_block(frame->at, 0); /* dx_root block is 0 */
1014 }
1015 hash = hinfo.hash;
1016 do {
1017 block = dx_get_block(frame->at);
Mingming Cao617ba132006-10-11 01:20:53 -07001018 if (!(bh = ext4_bread (NULL,dir, block, 0, err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001019 goto errout;
Mingming Cao617ba132006-10-11 01:20:53 -07001020 de = (struct ext4_dir_entry_2 *) bh->b_data;
1021 top = (struct ext4_dir_entry_2 *) ((char *) de + sb->s_blocksize -
1022 EXT4_DIR_REC_LEN(0));
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001023 for (; de < top; de = ext4_next_entry(de, sb->s_blocksize)) {
Duane Griffinf3b35f02008-07-11 19:27:31 -04001024 int off = (block << EXT4_BLOCK_SIZE_BITS(sb))
1025 + ((char *) de - bh->b_data);
1026
1027 if (!ext4_check_dir_entry(__func__, dir, de, bh, off)) {
1028 brelse(bh);
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001029 *err = ERR_BAD_DX_DIR;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001030 goto errout;
1031 }
Duane Griffinf3b35f02008-07-11 19:27:31 -04001032
1033 if (ext4_match(namelen, name, de)) {
1034 *res_dir = de;
1035 dx_release(frames);
1036 return bh;
1037 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001038 }
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001039 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001040 /* Check to see if we should continue to search */
Mingming Cao617ba132006-10-11 01:20:53 -07001041 retval = ext4_htree_next_block(dir, hash, frame,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001042 frames, NULL);
1043 if (retval < 0) {
Harvey Harrison46e665e2008-04-17 10:38:59 -04001044 ext4_warning(sb, __func__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001045 "error reading index page in directory #%lu",
1046 dir->i_ino);
1047 *err = retval;
1048 goto errout;
1049 }
1050 } while (retval == 1);
1051
1052 *err = -ENOENT;
1053errout:
Theodore Ts'o4776004f2008-09-08 23:00:52 -04001054 dxtrace(printk(KERN_DEBUG "%s not found\n", name));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001055 dx_release (frames);
1056 return NULL;
1057}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001058
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001059static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001060{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001061 struct inode *inode;
1062 struct ext4_dir_entry_2 *de;
1063 struct buffer_head *bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001064
Mingming Cao617ba132006-10-11 01:20:53 -07001065 if (dentry->d_name.len > EXT4_NAME_LEN)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001066 return ERR_PTR(-ENAMETOOLONG);
1067
Theodore Ts'of702ba02008-09-22 15:21:01 -04001068 bh = ext4_find_entry(dir, &dentry->d_name, &de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001069 inode = NULL;
1070 if (bh) {
Theodore Ts'o498e5f22008-11-05 00:14:04 -05001071 __u32 ino = le32_to_cpu(de->inode);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001072 brelse(bh);
Mingming Cao617ba132006-10-11 01:20:53 -07001073 if (!ext4_valid_inum(dir->i_sb, ino)) {
1074 ext4_error(dir->i_sb, "ext4_lookup",
Theodore Ts'o498e5f22008-11-05 00:14:04 -05001075 "bad inode number: %u", ino);
David Howells1d1fe1e2008-02-07 00:15:37 -08001076 return ERR_PTR(-EIO);
Vasily Averina6c15c22007-07-15 23:40:46 -07001077 }
David Howells1d1fe1e2008-02-07 00:15:37 -08001078 inode = ext4_iget(dir->i_sb, ino);
Bryan Donlane6f009b2009-02-22 21:20:25 -05001079 if (unlikely(IS_ERR(inode))) {
1080 if (PTR_ERR(inode) == -ESTALE) {
1081 ext4_error(dir->i_sb, __func__,
1082 "deleted inode referenced: %u",
1083 ino);
1084 return ERR_PTR(-EIO);
1085 } else {
1086 return ERR_CAST(inode);
1087 }
1088 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001089 }
1090 return d_splice_alias(inode, dentry);
1091}
1092
1093
Mingming Cao617ba132006-10-11 01:20:53 -07001094struct dentry *ext4_get_parent(struct dentry *child)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001095{
Theodore Ts'o498e5f22008-11-05 00:14:04 -05001096 __u32 ino;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001097 struct inode *inode;
Theodore Ts'of702ba02008-09-22 15:21:01 -04001098 static const struct qstr dotdot = {
1099 .name = "..",
1100 .len = 2,
1101 };
Mingming Cao617ba132006-10-11 01:20:53 -07001102 struct ext4_dir_entry_2 * de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001103 struct buffer_head *bh;
1104
Theodore Ts'of702ba02008-09-22 15:21:01 -04001105 bh = ext4_find_entry(child->d_inode, &dotdot, &de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001106 inode = NULL;
1107 if (!bh)
1108 return ERR_PTR(-ENOENT);
1109 ino = le32_to_cpu(de->inode);
1110 brelse(bh);
1111
Mingming Cao617ba132006-10-11 01:20:53 -07001112 if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1113 ext4_error(child->d_inode->i_sb, "ext4_get_parent",
Theodore Ts'o498e5f22008-11-05 00:14:04 -05001114 "bad inode number: %u", ino);
David Howells1d1fe1e2008-02-07 00:15:37 -08001115 return ERR_PTR(-EIO);
Vasily Averina6c15c22007-07-15 23:40:46 -07001116 }
1117
Christoph Hellwig44003722008-08-11 15:49:04 +02001118 return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001119}
1120
1121#define S_SHIFT 12
Mingming Cao617ba132006-10-11 01:20:53 -07001122static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1123 [S_IFREG >> S_SHIFT] = EXT4_FT_REG_FILE,
1124 [S_IFDIR >> S_SHIFT] = EXT4_FT_DIR,
1125 [S_IFCHR >> S_SHIFT] = EXT4_FT_CHRDEV,
1126 [S_IFBLK >> S_SHIFT] = EXT4_FT_BLKDEV,
1127 [S_IFIFO >> S_SHIFT] = EXT4_FT_FIFO,
1128 [S_IFSOCK >> S_SHIFT] = EXT4_FT_SOCK,
1129 [S_IFLNK >> S_SHIFT] = EXT4_FT_SYMLINK,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001130};
1131
Mingming Cao617ba132006-10-11 01:20:53 -07001132static inline void ext4_set_de_type(struct super_block *sb,
1133 struct ext4_dir_entry_2 *de,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001134 umode_t mode) {
Mingming Cao617ba132006-10-11 01:20:53 -07001135 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1136 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001137}
1138
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001139/*
1140 * Move count entries from end of map between two memory locations.
1141 * Returns pointer to last entry moved.
1142 */
Mingming Cao617ba132006-10-11 01:20:53 -07001143static struct ext4_dir_entry_2 *
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001144dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1145 unsigned blocksize)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001146{
1147 unsigned rec_len = 0;
1148
1149 while (count--) {
Toshiyuki Okajima9aee2282009-06-08 12:41:35 -04001150 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
1151 (from + (map->offs<<2));
Mingming Cao617ba132006-10-11 01:20:53 -07001152 rec_len = EXT4_DIR_REC_LEN(de->name_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001153 memcpy (to, de, rec_len);
Mingming Cao617ba132006-10-11 01:20:53 -07001154 ((struct ext4_dir_entry_2 *) to)->rec_len =
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001155 ext4_rec_len_to_disk(rec_len, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001156 de->inode = 0;
1157 map++;
1158 to += rec_len;
1159 }
Mingming Cao617ba132006-10-11 01:20:53 -07001160 return (struct ext4_dir_entry_2 *) (to - rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001161}
1162
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001163/*
1164 * Compact each dir entry in the range to the minimal rec_len.
1165 * Returns pointer to last entry in range.
1166 */
Theodore Ts'o8bad4592009-02-14 21:46:54 -05001167static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001168{
Mingming Cao617ba132006-10-11 01:20:53 -07001169 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001170 unsigned rec_len = 0;
1171
1172 prev = to = de;
Theodore Ts'o8bad4592009-02-14 21:46:54 -05001173 while ((char*)de < base + blocksize) {
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001174 next = ext4_next_entry(de, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001175 if (de->inode && de->name_len) {
Mingming Cao617ba132006-10-11 01:20:53 -07001176 rec_len = EXT4_DIR_REC_LEN(de->name_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001177 if (de > to)
1178 memmove(to, de, rec_len);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001179 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001180 prev = to;
Mingming Cao617ba132006-10-11 01:20:53 -07001181 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001182 }
1183 de = next;
1184 }
1185 return prev;
1186}
1187
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001188/*
1189 * Split a full leaf block to make room for a new dir entry.
