blob: 2a42cc04466f6858f55fea7e4a8beb2f1232319f [file] [log] [blame]
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001/*
Mingming Cao617ba132006-10-11 01:20:53 -07002 * linux/fs/ext4/namei.c
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003 *
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
8 *
9 * from
10 *
11 * linux/fs/minix/namei.c
12 *
13 * Copyright (C) 1991, 1992 Linus Torvalds
14 *
15 * Big-endian to little-endian byte-swapping/bitmaps by
16 * David S. Miller (davem@caip.rutgers.edu), 1995
17 * Directory entry file type support and forward compatibility hooks
18 * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
19 * Hash Tree Directory indexing (c)
20 * Daniel Phillips, 2001
21 * Hash Tree Directory indexing porting
22 * Christopher Li, 2002
23 * Hash Tree Directory indexing cleanup
24 * Theodore Ts'o, 2002
25 */
26
27#include <linux/fs.h>
28#include <linux/pagemap.h>
Mingming Caodab291a2006-10-11 01:21:01 -070029#include <linux/jbd2.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070030#include <linux/time.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070031#include <linux/fcntl.h>
32#include <linux/stat.h>
33#include <linux/string.h>
34#include <linux/quotaops.h>
35#include <linux/buffer_head.h>
36#include <linux/bio.h>
Christoph Hellwig3dcf5452008-04-29 18:13:32 -040037#include "ext4.h"
38#include "ext4_jbd2.h"
Dave Kleikampac27a0e2006-10-11 01:20:50 -070039
Dave Kleikampac27a0e2006-10-11 01:20:50 -070040#include "xattr.h"
41#include "acl.h"
42
Jiaying Zhang0562e0b2011-03-21 21:38:05 -040043#include <trace/events/ext4.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070044/*
45 * define how far ahead to read directories while searching them.
46 */
47#define NAMEI_RA_CHUNKS 2
48#define NAMEI_RA_BLOCKS 4
Dave Kleikamp8c55e202007-05-24 13:04:54 -040049#define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070050#define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b))
51
Mingming Cao617ba132006-10-11 01:20:53 -070052static struct buffer_head *ext4_append(handle_t *handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -070053 struct inode *inode,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -050054 ext4_lblk_t *block, int *err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -070055{
56 struct buffer_head *bh;
57
58 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
59
Akinobu Mitaa8716112008-04-17 10:38:59 -040060 bh = ext4_bread(handle, inode, *block, 1, err);
61 if (bh) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -070062 inode->i_size += inode->i_sb->s_blocksize;
Mingming Cao617ba132006-10-11 01:20:53 -070063 EXT4_I(inode)->i_disksize = inode->i_size;
Akinobu Mitaa8716112008-04-17 10:38:59 -040064 *err = ext4_journal_get_write_access(handle, bh);
65 if (*err) {
66 brelse(bh);
67 bh = NULL;
68 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -070069 }
70 return bh;
71}
72
73#ifndef assert
74#define assert(test) J_ASSERT(test)
75#endif
76
Dave Kleikampac27a0e2006-10-11 01:20:50 -070077#ifdef DX_DEBUG
78#define dxtrace(command) command
79#else
80#define dxtrace(command)
81#endif
82
83struct fake_dirent
84{
85 __le32 inode;
86 __le16 rec_len;
87 u8 name_len;
88 u8 file_type;
89};
90
91struct dx_countlimit
92{
93 __le16 limit;
94 __le16 count;
95};
96
97struct dx_entry
98{
99 __le32 hash;
100 __le32 block;
101};
102
103/*
104 * dx_root_info is laid out so that if it should somehow get overlaid by a
105 * dirent the two low bits of the hash version will be zero. Therefore, the
106 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
107 */
108
109struct dx_root
110{
111 struct fake_dirent dot;
112 char dot_name[4];
113 struct fake_dirent dotdot;
114 char dotdot_name[4];
115 struct dx_root_info
116 {
117 __le32 reserved_zero;
118 u8 hash_version;
119 u8 info_length; /* 8 */
120 u8 indirect_levels;
121 u8 unused_flags;
122 }
123 info;
124 struct dx_entry entries[0];
125};
126
127struct dx_node
128{
129 struct fake_dirent fake;
130 struct dx_entry entries[0];
131};
132
133
134struct dx_frame
135{
136 struct buffer_head *bh;
137 struct dx_entry *entries;
138 struct dx_entry *at;
139};
140
141struct dx_map_entry
142{
143 u32 hash;
Eric Sandeenef2b02d2007-09-18 22:46:42 -0700144 u16 offs;
145 u16 size;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700146};
147
Darrick J. Wonge6153912012-04-29 18:23:10 -0400148/*
149 * This goes at the end of each htree block.
150 */
151struct dx_tail {
152 u32 dt_reserved;
153 __le32 dt_checksum; /* crc32c(uuid+inum+dirblock) */
154};
155
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500156static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
157static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400158static inline unsigned dx_get_hash(struct dx_entry *entry);
159static void dx_set_hash(struct dx_entry *entry, unsigned value);
160static unsigned dx_get_count(struct dx_entry *entries);
161static unsigned dx_get_limit(struct dx_entry *entries);
162static void dx_set_count(struct dx_entry *entries, unsigned value);
163static void dx_set_limit(struct dx_entry *entries, unsigned value);
164static unsigned dx_root_limit(struct inode *dir, unsigned infosize);
165static unsigned dx_node_limit(struct inode *dir);
Theodore Ts'of702ba02008-09-22 15:21:01 -0400166static struct dx_frame *dx_probe(const struct qstr *d_name,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700167 struct inode *dir,
168 struct dx_hash_info *hinfo,
169 struct dx_frame *frame,
170 int *err);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400171static void dx_release(struct dx_frame *frames);
Theodore Ts'o8bad4592009-02-14 21:46:54 -0500172static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400173 struct dx_hash_info *hinfo, struct dx_map_entry map[]);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700174static void dx_sort_map(struct dx_map_entry *map, unsigned count);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400175static struct ext4_dir_entry_2 *dx_move_dirents(char *from, char *to,
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500176 struct dx_map_entry *offsets, int count, unsigned blocksize);
Theodore Ts'o8bad4592009-02-14 21:46:54 -0500177static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500178static void dx_insert_block(struct dx_frame *frame,
179 u32 hash, ext4_lblk_t block);
Mingming Cao617ba132006-10-11 01:20:53 -0700180static int ext4_htree_next_block(struct inode *dir, __u32 hash,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700181 struct dx_frame *frame,
182 struct dx_frame *frames,
183 __u32 *start_hash);
Theodore Ts'of702ba02008-09-22 15:21:01 -0400184static struct buffer_head * ext4_dx_find_entry(struct inode *dir,
185 const struct qstr *d_name,
186 struct ext4_dir_entry_2 **res_dir,
187 int *err);
Mingming Cao617ba132006-10-11 01:20:53 -0700188static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700189 struct inode *inode);
190
Darrick J. Wongdbe89442012-04-29 18:39:10 -0400191/* checksumming functions */
Darrick J. Wongb0336e82012-04-29 18:41:10 -0400192#define EXT4_DIRENT_TAIL(block, blocksize) \
193 ((struct ext4_dir_entry_tail *)(((void *)(block)) + \
194 ((blocksize) - \
195 sizeof(struct ext4_dir_entry_tail))))
196
197static void initialize_dirent_tail(struct ext4_dir_entry_tail *t,
198 unsigned int blocksize)
199{
200 memset(t, 0, sizeof(struct ext4_dir_entry_tail));
201 t->det_rec_len = ext4_rec_len_to_disk(
202 sizeof(struct ext4_dir_entry_tail), blocksize);
203 t->det_reserved_ft = EXT4_FT_DIR_CSUM;
204}
205
206/* Walk through a dirent block to find a checksum "dirent" at the tail */
207static struct ext4_dir_entry_tail *get_dirent_tail(struct inode *inode,
208 struct ext4_dir_entry *de)
209{
210 struct ext4_dir_entry_tail *t;
211
212#ifdef PARANOID
213 struct ext4_dir_entry *d, *top;
214
215 d = de;
216 top = (struct ext4_dir_entry *)(((void *)de) +
217 (EXT4_BLOCK_SIZE(inode->i_sb) -
218 sizeof(struct ext4_dir_entry_tail)));
219 while (d < top && d->rec_len)
220 d = (struct ext4_dir_entry *)(((void *)d) +
221 le16_to_cpu(d->rec_len));
222
223 if (d != top)
224 return NULL;
225
226 t = (struct ext4_dir_entry_tail *)d;
227#else
228 t = EXT4_DIRENT_TAIL(de, EXT4_BLOCK_SIZE(inode->i_sb));
229#endif
230
231 if (t->det_reserved_zero1 ||
232 le16_to_cpu(t->det_rec_len) != sizeof(struct ext4_dir_entry_tail) ||
233 t->det_reserved_zero2 ||
234 t->det_reserved_ft != EXT4_FT_DIR_CSUM)
235 return NULL;
236
237 return t;
238}
239
240static __le32 ext4_dirent_csum(struct inode *inode,
241 struct ext4_dir_entry *dirent, int size)
242{
243 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
244 struct ext4_inode_info *ei = EXT4_I(inode);
245 __u32 csum;
246
247 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
248 return cpu_to_le32(csum);
249}
250
251int ext4_dirent_csum_verify(struct inode *inode, struct ext4_dir_entry *dirent)
252{
253 struct ext4_dir_entry_tail *t;
254
255 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
256 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
257 return 1;
258
259 t = get_dirent_tail(inode, dirent);
260 if (!t) {
261 EXT4_ERROR_INODE(inode, "metadata_csum set but no space in dir "
262 "leaf for checksum. Please run e2fsck -D.");
263 return 0;
264 }
265
266 if (t->det_checksum != ext4_dirent_csum(inode, dirent,
267 (void *)t - (void *)dirent))
268 return 0;
269
270 return 1;
271}
272
273static void ext4_dirent_csum_set(struct inode *inode,
274 struct ext4_dir_entry *dirent)
275{
276 struct ext4_dir_entry_tail *t;
277
278 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
279 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
280 return;
281
282 t = get_dirent_tail(inode, dirent);
283 if (!t) {
284 EXT4_ERROR_INODE(inode, "metadata_csum set but no space in dir "
285 "leaf for checksum. Please run e2fsck -D.");
286 return;
287 }
288
289 t->det_checksum = ext4_dirent_csum(inode, dirent,
290 (void *)t - (void *)dirent);
291}
292
293static inline int ext4_handle_dirty_dirent_node(handle_t *handle,
294 struct inode *inode,
295 struct buffer_head *bh)
296{
297 ext4_dirent_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
298 return ext4_handle_dirty_metadata(handle, inode, bh);
299}
300
Darrick J. Wongdbe89442012-04-29 18:39:10 -0400301static struct dx_countlimit *get_dx_countlimit(struct inode *inode,
302 struct ext4_dir_entry *dirent,
303 int *offset)
304{
305 struct ext4_dir_entry *dp;
306 struct dx_root_info *root;
307 int count_offset;
308
309 if (le16_to_cpu(dirent->rec_len) == EXT4_BLOCK_SIZE(inode->i_sb))
310 count_offset = 8;
311 else if (le16_to_cpu(dirent->rec_len) == 12) {
312 dp = (struct ext4_dir_entry *)(((void *)dirent) + 12);
313 if (le16_to_cpu(dp->rec_len) !=
314 EXT4_BLOCK_SIZE(inode->i_sb) - 12)
315 return NULL;
316 root = (struct dx_root_info *)(((void *)dp + 12));
317 if (root->reserved_zero ||
318 root->info_length != sizeof(struct dx_root_info))
319 return NULL;
320 count_offset = 32;
321 } else
322 return NULL;
323
324 if (offset)
325 *offset = count_offset;
326 return (struct dx_countlimit *)(((void *)dirent) + count_offset);
327}
328
329static __le32 ext4_dx_csum(struct inode *inode, struct ext4_dir_entry *dirent,
330 int count_offset, int count, struct dx_tail *t)
331{
332 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
333 struct ext4_inode_info *ei = EXT4_I(inode);
334 __u32 csum, old_csum;
335 int size;
336
337 size = count_offset + (count * sizeof(struct dx_entry));
338 old_csum = t->dt_checksum;
339 t->dt_checksum = 0;
340 csum = ext4_chksum(sbi, ei->i_csum_seed, (__u8 *)dirent, size);
341 csum = ext4_chksum(sbi, csum, (__u8 *)t, sizeof(struct dx_tail));
342 t->dt_checksum = old_csum;
343
344 return cpu_to_le32(csum);
345}
346
347static int ext4_dx_csum_verify(struct inode *inode,
348 struct ext4_dir_entry *dirent)
349{
350 struct dx_countlimit *c;
351 struct dx_tail *t;
352 int count_offset, limit, count;
353
354 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
355 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
356 return 1;
357
358 c = get_dx_countlimit(inode, dirent, &count_offset);
359 if (!c) {
360 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
361 return 1;
362 }
363 limit = le16_to_cpu(c->limit);
364 count = le16_to_cpu(c->count);
365 if (count_offset + (limit * sizeof(struct dx_entry)) >
366 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
367 EXT4_ERROR_INODE(inode, "metadata_csum set but no space for "
368 "tree checksum found. Run e2fsck -D.");
369 return 1;
370 }
371 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
372
373 if (t->dt_checksum != ext4_dx_csum(inode, dirent, count_offset,
374 count, t))
375 return 0;
376 return 1;
377}
378
379static void ext4_dx_csum_set(struct inode *inode, struct ext4_dir_entry *dirent)
380{
381 struct dx_countlimit *c;
382 struct dx_tail *t;
383 int count_offset, limit, count;
384
385 if (!EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
386 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
387 return;
388
389 c = get_dx_countlimit(inode, dirent, &count_offset);
390 if (!c) {
391 EXT4_ERROR_INODE(inode, "dir seems corrupt? Run e2fsck -D.");
392 return;
393 }
394 limit = le16_to_cpu(c->limit);
395 count = le16_to_cpu(c->count);
396 if (count_offset + (limit * sizeof(struct dx_entry)) >
397 EXT4_BLOCK_SIZE(inode->i_sb) - sizeof(struct dx_tail)) {
398 EXT4_ERROR_INODE(inode, "metadata_csum set but no space for "
399 "tree checksum. Run e2fsck -D.");
400 return;
401 }
402 t = (struct dx_tail *)(((struct dx_entry *)c) + limit);
403
404 t->dt_checksum = ext4_dx_csum(inode, dirent, count_offset, count, t);
405}
406
407static inline int ext4_handle_dirty_dx_node(handle_t *handle,
408 struct inode *inode,
409 struct buffer_head *bh)
410{
411 ext4_dx_csum_set(inode, (struct ext4_dir_entry *)bh->b_data);
412 return ext4_handle_dirty_metadata(handle, inode, bh);
413}
414
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700415/*
Li Zefanf795e142008-07-11 19:27:31 -0400416 * p is at least 6 bytes before the end of page
417 */
418static inline struct ext4_dir_entry_2 *
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500419ext4_next_entry(struct ext4_dir_entry_2 *p, unsigned long blocksize)
Li Zefanf795e142008-07-11 19:27:31 -0400420{
421 return (struct ext4_dir_entry_2 *)((char *)p +
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500422 ext4_rec_len_from_disk(p->rec_len, blocksize));
Li Zefanf795e142008-07-11 19:27:31 -0400423}
424
425/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700426 * Future: use high four bits of block for coalesce-on-delete flags
427 * Mask them off for now.
