blob: fb673b14ccd5db7ea422253a63fb8e73b1a79159 [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>
Mingming Cao617ba132006-10-11 01:20:53 -070031#include <linux/ext4_fs.h>
Mingming Caodab291a2006-10-11 01:21:01 -070032#include <linux/ext4_jbd2.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070033#include <linux/fcntl.h>
34#include <linux/stat.h>
35#include <linux/string.h>
36#include <linux/quotaops.h>
37#include <linux/buffer_head.h>
38#include <linux/bio.h>
Dave Kleikampac27a0e2006-10-11 01:20:50 -070039
40#include "namei.h"
41#include "xattr.h"
42#include "acl.h"
43
44/*
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
Mingming Cao617ba132006-10-11 01:20:53 -070060 if ((bh = ext4_bread(handle, inode, *block, 1, err))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -070061 inode->i_size += inode->i_sb->s_blocksize;
Mingming Cao617ba132006-10-11 01:20:53 -070062 EXT4_I(inode)->i_disksize = inode->i_size;
63 ext4_journal_get_write_access(handle,bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -070064 }
65 return bh;
66}
67
68#ifndef assert
69#define assert(test) J_ASSERT(test)
70#endif
71
72#ifndef swap
73#define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0)
74#endif
75
76#ifdef DX_DEBUG
77#define dxtrace(command) command
78#else
79#define dxtrace(command)
80#endif
81
82struct fake_dirent
83{
84 __le32 inode;
85 __le16 rec_len;
86 u8 name_len;
87 u8 file_type;
88};
89
90struct dx_countlimit
91{
92 __le16 limit;
93 __le16 count;
94};
95
96struct dx_entry
97{
98 __le32 hash;
99 __le32 block;
100};
101
102/*
103 * dx_root_info is laid out so that if it should somehow get overlaid by a
104 * dirent the two low bits of the hash version will be zero. Therefore, the
105 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
106 */
107
108struct dx_root
109{
110 struct fake_dirent dot;
111 char dot_name[4];
112 struct fake_dirent dotdot;
113 char dotdot_name[4];
114 struct dx_root_info
115 {
116 __le32 reserved_zero;
117 u8 hash_version;
118 u8 info_length; /* 8 */
119 u8 indirect_levels;
120 u8 unused_flags;
121 }
122 info;
123 struct dx_entry entries[0];
124};
125
126struct dx_node
127{
128 struct fake_dirent fake;
129 struct dx_entry entries[0];
130};
131
132
133struct dx_frame
134{
135 struct buffer_head *bh;
136 struct dx_entry *entries;
137 struct dx_entry *at;
138};
139
140struct dx_map_entry
141{
142 u32 hash;
Eric Sandeenef2b02d2007-09-18 22:46:42 -0700143 u16 offs;
144 u16 size;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700145};
146
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500147static inline ext4_lblk_t dx_get_block(struct dx_entry *entry);
148static void dx_set_block(struct dx_entry *entry, ext4_lblk_t value);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700149static inline unsigned dx_get_hash (struct dx_entry *entry);
150static void dx_set_hash (struct dx_entry *entry, unsigned value);
151static unsigned dx_get_count (struct dx_entry *entries);
152static unsigned dx_get_limit (struct dx_entry *entries);
153static void dx_set_count (struct dx_entry *entries, unsigned value);
154static void dx_set_limit (struct dx_entry *entries, unsigned value);
155static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
156static unsigned dx_node_limit (struct inode *dir);
157static struct dx_frame *dx_probe(struct dentry *dentry,
158 struct inode *dir,
159 struct dx_hash_info *hinfo,
160 struct dx_frame *frame,
161 int *err);
162static void dx_release (struct dx_frame *frames);
Mingming Cao617ba132006-10-11 01:20:53 -0700163static int dx_make_map (struct ext4_dir_entry_2 *de, int size,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700164 struct dx_hash_info *hinfo, struct dx_map_entry map[]);
165static void dx_sort_map(struct dx_map_entry *map, unsigned count);
Mingming Cao617ba132006-10-11 01:20:53 -0700166static struct ext4_dir_entry_2 *dx_move_dirents (char *from, char *to,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700167 struct dx_map_entry *offsets, int count);
Mingming Cao617ba132006-10-11 01:20:53 -0700168static struct ext4_dir_entry_2* dx_pack_dirents (char *base, int size);
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500169static void dx_insert_block(struct dx_frame *frame,
170 u32 hash, ext4_lblk_t block);
Mingming Cao617ba132006-10-11 01:20:53 -0700171static int ext4_htree_next_block(struct inode *dir, __u32 hash,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700172 struct dx_frame *frame,
173 struct dx_frame *frames,
174 __u32 *start_hash);
Mingming Cao617ba132006-10-11 01:20:53 -0700175static struct buffer_head * ext4_dx_find_entry(struct dentry *dentry,
176 struct ext4_dir_entry_2 **res_dir, int *err);
177static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700178 struct inode *inode);
179
180/*
181 * Future: use high four bits of block for coalesce-on-delete flags
182 * Mask them off for now.