1190 * Allocate a new block, and move entries so that they are approx. equally full.
1191 * Returns pointer to de in block into which the new entry will be inserted.
1192 */
Mingming Cao617ba132006-10-11 01:20:53 -07001193static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001194 struct buffer_head **bh,struct dx_frame *frame,
1195 struct dx_hash_info *hinfo, int *error)
1196{
1197 unsigned blocksize = dir->i_sb->s_blocksize;
1198 unsigned count, continued;
1199 struct buffer_head *bh2;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001200 ext4_lblk_t newblock;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001201 u32 hash2;
1202 struct dx_map_entry *map;
1203 char *data1 = (*bh)->b_data, *data2;
Theodore Ts'o59e315b2008-12-06 16:58:39 -05001204 unsigned split, move, size;
Mingming Cao617ba132006-10-11 01:20:53 -07001205 struct ext4_dir_entry_2 *de = NULL, *de2;
Theodore Ts'o59e315b2008-12-06 16:58:39 -05001206 int err = 0, i;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001207
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001208 bh2 = ext4_append (handle, dir, &newblock, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001209 if (!(bh2)) {
1210 brelse(*bh);
1211 *bh = NULL;
1212 goto errout;
1213 }
1214
1215 BUFFER_TRACE(*bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001216 err = ext4_journal_get_write_access(handle, *bh);
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001217 if (err)
1218 goto journal_error;
1219
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001220 BUFFER_TRACE(frame->bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001221 err = ext4_journal_get_write_access(handle, frame->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001222 if (err)
1223 goto journal_error;
1224
1225 data2 = bh2->b_data;
1226
1227 /* create map in the end of data2 block */
1228 map = (struct dx_map_entry *) (data2 + blocksize);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001229 count = dx_make_map((struct ext4_dir_entry_2 *) data1,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001230 blocksize, hinfo, map);
1231 map -= count;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001232 dx_sort_map(map, count);
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001233 /* Split the existing block in the middle, size-wise */
1234 size = 0;
1235 move = 0;
1236 for (i = count-1; i >= 0; i--) {
1237 /* is more than half of this entry in 2nd half of the block? */
1238 if (size + map[i].size/2 > blocksize/2)
1239 break;
1240 size += map[i].size;
1241 move++;
1242 }
1243 /* map index at which we will split */
1244 split = count - move;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001245 hash2 = map[split].hash;
1246 continued = hash2 == map[split - 1].hash;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001247 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1248 (unsigned long)dx_get_block(frame->at),
1249 hash2, split, count-split));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001250
1251 /* Fancy dance to stay within two buffers */
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001252 de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001253 de = dx_pack_dirents(data1, blocksize);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001254 de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1255 blocksize);
1256 de2->rec_len = ext4_rec_len_to_disk(data2 + blocksize - (char *) de2,
1257 blocksize);
Mingming Cao617ba132006-10-11 01:20:53 -07001258 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1259 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001260
1261 /* Which block gets the new entry? */
1262 if (hinfo->hash >= hash2)
1263 {
1264 swap(*bh, bh2);
1265 de = de2;
1266 }
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001267 dx_insert_block(frame, hash2 + continued, newblock);
Frank Mayhar03901312009-01-07 00:06:22 -05001268 err = ext4_handle_dirty_metadata(handle, dir, bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001269 if (err)
1270 goto journal_error;
Frank Mayhar03901312009-01-07 00:06:22 -05001271 err = ext4_handle_dirty_metadata(handle, dir, frame->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001272 if (err)
1273 goto journal_error;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001274 brelse(bh2);
1275 dxtrace(dx_show_index("frame", frame->entries));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001276 return de;
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001277
1278journal_error:
1279 brelse(*bh);
1280 brelse(bh2);
1281 *bh = NULL;
1282 ext4_std_error(dir->i_sb, err);
1283errout:
1284 *error = err;
1285 return NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001286}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001287
1288/*
1289 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1290 * it points to a directory entry which is guaranteed to be large
1291 * enough for new directory entry. If de is NULL, then
1292 * add_dirent_to_buf will attempt search the directory block for
1293 * space. It will return -ENOSPC if no space is available, and -EIO
1294 * and -EEXIST if directory entry already exists.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001295 */
1296static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
Mingming Cao617ba132006-10-11 01:20:53 -07001297 struct inode *inode, struct ext4_dir_entry_2 *de,
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001298 struct buffer_head *bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001299{
1300 struct inode *dir = dentry->d_parent->d_inode;
1301 const char *name = dentry->d_name.name;
1302 int namelen = dentry->d_name.len;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05001303 unsigned int offset = 0;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001304 unsigned int blocksize = dir->i_sb->s_blocksize;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001305 unsigned short reclen;
1306 int nlen, rlen, err;
1307 char *top;
1308
Mingming Cao617ba132006-10-11 01:20:53 -07001309 reclen = EXT4_DIR_REC_LEN(namelen);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001310 if (!de) {
Mingming Cao617ba132006-10-11 01:20:53 -07001311 de = (struct ext4_dir_entry_2 *)bh->b_data;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001312 top = bh->b_data + blocksize - reclen;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001313 while ((char *) de <= top) {
Mingming Cao617ba132006-10-11 01:20:53 -07001314 if (!ext4_check_dir_entry("ext4_add_entry", dir, de,
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001315 bh, offset))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001316 return -EIO;
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001317 if (ext4_match(namelen, name, de))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001318 return -EEXIST;
Mingming Cao617ba132006-10-11 01:20:53 -07001319 nlen = EXT4_DIR_REC_LEN(de->name_len);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001320 rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001321 if ((de->inode? rlen - nlen: rlen) >= reclen)
1322 break;
Mingming Cao617ba132006-10-11 01:20:53 -07001323 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001324 offset += rlen;
1325 }
1326 if ((char *) de > top)
1327 return -ENOSPC;
1328 }
1329 BUFFER_TRACE(bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001330 err = ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001331 if (err) {
Mingming Cao617ba132006-10-11 01:20:53 -07001332 ext4_std_error(dir->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001333 return err;
1334 }
1335
1336 /* By now the buffer is marked for journaling */
Mingming Cao617ba132006-10-11 01:20:53 -07001337 nlen = EXT4_DIR_REC_LEN(de->name_len);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001338 rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001339 if (de->inode) {
Mingming Cao617ba132006-10-11 01:20:53 -07001340 struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001341 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, blocksize);
1342 de->rec_len = ext4_rec_len_to_disk(nlen, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001343 de = de1;
1344 }
Mingming Cao617ba132006-10-11 01:20:53 -07001345 de->file_type = EXT4_FT_UNKNOWN;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001346 if (inode) {
1347 de->inode = cpu_to_le32(inode->i_ino);
Mingming Cao617ba132006-10-11 01:20:53 -07001348 ext4_set_de_type(dir->i_sb, de, inode->i_mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001349 } else
1350 de->inode = 0;
1351 de->name_len = namelen;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001352 memcpy(de->name, name, namelen);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001353 /*
1354 * XXX shouldn't update any times until successful
1355 * completion of syscall, but too many callers depend
1356 * on this.
1357 *
1358 * XXX similarly, too many callers depend on
Mingming Cao617ba132006-10-11 01:20:53 -07001359 * ext4_new_inode() setting the times, but error
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001360 * recovery deletes the inode, so the worst that can
1361 * happen is that the times are slightly out of date
1362 * and/or different from the directory change time.
1363 */
Kalpak Shahef7f3832007-07-18 09:15:20 -04001364 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
Mingming Cao617ba132006-10-11 01:20:53 -07001365 ext4_update_dx_flag(dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001366 dir->i_version++;
Mingming Cao617ba132006-10-11 01:20:53 -07001367 ext4_mark_inode_dirty(handle, dir);
Frank Mayhar03901312009-01-07 00:06:22 -05001368 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1369 err = ext4_handle_dirty_metadata(handle, dir, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001370 if (err)
Mingming Cao617ba132006-10-11 01:20:53 -07001371 ext4_std_error(dir->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001372 return 0;
1373}
1374
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001375/*
1376 * This converts a one block unindexed directory to a 3 block indexed
1377 * directory, and adds the dentry to the indexed directory.