428 */
429
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500430static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700431{
432 return le32_to_cpu(entry->block) & 0x00ffffff;
433}
434
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500435static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700436{
437 entry->block = cpu_to_le32(value);
438}
439
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400440static inline unsigned dx_get_hash(struct dx_entry *entry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700441{
442 return le32_to_cpu(entry->hash);
443}
444
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400445static inline void dx_set_hash(struct dx_entry *entry, unsigned value)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700446{
447 entry->hash = cpu_to_le32(value);
448}
449
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400450static inline unsigned dx_get_count(struct dx_entry *entries)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700451{
452 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
453}
454
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400455static inline unsigned dx_get_limit(struct dx_entry *entries)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700456{
457 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
458}
459
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400460static inline void dx_set_count(struct dx_entry *entries, unsigned value)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700461{
462 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
463}
464
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400465static inline void dx_set_limit(struct dx_entry *entries, unsigned value)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700466{
467 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
468}
469
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400470static inline unsigned dx_root_limit(struct inode *dir, unsigned infosize)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700471{
Mingming Cao617ba132006-10-11 01:20:53 -0700472 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
473 EXT4_DIR_REC_LEN(2) - infosize;
Darrick J. Wongdbe89442012-04-29 18:39:10 -0400474
475 if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
476 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
477 entry_space -= sizeof(struct dx_tail);
Li Zefand9c769b2008-07-11 19:27:31 -0400478 return entry_space / sizeof(struct dx_entry);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700479}
480
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400481static inline unsigned dx_node_limit(struct inode *dir)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700482{
Mingming Cao617ba132006-10-11 01:20:53 -0700483 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
Darrick J. Wongdbe89442012-04-29 18:39:10 -0400484
485 if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
486 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
487 entry_space -= sizeof(struct dx_tail);
Li Zefand9c769b2008-07-11 19:27:31 -0400488 return entry_space / sizeof(struct dx_entry);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700489}
490
491/*
492 * Debug
493 */
494#ifdef DX_DEBUG
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400495static void dx_show_index(char * label, struct dx_entry *entries)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700496{
Andrew Morton63f57932006-10-11 01:21:24 -0700497 int i, n = dx_get_count (entries);
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400498 printk(KERN_DEBUG "%s index ", label);
Andrew Morton63f57932006-10-11 01:21:24 -0700499 for (i = 0; i < n; i++) {
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400500 printk("%x->%lu ", i ? dx_get_hash(entries + i) :
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500501 0, (unsigned long)dx_get_block(entries + i));
Andrew Morton63f57932006-10-11 01:21:24 -0700502 }
503 printk("\n");
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700504}
505
506struct stats
507{
508 unsigned names;
509 unsigned space;
510 unsigned bcount;
511};
512
Mingming Cao617ba132006-10-11 01:20:53 -0700513static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700514 int size, int show_names)
515{
516 unsigned names = 0, space = 0;
517 char *base = (char *) de;
518 struct dx_hash_info h = *hinfo;
519
520 printk("names: ");
521 while ((char *) de < base + size)
522 {
523 if (de->inode)
524 {
525 if (show_names)
526 {
527 int len = de->name_len;
528 char *name = de->name;
529 while (len--) printk("%c", *name++);
Mingming Cao617ba132006-10-11 01:20:53 -0700530 ext4fs_dirhash(de->name, de->name_len, &h);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700531 printk(":%x.%u ", h.hash,
Bernd Schubert265c6a02011-07-16 19:41:23 -0400532 (unsigned) ((char *) de - base));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700533 }
Mingming Cao617ba132006-10-11 01:20:53 -0700534 space += EXT4_DIR_REC_LEN(de->name_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700535 names++;
536 }
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500537 de = ext4_next_entry(de, size);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700538 }
539 printk("(%i)\n", names);
540 return (struct stats) { names, space, 1 };
541}
542
543struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
544 struct dx_entry *entries, int levels)
545{
546 unsigned blocksize = dir->i_sb->s_blocksize;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400547 unsigned count = dx_get_count(entries), names = 0, space = 0, i;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700548 unsigned bcount = 0;
549 struct buffer_head *bh;
550 int err;
551 printk("%i indexed blocks...\n", count);
552 for (i = 0; i < count; i++, entries++)
553 {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500554 ext4_lblk_t block = dx_get_block(entries);
555 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700556 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
557 struct stats stats;
558 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
Mingming Cao617ba132006-10-11 01:20:53 -0700559 if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700560 stats = levels?
561 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
Mingming Cao617ba132006-10-11 01:20:53 -0700562 dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700563 names += stats.names;
564 space += stats.space;
565 bcount += stats.bcount;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400566 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700567 }
568 if (bcount)
Theodore Ts'o60e66792010-05-17 07:00:00 -0400569 printk(KERN_DEBUG "%snames %u, fullness %u (%u%%)\n",
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400570 levels ? "" : " ", names, space/bcount,
571 (space/bcount)*100/blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700572 return (struct stats) { names, space, bcount};
573}
574#endif /* DX_DEBUG */
575
576/*
577 * Probe for a directory leaf block to search.
578 *
579 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
580 * error in the directory index, and the caller should fall back to
581 * searching the directory normally. The callers of dx_probe **MUST**
582 * check for this error code, and make sure it never gets reflected
583 * back to userspace.
584 */
585static struct dx_frame *
Theodore Ts'of702ba02008-09-22 15:21:01 -0400586dx_probe(const struct qstr *d_name, struct inode *dir,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700587 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
588{
589 unsigned count, indirect;
590 struct dx_entry *at, *entries, *p, *q, *m;
591 struct dx_root *root;
592 struct buffer_head *bh;
593 struct dx_frame *frame = frame_in;
594 u32 hash;
595
596 frame->bh = NULL;
Mingming Cao617ba132006-10-11 01:20:53 -0700597 if (!(bh = ext4_bread (NULL,dir, 0, 0, err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700598 goto fail;
599 root = (struct dx_root *) bh->b_data;
600 if (root->info.hash_version != DX_HASH_TEA &&
601 root->info.hash_version != DX_HASH_HALF_MD4 &&
602 root->info.hash_version != DX_HASH_LEGACY) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500603 ext4_warning(dir->i_sb, "Unrecognised inode hash code %d",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700604 root->info.hash_version);
605 brelse(bh);
606 *err = ERR_BAD_DX_DIR;
607 goto fail;
608 }
609 hinfo->hash_version = root->info.hash_version;
Theodore Ts'of99b2582008-10-28 13:21:44 -0400610 if (hinfo->hash_version <= DX_HASH_TEA)
611 hinfo->hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
Mingming Cao617ba132006-10-11 01:20:53 -0700612 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
Theodore Ts'of702ba02008-09-22 15:21:01 -0400613 if (d_name)
614 ext4fs_dirhash(d_name->name, d_name->len, hinfo);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700615 hash = hinfo->hash;
616
617 if (root->info.unused_flags & 1) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500618 ext4_warning(dir->i_sb, "Unimplemented inode hash flags: %#06x",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700619 root->info.unused_flags);
620 brelse(bh);
621 *err = ERR_BAD_DX_DIR;
622 goto fail;
623 }
624
625 if ((indirect = root->info.indirect_levels) > 1) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500626 ext4_warning(dir->i_sb, "Unimplemented inode hash depth: %#06x",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700627 root->info.indirect_levels);
628 brelse(bh);
629 *err = ERR_BAD_DX_DIR;
630 goto fail;
631 }
632
Darrick J. Wongdbe89442012-04-29 18:39:10 -0400633 if (!buffer_verified(bh) &&
634 !ext4_dx_csum_verify(dir, (struct ext4_dir_entry *)bh->b_data)) {
635 ext4_warning(dir->i_sb, "Root failed checksum");
636 brelse(bh);
637 *err = ERR_BAD_DX_DIR;
638 goto fail;
639 }
640 set_buffer_verified(bh);
641
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700642 entries = (struct dx_entry *) (((char *)&root->info) +
643 root->info.info_length);
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700644
645 if (dx_get_limit(entries) != dx_root_limit(dir,
646 root->info.info_length)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500647 ext4_warning(dir->i_sb, "dx entry: limit != root limit");
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700648 brelse(bh);
649 *err = ERR_BAD_DX_DIR;
650 goto fail;
651 }
652
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400653 dxtrace(printk("Look up %x", hash));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700654 while (1)
655 {
656 count = dx_get_count(entries);
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700657 if (!count || count > dx_get_limit(entries)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500658 ext4_warning(dir->i_sb,
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700659 "dx entry: no count or count > limit");
660 brelse(bh);
661 *err = ERR_BAD_DX_DIR;
662 goto fail2;
663 }
664
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700665 p = entries + 1;
666 q = entries + count - 1;
667 while (p <= q)
668 {
669 m = p + (q - p)/2;
670 dxtrace(printk("."));
671 if (dx_get_hash(m) > hash)
672 q = m - 1;
673 else
674 p = m + 1;
675 }
676
677 if (0) // linear search cross check
678 {
679 unsigned n = count - 1;
680 at = entries;
681 while (n--)
682 {
683 dxtrace(printk(","));
684 if (dx_get_hash(++at) > hash)
685 {
686 at--;
687 break;
688 }
689 }
690 assert (at == p - 1);
691 }
692
693 at = p - 1;
694 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
695 frame->bh = bh;
696 frame->entries = entries;
697 frame->at = at;
698 if (!indirect--) return frame;
Mingming Cao617ba132006-10-11 01:20:53 -0700699 if (!(bh = ext4_bread (NULL,dir, dx_get_block(at), 0, err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700700 goto fail2;
701 at = entries = ((struct dx_node *) bh->b_data)->entries;
Darrick J. Wongdbe89442012-04-29 18:39:10 -0400702
703 if (!buffer_verified(bh) &&
704 !ext4_dx_csum_verify(dir,
705 (struct ext4_dir_entry *)bh->b_data)) {
706 ext4_warning(dir->i_sb, "Node failed checksum");
707 brelse(bh);
708 *err = ERR_BAD_DX_DIR;
709 goto fail;
710 }
711 set_buffer_verified(bh);
712
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700713 if (dx_get_limit(entries) != dx_node_limit (dir)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -0500714 ext4_warning(dir->i_sb,
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700715 "dx entry: limit != node limit");
716 brelse(bh);
717 *err = ERR_BAD_DX_DIR;
718 goto fail2;
719 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700720 frame++;
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700721 frame->bh = NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700722 }
723fail2:
724 while (frame >= frame_in) {
725 brelse(frame->bh);
726 frame--;
727 }
728fail:
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700729 if (*err == ERR_BAD_DX_DIR)
Eric Sandeen12062dd2010-02-15 14:19:27 -0500730 ext4_warning(dir->i_sb,
Zheng Liu9ee49302012-02-20 23:09:36 -0500731 "Corrupt dir inode %lu, running e2fsck is "
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700732 "recommended.", dir->i_ino);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700733 return NULL;
734}
735
736static void dx_release (struct dx_frame *frames)
737{
738 if (frames[0].bh == NULL)
739 return;
740
741 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
742 brelse(frames[1].bh);
743 brelse(frames[0].bh);
744}
745
746/*
747 * This function increments the frame pointer to search the next leaf
748 * block, and reads in the necessary intervening nodes if the search
749 * should be necessary. Whether or not the search is necessary is
750 * controlled by the hash parameter. If the hash value is even, then
751 * the search is only continued if the next block starts with that
752 * hash value. This is used if we are searching for a specific file.
753 *
754 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
755 *
756 * This function returns 1 if the caller should continue to search,
757 * or 0 if it should not. If there is an error reading one of the
758 * index blocks, it will a negative error code.
759 *
760 * If start_hash is non-null, it will be filled in with the starting
761 * hash of the next page.
762 */
Mingming Cao617ba132006-10-11 01:20:53 -0700763static int ext4_htree_next_block(struct inode *dir, __u32 hash,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700764 struct dx_frame *frame,
765 struct dx_frame *frames,
766 __u32 *start_hash)
767{
768 struct dx_frame *p;
769 struct buffer_head *bh;
770 int err, num_frames = 0;
771 __u32 bhash;
772
773 p = frame;
774 /*
775 * Find the next leaf page by incrementing the frame pointer.
776 * If we run out of entries in the interior node, loop around and
777 * increment pointer in the parent node. When we break out of
778 * this loop, num_frames indicates the number of interior
779 * nodes need to be read.
780 */
781 while (1) {
782 if (++(p->at) < p->entries + dx_get_count(p->entries))
783 break;
784 if (p == frames)
785 return 0;
786 num_frames++;
787 p--;
788 }
789
790 /*
791 * If the hash is 1, then continue only if the next page has a
792 * continuation hash of any value. This is used for readdir
793 * handling. Otherwise, check to see if the hash matches the
794 * desired contiuation hash. If it doesn't, return since
795 * there's no point to read in the successive index pages.
796 */
797 bhash = dx_get_hash(p->at);
798 if (start_hash)
799 *start_hash = bhash;
800 if ((hash & 1) == 0) {
801 if ((bhash & ~1) != hash)
802 return 0;
803 }
804 /*
805 * If the hash is HASH_NB_ALWAYS, we always go to the next
806 * block so no check is necessary
807 */
808 while (num_frames--) {
Mingming Cao617ba132006-10-11 01:20:53 -0700809 if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at),
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700810 0, &err)))
811 return err; /* Failure */
Darrick J. Wongdbe89442012-04-29 18:39:10 -0400812
813 if (!buffer_verified(bh) &&
814 !ext4_dx_csum_verify(dir,
815 (struct ext4_dir_entry *)bh->b_data)) {
816 ext4_warning(dir->i_sb, "Node failed checksum");
817 return -EIO;
818 }
819 set_buffer_verified(bh);
820
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700821 p++;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400822 brelse(p->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700823 p->bh = bh;
824 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
825 }
826 return 1;
827}
828
829
830/*
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700831 * This function fills a red-black tree with information from a
832 * directory block. It returns the number directory entries loaded
833 * into the tree. If there is an error it is returned in err.
834 */
835static int htree_dirblock_to_tree(struct file *dir_file,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500836 struct inode *dir, ext4_lblk_t block,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700837 struct dx_hash_info *hinfo,
838 __u32 start_hash, __u32 start_minor_hash)
839{
840 struct buffer_head *bh;
Mingming Cao617ba132006-10-11 01:20:53 -0700841 struct ext4_dir_entry_2 *de, *top;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700842 int err, count = 0;
843
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500844 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
845 (unsigned long)block));
Mingming Cao617ba132006-10-11 01:20:53 -0700846 if (!(bh = ext4_bread (NULL, dir, block, 0, &err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700847 return err;
848
Darrick J. Wongb0336e82012-04-29 18:41:10 -0400849 if (!buffer_verified(bh) &&
850 !ext4_dirent_csum_verify(dir, (struct ext4_dir_entry *)bh->b_data))
851 return -EIO;
852 set_buffer_verified(bh);
853
Mingming Cao617ba132006-10-11 01:20:53 -0700854 de = (struct ext4_dir_entry_2 *) bh->b_data;
855 top = (struct ext4_dir_entry_2 *) ((char *) de +
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700856 dir->i_sb->s_blocksize -
Mingming Cao617ba132006-10-11 01:20:53 -0700857 EXT4_DIR_REC_LEN(0));
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500858 for (; de < top; de = ext4_next_entry(de, dir->i_sb->s_blocksize)) {
Theodore Ts'of7c21172011-01-10 12:10:55 -0500859 if (ext4_check_dir_entry(dir, NULL, de, bh,
Theodore Ts'ocad3f002010-12-19 22:07:02 -0500860 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
861 + ((char *)de - bh->b_data))) {
Eric Sandeene6c40212006-12-06 20:36:28 -0800862 /* On error, skip the f_pos to the next block. */
863 dir_file->f_pos = (dir_file->f_pos |
864 (dir->i_sb->s_blocksize - 1)) + 1;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -0400865 brelse(bh);
Eric Sandeene6c40212006-12-06 20:36:28 -0800866 return count;
867 }
Mingming Cao617ba132006-10-11 01:20:53 -0700868 ext4fs_dirhash(de->name, de->name_len, hinfo);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700869 if ((hinfo->hash < start_hash) ||
870 ((hinfo->hash == start_hash) &&
871 (hinfo->minor_hash < start_minor_hash)))
872 continue;
873 if (de->inode == 0)
874 continue;
Mingming Cao617ba132006-10-11 01:20:53 -0700875 if ((err = ext4_htree_store_dirent(dir_file,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700876 hinfo->hash, hinfo->minor_hash, de)) != 0) {
877 brelse(bh);
878 return err;
879 }
880 count++;
881 }
882 brelse(bh);
883 return count;
884}
885
886
887/*
888 * This function fills a red-black tree with information from a
889 * directory. We start scanning the directory in hash order, starting
890 * at start_hash and start_minor_hash.
891 *
892 * This function returns the number of entries inserted into the tree,
893 * or a negative error code.