183 */
184
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500185static inline ext4_lblk_t dx_get_block(struct dx_entry *entry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700186{
187 return le32_to_cpu(entry->block) & 0x00ffffff;
188}
189
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500190static inline void dx_set_block(struct dx_entry *entry, ext4_lblk_t value)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700191{
192 entry->block = cpu_to_le32(value);
193}
194
195static inline unsigned dx_get_hash (struct dx_entry *entry)
196{
197 return le32_to_cpu(entry->hash);
198}
199
200static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
201{
202 entry->hash = cpu_to_le32(value);
203}
204
205static inline unsigned dx_get_count (struct dx_entry *entries)
206{
207 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
208}
209
210static inline unsigned dx_get_limit (struct dx_entry *entries)
211{
212 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
213}
214
215static inline void dx_set_count (struct dx_entry *entries, unsigned value)
216{
217 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
218}
219
220static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
221{
222 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
223}
224
225static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
226{
Mingming Cao617ba132006-10-11 01:20:53 -0700227 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
228 EXT4_DIR_REC_LEN(2) - infosize;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700229 return 0? 20: entry_space / sizeof(struct dx_entry);
230}
231
232static inline unsigned dx_node_limit (struct inode *dir)
233{
Mingming Cao617ba132006-10-11 01:20:53 -0700234 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700235 return 0? 22: entry_space / sizeof(struct dx_entry);
236}
237
238/*
239 * Debug
240 */
241#ifdef DX_DEBUG
242static void dx_show_index (char * label, struct dx_entry *entries)
243{
Andrew Morton63f57932006-10-11 01:21:24 -0700244 int i, n = dx_get_count (entries);
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400245 printk("%s index ", label);
Andrew Morton63f57932006-10-11 01:21:24 -0700246 for (i = 0; i < n; i++) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500247 printk("%x->%lu ", i? dx_get_hash(entries + i) :
248 0, (unsigned long)dx_get_block(entries + i));
Andrew Morton63f57932006-10-11 01:21:24 -0700249 }
250 printk("\n");
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700251}
252
253struct stats
254{
255 unsigned names;
256 unsigned space;
257 unsigned bcount;
258};
259
Mingming Cao617ba132006-10-11 01:20:53 -0700260static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700261 int size, int show_names)
262{
263 unsigned names = 0, space = 0;
264 char *base = (char *) de;
265 struct dx_hash_info h = *hinfo;
266
267 printk("names: ");
268 while ((char *) de < base + size)
269 {
270 if (de->inode)
271 {
272 if (show_names)
273 {
274 int len = de->name_len;
275 char *name = de->name;
276 while (len--) printk("%c", *name++);
Mingming Cao617ba132006-10-11 01:20:53 -0700277 ext4fs_dirhash(de->name, de->name_len, &h);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700278 printk(":%x.%u ", h.hash,
279 ((char *) de - base));
280 }
Mingming Cao617ba132006-10-11 01:20:53 -0700281 space += EXT4_DIR_REC_LEN(de->name_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700282 names++;
283 }
Jan Karaa72d7f82008-01-28 23:58:27 -0500284 de = ext4_next_entry(de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700285 }
286 printk("(%i)\n", names);
287 return (struct stats) { names, space, 1 };
288}
289
290struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
291 struct dx_entry *entries, int levels)
292{
293 unsigned blocksize = dir->i_sb->s_blocksize;
294 unsigned count = dx_get_count (entries), names = 0, space = 0, i;
295 unsigned bcount = 0;
296 struct buffer_head *bh;
297 int err;
298 printk("%i indexed blocks...\n", count);
299 for (i = 0; i < count; i++, entries++)
300 {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500301 ext4_lblk_t block = dx_get_block(entries);
302 ext4_lblk_t hash = i ? dx_get_hash(entries): 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700303 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
304 struct stats stats;
305 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
Mingming Cao617ba132006-10-11 01:20:53 -0700306 if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700307 stats = levels?
308 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
Mingming Cao617ba132006-10-11 01:20:53 -0700309 dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700310 names += stats.names;
311 space += stats.space;
312 bcount += stats.bcount;
313 brelse (bh);
314 }
315 if (bcount)
316 printk("%snames %u, fullness %u (%u%%)\n", levels?"":" ",
317 names, space/bcount,(space/bcount)*100/blocksize);
318 return (struct stats) { names, space, bcount};
319}
320#endif /* DX_DEBUG */
321
322/*
323 * Probe for a directory leaf block to search.
324 *
325 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
326 * error in the directory index, and the caller should fall back to
327 * searching the directory normally. The callers of dx_probe **MUST**
328 * check for this error code, and make sure it never gets reflected
329 * back to userspace.