1378 */
1379static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1380 struct inode *inode, struct buffer_head *bh)
1381{
1382 struct inode *dir = dentry->d_parent->d_inode;
1383 const char *name = dentry->d_name.name;
1384 int namelen = dentry->d_name.len;
1385 struct buffer_head *bh2;
1386 struct dx_root *root;
1387 struct dx_frame frames[2], *frame;
1388 struct dx_entry *entries;
Mingming Cao617ba132006-10-11 01:20:53 -07001389 struct ext4_dir_entry_2 *de, *de2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001390 char *data1, *top;
1391 unsigned len;
1392 int retval;
1393 unsigned blocksize;
1394 struct dx_hash_info hinfo;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001395 ext4_lblk_t block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001396 struct fake_dirent *fde;
1397
1398 blocksize = dir->i_sb->s_blocksize;
Theodore Ts'oe6b8bc02009-01-16 11:13:40 -05001399 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
Mingming Cao617ba132006-10-11 01:20:53 -07001400 retval = ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001401 if (retval) {
Mingming Cao617ba132006-10-11 01:20:53 -07001402 ext4_std_error(dir->i_sb, retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001403 brelse(bh);
1404 return retval;
1405 }
1406 root = (struct dx_root *) bh->b_data;
1407
Theodore Ts'oe6b8bc02009-01-16 11:13:40 -05001408 /* The 0th block becomes the root, move the dirents out */
1409 fde = &root->dotdot;
1410 de = (struct ext4_dir_entry_2 *)((char *)fde +
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001411 ext4_rec_len_from_disk(fde->rec_len, blocksize));
Theodore Ts'oe6b8bc02009-01-16 11:13:40 -05001412 if ((char *) de >= (((char *) root) + blocksize)) {
1413 ext4_error(dir->i_sb, __func__,
1414 "invalid rec_len for '..' in inode %lu",
1415 dir->i_ino);
1416 brelse(bh);
1417 return -EIO;
1418 }
1419 len = ((char *) root) + blocksize - (char *) de;
1420
1421 /* Allocate new block for the 0th block's dirents */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001422 bh2 = ext4_append(handle, dir, &block, &retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001423 if (!(bh2)) {
1424 brelse(bh);
1425 return retval;
1426 }
Mingming Cao617ba132006-10-11 01:20:53 -07001427 EXT4_I(dir)->i_flags |= EXT4_INDEX_FL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001428 data1 = bh2->b_data;
1429
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001430 memcpy (data1, de, len);
Mingming Cao617ba132006-10-11 01:20:53 -07001431 de = (struct ext4_dir_entry_2 *) data1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001432 top = data1 + len;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001433 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001434 de = de2;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001435 de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de,
1436 blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001437 /* Initialize the root; the dot dirents already exist */
Mingming Cao617ba132006-10-11 01:20:53 -07001438 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001439 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1440 blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001441 memset (&root->info, 0, sizeof(root->info));
1442 root->info.info_length = sizeof(root->info);
Mingming Cao617ba132006-10-11 01:20:53 -07001443 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001444 entries = root->entries;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001445 dx_set_block(entries, 1);
1446 dx_set_count(entries, 1);
1447 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001448
1449 /* Initialize as for dx_probe */
1450 hinfo.hash_version = root->info.hash_version;
Theodore Ts'of99b2582008-10-28 13:21:44 -04001451 if (hinfo.hash_version <= DX_HASH_TEA)
1452 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
Mingming Cao617ba132006-10-11 01:20:53 -07001453 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1454 ext4fs_dirhash(name, namelen, &hinfo);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001455 frame = frames;
1456 frame->entries = entries;
1457 frame->at = entries;
1458 frame->bh = bh;
1459 bh = bh2;
1460 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1461 dx_release (frames);
1462 if (!(de))
1463 return retval;
1464
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001465 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1466 brelse(bh);
1467 return retval;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001468}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001469
1470/*
Mingming Cao617ba132006-10-11 01:20:53 -07001471 * ext4_add_entry()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001472 *
1473 * adds a file entry to the specified directory, using the same
Mingming Cao617ba132006-10-11 01:20:53 -07001474 * semantics as ext4_find_entry(). It returns NULL if it failed.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001475 *
1476 * NOTE!! The inode part of 'de' is left at 0 - which means you
1477 * may not sleep between calling this and putting something into
1478 * the entry, as someone else might have used it while you slept.
1479 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001480static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1481 struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001482{
1483 struct inode *dir = dentry->d_parent->d_inode;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001484 struct buffer_head *bh;
Mingming Cao617ba132006-10-11 01:20:53 -07001485 struct ext4_dir_entry_2 *de;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001486 struct super_block *sb;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001487 int retval;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001488 int dx_fallback=0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001489 unsigned blocksize;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001490 ext4_lblk_t block, blocks;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001491
1492 sb = dir->i_sb;
1493 blocksize = sb->s_blocksize;
1494 if (!dentry->d_name.len)
1495 return -EINVAL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001496 if (is_dx(dir)) {
Mingming Cao617ba132006-10-11 01:20:53 -07001497 retval = ext4_dx_add_entry(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001498 if (!retval || (retval != ERR_BAD_DX_DIR))
1499 return retval;
Mingming Cao617ba132006-10-11 01:20:53 -07001500 EXT4_I(dir)->i_flags &= ~EXT4_INDEX_FL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001501 dx_fallback++;
Mingming Cao617ba132006-10-11 01:20:53 -07001502 ext4_mark_inode_dirty(handle, dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001503 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001504 blocks = dir->i_size >> sb->s_blocksize_bits;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05001505 for (block = 0; block < blocks; block++) {
Mingming Cao617ba132006-10-11 01:20:53 -07001506 bh = ext4_bread(handle, dir, block, 0, &retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001507 if(!bh)
1508 return retval;
1509 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001510 if (retval != -ENOSPC) {
1511 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001512 return retval;
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001513 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001514
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001515 if (blocks == 1 && !dx_fallback &&
Theodore Ts'o1e424a32009-11-08 15:45:44 -05001516 EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1517 return make_indexed_dir(handle, dentry, inode, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001518 brelse(bh);
1519 }
Mingming Cao617ba132006-10-11 01:20:53 -07001520 bh = ext4_append(handle, dir, &block, &retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001521 if (!bh)
1522 return retval;
Mingming Cao617ba132006-10-11 01:20:53 -07001523 de = (struct ext4_dir_entry_2 *) bh->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001524 de->inode = 0;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001525 de->rec_len = ext4_rec_len_to_disk(blocksize, blocksize);
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001526 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1527 brelse(bh);
1528 return retval;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001529}
1530
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001531/*
1532 * Returns 0 for success, or a negative error value
1533 */
Mingming Cao617ba132006-10-11 01:20:53 -07001534static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001535 struct inode *inode)
1536{
1537 struct dx_frame frames[2], *frame;
1538 struct dx_entry *entries, *at;
1539 struct dx_hash_info hinfo;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001540 struct buffer_head *bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001541 struct inode *dir = dentry->d_parent->d_inode;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001542 struct super_block *sb = dir->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -07001543 struct ext4_dir_entry_2 *de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001544 int err;
1545
Theodore Ts'of702ba02008-09-22 15:21:01 -04001546 frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001547 if (!frame)
1548 return err;
1549 entries = frame->entries;
1550 at = frame->at;
1551
Mingming Cao617ba132006-10-11 01:20:53 -07001552 if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001553 goto cleanup;