894 */
Mingming Cao617ba132006-10-11 01:20:53 -0700895int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700896 __u32 start_minor_hash, __u32 *next_hash)
897{
898 struct dx_hash_info hinfo;
Mingming Cao617ba132006-10-11 01:20:53 -0700899 struct ext4_dir_entry_2 *de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700900 struct dx_frame frames[2], *frame;
901 struct inode *dir;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500902 ext4_lblk_t block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700903 int count = 0;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500904 int ret, err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700905 __u32 hashval;
906
Theodore Ts'o60e66792010-05-17 07:00:00 -0400907 dxtrace(printk(KERN_DEBUG "In htree_fill_tree, start hash: %x:%x\n",
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400908 start_hash, start_minor_hash));
Josef "Jeff" Sipek9d549892006-12-08 02:36:38 -0800909 dir = dir_file->f_path.dentry->d_inode;
Dmitry Monakhov12e9b892010-05-16 22:00:00 -0400910 if (!(ext4_test_inode_flag(dir, EXT4_INODE_INDEX))) {
Mingming Cao617ba132006-10-11 01:20:53 -0700911 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
Theodore Ts'of99b2582008-10-28 13:21:44 -0400912 if (hinfo.hash_version <= DX_HASH_TEA)
913 hinfo.hash_version +=
914 EXT4_SB(dir->i_sb)->s_hash_unsigned;
Mingming Cao617ba132006-10-11 01:20:53 -0700915 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700916 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
917 start_hash, start_minor_hash);
918 *next_hash = ~0;
919 return count;
920 }
921 hinfo.hash = start_hash;
922 hinfo.minor_hash = 0;
Theodore Ts'of702ba02008-09-22 15:21:01 -0400923 frame = dx_probe(NULL, dir, &hinfo, frames, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700924 if (!frame)
925 return err;
926
927 /* Add '.' and '..' from the htree header */
928 if (!start_hash && !start_minor_hash) {
Mingming Cao617ba132006-10-11 01:20:53 -0700929 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
930 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700931 goto errout;
932 count++;
933 }
934 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
Mingming Cao617ba132006-10-11 01:20:53 -0700935 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
Wei Yongjun3d0518f2009-02-14 23:01:36 -0500936 de = ext4_next_entry(de, dir->i_sb->s_blocksize);
Mingming Cao617ba132006-10-11 01:20:53 -0700937 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700938 goto errout;
939 count++;
940 }
941
942 while (1) {
943 block = dx_get_block(frame->at);
944 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
945 start_hash, start_minor_hash);
946 if (ret < 0) {
947 err = ret;
948 goto errout;
949 }
950 count += ret;
951 hashval = ~0;
Mingming Cao617ba132006-10-11 01:20:53 -0700952 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700953 frame, frames, &hashval);
954 *next_hash = hashval;
955 if (ret < 0) {
956 err = ret;
957 goto errout;
958 }
959 /*
960 * Stop if: (a) there are no more entries, or
961 * (b) we have inserted at least one entry and the
962 * next hash value is not a continuation
963 */
964 if ((ret == 0) ||
965 (count && ((hashval & 1) == 0)))
966 break;
967 }
968 dx_release(frames);
Theodore Ts'o4776004f2008-09-08 23:00:52 -0400969 dxtrace(printk(KERN_DEBUG "Fill tree: returned %d entries, "
970 "next hash: %x\n", count, *next_hash));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700971 return count;
972errout:
973 dx_release(frames);
974 return (err);
975}
976
977
978/*
979 * Directory block splitting, compacting
980 */
981
Eric Sandeenef2b02d2007-09-18 22:46:42 -0700982/*
983 * Create map of hash values, offsets, and sizes, stored at end of block.
984 * Returns number of entries mapped.
985 */
Theodore Ts'o8bad4592009-02-14 21:46:54 -0500986static int dx_make_map(struct ext4_dir_entry_2 *de, unsigned blocksize,
987 struct dx_hash_info *hinfo,
988 struct dx_map_entry *map_tail)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700989{
990 int count = 0;
991 char *base = (char *) de;
992 struct dx_hash_info h = *hinfo;
993
Theodore Ts'o8bad4592009-02-14 21:46:54 -0500994 while ((char *) de < base + blocksize) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700995 if (de->name_len && de->inode) {
Mingming Cao617ba132006-10-11 01:20:53 -0700996 ext4fs_dirhash(de->name, de->name_len, &h);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700997 map_tail--;
998 map_tail->hash = h.hash;
Toshiyuki Okajima9aee2282009-06-08 12:41:35 -0400999 map_tail->offs = ((char *) de - base)>>2;
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001000 map_tail->size = le16_to_cpu(de->rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001001 count++;
1002 cond_resched();
1003 }
1004 /* XXX: do we need to check rec_len == 0 case? -Chris */
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001005 de = ext4_next_entry(de, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001006 }
1007 return count;
1008}
1009
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001010/* Sort map by hash value */
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001011static void dx_sort_map (struct dx_map_entry *map, unsigned count)
1012{
Andrew Morton63f57932006-10-11 01:21:24 -07001013 struct dx_map_entry *p, *q, *top = map + count - 1;
1014 int more;
1015 /* Combsort until bubble sort doesn't suck */
1016 while (count > 2) {
1017 count = count*10/13;
1018 if (count - 9 < 2) /* 9, 10 -> 11 */
1019 count = 11;
1020 for (p = top, q = p - count; q >= map; p--, q--)
1021 if (p->hash < q->hash)
1022 swap(*p, *q);
1023 }
1024 /* Garden variety bubble sort */
1025 do {
1026 more = 0;
1027 q = top;
1028 while (q-- > map) {
1029 if (q[1].hash >= q[0].hash)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001030 continue;
Andrew Morton63f57932006-10-11 01:21:24 -07001031 swap(*(q+1), *q);
1032 more = 1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001033 }
1034 } while(more);
1035}
1036
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001037static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001038{
1039 struct dx_entry *entries = frame->entries;
1040 struct dx_entry *old = frame->at, *new = old + 1;
1041 int count = dx_get_count(entries);
1042
1043 assert(count < dx_get_limit(entries));
1044 assert(old < entries + count);
1045 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
1046 dx_set_hash(new, hash);
1047 dx_set_block(new, block);
1048 dx_set_count(entries, count + 1);
1049}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001050
Mingming Cao617ba132006-10-11 01:20:53 -07001051static void ext4_update_dx_flag(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001052{
Mingming Cao617ba132006-10-11 01:20:53 -07001053 if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
1054 EXT4_FEATURE_COMPAT_DIR_INDEX))
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04001055 ext4_clear_inode_flag(inode, EXT4_INODE_INDEX);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001056}
1057
1058/*
Mingming Cao617ba132006-10-11 01:20:53 -07001059 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001060 *
Mingming Cao617ba132006-10-11 01:20:53 -07001061 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001062 * `de != NULL' is guaranteed by caller.
1063 */
Mingming Cao617ba132006-10-11 01:20:53 -07001064static inline int ext4_match (int len, const char * const name,
1065 struct ext4_dir_entry_2 * de)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001066{
1067 if (len != de->name_len)
1068 return 0;
1069 if (!de->inode)
1070 return 0;
1071 return !memcmp(name, de->name, len);
1072}
1073
1074/*
1075 * Returns 0 if not found, -1 on failure, and 1 on success
1076 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001077static inline int search_dirblock(struct buffer_head *bh,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001078 struct inode *dir,
Theodore Ts'of702ba02008-09-22 15:21:01 -04001079 const struct qstr *d_name,
Theodore Ts'o498e5f22008-11-05 00:14:04 -05001080 unsigned int offset,
Mingming Cao617ba132006-10-11 01:20:53 -07001081 struct ext4_dir_entry_2 ** res_dir)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001082{
Mingming Cao617ba132006-10-11 01:20:53 -07001083 struct ext4_dir_entry_2 * de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001084 char * dlimit;
1085 int de_len;
Theodore Ts'of702ba02008-09-22 15:21:01 -04001086 const char *name = d_name->name;
1087 int namelen = d_name->len;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001088
Mingming Cao617ba132006-10-11 01:20:53 -07001089 de = (struct ext4_dir_entry_2 *) bh->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001090 dlimit = bh->b_data + dir->i_sb->s_blocksize;
1091 while ((char *) de < dlimit) {
1092 /* this code is executed quadratically often */
1093 /* do minimal checking `by hand' */
1094
1095 if ((char *) de + namelen <= dlimit &&
Mingming Cao617ba132006-10-11 01:20:53 -07001096 ext4_match (namelen, name, de)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001097 /* found a match - just to be sure, do a full check */
Theodore Ts'of7c21172011-01-10 12:10:55 -05001098 if (ext4_check_dir_entry(dir, NULL, de, bh, offset))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001099 return -1;
1100 *res_dir = de;
1101 return 1;
1102 }
1103 /* prevent looping on a bad block */
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001104 de_len = ext4_rec_len_from_disk(de->rec_len,
1105 dir->i_sb->s_blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001106 if (de_len <= 0)
1107 return -1;
1108 offset += de_len;
Mingming Cao617ba132006-10-11 01:20:53 -07001109 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001110 }
1111 return 0;
1112}
1113
1114
1115/*
Mingming Cao617ba132006-10-11 01:20:53 -07001116 * ext4_find_entry()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001117 *
1118 * finds an entry in the specified directory with the wanted name. It
1119 * returns the cache buffer in which the entry was found, and the entry
1120 * itself (as a parameter - res_dir). It does NOT read the inode of the
1121 * entry - you'll have to do that yourself if you want to.
1122 *
1123 * The returned buffer_head has ->b_count elevated. The caller is expected
1124 * to brelse() it when appropriate.
1125 */
Theodore Ts'of702ba02008-09-22 15:21:01 -04001126static struct buffer_head * ext4_find_entry (struct inode *dir,
1127 const struct qstr *d_name,
Mingming Cao617ba132006-10-11 01:20:53 -07001128 struct ext4_dir_entry_2 ** res_dir)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001129{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001130 struct super_block *sb;
1131 struct buffer_head *bh_use[NAMEI_RA_SIZE];
1132 struct buffer_head *bh, *ret = NULL;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001133 ext4_lblk_t start, block, b;
Theodore Ts'o8941ec82010-10-27 21:30:08 -04001134 const u8 *name = d_name->name;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001135 int ra_max = 0; /* Number of bh's in the readahead
1136 buffer, bh_use[] */
1137 int ra_ptr = 0; /* Current index into readahead
1138 buffer */
1139 int num = 0;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001140 ext4_lblk_t nblocks;
1141 int i, err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001142 int namelen;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001143
1144 *res_dir = NULL;
1145 sb = dir->i_sb;
Theodore Ts'of702ba02008-09-22 15:21:01 -04001146 namelen = d_name->len;
Mingming Cao617ba132006-10-11 01:20:53 -07001147 if (namelen > EXT4_NAME_LEN)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001148 return NULL;
Theodore Ts'o8941ec82010-10-27 21:30:08 -04001149 if ((namelen <= 2) && (name[0] == '.') &&
Aaro Koskinen6d5c3aa2010-12-14 21:45:31 -05001150 (name[1] == '.' || name[1] == '\0')) {
Theodore Ts'o8941ec82010-10-27 21:30:08 -04001151 /*
1152 * "." or ".." will only be in the first block
1153 * NFS may look up ".."; "." should be handled by the VFS
1154 */
1155 block = start = 0;
1156 nblocks = 1;
1157 goto restart;
1158 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001159 if (is_dx(dir)) {
Theodore Ts'of702ba02008-09-22 15:21:01 -04001160 bh = ext4_dx_find_entry(dir, d_name, res_dir, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001161 /*
1162 * On success, or if the error was file not found,
1163 * return. Otherwise, fall back to doing a search the
1164 * old fashioned way.
1165 */
1166 if (bh || (err != ERR_BAD_DX_DIR))
1167 return bh;
Theodore Ts'o4776004f2008-09-08 23:00:52 -04001168 dxtrace(printk(KERN_DEBUG "ext4_find_entry: dx failed, "
1169 "falling back\n"));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001170 }
Mingming Cao617ba132006-10-11 01:20:53 -07001171 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
1172 start = EXT4_I(dir)->i_dir_start_lookup;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001173 if (start >= nblocks)
1174 start = 0;
1175 block = start;
1176restart:
1177 do {
1178 /*
1179 * We deal with the read-ahead logic here.
1180 */
1181 if (ra_ptr >= ra_max) {
1182 /* Refill the readahead buffer */
1183 ra_ptr = 0;
1184 b = block;
1185 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1186 /*
1187 * Terminate if we reach the end of the
1188 * directory and must wrap, or if our
1189 * search has finished at this block.
1190 */
1191 if (b >= nblocks || (num && block == start)) {
1192 bh_use[ra_max] = NULL;
1193 break;
1194 }
1195 num++;
Mingming Cao617ba132006-10-11 01:20:53 -07001196 bh = ext4_getblk(NULL, dir, b++, 0, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001197 bh_use[ra_max] = bh;
1198 if (bh)
Christoph Hellwig65299a32011-08-23 14:50:29 +02001199 ll_rw_block(READ | REQ_META | REQ_PRIO,
1200 1, &bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001201 }
1202 }
1203 if ((bh = bh_use[ra_ptr++]) == NULL)
1204 goto next;
1205 wait_on_buffer(bh);
1206 if (!buffer_uptodate(bh)) {
1207 /* read error, skip block & hope for the best */
Theodore Ts'o24676da2010-05-16 21:00:00 -04001208 EXT4_ERROR_INODE(dir, "reading directory lblock %lu",
1209 (unsigned long) block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001210 brelse(bh);
1211 goto next;
1212 }
Darrick J. Wongb0336e82012-04-29 18:41:10 -04001213 if (!buffer_verified(bh) &&
1214 !ext4_dirent_csum_verify(dir,
1215 (struct ext4_dir_entry *)bh->b_data)) {
1216 EXT4_ERROR_INODE(dir, "checksumming directory "
1217 "block %lu", (unsigned long)block);
1218 brelse(bh);
1219 goto next;
1220 }
1221 set_buffer_verified(bh);
Theodore Ts'of702ba02008-09-22 15:21:01 -04001222 i = search_dirblock(bh, dir, d_name,
Mingming Cao617ba132006-10-11 01:20:53 -07001223 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001224 if (i == 1) {
Mingming Cao617ba132006-10-11 01:20:53 -07001225 EXT4_I(dir)->i_dir_start_lookup = block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001226 ret = bh;
1227 goto cleanup_and_exit;
1228 } else {
1229 brelse(bh);
1230 if (i < 0)
1231 goto cleanup_and_exit;
1232 }
1233 next:
1234 if (++block >= nblocks)
1235 block = 0;
1236 } while (block != start);
1237
1238 /*
1239 * If the directory has grown while we were searching, then
1240 * search the last part of the directory before giving up.