1554
1555 BUFFER_TRACE(bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001556 err = ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001557 if (err)
1558 goto journal_error;
1559
1560 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001561 if (err != -ENOSPC)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001562 goto cleanup;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001563
1564 /* Block full, should compress but for now just split */
Theodore Ts'o4776004f2008-09-08 23:00:52 -04001565 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001566 dx_get_count(entries), dx_get_limit(entries)));
1567 /* Need to split index? */
1568 if (dx_get_count(entries) == dx_get_limit(entries)) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001569 ext4_lblk_t newblock;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001570 unsigned icount = dx_get_count(entries);
1571 int levels = frame - frames;
1572 struct dx_entry *entries2;
1573 struct dx_node *node2;
1574 struct buffer_head *bh2;
1575
1576 if (levels && (dx_get_count(frames->entries) ==
1577 dx_get_limit(frames->entries))) {
Harvey Harrison46e665e2008-04-17 10:38:59 -04001578 ext4_warning(sb, __func__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001579 "Directory index full!");
1580 err = -ENOSPC;
1581 goto cleanup;
1582 }
Mingming Cao617ba132006-10-11 01:20:53 -07001583 bh2 = ext4_append (handle, dir, &newblock, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001584 if (!(bh2))
1585 goto cleanup;
1586 node2 = (struct dx_node *)(bh2->b_data);
1587 entries2 = node2->entries;
Andreas Schlick1f7bebb2009-09-10 23:16:07 -04001588 memset(&node2->fake, 0, sizeof(struct fake_dirent));
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001589 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
1590 sb->s_blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001591 BUFFER_TRACE(frame->bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001592 err = ext4_journal_get_write_access(handle, frame->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001593 if (err)
1594 goto journal_error;
1595 if (levels) {
1596 unsigned icount1 = icount/2, icount2 = icount - icount1;
1597 unsigned hash2 = dx_get_hash(entries + icount1);
Theodore Ts'o4776004f2008-09-08 23:00:52 -04001598 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
1599 icount1, icount2));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001600
1601 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
Mingming Cao617ba132006-10-11 01:20:53 -07001602 err = ext4_journal_get_write_access(handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001603 frames[0].bh);
1604 if (err)
1605 goto journal_error;
1606
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001607 memcpy((char *) entries2, (char *) (entries + icount1),
1608 icount2 * sizeof(struct dx_entry));
1609 dx_set_count(entries, icount1);
1610 dx_set_count(entries2, icount2);
1611 dx_set_limit(entries2, dx_node_limit(dir));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001612
1613 /* Which index block gets the new entry? */
1614 if (at - entries >= icount1) {
1615 frame->at = at = at - entries - icount1 + entries2;
1616 frame->entries = entries = entries2;
1617 swap(frame->bh, bh2);
1618 }
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001619 dx_insert_block(frames + 0, hash2, newblock);
1620 dxtrace(dx_show_index("node", frames[1].entries));
1621 dxtrace(dx_show_index("node",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001622 ((struct dx_node *) bh2->b_data)->entries));
Frank Mayhar03901312009-01-07 00:06:22 -05001623 err = ext4_handle_dirty_metadata(handle, inode, bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001624 if (err)
1625 goto journal_error;
1626 brelse (bh2);
1627 } else {
Theodore Ts'o4776004f2008-09-08 23:00:52 -04001628 dxtrace(printk(KERN_DEBUG
1629 "Creating second level index...\n"));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001630 memcpy((char *) entries2, (char *) entries,
1631 icount * sizeof(struct dx_entry));
1632 dx_set_limit(entries2, dx_node_limit(dir));
1633
1634 /* Set up root */
1635 dx_set_count(entries, 1);
1636 dx_set_block(entries + 0, newblock);
1637 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1638
1639 /* Add new access path frame */
1640 frame = frames + 1;
1641 frame->at = at = at - entries + entries2;
1642 frame->entries = entries = entries2;
1643 frame->bh = bh2;
Mingming Cao617ba132006-10-11 01:20:53 -07001644 err = ext4_journal_get_write_access(handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001645 frame->bh);
1646 if (err)
1647 goto journal_error;
1648 }
Frank Mayhar03901312009-01-07 00:06:22 -05001649 ext4_handle_dirty_metadata(handle, inode, frames[0].bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001650 }
1651 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1652 if (!de)
1653 goto cleanup;
1654 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001655 goto cleanup;
1656
1657journal_error:
Mingming Cao617ba132006-10-11 01:20:53 -07001658 ext4_std_error(dir->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001659cleanup:
1660 if (bh)
1661 brelse(bh);
1662 dx_release(frames);
1663 return err;
1664}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001665
1666/*
Mingming Cao617ba132006-10-11 01:20:53 -07001667 * ext4_delete_entry deletes a directory entry by merging it with the
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001668 * previous entry
1669 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001670static int ext4_delete_entry(handle_t *handle,
1671 struct inode *dir,
1672 struct ext4_dir_entry_2 *de_del,
1673 struct buffer_head *bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001674{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001675 struct ext4_dir_entry_2 *de, *pde;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001676 unsigned int blocksize = dir->i_sb->s_blocksize;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001677 int i;
1678
1679 i = 0;
1680 pde = NULL;
Mingming Cao617ba132006-10-11 01:20:53 -07001681 de = (struct ext4_dir_entry_2 *) bh->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001682 while (i < bh->b_size) {
Mingming Cao617ba132006-10-11 01:20:53 -07001683 if (!ext4_check_dir_entry("ext4_delete_entry", dir, de, bh, i))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001684 return -EIO;
1685 if (de == de_del) {
1686 BUFFER_TRACE(bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001687 ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001688 if (pde)
Jan Karaa72d7f82008-01-28 23:58:27 -05001689 pde->rec_len = ext4_rec_len_to_disk(
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001690 ext4_rec_len_from_disk(pde->rec_len,
1691 blocksize) +
1692 ext4_rec_len_from_disk(de->rec_len,
1693 blocksize),
1694 blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001695 else
1696 de->inode = 0;
1697 dir->i_version++;
Frank Mayhar03901312009-01-07 00:06:22 -05001698 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
1699 ext4_handle_dirty_metadata(handle, dir, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001700 return 0;
1701 }
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001702 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001703 pde = de;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001704 de = ext4_next_entry(de, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001705 }
1706 return -ENOENT;
1707}
1708
Andreas Dilgerf8628a12007-07-18 08:38:01 -04001709/*
1710 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
1711 * since this indicates that nlinks count was previously 1.
1712 */
1713static void ext4_inc_count(handle_t *handle, struct inode *inode)
1714{
1715 inc_nlink(inode);
1716 if (is_dx(inode) && inode->i_nlink > 1) {
1717 /* limit is 16-bit i_links_count */
1718 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
1719 inode->i_nlink = 1;
1720 EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
1721 EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
1722 }
1723 }
1724}
1725
1726/*
1727 * If a directory had nlink == 1, then we should let it be 1. This indicates
1728 * directory has >EXT4_LINK_MAX subdirs.
1729 */
1730static void ext4_dec_count(handle_t *handle, struct inode *inode)
1731{
1732 drop_nlink(inode);
1733 if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
1734 inc_nlink(inode);
1735}
1736
1737
Mingming Cao617ba132006-10-11 01:20:53 -07001738static int ext4_add_nondir(handle_t *handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001739 struct dentry *dentry, struct inode *inode)
1740{
Mingming Cao617ba132006-10-11 01:20:53 -07001741 int err = ext4_add_entry(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001742 if (!err) {
Mingming Cao617ba132006-10-11 01:20:53 -07001743 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001744 d_instantiate(dentry, inode);
Al Viro6b38e842008-12-30 02:03:31 -05001745 unlock_new_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001746 return 0;
1747 }
Eric Sandeen731b9a52007-02-10 01:46:16 -08001748 drop_nlink(inode);
Al Viro6b38e842008-12-30 02:03:31 -05001749 unlock_new_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001750 iput(inode);
1751 return err;
1752}
1753
1754/*
1755 * By the time this is called, we already have created
1756 * the directory cache entry for the new file, but it
1757 * is so far negative - it has no inode.
1758 *
1759 * If the create succeeds, we fill in the inode information
1760 * with d_instantiate().
1761 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001762static int ext4_create(struct inode *dir, struct dentry *dentry, int mode,
1763 struct nameidata *nd)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001764{
1765 handle_t *handle;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001766 struct inode *inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001767 int err, retries = 0;
1768
1769retry:
Mingming Cao617ba132006-10-11 01:20:53 -07001770 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1771 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
Dmitry Monakhov5aca07e2009-12-08 22:42:15 -05001772 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001773 if (IS_ERR(handle))
1774 return PTR_ERR(handle);
1775
1776 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05001777 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001778
Andreas Dilger11013912009-06-13 11:45:35 -04001779 inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001780 err = PTR_ERR(inode);
1781 if (!IS_ERR(inode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07001782 inode->i_op = &ext4_file_inode_operations;
1783 inode->i_fop = &ext4_file_operations;
1784 ext4_set_aops(inode);
1785 err = ext4_add_nondir(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001786 }
Mingming Cao617ba132006-10-11 01:20:53 -07001787 ext4_journal_stop(handle);
1788 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001789 goto retry;
1790 return err;
1791}
1792
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001793static int ext4_mknod(struct inode *dir, struct dentry *dentry,
1794 int mode, dev_t rdev)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001795{
1796 handle_t *handle;
1797 struct inode *inode;
1798 int err, retries = 0;
1799
1800 if (!new_valid_dev(rdev))
1801 return -EINVAL;
1802
1803retry:
Mingming Cao617ba132006-10-11 01:20:53 -07001804 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1805 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
Dmitry Monakhov5aca07e2009-12-08 22:42:15 -05001806 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001807 if (IS_ERR(handle))
1808 return PTR_ERR(handle);
1809
1810 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05001811 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001812
Andreas Dilger11013912009-06-13 11:45:35 -04001813 inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001814 err = PTR_ERR(inode);
1815 if (!IS_ERR(inode)) {
1816 init_special_inode(inode, inode->i_mode, rdev);
Theodore Ts'o03010a32008-10-10 20:02:48 -04001817#ifdef CONFIG_EXT4_FS_XATTR
Mingming Cao617ba132006-10-11 01:20:53 -07001818 inode->i_op = &ext4_special_inode_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001819#endif
Mingming Cao617ba132006-10-11 01:20:53 -07001820 err = ext4_add_nondir(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001821 }
Mingming Cao617ba132006-10-11 01:20:53 -07001822 ext4_journal_stop(handle);
1823 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001824 goto retry;
1825 return err;
1826}
1827
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001828static int ext4_mkdir(struct inode *dir, struct dentry *dentry, int mode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001829{
1830 handle_t *handle;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001831 struct inode *inode;
1832 struct buffer_head *dir_block;
1833 struct ext4_dir_entry_2 *de;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001834 unsigned int blocksize = dir->i_sb->s_blocksize;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001835 int err, retries = 0;
1836
Andreas Dilgerf8628a12007-07-18 08:38:01 -04001837 if (EXT4_DIR_LINK_MAX(dir))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001838 return -EMLINK;
1839
1840retry:
Mingming Cao617ba132006-10-11 01:20:53 -07001841 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1842 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
Dmitry Monakhov5aca07e2009-12-08 22:42:15 -05001843 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001844 if (IS_ERR(handle))
1845 return PTR_ERR(handle);
1846
1847 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05001848 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001849
Andreas Dilger11013912009-06-13 11:45:35 -04001850 inode = ext4_new_inode(handle, dir, S_IFDIR | mode,
1851 &dentry->d_name, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001852 err = PTR_ERR(inode);
1853 if (IS_ERR(inode))
1854 goto out_stop;
1855
Mingming Cao617ba132006-10-11 01:20:53 -07001856 inode->i_op = &ext4_dir_inode_operations;
1857 inode->i_fop = &ext4_dir_operations;
1858 inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001859 dir_block = ext4_bread(handle, inode, 0, 1, &err);
Aneesh Kumar K.V4cdeed82008-02-22 06:17:31 -05001860 if (!dir_block)
1861 goto out_clear_inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001862 BUFFER_TRACE(dir_block, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001863 ext4_journal_get_write_access(handle, dir_block);
1864 de = (struct ext4_dir_entry_2 *) dir_block->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001865 de->inode = cpu_to_le32(inode->i_ino);
1866 de->name_len = 1;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001867 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
1868 blocksize);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001869 strcpy(de->name, ".");
Mingming Cao617ba132006-10-11 01:20:53 -07001870 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001871 de = ext4_next_entry(de, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001872 de->inode = cpu_to_le32(dir->i_ino);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001873 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(1),
1874 blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001875 de->name_len = 2;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001876 strcpy(de->name, "..");
Mingming Cao617ba132006-10-11 01:20:53 -07001877 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001878 inode->i_nlink = 2;
Frank Mayhar03901312009-01-07 00:06:22 -05001879 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
1880 ext4_handle_dirty_metadata(handle, dir, dir_block);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001881 brelse(dir_block);
Mingming Cao617ba132006-10-11 01:20:53 -07001882 ext4_mark_inode_dirty(handle, inode);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001883 err = ext4_add_entry(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001884 if (err) {
Aneesh Kumar K.V4cdeed82008-02-22 06:17:31 -05001885out_clear_inode:
1886 clear_nlink(inode);
Al Viro6b38e842008-12-30 02:03:31 -05001887 unlock_new_inode(inode);
Mingming Cao617ba132006-10-11 01:20:53 -07001888 ext4_mark_inode_dirty(handle, inode);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001889 iput(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001890 goto out_stop;
1891 }
Andreas Dilgerf8628a12007-07-18 08:38:01 -04001892 ext4_inc_count(handle, dir);
Mingming Cao617ba132006-10-11 01:20:53 -07001893 ext4_update_dx_flag(dir);
1894 ext4_mark_inode_dirty(handle, dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001895 d_instantiate(dentry, inode);
Al Viro6b38e842008-12-30 02:03:31 -05001896 unlock_new_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001897out_stop:
Mingming Cao617ba132006-10-11 01:20:53 -07001898 ext4_journal_stop(handle);
1899 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001900 goto retry;
1901 return err;
1902}
1903
1904/*
1905 * routine to check that the specified directory is empty (for rmdir)
1906 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001907static int empty_dir(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001908{
Theodore Ts'o498e5f22008-11-05 00:14:04 -05001909 unsigned int offset;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001910 struct buffer_head *bh;
1911 struct ext4_dir_entry_2 *de, *de1;
1912 struct super_block *sb;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001913 int err = 0;
1914
1915 sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -07001916 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001917 !(bh = ext4_bread(NULL, inode, 0, 0, &err))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001918 if (err)
Harvey Harrison46e665e2008-04-17 10:38:59 -04001919 ext4_error(inode->i_sb, __func__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001920 "error %d reading directory #%lu offset 0",
1921 err, inode->i_ino);
1922 else
Harvey Harrison46e665e2008-04-17 10:38:59 -04001923 ext4_warning(inode->i_sb, __func__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001924 "bad directory (dir #%lu) - no data block",
1925 inode->i_ino);
1926 return 1;
1927 }
Mingming Cao617ba132006-10-11 01:20:53 -07001928 de = (struct ext4_dir_entry_2 *) bh->b_data;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001929 de1 = ext4_next_entry(de, sb->s_blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001930 if (le32_to_cpu(de->inode) != inode->i_ino ||
1931 !le32_to_cpu(de1->inode) ||
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001932 strcmp(".", de->name) ||
1933 strcmp("..", de1->name)) {
1934 ext4_warning(inode->i_sb, "empty_dir",
1935 "bad directory (dir #%lu) - no `.' or `..'",
1936 inode->i_ino);
1937 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001938 return 1;
1939 }
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001940 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
1941 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
1942 de = ext4_next_entry(de1, sb->s_blocksize);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001943 while (offset < inode->i_size) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001944 if (!bh ||
1945 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1946 err = 0;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001947 brelse(bh);
1948 bh = ext4_bread(NULL, inode,
Mingming Cao617ba132006-10-11 01:20:53 -07001949 offset >> EXT4_BLOCK_SIZE_BITS(sb), 0, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001950 if (!bh) {
1951 if (err)
Harvey Harrison46e665e2008-04-17 10:38:59 -04001952 ext4_error(sb, __func__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001953 "error %d reading directory"
Theodore Ts'o498e5f22008-11-05 00:14:04 -05001954 " #%lu offset %u",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001955 err, inode->i_ino, offset);
1956 offset += sb->s_blocksize;
1957 continue;
1958 }
Mingming Cao617ba132006-10-11 01:20:53 -07001959 de = (struct ext4_dir_entry_2 *) bh->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001960 }
Mingming Cao617ba132006-10-11 01:20:53 -07001961 if (!ext4_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1962 de = (struct ext4_dir_entry_2 *)(bh->b_data +
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001963 sb->s_blocksize);
1964 offset = (offset | (sb->s_blocksize - 1)) + 1;
1965 continue;
1966 }
1967 if (le32_to_cpu(de->inode)) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001968 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001969 return 0;
1970 }
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001971 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
1972 de = ext4_next_entry(de, sb->s_blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001973 }
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001974 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001975 return 1;
1976}
1977
Mingming Cao617ba132006-10-11 01:20:53 -07001978/* ext4_orphan_add() links an unlinked or truncated inode into a list of
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001979 * such inodes, starting at the superblock, in case we crash before the
1980 * file is closed/deleted, or in case the inode truncate spans multiple
1981 * transactions and the last transaction is not recovered after a crash.