1241 */
1242 block = nblocks;
Mingming Cao617ba132006-10-11 01:20:53 -07001243 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001244 if (block < nblocks) {
1245 start = 0;
1246 goto restart;
1247 }
1248
1249cleanup_and_exit:
1250 /* Clean up the read-ahead blocks */
1251 for (; ra_ptr < ra_max; ra_ptr++)
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001252 brelse(bh_use[ra_ptr]);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001253 return ret;
1254}
1255
Theodore Ts'of702ba02008-09-22 15:21:01 -04001256static struct buffer_head * ext4_dx_find_entry(struct inode *dir, const struct qstr *d_name,
Mingming Cao617ba132006-10-11 01:20:53 -07001257 struct ext4_dir_entry_2 **res_dir, int *err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001258{
Theodore Ts'o8941ec82010-10-27 21:30:08 -04001259 struct super_block * sb = dir->i_sb;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001260 struct dx_hash_info hinfo;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001261 struct dx_frame frames[2], *frame;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001262 struct buffer_head *bh;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001263 ext4_lblk_t block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001264 int retval;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001265
Theodore Ts'o8941ec82010-10-27 21:30:08 -04001266 if (!(frame = dx_probe(d_name, dir, &hinfo, frames, err)))
1267 return NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001268 do {
1269 block = dx_get_block(frame->at);
Theodore Ts'o7845c042010-10-27 21:30:08 -04001270 if (!(bh = ext4_bread(NULL, dir, block, 0, err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001271 goto errout;
Duane Griffinf3b35f02008-07-11 19:27:31 -04001272
Darrick J. Wongb0336e82012-04-29 18:41:10 -04001273 if (!buffer_verified(bh) &&
1274 !ext4_dirent_csum_verify(dir,
1275 (struct ext4_dir_entry *)bh->b_data)) {
1276 EXT4_ERROR_INODE(dir, "checksumming directory "
1277 "block %lu", (unsigned long)block);
1278 brelse(bh);
1279 *err = -EIO;
1280 goto errout;
1281 }
1282 set_buffer_verified(bh);
Theodore Ts'o7845c042010-10-27 21:30:08 -04001283 retval = search_dirblock(bh, dir, d_name,
1284 block << EXT4_BLOCK_SIZE_BITS(sb),
1285 res_dir);
1286 if (retval == 1) { /* Success! */
1287 dx_release(frames);
1288 return bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001289 }
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001290 brelse(bh);
Theodore Ts'o7845c042010-10-27 21:30:08 -04001291 if (retval == -1) {
1292 *err = ERR_BAD_DX_DIR;
1293 goto errout;
1294 }
1295
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001296 /* Check to see if we should continue to search */
Theodore Ts'o8941ec82010-10-27 21:30:08 -04001297 retval = ext4_htree_next_block(dir, hinfo.hash, frame,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001298 frames, NULL);
1299 if (retval < 0) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001300 ext4_warning(sb,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001301 "error reading index page in directory #%lu",
1302 dir->i_ino);
1303 *err = retval;
1304 goto errout;
1305 }
1306 } while (retval == 1);
1307
1308 *err = -ENOENT;
1309errout:
Bernd Schubert265c6a02011-07-16 19:41:23 -04001310 dxtrace(printk(KERN_DEBUG "%s not found\n", d_name->name));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001311 dx_release (frames);
1312 return NULL;
1313}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001314
Al Viro00cd8dd2012-06-10 17:13:09 -04001315static struct dentry *ext4_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001316{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001317 struct inode *inode;
1318 struct ext4_dir_entry_2 *de;
1319 struct buffer_head *bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001320
Mingming Cao617ba132006-10-11 01:20:53 -07001321 if (dentry->d_name.len > EXT4_NAME_LEN)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001322 return ERR_PTR(-ENAMETOOLONG);
1323
Theodore Ts'of702ba02008-09-22 15:21:01 -04001324 bh = ext4_find_entry(dir, &dentry->d_name, &de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001325 inode = NULL;
1326 if (bh) {
Theodore Ts'o498e5f22008-11-05 00:14:04 -05001327 __u32 ino = le32_to_cpu(de->inode);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001328 brelse(bh);
Mingming Cao617ba132006-10-11 01:20:53 -07001329 if (!ext4_valid_inum(dir->i_sb, ino)) {
Theodore Ts'o24676da2010-05-16 21:00:00 -04001330 EXT4_ERROR_INODE(dir, "bad inode number: %u", ino);
David Howells1d1fe1e2008-02-07 00:15:37 -08001331 return ERR_PTR(-EIO);
Vasily Averina6c15c22007-07-15 23:40:46 -07001332 }
Andreas Dilger7e936b72012-05-28 17:02:25 -04001333 if (unlikely(ino == dir->i_ino)) {
1334 EXT4_ERROR_INODE(dir, "'%.*s' linked to parent dir",
1335 dentry->d_name.len,
1336 dentry->d_name.name);
1337 return ERR_PTR(-EIO);
1338 }
David Howells1d1fe1e2008-02-07 00:15:37 -08001339 inode = ext4_iget(dir->i_sb, ino);
Al Viroa9049372011-07-08 21:20:11 -04001340 if (inode == ERR_PTR(-ESTALE)) {
1341 EXT4_ERROR_INODE(dir,
1342 "deleted inode referenced: %u",
1343 ino);
1344 return ERR_PTR(-EIO);
Bryan Donlane6f009b2009-02-22 21:20:25 -05001345 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001346 }
1347 return d_splice_alias(inode, dentry);
1348}
1349
1350
Mingming Cao617ba132006-10-11 01:20:53 -07001351struct dentry *ext4_get_parent(struct dentry *child)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001352{
Theodore Ts'o498e5f22008-11-05 00:14:04 -05001353 __u32 ino;
Linus Torvalds26fe5752012-05-10 13:14:12 -07001354 static const struct qstr dotdot = QSTR_INIT("..", 2);
Mingming Cao617ba132006-10-11 01:20:53 -07001355 struct ext4_dir_entry_2 * de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001356 struct buffer_head *bh;
1357
Theodore Ts'of702ba02008-09-22 15:21:01 -04001358 bh = ext4_find_entry(child->d_inode, &dotdot, &de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001359 if (!bh)
1360 return ERR_PTR(-ENOENT);
1361 ino = le32_to_cpu(de->inode);
1362 brelse(bh);
1363
Mingming Cao617ba132006-10-11 01:20:53 -07001364 if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
Theodore Ts'o24676da2010-05-16 21:00:00 -04001365 EXT4_ERROR_INODE(child->d_inode,
1366 "bad parent inode number: %u", ino);
David Howells1d1fe1e2008-02-07 00:15:37 -08001367 return ERR_PTR(-EIO);
Vasily Averina6c15c22007-07-15 23:40:46 -07001368 }
1369
Christoph Hellwig44003722008-08-11 15:49:04 +02001370 return d_obtain_alias(ext4_iget(child->d_inode->i_sb, ino));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001371}
1372
1373#define S_SHIFT 12
Mingming Cao617ba132006-10-11 01:20:53 -07001374static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1375 [S_IFREG >> S_SHIFT] = EXT4_FT_REG_FILE,
1376 [S_IFDIR >> S_SHIFT] = EXT4_FT_DIR,
1377 [S_IFCHR >> S_SHIFT] = EXT4_FT_CHRDEV,
1378 [S_IFBLK >> S_SHIFT] = EXT4_FT_BLKDEV,
1379 [S_IFIFO >> S_SHIFT] = EXT4_FT_FIFO,
1380 [S_IFSOCK >> S_SHIFT] = EXT4_FT_SOCK,
1381 [S_IFLNK >> S_SHIFT] = EXT4_FT_SYMLINK,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001382};
1383
Mingming Cao617ba132006-10-11 01:20:53 -07001384static inline void ext4_set_de_type(struct super_block *sb,
1385 struct ext4_dir_entry_2 *de,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001386 umode_t mode) {
Mingming Cao617ba132006-10-11 01:20:53 -07001387 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1388 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001389}
1390
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001391/*
1392 * Move count entries from end of map between two memory locations.
1393 * Returns pointer to last entry moved.
1394 */
Mingming Cao617ba132006-10-11 01:20:53 -07001395static struct ext4_dir_entry_2 *
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001396dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count,
1397 unsigned blocksize)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001398{
1399 unsigned rec_len = 0;
1400
1401 while (count--) {
Theodore Ts'o60e66792010-05-17 07:00:00 -04001402 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *)
Toshiyuki Okajima9aee2282009-06-08 12:41:35 -04001403 (from + (map->offs<<2));
Mingming Cao617ba132006-10-11 01:20:53 -07001404 rec_len = EXT4_DIR_REC_LEN(de->name_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001405 memcpy (to, de, rec_len);
Mingming Cao617ba132006-10-11 01:20:53 -07001406 ((struct ext4_dir_entry_2 *) to)->rec_len =
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001407 ext4_rec_len_to_disk(rec_len, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001408 de->inode = 0;
1409 map++;
1410 to += rec_len;
1411 }
Mingming Cao617ba132006-10-11 01:20:53 -07001412 return (struct ext4_dir_entry_2 *) (to - rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001413}
1414
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001415/*
1416 * Compact each dir entry in the range to the minimal rec_len.
1417 * Returns pointer to last entry in range.
1418 */
Theodore Ts'o8bad4592009-02-14 21:46:54 -05001419static struct ext4_dir_entry_2* dx_pack_dirents(char *base, unsigned blocksize)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001420{
Mingming Cao617ba132006-10-11 01:20:53 -07001421 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001422 unsigned rec_len = 0;
1423
1424 prev = to = de;
Theodore Ts'o8bad4592009-02-14 21:46:54 -05001425 while ((char*)de < base + blocksize) {
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001426 next = ext4_next_entry(de, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001427 if (de->inode && de->name_len) {
Mingming Cao617ba132006-10-11 01:20:53 -07001428 rec_len = EXT4_DIR_REC_LEN(de->name_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001429 if (de > to)
1430 memmove(to, de, rec_len);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001431 to->rec_len = ext4_rec_len_to_disk(rec_len, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001432 prev = to;
Mingming Cao617ba132006-10-11 01:20:53 -07001433 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001434 }
1435 de = next;
1436 }
1437 return prev;
1438}
1439
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001440/*
1441 * Split a full leaf block to make room for a new dir entry.
1442 * Allocate a new block, and move entries so that they are approx. equally full.
1443 * Returns pointer to de in block into which the new entry will be inserted.
1444 */
Mingming Cao617ba132006-10-11 01:20:53 -07001445static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001446 struct buffer_head **bh,struct dx_frame *frame,
1447 struct dx_hash_info *hinfo, int *error)
1448{
1449 unsigned blocksize = dir->i_sb->s_blocksize;
1450 unsigned count, continued;
1451 struct buffer_head *bh2;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001452 ext4_lblk_t newblock;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001453 u32 hash2;
1454 struct dx_map_entry *map;
1455 char *data1 = (*bh)->b_data, *data2;
Theodore Ts'o59e315b2008-12-06 16:58:39 -05001456 unsigned split, move, size;
Mingming Cao617ba132006-10-11 01:20:53 -07001457 struct ext4_dir_entry_2 *de = NULL, *de2;
Darrick J. Wongb0336e82012-04-29 18:41:10 -04001458 struct ext4_dir_entry_tail *t;
1459 int csum_size = 0;
Theodore Ts'o59e315b2008-12-06 16:58:39 -05001460 int err = 0, i;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001461
Darrick J. Wongb0336e82012-04-29 18:41:10 -04001462 if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
1463 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1464 csum_size = sizeof(struct ext4_dir_entry_tail);
1465
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001466 bh2 = ext4_append (handle, dir, &newblock, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001467 if (!(bh2)) {
1468 brelse(*bh);
1469 *bh = NULL;
1470 goto errout;
1471 }
1472
1473 BUFFER_TRACE(*bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001474 err = ext4_journal_get_write_access(handle, *bh);
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001475 if (err)
1476 goto journal_error;
1477
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001478 BUFFER_TRACE(frame->bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001479 err = ext4_journal_get_write_access(handle, frame->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001480 if (err)
1481 goto journal_error;
1482
1483 data2 = bh2->b_data;
1484
1485 /* create map in the end of data2 block */
1486 map = (struct dx_map_entry *) (data2 + blocksize);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001487 count = dx_make_map((struct ext4_dir_entry_2 *) data1,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001488 blocksize, hinfo, map);
1489 map -= count;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001490 dx_sort_map(map, count);
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001491 /* Split the existing block in the middle, size-wise */
1492 size = 0;
1493 move = 0;
1494 for (i = count-1; i >= 0; i--) {
1495 /* is more than half of this entry in 2nd half of the block? */
1496 if (size + map[i].size/2 > blocksize/2)
1497 break;
1498 size += map[i].size;
1499 move++;
1500 }
1501 /* map index at which we will split */
1502 split = count - move;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001503 hash2 = map[split].hash;
1504 continued = hash2 == map[split - 1].hash;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001505 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1506 (unsigned long)dx_get_block(frame->at),
1507 hash2, split, count-split));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001508
1509 /* Fancy dance to stay within two buffers */
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001510 de2 = dx_move_dirents(data1, data2, map + split, count - split, blocksize);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001511 de = dx_pack_dirents(data1, blocksize);
Darrick J. Wongb0336e82012-04-29 18:41:10 -04001512 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1513 (char *) de,
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001514 blocksize);
Darrick J. Wongb0336e82012-04-29 18:41:10 -04001515 de2->rec_len = ext4_rec_len_to_disk(data2 + (blocksize - csum_size) -
1516 (char *) de2,
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001517 blocksize);
Darrick J. Wongb0336e82012-04-29 18:41:10 -04001518 if (csum_size) {
1519 t = EXT4_DIRENT_TAIL(data2, blocksize);
1520 initialize_dirent_tail(t, blocksize);
1521
1522 t = EXT4_DIRENT_TAIL(data1, blocksize);
1523 initialize_dirent_tail(t, blocksize);
1524 }
1525
Mingming Cao617ba132006-10-11 01:20:53 -07001526 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1527 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001528
1529 /* Which block gets the new entry? */
1530 if (hinfo->hash >= hash2)
1531 {
1532 swap(*bh, bh2);
1533 de = de2;
1534 }
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001535 dx_insert_block(frame, hash2 + continued, newblock);
Darrick J. Wongb0336e82012-04-29 18:41:10 -04001536 err = ext4_handle_dirty_dirent_node(handle, dir, bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001537 if (err)
1538 goto journal_error;
Darrick J. Wongdbe89442012-04-29 18:39:10 -04001539 err = ext4_handle_dirty_dx_node(handle, dir, frame->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001540 if (err)
1541 goto journal_error;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001542 brelse(bh2);
1543 dxtrace(dx_show_index("frame", frame->entries));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001544 return de;
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001545
1546journal_error:
1547 brelse(*bh);
1548 brelse(bh2);
1549 *bh = NULL;
1550 ext4_std_error(dir->i_sb, err);
1551errout:
1552 *error = err;
1553 return NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001554}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001555
1556/*
1557 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1558 * it points to a directory entry which is guaranteed to be large
1559 * enough for new directory entry. If de is NULL, then
1560 * add_dirent_to_buf will attempt search the directory block for
1561 * space. It will return -ENOSPC if no space is available, and -EIO
1562 * and -EEXIST if directory entry already exists.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001563 */
1564static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
Mingming Cao617ba132006-10-11 01:20:53 -07001565 struct inode *inode, struct ext4_dir_entry_2 *de,
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001566 struct buffer_head *bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001567{
1568 struct inode *dir = dentry->d_parent->d_inode;
1569 const char *name = dentry->d_name.name;
1570 int namelen = dentry->d_name.len;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05001571 unsigned int offset = 0;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001572 unsigned int blocksize = dir->i_sb->s_blocksize;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001573 unsigned short reclen;
1574 int nlen, rlen, err;
1575 char *top;
Darrick J. Wongb0336e82012-04-29 18:41:10 -04001576 int csum_size = 0;
1577
1578 if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1579 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1580 csum_size = sizeof(struct ext4_dir_entry_tail);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001581
Mingming Cao617ba132006-10-11 01:20:53 -07001582 reclen = EXT4_DIR_REC_LEN(namelen);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001583 if (!de) {
Mingming Cao617ba132006-10-11 01:20:53 -07001584 de = (struct ext4_dir_entry_2 *)bh->b_data;
Darrick J. Wongb0336e82012-04-29 18:41:10 -04001585 top = bh->b_data + (blocksize - csum_size) - reclen;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001586 while ((char *) de <= top) {
Theodore Ts'of7c21172011-01-10 12:10:55 -05001587 if (ext4_check_dir_entry(dir, NULL, de, bh, offset))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001588 return -EIO;
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001589 if (ext4_match(namelen, name, de))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001590 return -EEXIST;
Mingming Cao617ba132006-10-11 01:20:53 -07001591 nlen = EXT4_DIR_REC_LEN(de->name_len);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001592 rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001593 if ((de->inode? rlen - nlen: rlen) >= reclen)
1594 break;
Mingming Cao617ba132006-10-11 01:20:53 -07001595 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001596 offset += rlen;
1597 }
1598 if ((char *) de > top)
1599 return -ENOSPC;
1600 }
1601 BUFFER_TRACE(bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001602 err = ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001603 if (err) {
Mingming Cao617ba132006-10-11 01:20:53 -07001604 ext4_std_error(dir->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001605 return err;
1606 }
1607
1608 /* By now the buffer is marked for journaling */
Mingming Cao617ba132006-10-11 01:20:53 -07001609 nlen = EXT4_DIR_REC_LEN(de->name_len);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001610 rlen = ext4_rec_len_from_disk(de->rec_len, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001611 if (de->inode) {
Mingming Cao617ba132006-10-11 01:20:53 -07001612 struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001613 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen, blocksize);
1614 de->rec_len = ext4_rec_len_to_disk(nlen, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001615 de = de1;
1616 }
Mingming Cao617ba132006-10-11 01:20:53 -07001617 de->file_type = EXT4_FT_UNKNOWN;
Theodore Ts'ob09de7f2012-04-30 07:40:00 -04001618 de->inode = cpu_to_le32(inode->i_ino);
1619 ext4_set_de_type(dir->i_sb, de, inode->i_mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001620 de->name_len = namelen;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001621 memcpy(de->name, name, namelen);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001622 /*
1623 * XXX shouldn't update any times until successful
1624 * completion of syscall, but too many callers depend
1625 * on this.
1626 *
1627 * XXX similarly, too many callers depend on
Mingming Cao617ba132006-10-11 01:20:53 -07001628 * ext4_new_inode() setting the times, but error
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001629 * recovery deletes the inode, so the worst that can
1630 * happen is that the times are slightly out of date
1631 * and/or different from the directory change time.
1632 */
Kalpak Shahef7f3832007-07-18 09:15:20 -04001633 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
Mingming Cao617ba132006-10-11 01:20:53 -07001634 ext4_update_dx_flag(dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001635 dir->i_version++;
Mingming Cao617ba132006-10-11 01:20:53 -07001636 ext4_mark_inode_dirty(handle, dir);
Frank Mayhar03901312009-01-07 00:06:22 -05001637 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
Darrick J. Wongb0336e82012-04-29 18:41:10 -04001638 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001639 if (err)
Mingming Cao617ba132006-10-11 01:20:53 -07001640 ext4_std_error(dir->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001641 return 0;
1642}
1643
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001644/*
1645 * This converts a one block unindexed directory to a 3 block indexed
1646 * directory, and adds the dentry to the indexed directory.
1647 */
1648static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1649 struct inode *inode, struct buffer_head *bh)
1650{
1651 struct inode *dir = dentry->d_parent->d_inode;
1652 const char *name = dentry->d_name.name;
1653 int namelen = dentry->d_name.len;
1654 struct buffer_head *bh2;
1655 struct dx_root *root;
1656 struct dx_frame frames[2], *frame;
1657 struct dx_entry *entries;
Mingming Cao617ba132006-10-11 01:20:53 -07001658 struct ext4_dir_entry_2 *de, *de2;
Darrick J. Wongb0336e82012-04-29 18:41:10 -04001659 struct ext4_dir_entry_tail *t;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001660 char *data1, *top;
1661 unsigned len;
1662 int retval;
1663 unsigned blocksize;
1664 struct dx_hash_info hinfo;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001665 ext4_lblk_t block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001666 struct fake_dirent *fde;
Darrick J. Wongb0336e82012-04-29 18:41:10 -04001667 int csum_size = 0;
1668
1669 if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1670 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1671 csum_size = sizeof(struct ext4_dir_entry_tail);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001672
1673 blocksize = dir->i_sb->s_blocksize;
Theodore Ts'oe6b8bc02009-01-16 11:13:40 -05001674 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
Mingming Cao617ba132006-10-11 01:20:53 -07001675 retval = ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001676 if (retval) {
Mingming Cao617ba132006-10-11 01:20:53 -07001677 ext4_std_error(dir->i_sb, retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001678 brelse(bh);
1679 return retval;
1680 }
1681 root = (struct dx_root *) bh->b_data;
1682
Theodore Ts'oe6b8bc02009-01-16 11:13:40 -05001683 /* The 0th block becomes the root, move the dirents out */
1684 fde = &root->dotdot;
1685 de = (struct ext4_dir_entry_2 *)((char *)fde +
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001686 ext4_rec_len_from_disk(fde->rec_len, blocksize));
Theodore Ts'oe6b8bc02009-01-16 11:13:40 -05001687 if ((char *) de >= (((char *) root) + blocksize)) {
Theodore Ts'o24676da2010-05-16 21:00:00 -04001688 EXT4_ERROR_INODE(dir, "invalid rec_len for '..'");
Theodore Ts'oe6b8bc02009-01-16 11:13:40 -05001689 brelse(bh);
1690 return -EIO;
1691 }
Darrick J. Wongb0336e82012-04-29 18:41:10 -04001692 len = ((char *) root) + (blocksize - csum_size) - (char *) de;
Theodore Ts'oe6b8bc02009-01-16 11:13:40 -05001693
1694 /* Allocate new block for the 0th block's dirents */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001695 bh2 = ext4_append(handle, dir, &block, &retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001696 if (!(bh2)) {
1697 brelse(bh);
1698 return retval;
1699 }
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04001700 ext4_set_inode_flag(dir, EXT4_INODE_INDEX);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001701 data1 = bh2->b_data;
1702
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001703 memcpy (data1, de, len);
Mingming Cao617ba132006-10-11 01:20:53 -07001704 de = (struct ext4_dir_entry_2 *) data1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001705 top = data1 + len;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001706 while ((char *)(de2 = ext4_next_entry(de, blocksize)) < top)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001707 de = de2;
Darrick J. Wongb0336e82012-04-29 18:41:10 -04001708 de->rec_len = ext4_rec_len_to_disk(data1 + (blocksize - csum_size) -
1709 (char *) de,
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001710 blocksize);
Darrick J. Wongb0336e82012-04-29 18:41:10 -04001711
1712 if (csum_size) {
1713 t = EXT4_DIRENT_TAIL(data1, blocksize);
1714 initialize_dirent_tail(t, blocksize);
1715 }
1716
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001717 /* Initialize the root; the dot dirents already exist */
Mingming Cao617ba132006-10-11 01:20:53 -07001718 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001719 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2),
1720 blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001721 memset (&root->info, 0, sizeof(root->info));
1722 root->info.info_length = sizeof(root->info);
Mingming Cao617ba132006-10-11 01:20:53 -07001723 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001724 entries = root->entries;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001725 dx_set_block(entries, 1);
1726 dx_set_count(entries, 1);
1727 dx_set_limit(entries, dx_root_limit(dir, sizeof(root->info)));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001728
1729 /* Initialize as for dx_probe */
1730 hinfo.hash_version = root->info.hash_version;
Theodore Ts'of99b2582008-10-28 13:21:44 -04001731 if (hinfo.hash_version <= DX_HASH_TEA)
1732 hinfo.hash_version += EXT4_SB(dir->i_sb)->s_hash_unsigned;
Mingming Cao617ba132006-10-11 01:20:53 -07001733 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1734 ext4fs_dirhash(name, namelen, &hinfo);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001735 frame = frames;
1736 frame->entries = entries;
1737 frame->at = entries;
1738 frame->bh = bh;
1739 bh = bh2;
Allison Henderson6976a6f2011-05-15 00:19:41 -04001740
Darrick J. Wongdbe89442012-04-29 18:39:10 -04001741 ext4_handle_dirty_dx_node(handle, dir, frame->bh);
Darrick J. Wongb0336e82012-04-29 18:41:10 -04001742 ext4_handle_dirty_dirent_node(handle, dir, bh);
Allison Henderson6976a6f2011-05-15 00:19:41 -04001743
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001744 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
Jan Kara7ad8e4e2011-05-03 11:05:55 -04001745 if (!de) {
1746 /*
1747 * Even if the block split failed, we have to properly write
1748 * out all the changes we did so far. Otherwise we can end up
1749 * with corrupted filesystem.
1750 */
1751 ext4_mark_inode_dirty(handle, dir);
Jan Kara7ad8e4e2011-05-03 11:05:55 -04001752 dx_release(frames);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001753 return retval;
Jan Kara7ad8e4e2011-05-03 11:05:55 -04001754 }
1755 dx_release(frames);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001756
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001757 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1758 brelse(bh);
1759 return retval;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001760}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001761
1762/*
Mingming Cao617ba132006-10-11 01:20:53 -07001763 * ext4_add_entry()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001764 *
1765 * adds a file entry to the specified directory, using the same
Mingming Cao617ba132006-10-11 01:20:53 -07001766 * semantics as ext4_find_entry(). It returns NULL if it failed.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001767 *
1768 * NOTE!! The inode part of 'de' is left at 0 - which means you
1769 * may not sleep between calling this and putting something into
1770 * the entry, as someone else might have used it while you slept.
1771 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001772static int ext4_add_entry(handle_t *handle, struct dentry *dentry,
1773 struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001774{
1775 struct inode *dir = dentry->d_parent->d_inode;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001776 struct buffer_head *bh;
Mingming Cao617ba132006-10-11 01:20:53 -07001777 struct ext4_dir_entry_2 *de;
Darrick J. Wongb0336e82012-04-29 18:41:10 -04001778 struct ext4_dir_entry_tail *t;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001779 struct super_block *sb;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001780 int retval;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001781 int dx_fallback=0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001782 unsigned blocksize;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001783 ext4_lblk_t block, blocks;
Darrick J. Wongb0336e82012-04-29 18:41:10 -04001784 int csum_size = 0;
1785
1786 if (EXT4_HAS_RO_COMPAT_FEATURE(inode->i_sb,
1787 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
1788 csum_size = sizeof(struct ext4_dir_entry_tail);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001789
1790 sb = dir->i_sb;
1791 blocksize = sb->s_blocksize;
1792 if (!dentry->d_name.len)
1793 return -EINVAL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001794 if (is_dx(dir)) {
Mingming Cao617ba132006-10-11 01:20:53 -07001795 retval = ext4_dx_add_entry(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001796 if (!retval || (retval != ERR_BAD_DX_DIR))
1797 return retval;
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04001798 ext4_clear_inode_flag(dir, EXT4_INODE_INDEX);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001799 dx_fallback++;
Mingming Cao617ba132006-10-11 01:20:53 -07001800 ext4_mark_inode_dirty(handle, dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001801 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001802 blocks = dir->i_size >> sb->s_blocksize_bits;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05001803 for (block = 0; block < blocks; block++) {
Mingming Cao617ba132006-10-11 01:20:53 -07001804 bh = ext4_bread(handle, dir, block, 0, &retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001805 if(!bh)
1806 return retval;
Darrick J. Wongb0336e82012-04-29 18:41:10 -04001807 if (!buffer_verified(bh) &&
1808 !ext4_dirent_csum_verify(dir,
1809 (struct ext4_dir_entry *)bh->b_data))
1810 return -EIO;
1811 set_buffer_verified(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001812 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001813 if (retval != -ENOSPC) {
1814 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001815 return retval;
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001816 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001817
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001818 if (blocks == 1 && !dx_fallback &&
Theodore Ts'o1e424a32009-11-08 15:45:44 -05001819 EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
1820 return make_indexed_dir(handle, dentry, inode, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001821 brelse(bh);
1822 }
Mingming Cao617ba132006-10-11 01:20:53 -07001823 bh = ext4_append(handle, dir, &block, &retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001824 if (!bh)
1825 return retval;
Mingming Cao617ba132006-10-11 01:20:53 -07001826 de = (struct ext4_dir_entry_2 *) bh->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001827 de->inode = 0;
Darrick J. Wongb0336e82012-04-29 18:41:10 -04001828 de->rec_len = ext4_rec_len_to_disk(blocksize - csum_size, blocksize);
1829
1830 if (csum_size) {
1831 t = EXT4_DIRENT_TAIL(bh->b_data, blocksize);
1832 initialize_dirent_tail(t, blocksize);
1833 }
1834
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001835 retval = add_dirent_to_buf(handle, dentry, inode, de, bh);
1836 brelse(bh);
Frank Mayhar14ece102010-05-17 08:00:00 -04001837 if (retval == 0)
1838 ext4_set_inode_state(inode, EXT4_STATE_NEWENTRY);
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001839 return retval;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001840}
1841
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001842/*
1843 * Returns 0 for success, or a negative error value
1844 */
Mingming Cao617ba132006-10-11 01:20:53 -07001845static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001846 struct inode *inode)
1847{
1848 struct dx_frame frames[2], *frame;
1849 struct dx_entry *entries, *at;
1850 struct dx_hash_info hinfo;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001851 struct buffer_head *bh;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001852 struct inode *dir = dentry->d_parent->d_inode;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001853 struct super_block *sb = dir->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -07001854 struct ext4_dir_entry_2 *de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001855 int err;
1856
Theodore Ts'of702ba02008-09-22 15:21:01 -04001857 frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001858 if (!frame)
1859 return err;
1860 entries = frame->entries;
1861 at = frame->at;
1862
Mingming Cao617ba132006-10-11 01:20:53 -07001863 if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001864 goto cleanup;
1865
Darrick J. Wongb0336e82012-04-29 18:41:10 -04001866 if (!buffer_verified(bh) &&
1867 !ext4_dirent_csum_verify(dir, (struct ext4_dir_entry *)bh->b_data))
1868 goto journal_error;
1869 set_buffer_verified(bh);
1870
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001871 BUFFER_TRACE(bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001872 err = ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001873 if (err)
1874 goto journal_error;
1875
1876 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
Theodore Ts'o2de770a2009-11-23 07:25:49 -05001877 if (err != -ENOSPC)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001878 goto cleanup;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001879
1880 /* Block full, should compress but for now just split */
Theodore Ts'o4776004f2008-09-08 23:00:52 -04001881 dxtrace(printk(KERN_DEBUG "using %u of %u node entries\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001882 dx_get_count(entries), dx_get_limit(entries)));
1883 /* Need to split index? */
1884 if (dx_get_count(entries) == dx_get_limit(entries)) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001885 ext4_lblk_t newblock;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001886 unsigned icount = dx_get_count(entries);
1887 int levels = frame - frames;
1888 struct dx_entry *entries2;
1889 struct dx_node *node2;
1890 struct buffer_head *bh2;
1891
1892 if (levels && (dx_get_count(frames->entries) ==
1893 dx_get_limit(frames->entries))) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05001894 ext4_warning(sb, "Directory index full!");
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001895 err = -ENOSPC;
1896 goto cleanup;
1897 }
Mingming Cao617ba132006-10-11 01:20:53 -07001898 bh2 = ext4_append (handle, dir, &newblock, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001899 if (!(bh2))
1900 goto cleanup;
1901 node2 = (struct dx_node *)(bh2->b_data);
1902 entries2 = node2->entries;
Andreas Schlick1f7bebb2009-09-10 23:16:07 -04001903 memset(&node2->fake, 0, sizeof(struct fake_dirent));
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001904 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize,
1905 sb->s_blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001906 BUFFER_TRACE(frame->bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001907 err = ext4_journal_get_write_access(handle, frame->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001908 if (err)
1909 goto journal_error;
1910 if (levels) {
1911 unsigned icount1 = icount/2, icount2 = icount - icount1;
1912 unsigned hash2 = dx_get_hash(entries + icount1);
Theodore Ts'o4776004f2008-09-08 23:00:52 -04001913 dxtrace(printk(KERN_DEBUG "Split index %i/%i\n",
1914 icount1, icount2));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001915
1916 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
Mingming Cao617ba132006-10-11 01:20:53 -07001917 err = ext4_journal_get_write_access(handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001918 frames[0].bh);
1919 if (err)
1920 goto journal_error;
1921
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001922 memcpy((char *) entries2, (char *) (entries + icount1),
1923 icount2 * sizeof(struct dx_entry));
1924 dx_set_count(entries, icount1);
1925 dx_set_count(entries2, icount2);
1926 dx_set_limit(entries2, dx_node_limit(dir));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001927
1928 /* Which index block gets the new entry? */
1929 if (at - entries >= icount1) {
1930 frame->at = at = at - entries - icount1 + entries2;
1931 frame->entries = entries = entries2;
1932 swap(frame->bh, bh2);
1933 }
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001934 dx_insert_block(frames + 0, hash2, newblock);
1935 dxtrace(dx_show_index("node", frames[1].entries));
1936 dxtrace(dx_show_index("node",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001937 ((struct dx_node *) bh2->b_data)->entries));
Darrick J. Wongdbe89442012-04-29 18:39:10 -04001938 err = ext4_handle_dirty_dx_node(handle, dir, bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001939 if (err)
1940 goto journal_error;
1941 brelse (bh2);
1942 } else {
Theodore Ts'o4776004f2008-09-08 23:00:52 -04001943 dxtrace(printk(KERN_DEBUG
1944 "Creating second level index...\n"));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001945 memcpy((char *) entries2, (char *) entries,
1946 icount * sizeof(struct dx_entry));
1947 dx_set_limit(entries2, dx_node_limit(dir));
1948
1949 /* Set up root */
1950 dx_set_count(entries, 1);
1951 dx_set_block(entries + 0, newblock);
1952 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1953
1954 /* Add new access path frame */
1955 frame = frames + 1;
1956 frame->at = at = at - entries + entries2;
1957 frame->entries = entries = entries2;
1958 frame->bh = bh2;
Mingming Cao617ba132006-10-11 01:20:53 -07001959 err = ext4_journal_get_write_access(handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001960 frame->bh);
1961 if (err)
1962 goto journal_error;
1963 }
Darrick J. Wongdbe89442012-04-29 18:39:10 -04001964 err = ext4_handle_dirty_dx_node(handle, dir, frames[0].bh);
Theodore Ts'ob4097142011-01-10 12:46:59 -05001965 if (err) {
1966 ext4_std_error(inode->i_sb, err);
1967 goto cleanup;
1968 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001969 }
1970 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1971 if (!de)
1972 goto cleanup;
1973 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001974 goto cleanup;
1975
1976journal_error:
Mingming Cao617ba132006-10-11 01:20:53 -07001977 ext4_std_error(dir->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001978cleanup:
1979 if (bh)
1980 brelse(bh);
1981 dx_release(frames);
1982 return err;
1983}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001984
1985/*
Mingming Cao617ba132006-10-11 01:20:53 -07001986 * ext4_delete_entry deletes a directory entry by merging it with the
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001987 * previous entry
1988 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001989static int ext4_delete_entry(handle_t *handle,
1990 struct inode *dir,
1991 struct ext4_dir_entry_2 *de_del,
1992 struct buffer_head *bh)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001993{
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04001994 struct ext4_dir_entry_2 *de, *pde;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05001995 unsigned int blocksize = dir->i_sb->s_blocksize;
Darrick J. Wongb0336e82012-04-29 18:41:10 -04001996 int csum_size = 0;
Theodore Ts'ob4097142011-01-10 12:46:59 -05001997 int i, err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001998
Darrick J. Wongb0336e82012-04-29 18:41:10 -04001999 if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
2000 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
2001 csum_size = sizeof(struct ext4_dir_entry_tail);
2002
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002003 i = 0;
2004 pde = NULL;
Mingming Cao617ba132006-10-11 01:20:53 -07002005 de = (struct ext4_dir_entry_2 *) bh->b_data;
Darrick J. Wongb0336e82012-04-29 18:41:10 -04002006 while (i < bh->b_size - csum_size) {
Theodore Ts'of7c21172011-01-10 12:10:55 -05002007 if (ext4_check_dir_entry(dir, NULL, de, bh, i))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002008 return -EIO;
2009 if (de == de_del) {
2010 BUFFER_TRACE(bh, "get_write_access");
Theodore Ts'ob4097142011-01-10 12:46:59 -05002011 err = ext4_journal_get_write_access(handle, bh);
2012 if (unlikely(err)) {
2013 ext4_std_error(dir->i_sb, err);
2014 return err;
2015 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002016 if (pde)
Jan Karaa72d7f82008-01-28 23:58:27 -05002017 pde->rec_len = ext4_rec_len_to_disk(
Wei Yongjun3d0518f2009-02-14 23:01:36 -05002018 ext4_rec_len_from_disk(pde->rec_len,
2019 blocksize) +
2020 ext4_rec_len_from_disk(de->rec_len,
2021 blocksize),
2022 blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002023 else
2024 de->inode = 0;
2025 dir->i_version++;
Frank Mayhar03901312009-01-07 00:06:22 -05002026 BUFFER_TRACE(bh, "call ext4_handle_dirty_metadata");
Darrick J. Wongb0336e82012-04-29 18:41:10 -04002027 err = ext4_handle_dirty_dirent_node(handle, dir, bh);
Theodore Ts'ob4097142011-01-10 12:46:59 -05002028 if (unlikely(err)) {
2029 ext4_std_error(dir->i_sb, err);
2030 return err;
2031 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002032 return 0;
2033 }
Wei Yongjun3d0518f2009-02-14 23:01:36 -05002034 i += ext4_rec_len_from_disk(de->rec_len, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002035 pde = de;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05002036 de = ext4_next_entry(de, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002037 }
2038 return -ENOENT;
2039}
2040
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002041/*
2042 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
2043 * since this indicates that nlinks count was previously 1.