1982 *
1983 * At filesystem recovery time, we walk this list deleting unlinked
Mingming Cao617ba132006-10-11 01:20:53 -07001984 * inodes and truncating linked inodes in ext4_orphan_cleanup().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001985 */
Mingming Cao617ba132006-10-11 01:20:53 -07001986int ext4_orphan_add(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001987{
1988 struct super_block *sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -07001989 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001990 int err = 0, rc;
1991
Frank Mayhar03901312009-01-07 00:06:22 -05001992 if (!ext4_handle_valid(handle))
1993 return 0;
1994
Theodore Ts'o3b9d4ed2009-04-25 22:54:04 -04001995 mutex_lock(&EXT4_SB(sb)->s_orphan_lock);
Mingming Cao617ba132006-10-11 01:20:53 -07001996 if (!list_empty(&EXT4_I(inode)->i_orphan))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001997 goto out_unlock;
1998
1999 /* Orphan handling is only valid for files with data blocks
2000 * being truncated, or files being unlinked. */
2001
2002 /* @@@ FIXME: Observation from aviro:
Mingming Cao617ba132006-10-11 01:20:53 -07002003 * I think I can trigger J_ASSERT in ext4_orphan_add(). We block
Theodore Ts'o3b9d4ed2009-04-25 22:54:04 -04002004 * here (on s_orphan_lock), so race with ext4_link() which might bump
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002005 * ->i_nlink. For, say it, character device. Not a regular file,
2006 * not a directory, not a symlink and ->i_nlink > 0.
Theodore Ts'o3b9d4ed2009-04-25 22:54:04 -04002007 *
2008 * tytso, 4/25/2009: I'm not sure how that could happen;
2009 * shouldn't the fs core protect us from these sort of
2010 * unlink()/link() races?
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002011 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002012 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2013 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002014
Mingming Cao617ba132006-10-11 01:20:53 -07002015 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
2016 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002017 if (err)
2018 goto out_unlock;
2019
Mingming Cao617ba132006-10-11 01:20:53 -07002020 err = ext4_reserve_inode_write(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002021 if (err)
2022 goto out_unlock;
2023
2024 /* Insert this inode at the head of the on-disk orphan list... */
Mingming Cao617ba132006-10-11 01:20:53 -07002025 NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
2026 EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
Frank Mayhar03901312009-01-07 00:06:22 -05002027 err = ext4_handle_dirty_metadata(handle, inode, EXT4_SB(sb)->s_sbh);
Mingming Cao617ba132006-10-11 01:20:53 -07002028 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002029 if (!err)
2030 err = rc;
2031
2032 /* Only add to the head of the in-memory list if all the
2033 * previous operations succeeded. If the orphan_add is going to
2034 * fail (possibly taking the journal offline), we can't risk
2035 * leaving the inode on the orphan list: stray orphan-list
2036 * entries can cause panics at unmount time.
2037 *
2038 * This is safe: on error we're going to ignore the orphan list
2039 * anyway on the next recovery. */
2040 if (!err)
Mingming Cao617ba132006-10-11 01:20:53 -07002041 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002042
2043 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2044 jbd_debug(4, "orphan inode %lu will point to %d\n",
2045 inode->i_ino, NEXT_ORPHAN(inode));
2046out_unlock:
Theodore Ts'o3b9d4ed2009-04-25 22:54:04 -04002047 mutex_unlock(&EXT4_SB(sb)->s_orphan_lock);
Mingming Cao617ba132006-10-11 01:20:53 -07002048 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002049 return err;
2050}
2051
2052/*
Mingming Cao617ba132006-10-11 01:20:53 -07002053 * ext4_orphan_del() removes an unlinked or truncated inode from the list
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002054 * of such inodes stored on disk, because it is finally being cleaned up.
2055 */
Mingming Cao617ba132006-10-11 01:20:53 -07002056int ext4_orphan_del(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002057{
2058 struct list_head *prev;
Mingming Cao617ba132006-10-11 01:20:53 -07002059 struct ext4_inode_info *ei = EXT4_I(inode);
2060 struct ext4_sb_info *sbi;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05002061 __u32 ino_next;
Mingming Cao617ba132006-10-11 01:20:53 -07002062 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002063 int err = 0;
2064
Curt Wohlgemuthd3d1faf2009-09-29 11:01:03 -04002065 /* ext4_handle_valid() assumes a valid handle_t pointer */
2066 if (handle && !ext4_handle_valid(handle))
Frank Mayhar03901312009-01-07 00:06:22 -05002067 return 0;
2068
Theodore Ts'o3b9d4ed2009-04-25 22:54:04 -04002069 mutex_lock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2070 if (list_empty(&ei->i_orphan))
2071 goto out;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002072
2073 ino_next = NEXT_ORPHAN(inode);
2074 prev = ei->i_orphan.prev;
Mingming Cao617ba132006-10-11 01:20:53 -07002075 sbi = EXT4_SB(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002076
2077 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2078
2079 list_del_init(&ei->i_orphan);
2080
2081 /* If we're on an error path, we may not have a valid
2082 * transaction handle with which to update the orphan list on
2083 * disk, but we still need to remove the inode from the linked
2084 * list in memory. */
Frank Mayhar03901312009-01-07 00:06:22 -05002085 if (sbi->s_journal && !handle)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002086 goto out;
2087
Mingming Cao617ba132006-10-11 01:20:53 -07002088 err = ext4_reserve_inode_write(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002089 if (err)
2090 goto out_err;
2091
2092 if (prev == &sbi->s_orphan) {
Theodore Ts'o498e5f22008-11-05 00:14:04 -05002093 jbd_debug(4, "superblock will point to %u\n", ino_next);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002094 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07002095 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002096 if (err)
2097 goto out_brelse;
2098 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
Frank Mayhar03901312009-01-07 00:06:22 -05002099 err = ext4_handle_dirty_metadata(handle, inode, sbi->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002100 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07002101 struct ext4_iloc iloc2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002102 struct inode *i_prev =
Mingming Cao617ba132006-10-11 01:20:53 -07002103 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002104
Theodore Ts'o498e5f22008-11-05 00:14:04 -05002105 jbd_debug(4, "orphan inode %lu will point to %u\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002106 i_prev->i_ino, ino_next);
Mingming Cao617ba132006-10-11 01:20:53 -07002107 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002108 if (err)
2109 goto out_brelse;
2110 NEXT_ORPHAN(i_prev) = ino_next;
Mingming Cao617ba132006-10-11 01:20:53 -07002111 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002112 }
2113 if (err)
2114 goto out_brelse;
2115 NEXT_ORPHAN(inode) = 0;
Mingming Cao617ba132006-10-11 01:20:53 -07002116 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002117
2118out_err:
Mingming Cao617ba132006-10-11 01:20:53 -07002119 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002120out:
Theodore Ts'o3b9d4ed2009-04-25 22:54:04 -04002121 mutex_unlock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002122 return err;
2123
2124out_brelse:
2125 brelse(iloc.bh);
2126 goto out_err;
2127}
2128
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002129static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002130{
2131 int retval;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002132 struct inode *inode;
2133 struct buffer_head *bh;
2134 struct ext4_dir_entry_2 *de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002135 handle_t *handle;
2136
2137 /* Initialize quotas before so that eventual writes go in
2138 * separate transaction */
Jan Karaa269eb12009-01-26 17:04:39 +01002139 vfs_dq_init(dentry->d_inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002140 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002141 if (IS_ERR(handle))
2142 return PTR_ERR(handle);
2143
2144 retval = -ENOENT;
Theodore Ts'of702ba02008-09-22 15:21:01 -04002145 bh = ext4_find_entry(dir, &dentry->d_name, &de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002146 if (!bh)
2147 goto end_rmdir;
2148
2149 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05002150 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002151
2152 inode = dentry->d_inode;
2153
2154 retval = -EIO;
2155 if (le32_to_cpu(de->inode) != inode->i_ino)
2156 goto end_rmdir;
2157
2158 retval = -ENOTEMPTY;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002159 if (!empty_dir(inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002160 goto end_rmdir;
2161
Mingming Cao617ba132006-10-11 01:20:53 -07002162 retval = ext4_delete_entry(handle, dir, de, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002163 if (retval)
2164 goto end_rmdir;
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002165 if (!EXT4_DIR_LINK_EMPTY(inode))
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002166 ext4_warning(inode->i_sb, "ext4_rmdir",
2167 "empty directory has too many links (%d)",
2168 inode->i_nlink);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002169 inode->i_version++;
2170 clear_nlink(inode);
2171 /* There's no need to set i_disksize: the fact that i_nlink is
2172 * zero will ensure that the right thing happens during any
2173 * recovery. */
2174 inode->i_size = 0;
Mingming Cao617ba132006-10-11 01:20:53 -07002175 ext4_orphan_add(handle, inode);
Kalpak Shahef7f3832007-07-18 09:15:20 -04002176 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002177 ext4_mark_inode_dirty(handle, inode);