2044 */
2045static void ext4_inc_count(handle_t *handle, struct inode *inode)
2046{
2047 inc_nlink(inode);
2048 if (is_dx(inode) && inode->i_nlink > 1) {
2049 /* limit is 16-bit i_links_count */
2050 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
Miklos Szeredibfe86842011-10-28 14:13:29 +02002051 set_nlink(inode, 1);
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002052 EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
2053 EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
2054 }
2055 }
2056}
2057
2058/*
2059 * If a directory had nlink == 1, then we should let it be 1. This indicates
2060 * directory has >EXT4_LINK_MAX subdirs.
2061 */
2062static void ext4_dec_count(handle_t *handle, struct inode *inode)
2063{
Andreas Dilger909a4cf2011-10-26 03:22:31 -04002064 if (!S_ISDIR(inode->i_mode) || inode->i_nlink > 2)
2065 drop_nlink(inode);
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002066}
2067
2068
Mingming Cao617ba132006-10-11 01:20:53 -07002069static int ext4_add_nondir(handle_t *handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002070 struct dentry *dentry, struct inode *inode)
2071{
Mingming Cao617ba132006-10-11 01:20:53 -07002072 int err = ext4_add_entry(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002073 if (!err) {
Mingming Cao617ba132006-10-11 01:20:53 -07002074 ext4_mark_inode_dirty(handle, inode);
Al Viro6b38e842008-12-30 02:03:31 -05002075 unlock_new_inode(inode);
Al Viro8fc37ec2012-07-19 09:18:15 +04002076 d_instantiate(dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002077 return 0;
2078 }
Eric Sandeen731b9a52007-02-10 01:46:16 -08002079 drop_nlink(inode);
Al Viro6b38e842008-12-30 02:03:31 -05002080 unlock_new_inode(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002081 iput(inode);
2082 return err;
2083}
2084
2085/*
2086 * By the time this is called, we already have created
2087 * the directory cache entry for the new file, but it
2088 * is so far negative - it has no inode.
2089 *
2090 * If the create succeeds, we fill in the inode information
2091 * with d_instantiate().
2092 */
Al Viro4acdaf22011-07-26 01:42:34 -04002093static int ext4_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viroebfc3b42012-06-10 18:05:36 -04002094 bool excl)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002095{
2096 handle_t *handle;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002097 struct inode *inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002098 int err, retries = 0;
2099
Christoph Hellwig871a2932010-03-03 09:05:07 -05002100 dquot_initialize(dir);
Christoph Hellwig907f4552010-03-03 09:05:06 -05002101
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002102retry:
Mingming Cao617ba132006-10-11 01:20:53 -07002103 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2104 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
Dmitry Monakhov5aca07e2009-12-08 22:42:15 -05002105 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002106 if (IS_ERR(handle))
2107 return PTR_ERR(handle);
2108
2109 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05002110 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002111
Dmitry Monakhov5cb81da2011-10-29 09:05:00 -04002112 inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002113 err = PTR_ERR(inode);
2114 if (!IS_ERR(inode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07002115 inode->i_op = &ext4_file_inode_operations;
2116 inode->i_fop = &ext4_file_operations;
2117 ext4_set_aops(inode);
2118 err = ext4_add_nondir(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002119 }
Mingming Cao617ba132006-10-11 01:20:53 -07002120 ext4_journal_stop(handle);
2121 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002122 goto retry;
2123 return err;
2124}
2125
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002126static int ext4_mknod(struct inode *dir, struct dentry *dentry,
Al Viro1a67aaf2011-07-26 01:52:52 -04002127 umode_t mode, dev_t rdev)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002128{
2129 handle_t *handle;
2130 struct inode *inode;
2131 int err, retries = 0;
2132
2133 if (!new_valid_dev(rdev))
2134 return -EINVAL;
2135
Christoph Hellwig871a2932010-03-03 09:05:07 -05002136 dquot_initialize(dir);
Christoph Hellwig907f4552010-03-03 09:05:06 -05002137
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002138retry:
Mingming Cao617ba132006-10-11 01:20:53 -07002139 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2140 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
Dmitry Monakhov5aca07e2009-12-08 22:42:15 -05002141 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002142 if (IS_ERR(handle))
2143 return PTR_ERR(handle);
2144
2145 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05002146 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002147
Dmitry Monakhov5cb81da2011-10-29 09:05:00 -04002148 inode = ext4_new_inode(handle, dir, mode, &dentry->d_name, 0, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002149 err = PTR_ERR(inode);
2150 if (!IS_ERR(inode)) {
2151 init_special_inode(inode, inode->i_mode, rdev);
Theodore Ts'o03010a32008-10-10 20:02:48 -04002152#ifdef CONFIG_EXT4_FS_XATTR
Mingming Cao617ba132006-10-11 01:20:53 -07002153 inode->i_op = &ext4_special_inode_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002154#endif
Mingming Cao617ba132006-10-11 01:20:53 -07002155 err = ext4_add_nondir(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002156 }
Mingming Cao617ba132006-10-11 01:20:53 -07002157 ext4_journal_stop(handle);
2158 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002159 goto retry;
2160 return err;
2161}
2162
Al Viro18bb1db2011-07-26 01:41:39 -04002163static int ext4_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002164{
2165 handle_t *handle;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002166 struct inode *inode;
Namhyung Kimdabd9912011-01-10 12:11:16 -05002167 struct buffer_head *dir_block = NULL;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002168 struct ext4_dir_entry_2 *de;
Darrick J. Wongb0336e82012-04-29 18:41:10 -04002169 struct ext4_dir_entry_tail *t;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05002170 unsigned int blocksize = dir->i_sb->s_blocksize;
Darrick J. Wongb0336e82012-04-29 18:41:10 -04002171 int csum_size = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002172 int err, retries = 0;
2173
Darrick J. Wongb0336e82012-04-29 18:41:10 -04002174 if (EXT4_HAS_RO_COMPAT_FEATURE(dir->i_sb,
2175 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
2176 csum_size = sizeof(struct ext4_dir_entry_tail);
2177
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002178 if (EXT4_DIR_LINK_MAX(dir))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002179 return -EMLINK;
2180
Christoph Hellwig871a2932010-03-03 09:05:07 -05002181 dquot_initialize(dir);
Christoph Hellwig907f4552010-03-03 09:05:06 -05002182
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002183retry:
Mingming Cao617ba132006-10-11 01:20:53 -07002184 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2185 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
Dmitry Monakhov5aca07e2009-12-08 22:42:15 -05002186 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002187 if (IS_ERR(handle))
2188 return PTR_ERR(handle);
2189
2190 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05002191 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002192
Andreas Dilger11013912009-06-13 11:45:35 -04002193 inode = ext4_new_inode(handle, dir, S_IFDIR | mode,
Dmitry Monakhov5cb81da2011-10-29 09:05:00 -04002194 &dentry->d_name, 0, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002195 err = PTR_ERR(inode);
2196 if (IS_ERR(inode))
2197 goto out_stop;
2198
Mingming Cao617ba132006-10-11 01:20:53 -07002199 inode->i_op = &ext4_dir_inode_operations;
2200 inode->i_fop = &ext4_dir_operations;
2201 inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002202 dir_block = ext4_bread(handle, inode, 0, 1, &err);
Aneesh Kumar K.V4cdeed82008-02-22 06:17:31 -05002203 if (!dir_block)
2204 goto out_clear_inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002205 BUFFER_TRACE(dir_block, "get_write_access");
Namhyung Kimdabd9912011-01-10 12:11:16 -05002206 err = ext4_journal_get_write_access(handle, dir_block);
2207 if (err)
2208 goto out_clear_inode;
Mingming Cao617ba132006-10-11 01:20:53 -07002209 de = (struct ext4_dir_entry_2 *) dir_block->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002210 de->inode = cpu_to_le32(inode->i_ino);
2211 de->name_len = 1;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05002212 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len),
2213 blocksize);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002214 strcpy(de->name, ".");
Mingming Cao617ba132006-10-11 01:20:53 -07002215 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05002216 de = ext4_next_entry(de, blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002217 de->inode = cpu_to_le32(dir->i_ino);
Darrick J. Wongb0336e82012-04-29 18:41:10 -04002218 de->rec_len = ext4_rec_len_to_disk(blocksize -
2219 (csum_size + EXT4_DIR_REC_LEN(1)),
Wei Yongjun3d0518f2009-02-14 23:01:36 -05002220 blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002221 de->name_len = 2;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002222 strcpy(de->name, "..");
Mingming Cao617ba132006-10-11 01:20:53 -07002223 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
Miklos Szeredibfe86842011-10-28 14:13:29 +02002224 set_nlink(inode, 2);
Darrick J. Wongb0336e82012-04-29 18:41:10 -04002225
2226 if (csum_size) {
2227 t = EXT4_DIRENT_TAIL(dir_block->b_data, blocksize);
2228 initialize_dirent_tail(t, blocksize);
2229 }
2230
Frank Mayhar03901312009-01-07 00:06:22 -05002231 BUFFER_TRACE(dir_block, "call ext4_handle_dirty_metadata");
Darrick J. Wongb0336e82012-04-29 18:41:10 -04002232 err = ext4_handle_dirty_dirent_node(handle, inode, dir_block);
Namhyung Kimdabd9912011-01-10 12:11:16 -05002233 if (err)
2234 goto out_clear_inode;
Darrick J. Wongb0336e82012-04-29 18:41:10 -04002235 set_buffer_verified(dir_block);
Namhyung Kimdabd9912011-01-10 12:11:16 -05002236 err = ext4_mark_inode_dirty(handle, inode);
2237 if (!err)
2238 err = ext4_add_entry(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002239 if (err) {
Aneesh Kumar K.V4cdeed82008-02-22 06:17:31 -05002240out_clear_inode:
2241 clear_nlink(inode);
Al Viro6b38e842008-12-30 02:03:31 -05002242 unlock_new_inode(inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002243 ext4_mark_inode_dirty(handle, inode);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002244 iput(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002245 goto out_stop;
2246 }
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002247 ext4_inc_count(handle, dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002248 ext4_update_dx_flag(dir);
Namhyung Kimdabd9912011-01-10 12:11:16 -05002249 err = ext4_mark_inode_dirty(handle, dir);
2250 if (err)
2251 goto out_clear_inode;
Al Viro6b38e842008-12-30 02:03:31 -05002252 unlock_new_inode(inode);
Al Viro8fc37ec2012-07-19 09:18:15 +04002253 d_instantiate(dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002254out_stop:
Namhyung Kimdabd9912011-01-10 12:11:16 -05002255 brelse(dir_block);
Mingming Cao617ba132006-10-11 01:20:53 -07002256 ext4_journal_stop(handle);
2257 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002258 goto retry;
2259 return err;
2260}
2261
2262/*
2263 * routine to check that the specified directory is empty (for rmdir)
2264 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002265static int empty_dir(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002266{
Theodore Ts'o498e5f22008-11-05 00:14:04 -05002267 unsigned int offset;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002268 struct buffer_head *bh;
2269 struct ext4_dir_entry_2 *de, *de1;
2270 struct super_block *sb;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002271 int err = 0;
2272
2273 sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -07002274 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002275 !(bh = ext4_bread(NULL, inode, 0, 0, &err))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002276 if (err)
Theodore Ts'o24676da2010-05-16 21:00:00 -04002277 EXT4_ERROR_INODE(inode,
2278 "error %d reading directory lblock 0", err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002279 else
Eric Sandeen12062dd2010-02-15 14:19:27 -05002280 ext4_warning(inode->i_sb,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002281 "bad directory (dir #%lu) - no data block",
2282 inode->i_ino);
2283 return 1;
2284 }
Darrick J. Wongb0336e82012-04-29 18:41:10 -04002285 if (!buffer_verified(bh) &&
2286 !ext4_dirent_csum_verify(inode,
2287 (struct ext4_dir_entry *)bh->b_data)) {
2288 EXT4_ERROR_INODE(inode, "checksum error reading directory "
2289 "lblock 0");
2290 return -EIO;
2291 }
2292 set_buffer_verified(bh);
Mingming Cao617ba132006-10-11 01:20:53 -07002293 de = (struct ext4_dir_entry_2 *) bh->b_data;
Wei Yongjun3d0518f2009-02-14 23:01:36 -05002294 de1 = ext4_next_entry(de, sb->s_blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002295 if (le32_to_cpu(de->inode) != inode->i_ino ||
2296 !le32_to_cpu(de1->inode) ||
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002297 strcmp(".", de->name) ||
2298 strcmp("..", de1->name)) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05002299 ext4_warning(inode->i_sb,
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002300 "bad directory (dir #%lu) - no `.' or `..'",
2301 inode->i_ino);
2302 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002303 return 1;
2304 }
Wei Yongjun3d0518f2009-02-14 23:01:36 -05002305 offset = ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize) +
2306 ext4_rec_len_from_disk(de1->rec_len, sb->s_blocksize);
2307 de = ext4_next_entry(de1, sb->s_blocksize);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002308 while (offset < inode->i_size) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002309 if (!bh ||
Theodore Ts'o24676da2010-05-16 21:00:00 -04002310 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
2311 unsigned int lblock;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002312 err = 0;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002313 brelse(bh);
Theodore Ts'o24676da2010-05-16 21:00:00 -04002314 lblock = offset >> EXT4_BLOCK_SIZE_BITS(sb);
2315 bh = ext4_bread(NULL, inode, lblock, 0, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002316 if (!bh) {
2317 if (err)
Theodore Ts'o24676da2010-05-16 21:00:00 -04002318 EXT4_ERROR_INODE(inode,
2319 "error %d reading directory "
2320 "lblock %u", err, lblock);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002321 offset += sb->s_blocksize;
2322 continue;
2323 }
Darrick J. Wongb0336e82012-04-29 18:41:10 -04002324 if (!buffer_verified(bh) &&
2325 !ext4_dirent_csum_verify(inode,
2326 (struct ext4_dir_entry *)bh->b_data)) {
2327 EXT4_ERROR_INODE(inode, "checksum error "
2328 "reading directory lblock 0");
2329 return -EIO;
2330 }
2331 set_buffer_verified(bh);
Mingming Cao617ba132006-10-11 01:20:53 -07002332 de = (struct ext4_dir_entry_2 *) bh->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002333 }
Theodore Ts'of7c21172011-01-10 12:10:55 -05002334 if (ext4_check_dir_entry(inode, NULL, de, bh, offset)) {
Mingming Cao617ba132006-10-11 01:20:53 -07002335 de = (struct ext4_dir_entry_2 *)(bh->b_data +
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002336 sb->s_blocksize);
2337 offset = (offset | (sb->s_blocksize - 1)) + 1;
2338 continue;
2339 }
2340 if (le32_to_cpu(de->inode)) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002341 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002342 return 0;
2343 }
Wei Yongjun3d0518f2009-02-14 23:01:36 -05002344 offset += ext4_rec_len_from_disk(de->rec_len, sb->s_blocksize);
2345 de = ext4_next_entry(de, sb->s_blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002346 }
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002347 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002348 return 1;
2349}
2350
Mingming Cao617ba132006-10-11 01:20:53 -07002351/* ext4_orphan_add() links an unlinked or truncated inode into a list of
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002352 * such inodes, starting at the superblock, in case we crash before the
2353 * file is closed/deleted, or in case the inode truncate spans multiple
2354 * transactions and the last transaction is not recovered after a crash.
2355 *
2356 * At filesystem recovery time, we walk this list deleting unlinked
Mingming Cao617ba132006-10-11 01:20:53 -07002357 * inodes and truncating linked inodes in ext4_orphan_cleanup().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002358 */
Mingming Cao617ba132006-10-11 01:20:53 -07002359int ext4_orphan_add(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002360{
2361 struct super_block *sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -07002362 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002363 int err = 0, rc;
2364
Frank Mayhar03901312009-01-07 00:06:22 -05002365 if (!ext4_handle_valid(handle))
2366 return 0;
2367
Theodore Ts'o3b9d4ed2009-04-25 22:54:04 -04002368 mutex_lock(&EXT4_SB(sb)->s_orphan_lock);
Mingming Cao617ba132006-10-11 01:20:53 -07002369 if (!list_empty(&EXT4_I(inode)->i_orphan))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002370 goto out_unlock;
2371
Lukas Czernerafb86172011-07-11 18:47:04 -04002372 /*
2373 * Orphan handling is only valid for files with data blocks
2374 * being truncated, or files being unlinked. Note that we either
2375 * hold i_mutex, or the inode can not be referenced from outside,
2376 * so i_nlink should not be bumped due to race
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002377 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002378 J_ASSERT((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2379 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002380
Mingming Cao617ba132006-10-11 01:20:53 -07002381 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
2382 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002383 if (err)
2384 goto out_unlock;
2385
Mingming Cao617ba132006-10-11 01:20:53 -07002386 err = ext4_reserve_inode_write(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002387 if (err)
2388 goto out_unlock;
Dmitry Monakhov6e3617e2010-03-01 23:29:39 -05002389 /*
2390 * Due to previous errors inode may be already a part of on-disk
2391 * orphan list. If so skip on-disk list modification.
2392 */
2393 if (NEXT_ORPHAN(inode) && NEXT_ORPHAN(inode) <=
2394 (le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count)))
2395 goto mem_insert;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002396
2397 /* Insert this inode at the head of the on-disk orphan list... */
Mingming Cao617ba132006-10-11 01:20:53 -07002398 NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
2399 EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
Artem Bityutskiyb50924c2012-07-22 20:37:31 -04002400 err = ext4_handle_dirty_super(handle, sb);
Mingming Cao617ba132006-10-11 01:20:53 -07002401 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002402 if (!err)
2403 err = rc;
2404
2405 /* Only add to the head of the in-memory list if all the
2406 * previous operations succeeded. If the orphan_add is going to
2407 * fail (possibly taking the journal offline), we can't risk
2408 * leaving the inode on the orphan list: stray orphan-list
2409 * entries can cause panics at unmount time.
2410 *
2411 * This is safe: on error we're going to ignore the orphan list
2412 * anyway on the next recovery. */
Dmitry Monakhov6e3617e2010-03-01 23:29:39 -05002413mem_insert:
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002414 if (!err)
Mingming Cao617ba132006-10-11 01:20:53 -07002415 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002416
2417 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
2418 jbd_debug(4, "orphan inode %lu will point to %d\n",
2419 inode->i_ino, NEXT_ORPHAN(inode));
2420out_unlock:
Theodore Ts'o3b9d4ed2009-04-25 22:54:04 -04002421 mutex_unlock(&EXT4_SB(sb)->s_orphan_lock);
Mingming Cao617ba132006-10-11 01:20:53 -07002422 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002423 return err;
2424}
2425
2426/*
Mingming Cao617ba132006-10-11 01:20:53 -07002427 * ext4_orphan_del() removes an unlinked or truncated inode from the list
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002428 * of such inodes stored on disk, because it is finally being cleaned up.
2429 */
Mingming Cao617ba132006-10-11 01:20:53 -07002430int ext4_orphan_del(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002431{
2432 struct list_head *prev;
Mingming Cao617ba132006-10-11 01:20:53 -07002433 struct ext4_inode_info *ei = EXT4_I(inode);
2434 struct ext4_sb_info *sbi;
Theodore Ts'o498e5f22008-11-05 00:14:04 -05002435 __u32 ino_next;
Mingming Cao617ba132006-10-11 01:20:53 -07002436 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002437 int err = 0;
2438
Curt Wohlgemuthd3d1faf2009-09-29 11:01:03 -04002439 /* ext4_handle_valid() assumes a valid handle_t pointer */
2440 if (handle && !ext4_handle_valid(handle))
Frank Mayhar03901312009-01-07 00:06:22 -05002441 return 0;
2442
Theodore Ts'o3b9d4ed2009-04-25 22:54:04 -04002443 mutex_lock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
2444 if (list_empty(&ei->i_orphan))
2445 goto out;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002446
2447 ino_next = NEXT_ORPHAN(inode);
2448 prev = ei->i_orphan.prev;
Mingming Cao617ba132006-10-11 01:20:53 -07002449 sbi = EXT4_SB(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002450
2451 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2452
2453 list_del_init(&ei->i_orphan);
2454
2455 /* If we're on an error path, we may not have a valid
2456 * transaction handle with which to update the orphan list on
2457 * disk, but we still need to remove the inode from the linked
2458 * list in memory. */
Frank Mayhar03901312009-01-07 00:06:22 -05002459 if (sbi->s_journal && !handle)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002460 goto out;
2461
Mingming Cao617ba132006-10-11 01:20:53 -07002462 err = ext4_reserve_inode_write(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002463 if (err)
2464 goto out_err;
2465
2466 if (prev == &sbi->s_orphan) {
Theodore Ts'o498e5f22008-11-05 00:14:04 -05002467 jbd_debug(4, "superblock will point to %u\n", ino_next);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002468 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07002469 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002470 if (err)
2471 goto out_brelse;
2472 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
Artem Bityutskiyb50924c2012-07-22 20:37:31 -04002473 err = ext4_handle_dirty_super(handle, inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002474 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07002475 struct ext4_iloc iloc2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002476 struct inode *i_prev =
Mingming Cao617ba132006-10-11 01:20:53 -07002477 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002478
Theodore Ts'o498e5f22008-11-05 00:14:04 -05002479 jbd_debug(4, "orphan inode %lu will point to %u\n",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002480 i_prev->i_ino, ino_next);
Mingming Cao617ba132006-10-11 01:20:53 -07002481 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002482 if (err)
2483 goto out_brelse;
2484 NEXT_ORPHAN(i_prev) = ino_next;
Mingming Cao617ba132006-10-11 01:20:53 -07002485 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002486 }
2487 if (err)
2488 goto out_brelse;
2489 NEXT_ORPHAN(inode) = 0;
Mingming Cao617ba132006-10-11 01:20:53 -07002490 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002491
2492out_err:
Mingming Cao617ba132006-10-11 01:20:53 -07002493 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002494out:
Theodore Ts'o3b9d4ed2009-04-25 22:54:04 -04002495 mutex_unlock(&EXT4_SB(inode->i_sb)->s_orphan_lock);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002496 return err;
2497
2498out_brelse:
2499 brelse(iloc.bh);
2500 goto out_err;
2501}
2502
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002503static int ext4_rmdir(struct inode *dir, struct dentry *dentry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002504{
2505 int retval;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002506 struct inode *inode;
2507 struct buffer_head *bh;
2508 struct ext4_dir_entry_2 *de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002509 handle_t *handle;
2510
2511 /* Initialize quotas before so that eventual writes go in
2512 * separate transaction */
Christoph Hellwig871a2932010-03-03 09:05:07 -05002513 dquot_initialize(dir);
2514 dquot_initialize(dentry->d_inode);
Christoph Hellwig907f4552010-03-03 09:05:06 -05002515
Mingming Cao617ba132006-10-11 01:20:53 -07002516 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002517 if (IS_ERR(handle))
2518 return PTR_ERR(handle);
2519
2520 retval = -ENOENT;
Theodore Ts'of702ba02008-09-22 15:21:01 -04002521 bh = ext4_find_entry(dir, &dentry->d_name, &de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002522 if (!bh)
2523 goto end_rmdir;
2524
2525 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05002526 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002527
2528 inode = dentry->d_inode;
2529
2530 retval = -EIO;
2531 if (le32_to_cpu(de->inode) != inode->i_ino)
2532 goto end_rmdir;
2533
2534 retval = -ENOTEMPTY;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002535 if (!empty_dir(inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002536 goto end_rmdir;
2537
Mingming Cao617ba132006-10-11 01:20:53 -07002538 retval = ext4_delete_entry(handle, dir, de, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002539 if (retval)
2540 goto end_rmdir;
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002541 if (!EXT4_DIR_LINK_EMPTY(inode))
Eric Sandeen12062dd2010-02-15 14:19:27 -05002542 ext4_warning(inode->i_sb,
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002543 "empty directory has too many links (%d)",
2544 inode->i_nlink);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002545 inode->i_version++;
2546 clear_nlink(inode);
2547 /* There's no need to set i_disksize: the fact that i_nlink is
2548 * zero will ensure that the right thing happens during any
2549 * recovery. */
2550 inode->i_size = 0;
Mingming Cao617ba132006-10-11 01:20:53 -07002551 ext4_orphan_add(handle, inode);
Kalpak Shahef7f3832007-07-18 09:15:20 -04002552 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002553 ext4_mark_inode_dirty(handle, inode);
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002554 ext4_dec_count(handle, dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002555 ext4_update_dx_flag(dir);
2556 ext4_mark_inode_dirty(handle, dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002557
2558end_rmdir:
Mingming Cao617ba132006-10-11 01:20:53 -07002559 ext4_journal_stop(handle);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002560 brelse(bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002561 return retval;
2562}
2563
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002564static int ext4_unlink(struct inode *dir, struct dentry *dentry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002565{
2566 int retval;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002567 struct inode *inode;
2568 struct buffer_head *bh;
2569 struct ext4_dir_entry_2 *de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002570 handle_t *handle;
2571
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04002572 trace_ext4_unlink_enter(dir, dentry);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002573 /* Initialize quotas before so that eventual writes go
2574 * in separate transaction */
Christoph Hellwig871a2932010-03-03 09:05:07 -05002575 dquot_initialize(dir);
2576 dquot_initialize(dentry->d_inode);
Christoph Hellwig907f4552010-03-03 09:05:06 -05002577
Mingming Cao617ba132006-10-11 01:20:53 -07002578 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002579 if (IS_ERR(handle))
2580 return PTR_ERR(handle);
2581
2582 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05002583 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002584
2585 retval = -ENOENT;
Theodore Ts'of702ba02008-09-22 15:21:01 -04002586 bh = ext4_find_entry(dir, &dentry->d_name, &de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002587 if (!bh)
2588 goto end_unlink;
2589
2590 inode = dentry->d_inode;
2591
2592 retval = -EIO;
2593 if (le32_to_cpu(de->inode) != inode->i_ino)
2594 goto end_unlink;
2595
2596 if (!inode->i_nlink) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05002597 ext4_warning(inode->i_sb,
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002598 "Deleting nonexistent file (%lu), %d",
2599 inode->i_ino, inode->i_nlink);
Miklos Szeredibfe86842011-10-28 14:13:29 +02002600 set_nlink(inode, 1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002601 }
Mingming Cao617ba132006-10-11 01:20:53 -07002602 retval = ext4_delete_entry(handle, dir, de, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002603 if (retval)
2604 goto end_unlink;
Kalpak Shahef7f3832007-07-18 09:15:20 -04002605 dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002606 ext4_update_dx_flag(dir);
2607 ext4_mark_inode_dirty(handle, dir);
Theodore Ts'o825f1482008-02-15 15:00:38 -05002608 drop_nlink(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002609 if (!inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07002610 ext4_orphan_add(handle, inode);
Kalpak Shahef7f3832007-07-18 09:15:20 -04002611 inode->i_ctime = ext4_current_time(inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002612 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002613 retval = 0;
2614
2615end_unlink:
Mingming Cao617ba132006-10-11 01:20:53 -07002616 ext4_journal_stop(handle);
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002617 brelse(bh);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -04002618 trace_ext4_unlink_exit(dentry, retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002619 return retval;
2620}
2621
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002622static int ext4_symlink(struct inode *dir,
2623 struct dentry *dentry, const char *symname)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002624{
2625 handle_t *handle;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002626 struct inode *inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002627 int l, err, retries = 0;
Jan Karadf5e6222011-05-03 11:12:58 -04002628 int credits;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002629
2630 l = strlen(symname)+1;
2631 if (l > dir->i_sb->s_blocksize)
2632 return -ENAMETOOLONG;
2633
Christoph Hellwig871a2932010-03-03 09:05:07 -05002634 dquot_initialize(dir);
Christoph Hellwig907f4552010-03-03 09:05:06 -05002635
Jan Karadf5e6222011-05-03 11:12:58 -04002636 if (l > EXT4_N_BLOCKS * 4) {
2637 /*
2638 * For non-fast symlinks, we just allocate inode and put it on
2639 * orphan list in the first transaction => we need bitmap,
Eric Sandeen8c208712011-08-11 09:54:31 -05002640 * group descriptor, sb, inode block, quota blocks, and
2641 * possibly selinux xattr blocks.
Jan Karadf5e6222011-05-03 11:12:58 -04002642 */
Eric Sandeen8c208712011-08-11 09:54:31 -05002643 credits = 4 + EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb) +
2644 EXT4_XATTR_TRANS_BLOCKS;
Jan Karadf5e6222011-05-03 11:12:58 -04002645 } else {
2646 /*
2647 * Fast symlink. We have to add entry to directory
2648 * (EXT4_DATA_TRANS_BLOCKS + EXT4_INDEX_EXTRA_TRANS_BLOCKS),
2649 * allocate new inode (bitmap, group descriptor, inode block,
2650 * quota blocks, sb is already counted in previous macros).
2651 */
2652 credits = EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2653 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
2654 EXT4_MAXQUOTAS_INIT_BLOCKS(dir->i_sb);
2655 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002656retry:
Jan Karadf5e6222011-05-03 11:12:58 -04002657 handle = ext4_journal_start(dir, credits);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002658 if (IS_ERR(handle))
2659 return PTR_ERR(handle);
2660
2661 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05002662 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002663
Andreas Dilger11013912009-06-13 11:45:35 -04002664 inode = ext4_new_inode(handle, dir, S_IFLNK|S_IRWXUGO,
Dmitry Monakhov5cb81da2011-10-29 09:05:00 -04002665 &dentry->d_name, 0, NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002666 err = PTR_ERR(inode);
2667 if (IS_ERR(inode))
2668 goto out_stop;
2669
Jan Karadf5e6222011-05-03 11:12:58 -04002670 if (l > EXT4_N_BLOCKS * 4) {
Mingming Cao617ba132006-10-11 01:20:53 -07002671 inode->i_op = &ext4_symlink_inode_operations;
2672 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002673 /*
Jan Karadf5e6222011-05-03 11:12:58 -04002674 * We cannot call page_symlink() with transaction started
2675 * because it calls into ext4_write_begin() which can wait
2676 * for transaction commit if we are running out of space
2677 * and thus we deadlock. So we have to stop transaction now
2678 * and restart it when symlink contents is written.