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002178 ext4_dec_count(handle, dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002179 ext4_update_dx_flag(dir);
2180 ext4_mark_inode_dirty(handle, dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002181
2182end_rmdir:
Mingming Cao617ba132006-10-11 01:20:53 -07002183 ext4_journal_stop(handle);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002184 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002185 return retval;
2186}
2187
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002188static int ext4_unlink(struct inode *dir, struct dentry *dentry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002189{
2190 int retval;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002191 struct inode *inode;
2192 struct buffer_head *bh;
2193 struct ext4_dir_entry_2 *de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002194 handle_t *handle;
2195
2196 /* Initialize quotas before so that eventual writes go
2197 * in separate transaction */
Jan Karaa269eb12009-01-26 17:04:39 +01002198 vfs_dq_init(dentry->d_inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002199 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002200 if (IS_ERR(handle))
2201 return PTR_ERR(handle);
2202
2203 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05002204 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002205
2206 retval = -ENOENT;
Theodore Ts'of702ba02008-09-22 15:21:01 -04002207 bh = ext4_find_entry(dir, &dentry->d_name, &de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002208 if (!bh)
2209 goto end_unlink;
2210
2211 inode = dentry->d_inode;
2212
2213 retval = -EIO;
2214 if (le32_to_cpu(de->inode) != inode->i_ino)
2215 goto end_unlink;
2216
2217 if (!inode->i_nlink) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002218 ext4_warning(inode->i_sb, "ext4_unlink",
2219 "Deleting nonexistent file (%lu), %d",
2220 inode->i_ino, inode->i_nlink);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002221 inode->i_nlink = 1;
2222 }
Mingming Cao617ba132006-10-11 01:20:53 -07002223 retval = ext4_delete_entry(handle, dir, de, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002224 if (retval)
2225 goto end_unlink;
Kalpak Shahef7f3832007-07-18 09:15:20 -04002226 dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002227 ext4_update_dx_flag(dir);
2228 ext4_mark_inode_dirty(handle, dir);
Theodore Ts'o825f1482008-02-15 15:00:38 -05002229 drop_nlink(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002230 if (!inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07002231 ext4_orphan_add(handle, inode);
Kalpak Shahef7f3832007-07-18 09:15:20 -04002232 inode->i_ctime = ext4_current_time(inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002233 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002234 retval = 0;
2235
2236end_unlink:
Mingming Cao617ba132006-10-11 01:20:53 -07002237 ext4_journal_stop(handle);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002238 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002239 return retval;
2240}
2241
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002242static int ext4_symlink(struct inode *dir,
2243 struct dentry *dentry, const char *symname)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002244{
2245 handle_t *handle;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002246 struct inode *inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002247 int l, err, retries = 0;
2248
2249 l = strlen(symname)+1;
2250 if (l > dir->i_sb->s_blocksize)
2251 return -ENAMETOOLONG;
2252
2253retry:
Mingming Cao617ba132006-10-11 01:20:53 -07002254 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2255 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 5 +
Dmitry Monakhov5aca07e2009-12-08 22:42:15 -05002256 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002257 if (IS_ERR(handle))
2258 return PTR_ERR(handle);
2259
2260 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05002261 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002262
Andreas Dilger11013912009-06-13 11:45:35 -04002263 inode = ext4_new_inode(handle, dir, S_IFLNK|S_IRWXUGO,
2264 &dentry->d_name, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002265 err = PTR_ERR(inode);
2266 if (IS_ERR(inode))
2267 goto out_stop;
2268
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002269 if (l > sizeof(EXT4_I(inode)->i_data)) {
Mingming Cao617ba132006-10-11 01:20:53 -07002270 inode->i_op = &ext4_symlink_inode_operations;
2271 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002272 /*
Mingming Cao617ba132006-10-11 01:20:53 -07002273 * page_symlink() calls into ext4_prepare/commit_write.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002274 * We have a transaction open. All is sweetness. It also sets
2275 * i_size in generic_commit_write().
2276 */
Nick Piggin54566b22009-01-04 12:00:53 -08002277 err = __page_symlink(inode, symname, l, 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002278 if (err) {
Theodore Ts'o825f1482008-02-15 15:00:38 -05002279 clear_nlink(inode);
Al Viro6b38e842008-12-30 02:03:31 -05002280 unlock_new_inode(inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002281 ext4_mark_inode_dirty(handle, inode);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002282 iput(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002283 goto out_stop;
2284 }
2285 } else {
Aneesh Kumar K.Ve65187e2008-04-29 08:11:12 -04002286 /* clear the extent format for fast symlink */
2287 EXT4_I(inode)->i_flags &= ~EXT4_EXTENTS_FL;
Mingming Cao617ba132006-10-11 01:20:53 -07002288 inode->i_op = &ext4_fast_symlink_inode_operations;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002289 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002290 inode->i_size = l-1;
2291 }
Mingming Cao617ba132006-10-11 01:20:53 -07002292 EXT4_I(inode)->i_disksize = inode->i_size;
2293 err = ext4_add_nondir(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002294out_stop:
Mingming Cao617ba132006-10-11 01:20:53 -07002295 ext4_journal_stop(handle);
2296 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002297 goto retry;
2298 return err;
2299}
2300
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002301static int ext4_link(struct dentry *old_dentry,
2302 struct inode *dir, struct dentry *dentry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002303{
2304 handle_t *handle;
2305 struct inode *inode = old_dentry->d_inode;
2306 int err, retries = 0;
2307
Theodore Ts'ob05ab1d2009-08-29 21:08:08 -04002308 if (inode->i_nlink >= EXT4_LINK_MAX)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002309 return -EMLINK;
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002310
Eric Sandeen2988a772007-02-10 01:46:16 -08002311 /*
2312 * Return -ENOENT if we've raced with unlink and i_nlink is 0. Doing
2313 * otherwise has the potential to corrupt the orphan inode list.
2314 */
2315 if (inode->i_nlink == 0)
2316 return -ENOENT;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002317
2318retry:
Mingming Cao617ba132006-10-11 01:20:53 -07002319 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2320 EXT4_INDEX_EXTRA_TRANS_BLOCKS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002321 if (IS_ERR(handle))
2322 return PTR_ERR(handle);
2323
2324 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05002325 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002326
Kalpak Shahef7f3832007-07-18 09:15:20 -04002327 inode->i_ctime = ext4_current_time(inode);
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002328 ext4_inc_count(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002329 atomic_inc(&inode->i_count);
2330
Al Viro6b38e842008-12-30 02:03:31 -05002331 err = ext4_add_entry(handle, dentry, inode);
2332 if (!err) {
2333 ext4_mark_inode_dirty(handle, inode);
2334 d_instantiate(dentry, inode);
2335 } else {
2336 drop_nlink(inode);
2337 iput(inode);
2338 }
Mingming Cao617ba132006-10-11 01:20:53 -07002339 ext4_journal_stop(handle);
2340 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002341 goto retry;
2342 return err;
2343}
2344
Wei Yongjun3d0518f2009-02-14 23:01:36 -05002345#define PARENT_INO(buffer, size) \
2346 (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer), size)->inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002347
2348/*
2349 * Anybody can rename anything with this: the permission checks are left to the
2350 * higher-level routines.
2351 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002352static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
2353 struct inode *new_dir, struct dentry *new_dentry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002354{
2355 handle_t *handle;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002356 struct inode *old_inode, *new_inode;
2357 struct buffer_head *old_bh, *new_bh, *dir_bh;
2358 struct ext4_dir_entry_2 *old_de, *new_de;
Theodore Ts'o8750c6d2009-02-23 23:05:27 -05002359 int retval, force_da_alloc = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002360
2361 old_bh = new_bh = dir_bh = NULL;
2362
2363 /* Initialize quotas before so that eventual writes go
2364 * in separate transaction */
2365 if (new_dentry->d_inode)
Jan Karaa269eb12009-01-26 17:04:39 +01002366 vfs_dq_init(new_dentry->d_inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002367 handle = ext4_journal_start(old_dir, 2 *
2368 EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2369 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002370 if (IS_ERR(handle))
2371 return PTR_ERR(handle);
2372
2373 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
Frank Mayhar03901312009-01-07 00:06:22 -05002374 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002375
Theodore Ts'of702ba02008-09-22 15:21:01 -04002376 old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002377 /*
2378 * Check for inode number is _not_ due to possible IO errors.