2679 *
2680 * To keep fs consistent in case of crash, we have to put inode
2681 * to orphan list in the mean time.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002682 */
Jan Karadf5e6222011-05-03 11:12:58 -04002683 drop_nlink(inode);
2684 err = ext4_orphan_add(handle, inode);
2685 ext4_journal_stop(handle);
2686 if (err)
2687 goto err_drop_inode;
Nick Piggin54566b22009-01-04 12:00:53 -08002688 err = __page_symlink(inode, symname, l, 1);
Jan Karadf5e6222011-05-03 11:12:58 -04002689 if (err)
2690 goto err_drop_inode;
2691 /*
2692 * Now inode is being linked into dir (EXT4_DATA_TRANS_BLOCKS
2693 * + EXT4_INDEX_EXTRA_TRANS_BLOCKS), inode is also modified
2694 */
2695 handle = ext4_journal_start(dir,
2696 EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2697 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 1);
2698 if (IS_ERR(handle)) {
2699 err = PTR_ERR(handle);
2700 goto err_drop_inode;
2701 }
Al Viro0ce8c0102012-01-08 19:50:23 -05002702 set_nlink(inode, 1);
Jan Karadf5e6222011-05-03 11:12:58 -04002703 err = ext4_orphan_del(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002704 if (err) {
Jan Karadf5e6222011-05-03 11:12:58 -04002705 ext4_journal_stop(handle);
Theodore Ts'o825f1482008-02-15 15:00:38 -05002706 clear_nlink(inode);
Jan Karadf5e6222011-05-03 11:12:58 -04002707 goto err_drop_inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002708 }
2709 } else {
Aneesh Kumar K.Ve65187e2008-04-29 08:11:12 -04002710 /* clear the extent format for fast symlink */
Dmitry Monakhov12e9b892010-05-16 22:00:00 -04002711 ext4_clear_inode_flag(inode, EXT4_INODE_EXTENTS);
Mingming Cao617ba132006-10-11 01:20:53 -07002712 inode->i_op = &ext4_fast_symlink_inode_operations;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002713 memcpy((char *)&EXT4_I(inode)->i_data, symname, l);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002714 inode->i_size = l-1;
2715 }
Mingming Cao617ba132006-10-11 01:20:53 -07002716 EXT4_I(inode)->i_disksize = inode->i_size;
2717 err = ext4_add_nondir(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002718out_stop:
Mingming Cao617ba132006-10-11 01:20:53 -07002719 ext4_journal_stop(handle);
2720 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002721 goto retry;
2722 return err;
Jan Karadf5e6222011-05-03 11:12:58 -04002723err_drop_inode:
2724 unlock_new_inode(inode);
2725 iput(inode);
2726 return err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002727}
2728
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002729static int ext4_link(struct dentry *old_dentry,
2730 struct inode *dir, struct dentry *dentry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002731{
2732 handle_t *handle;
2733 struct inode *inode = old_dentry->d_inode;
2734 int err, retries = 0;
2735
Theodore Ts'ob05ab1d2009-08-29 21:08:08 -04002736 if (inode->i_nlink >= EXT4_LINK_MAX)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002737 return -EMLINK;
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002738
Christoph Hellwig871a2932010-03-03 09:05:07 -05002739 dquot_initialize(dir);
Christoph Hellwig907f4552010-03-03 09:05:06 -05002740
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002741retry:
Mingming Cao617ba132006-10-11 01:20:53 -07002742 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2743 EXT4_INDEX_EXTRA_TRANS_BLOCKS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002744 if (IS_ERR(handle))
2745 return PTR_ERR(handle);
2746
2747 if (IS_DIRSYNC(dir))
Frank Mayhar03901312009-01-07 00:06:22 -05002748 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002749
Kalpak Shahef7f3832007-07-18 09:15:20 -04002750 inode->i_ctime = ext4_current_time(inode);
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002751 ext4_inc_count(handle, inode);
Al Viro7de9c6ee2010-10-23 11:11:40 -04002752 ihold(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002753
Al Viro6b38e842008-12-30 02:03:31 -05002754 err = ext4_add_entry(handle, dentry, inode);
2755 if (!err) {
2756 ext4_mark_inode_dirty(handle, inode);
2757 d_instantiate(dentry, inode);
2758 } else {
2759 drop_nlink(inode);
2760 iput(inode);
2761 }
Mingming Cao617ba132006-10-11 01:20:53 -07002762 ext4_journal_stop(handle);
2763 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002764 goto retry;
2765 return err;
2766}
2767
Wei Yongjun3d0518f2009-02-14 23:01:36 -05002768#define PARENT_INO(buffer, size) \
2769 (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer), size)->inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002770
2771/*
2772 * Anybody can rename anything with this: the permission checks are left to the
2773 * higher-level routines.
2774 */
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002775static int ext4_rename(struct inode *old_dir, struct dentry *old_dentry,
2776 struct inode *new_dir, struct dentry *new_dentry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002777{
2778 handle_t *handle;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002779 struct inode *old_inode, *new_inode;
2780 struct buffer_head *old_bh, *new_bh, *dir_bh;
2781 struct ext4_dir_entry_2 *old_de, *new_de;
Theodore Ts'o8750c6d2009-02-23 23:05:27 -05002782 int retval, force_da_alloc = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002783
Christoph Hellwig871a2932010-03-03 09:05:07 -05002784 dquot_initialize(old_dir);
2785 dquot_initialize(new_dir);
Christoph Hellwig907f4552010-03-03 09:05:06 -05002786
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002787 old_bh = new_bh = dir_bh = NULL;
2788
2789 /* Initialize quotas before so that eventual writes go
2790 * in separate transaction */
2791 if (new_dentry->d_inode)
Christoph Hellwig871a2932010-03-03 09:05:07 -05002792 dquot_initialize(new_dentry->d_inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002793 handle = ext4_journal_start(old_dir, 2 *
2794 EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2795 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002796 if (IS_ERR(handle))
2797 return PTR_ERR(handle);
2798
2799 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
Frank Mayhar03901312009-01-07 00:06:22 -05002800 ext4_handle_sync(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002801
Theodore Ts'of702ba02008-09-22 15:21:01 -04002802 old_bh = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002803 /*
2804 * Check for inode number is _not_ due to possible IO errors.
2805 * We might rmdir the source, keep it as pwd of some process
2806 * and merrily kill the link to whatever was created under the
2807 * same name. Goodbye sticky bit ;-<
2808 */
2809 old_inode = old_dentry->d_inode;
2810 retval = -ENOENT;
2811 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2812 goto end_rename;
2813
2814 new_inode = new_dentry->d_inode;
Theodore Ts'of702ba02008-09-22 15:21:01 -04002815 new_bh = ext4_find_entry(new_dir, &new_dentry->d_name, &new_de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002816 if (new_bh) {
2817 if (!new_inode) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002818 brelse(new_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002819 new_bh = NULL;
2820 }
2821 }
2822 if (S_ISDIR(old_inode->i_mode)) {
2823 if (new_inode) {
2824 retval = -ENOTEMPTY;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002825 if (!empty_dir(new_inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002826 goto end_rename;
2827 }
2828 retval = -EIO;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002829 dir_bh = ext4_bread(handle, old_inode, 0, 0, &retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002830 if (!dir_bh)
2831 goto end_rename;
Darrick J. Wongb0336e82012-04-29 18:41:10 -04002832 if (!buffer_verified(dir_bh) &&
2833 !ext4_dirent_csum_verify(old_inode,
2834 (struct ext4_dir_entry *)dir_bh->b_data))
2835 goto end_rename;
2836 set_buffer_verified(dir_bh);
Wei Yongjun3d0518f2009-02-14 23:01:36 -05002837 if (le32_to_cpu(PARENT_INO(dir_bh->b_data,
2838 old_dir->i_sb->s_blocksize)) != old_dir->i_ino)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002839 goto end_rename;
2840 retval = -EMLINK;
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002841 if (!new_inode && new_dir != old_dir &&
Aneesh Kumar K.V2c94eb82009-08-28 21:43:15 -04002842 EXT4_DIR_LINK_MAX(new_dir))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002843 goto end_rename;
Amir Goldsteinef607892011-03-20 21:18:44 -04002844 BUFFER_TRACE(dir_bh, "get_write_access");
2845 retval = ext4_journal_get_write_access(handle, dir_bh);
2846 if (retval)
2847 goto end_rename;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002848 }
2849 if (!new_bh) {
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002850 retval = ext4_add_entry(handle, new_dentry, old_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002851 if (retval)
2852 goto end_rename;
2853 } else {
2854 BUFFER_TRACE(new_bh, "get write access");
Amir Goldsteinef607892011-03-20 21:18:44 -04002855 retval = ext4_journal_get_write_access(handle, new_bh);
2856 if (retval)
2857 goto end_rename;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002858 new_de->inode = cpu_to_le32(old_inode->i_ino);
Mingming Cao617ba132006-10-11 01:20:53 -07002859 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2860 EXT4_FEATURE_INCOMPAT_FILETYPE))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002861 new_de->file_type = old_de->file_type;
2862 new_dir->i_version++;
Jan Kara53b7e9f2008-04-29 22:02:11 -04002863 new_dir->i_ctime = new_dir->i_mtime =
2864 ext4_current_time(new_dir);
2865 ext4_mark_inode_dirty(handle, new_dir);
Frank Mayhar03901312009-01-07 00:06:22 -05002866 BUFFER_TRACE(new_bh, "call ext4_handle_dirty_metadata");
Darrick J. Wongb0336e82012-04-29 18:41:10 -04002867 retval = ext4_handle_dirty_dirent_node(handle, new_dir, new_bh);
Theodore Ts'ob4097142011-01-10 12:46:59 -05002868 if (unlikely(retval)) {
2869 ext4_std_error(new_dir->i_sb, retval);
2870 goto end_rename;
2871 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002872 brelse(new_bh);
2873 new_bh = NULL;
2874 }
2875
2876 /*
2877 * Like most other Unix systems, set the ctime for inodes on a
2878 * rename.
2879 */
Kalpak Shahef7f3832007-07-18 09:15:20 -04002880 old_inode->i_ctime = ext4_current_time(old_inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002881 ext4_mark_inode_dirty(handle, old_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002882
2883 /*
2884 * ok, that's it
2885 */
2886 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2887 old_de->name_len != old_dentry->d_name.len ||
2888 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
Mingming Cao617ba132006-10-11 01:20:53 -07002889 (retval = ext4_delete_entry(handle, old_dir,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002890 old_de, old_bh)) == -ENOENT) {
2891 /* old_de could have moved from under us during htree split, so
2892 * make sure that we are deleting the right entry. We might
2893 * also be pointing to a stale entry in the unused part of
2894 * old_bh so just checking inum and the name isn't enough. */
2895 struct buffer_head *old_bh2;
Mingming Cao617ba132006-10-11 01:20:53 -07002896 struct ext4_dir_entry_2 *old_de2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002897
Theodore Ts'of702ba02008-09-22 15:21:01 -04002898 old_bh2 = ext4_find_entry(old_dir, &old_dentry->d_name, &old_de2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002899 if (old_bh2) {
Mingming Cao617ba132006-10-11 01:20:53 -07002900 retval = ext4_delete_entry(handle, old_dir,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002901 old_de2, old_bh2);
2902 brelse(old_bh2);
2903 }
2904 }
2905 if (retval) {
Eric Sandeen12062dd2010-02-15 14:19:27 -05002906 ext4_warning(old_dir->i_sb,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002907 "Deleting old file (%lu), %d, error=%d",
2908 old_dir->i_ino, old_dir->i_nlink, retval);
2909 }
2910
2911 if (new_inode) {
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002912 ext4_dec_count(handle, new_inode);
Kalpak Shahef7f3832007-07-18 09:15:20 -04002913 new_inode->i_ctime = ext4_current_time(new_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002914 }
Kalpak Shahef7f3832007-07-18 09:15:20 -04002915 old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002916 ext4_update_dx_flag(old_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002917 if (dir_bh) {
Wei Yongjun3d0518f2009-02-14 23:01:36 -05002918 PARENT_INO(dir_bh->b_data, new_dir->i_sb->s_blocksize) =
2919 cpu_to_le32(new_dir->i_ino);
Frank Mayhar03901312009-01-07 00:06:22 -05002920 BUFFER_TRACE(dir_bh, "call ext4_handle_dirty_metadata");
Tao Maef58f692012-07-09 16:29:05 -04002921 if (is_dx(old_inode)) {
2922 retval = ext4_handle_dirty_dx_node(handle,
2923 old_inode,
2924 dir_bh);
2925 } else {
2926 retval = ext4_handle_dirty_dirent_node(handle,
2927 old_inode,
2928 dir_bh);
2929 }
Theodore Ts'ob4097142011-01-10 12:46:59 -05002930 if (retval) {
2931 ext4_std_error(old_dir->i_sb, retval);
2932 goto end_rename;
2933 }
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002934 ext4_dec_count(handle, old_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002935 if (new_inode) {
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002936 /* checked empty_dir above, can't have another parent,
Theodore Ts'o825f1482008-02-15 15:00:38 -05002937 * ext4_dec_count() won't work for many-linked dirs */
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +02002938 clear_nlink(new_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002939 } else {
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002940 ext4_inc_count(handle, new_dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002941 ext4_update_dx_flag(new_dir);
2942 ext4_mark_inode_dirty(handle, new_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002943 }
2944 }
Mingming Cao617ba132006-10-11 01:20:53 -07002945 ext4_mark_inode_dirty(handle, old_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002946 if (new_inode) {
Mingming Cao617ba132006-10-11 01:20:53 -07002947 ext4_mark_inode_dirty(handle, new_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002948 if (!new_inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07002949 ext4_orphan_add(handle, new_inode);
Theodore Ts'oafd46722009-03-16 23:12:23 -04002950 if (!test_opt(new_dir->i_sb, NO_AUTO_DA_ALLOC))
2951 force_da_alloc = 1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002952 }
2953 retval = 0;
2954
2955end_rename:
Theodore Ts'oaf5bc922008-09-08 22:25:24 -04002956 brelse(dir_bh);
2957 brelse(old_bh);
2958 brelse(new_bh);
Mingming Cao617ba132006-10-11 01:20:53 -07002959 ext4_journal_stop(handle);
Theodore Ts'o8750c6d2009-02-23 23:05:27 -05002960 if (retval == 0 && force_da_alloc)
2961 ext4_alloc_da_blocks(old_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002962 return retval;
2963}
2964
2965/*
2966 * directories can handle most operations...
2967 */
Arjan van de Ven754661f2007-02-12 00:55:38 -08002968const struct inode_operations ext4_dir_inode_operations = {
Mingming Cao617ba132006-10-11 01:20:53 -07002969 .create = ext4_create,
2970 .lookup = ext4_lookup,
2971 .link = ext4_link,
2972 .unlink = ext4_unlink,
2973 .symlink = ext4_symlink,
2974 .mkdir = ext4_mkdir,
2975 .rmdir = ext4_rmdir,
2976 .mknod = ext4_mknod,
2977 .rename = ext4_rename,
2978 .setattr = ext4_setattr,
Theodore Ts'o03010a32008-10-10 20:02:48 -04002979#ifdef CONFIG_EXT4_FS_XATTR
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002980 .setxattr = generic_setxattr,
2981 .getxattr = generic_getxattr,
Mingming Cao617ba132006-10-11 01:20:53 -07002982 .listxattr = ext4_listxattr,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002983 .removexattr = generic_removexattr,
2984#endif
Christoph Hellwig4e34e712011-07-23 17:37:31 +02002985 .get_acl = ext4_get_acl,
Aneesh Kumar K.Vabc87462009-05-02 22:54:32 -04002986 .fiemap = ext4_fiemap,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002987};
2988
Arjan van de Ven754661f2007-02-12 00:55:38 -08002989const struct inode_operations ext4_special_inode_operations = {
Mingming Cao617ba132006-10-11 01:20:53 -07002990 .setattr = ext4_setattr,
Theodore Ts'o03010a32008-10-10 20:02:48 -04002991#ifdef CONFIG_EXT4_FS_XATTR
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002992 .setxattr = generic_setxattr,
2993 .getxattr = generic_getxattr,
Mingming Cao617ba132006-10-11 01:20:53 -07002994 .listxattr = ext4_listxattr,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002995 .removexattr = generic_removexattr,
2996#endif
Christoph Hellwig4e34e712011-07-23 17:37:31 +02002997 .get_acl = ext4_get_acl,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002998};