2379 * We might rmdir the source, keep it as pwd of some process
2380 * and merrily kill the link to whatever was created under the
2381 * same name. Goodbye sticky bit ;-<
2382 */
2383 old_inode = old_dentry->d_inode;
2384 retval = -ENOENT;
2385 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2386 goto end_rename;
2387
2388 new_inode = new_dentry->d_inode;
Theodore Ts'of702ba02008-09-22 15:21:01 -04002389 new_bh = ext4_find_entry(new_dir, &new_dentry->d_name, &new_de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002390 if (new_bh) {
2391 if (!new_inode) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002392 brelse(new_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002393 new_bh = NULL;
2394 }
2395 }
2396 if (S_ISDIR(old_inode->i_mode)) {
2397 if (new_inode) {
2398 retval = -ENOTEMPTY;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002399 if (!empty_dir(new_inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002400 goto end_rename;
2401 }
2402 retval = -EIO;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002403 dir_bh = ext4_bread(handle, old_inode, 0, 0, &retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002404 if (!dir_bh)
2405 goto end_rename;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05002406 if (le32_to_cpu(PARENT_INO(dir_bh->b_data,
2407 old_dir->i_sb->s_blocksize)) != old_dir->i_ino)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002408 goto end_rename;
2409 retval = -EMLINK;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002410 if (!new_inode && new_dir != old_dir &&
Aneesh Kumar K.V2c94eb82009-08-28 21:43:15 -04002411 EXT4_DIR_LINK_MAX(new_dir))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002412 goto end_rename;
2413 }
2414 if (!new_bh) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002415 retval = ext4_add_entry(handle, new_dentry, old_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002416 if (retval)
2417 goto end_rename;
2418 } else {
2419 BUFFER_TRACE(new_bh, "get write access");
Mingming Cao617ba132006-10-11 01:20:53 -07002420 ext4_journal_get_write_access(handle, new_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002421 new_de->inode = cpu_to_le32(old_inode->i_ino);
Mingming Cao617ba132006-10-11 01:20:53 -07002422 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2423 EXT4_FEATURE_INCOMPAT_FILETYPE))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002424 new_de->file_type = old_de->file_type;
2425 new_dir->i_version++;
Jan Kara53b7e9f2008-04-29 22:02:11 -04002426 new_dir->i_ctime = new_dir->i_mtime =
2427 ext4_current_time(new_dir);
2428 ext4_mark_inode_dirty(handle, new_dir);
Frank Mayhar03901312009-01-07 00:06:22 -05002429 BUFFER_TRACE(new_bh, "call ext4_handle_dirty_metadata");
2430 ext4_handle_dirty_metadata(handle, new_dir, new_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002431 brelse(new_bh);
2432 new_bh = NULL;
2433 }
2434
2435 /*
2436 * Like most other Unix systems, set the ctime for inodes on a
2437 * rename.
2438 */
Kalpak Shahef7f3832007-07-18 09:15:20 -04002439 old_inode->i_ctime = ext4_current_time(old_inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002440 ext4_mark_inode_dirty(handle, old_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002441
2442 /*
2443 * ok, that's it
2444 */
2445 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2446 old_de->name_len != old_dentry->d_name.len ||
2447 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
Mingming Cao617ba132006-10-11 01:20:53 -07002448 (retval = ext4_delete_entry(handle, old_dir,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002449 old_de, old_bh)) == -ENOENT) {
2450 /* old_de could have moved from under us during htree split, so
2451 * make sure that we are deleting the right entry. We might
2452 * also be pointing to a stale entry in the unused part of
2453 * old_bh so just checking inum and the name isn't enough. */
2454 struct buffer_head *old_bh2;
Mingming Cao617ba132006-10-11 01:20:53 -07002455 struct ext4_dir_entry_2 *old_de2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002456
Theodore Ts'of702ba02008-09-22 15:21:01 -04002457 old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002458 if (old_bh2) {
Mingming Cao617ba132006-10-11 01:20:53 -07002459 retval = ext4_delete_entry(handle, old_dir,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002460 old_de2, old_bh2);
2461 brelse(old_bh2);
2462 }
2463 }
2464 if (retval) {
Mingming Cao617ba132006-10-11 01:20:53 -07002465 ext4_warning(old_dir->i_sb, "ext4_rename",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002466 "Deleting old file (%lu), %d, error=%d",
2467 old_dir->i_ino, old_dir->i_nlink, retval);
2468 }
2469
2470 if (new_inode) {
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002471 ext4_dec_count(handle, new_inode);
Kalpak Shahef7f3832007-07-18 09:15:20 -04002472 new_inode->i_ctime = ext4_current_time(new_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002473 }
Kalpak Shahef7f3832007-07-18 09:15:20 -04002474 old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002475 ext4_update_dx_flag(old_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002476 if (dir_bh) {
2477 BUFFER_TRACE(dir_bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07002478 ext4_journal_get_write_access(handle, dir_bh);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05002479 PARENT_INO(dir_bh->b_data, new_dir->i_sb->s_blocksize) =
2480 cpu_to_le32(new_dir->i_ino);
Frank Mayhar03901312009-01-07 00:06:22 -05002481 BUFFER_TRACE(dir_bh, "call ext4_handle_dirty_metadata");
2482 ext4_handle_dirty_metadata(handle, old_dir, dir_bh);
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002483 ext4_dec_count(handle, old_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002484 if (new_inode) {
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002485 /* checked empty_dir above, can't have another parent,
Theodore Ts'o825f1482008-02-15 15:00:38 -05002486 * ext4_dec_count() won't work for many-linked dirs */
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002487 new_inode->i_nlink = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002488 } else {
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002489 ext4_inc_count(handle, new_dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002490 ext4_update_dx_flag(new_dir);
2491 ext4_mark_inode_dirty(handle, new_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002492 }
2493 }
Mingming Cao617ba132006-10-11 01:20:53 -07002494 ext4_mark_inode_dirty(handle, old_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002495 if (new_inode) {
Mingming Cao617ba132006-10-11 01:20:53 -07002496 ext4_mark_inode_dirty(handle, new_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002497 if (!new_inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07002498 ext4_orphan_add(handle, new_inode);
Theodore Ts'oafd46722009-03-16 23:12:23 -04002499 if (!test_opt(new_dir->i_sb, NO_AUTO_DA_ALLOC))
2500 force_da_alloc = 1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002501 }
2502 retval = 0;
2503
2504end_rename:
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002505 brelse(dir_bh);
2506 brelse(old_bh);
2507 brelse(new_bh);
Mingming Cao617ba132006-10-11 01:20:53 -07002508 ext4_journal_stop(handle);
Theodore Ts'o8750c6d2009-02-23 23:05:27 -05002509 if (retval == 0 && force_da_alloc)
2510 ext4_alloc_da_blocks(old_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002511 return retval;
2512}
2513
2514/*
2515 * directories can handle most operations...
2516 */
Arjan van de Ven754661f2007-02-12 00:55:38 -08002517const struct inode_operations ext4_dir_inode_operations = {
Mingming Cao617ba132006-10-11 01:20:53 -07002518 .create = ext4_create,
2519 .lookup = ext4_lookup,
2520 .link = ext4_link,
2521 .unlink = ext4_unlink,
2522 .symlink = ext4_symlink,
2523 .mkdir = ext4_mkdir,
2524 .rmdir = ext4_rmdir,
2525 .mknod = ext4_mknod,
2526 .rename = ext4_rename,
2527 .setattr = ext4_setattr,
Theodore Ts'o03010a32008-10-10 20:02:48 -04002528#ifdef CONFIG_EXT4_FS_XATTR
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002529 .setxattr = generic_setxattr,
2530 .getxattr = generic_getxattr,
Mingming Cao617ba132006-10-11 01:20:53 -07002531 .listxattr = ext4_listxattr,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002532 .removexattr = generic_removexattr,
2533#endif
Linus Torvalds1d5ccd12009-08-28 12:12:24 -07002534 .check_acl = ext4_check_acl,
Aneesh Kumar K.Vabc87462009-05-02 22:54:32 -04002535 .fiemap = ext4_fiemap,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002536};
2537
Arjan van de Ven754661f2007-02-12 00:55:38 -08002538const struct inode_operations ext4_special_inode_operations = {
Mingming Cao617ba132006-10-11 01:20:53 -07002539 .setattr = ext4_setattr,
Theodore Ts'o03010a32008-10-10 20:02:48 -04002540#ifdef CONFIG_EXT4_FS_XATTR
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002541 .setxattr = generic_setxattr,
2542 .getxattr = generic_getxattr,
Mingming Cao617ba132006-10-11 01:20:53 -07002543 .listxattr = ext4_listxattr,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002544 .removexattr = generic_removexattr,
2545#endif
Linus Torvalds1d5ccd12009-08-28 12:12:24 -07002546 .check_acl = ext4_check_acl,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002547};