330 */
331static struct dx_frame *
332dx_probe(struct dentry *dentry, struct inode *dir,
333 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
334{
335 unsigned count, indirect;
336 struct dx_entry *at, *entries, *p, *q, *m;
337 struct dx_root *root;
338 struct buffer_head *bh;
339 struct dx_frame *frame = frame_in;
340 u32 hash;
341
342 frame->bh = NULL;
343 if (dentry)
344 dir = dentry->d_parent->d_inode;
Mingming Cao617ba132006-10-11 01:20:53 -0700345 if (!(bh = ext4_bread (NULL,dir, 0, 0, err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700346 goto fail;
347 root = (struct dx_root *) bh->b_data;
348 if (root->info.hash_version != DX_HASH_TEA &&
349 root->info.hash_version != DX_HASH_HALF_MD4 &&
350 root->info.hash_version != DX_HASH_LEGACY) {
Mingming Cao617ba132006-10-11 01:20:53 -0700351 ext4_warning(dir->i_sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700352 "Unrecognised inode hash code %d",
353 root->info.hash_version);
354 brelse(bh);
355 *err = ERR_BAD_DX_DIR;
356 goto fail;
357 }
358 hinfo->hash_version = root->info.hash_version;
Mingming Cao617ba132006-10-11 01:20:53 -0700359 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700360 if (dentry)
Mingming Cao617ba132006-10-11 01:20:53 -0700361 ext4fs_dirhash(dentry->d_name.name, dentry->d_name.len, hinfo);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700362 hash = hinfo->hash;
363
364 if (root->info.unused_flags & 1) {
Mingming Cao617ba132006-10-11 01:20:53 -0700365 ext4_warning(dir->i_sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700366 "Unimplemented inode hash flags: %#06x",
367 root->info.unused_flags);
368 brelse(bh);
369 *err = ERR_BAD_DX_DIR;
370 goto fail;
371 }
372
373 if ((indirect = root->info.indirect_levels) > 1) {
Mingming Cao617ba132006-10-11 01:20:53 -0700374 ext4_warning(dir->i_sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700375 "Unimplemented inode hash depth: %#06x",
376 root->info.indirect_levels);
377 brelse(bh);
378 *err = ERR_BAD_DX_DIR;
379 goto fail;
380 }
381
382 entries = (struct dx_entry *) (((char *)&root->info) +
383 root->info.info_length);
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700384
385 if (dx_get_limit(entries) != dx_root_limit(dir,
386 root->info.info_length)) {
387 ext4_warning(dir->i_sb, __FUNCTION__,
388 "dx entry: limit != root limit");
389 brelse(bh);
390 *err = ERR_BAD_DX_DIR;
391 goto fail;
392 }
393
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700394 dxtrace (printk("Look up %x", hash));
395 while (1)
396 {
397 count = dx_get_count(entries);
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700398 if (!count || count > dx_get_limit(entries)) {
399 ext4_warning(dir->i_sb, __FUNCTION__,
400 "dx entry: no count or count > limit");
401 brelse(bh);
402 *err = ERR_BAD_DX_DIR;
403 goto fail2;
404 }
405
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700406 p = entries + 1;
407 q = entries + count - 1;
408 while (p <= q)
409 {
410 m = p + (q - p)/2;
411 dxtrace(printk("."));
412 if (dx_get_hash(m) > hash)
413 q = m - 1;
414 else
415 p = m + 1;
416 }
417
418 if (0) // linear search cross check
419 {
420 unsigned n = count - 1;
421 at = entries;
422 while (n--)
423 {
424 dxtrace(printk(","));
425 if (dx_get_hash(++at) > hash)
426 {
427 at--;
428 break;
429 }
430 }
431 assert (at == p - 1);
432 }
433
434 at = p - 1;
435 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
436 frame->bh = bh;
437 frame->entries = entries;
438 frame->at = at;
439 if (!indirect--) return frame;
Mingming Cao617ba132006-10-11 01:20:53 -0700440 if (!(bh = ext4_bread (NULL,dir, dx_get_block(at), 0, err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700441 goto fail2;
442 at = entries = ((struct dx_node *) bh->b_data)->entries;
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700443 if (dx_get_limit(entries) != dx_node_limit (dir)) {
444 ext4_warning(dir->i_sb, __FUNCTION__,
445 "dx entry: limit != node limit");
446 brelse(bh);
447 *err = ERR_BAD_DX_DIR;
448 goto fail2;
449 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700450 frame++;
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700451 frame->bh = NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700452 }
453fail2:
454 while (frame >= frame_in) {
455 brelse(frame->bh);
456 frame--;
457 }
458fail:
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700459 if (*err == ERR_BAD_DX_DIR)
460 ext4_warning(dir->i_sb, __FUNCTION__,
461 "Corrupt dir inode %ld, running e2fsck is "
462 "recommended.", dir->i_ino);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700463 return NULL;
464}
465
466static void dx_release (struct dx_frame *frames)
467{
468 if (frames[0].bh == NULL)
469 return;
470
471 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
472 brelse(frames[1].bh);
473 brelse(frames[0].bh);
474}
475
476/*
477 * This function increments the frame pointer to search the next leaf
478 * block, and reads in the necessary intervening nodes if the search
479 * should be necessary. Whether or not the search is necessary is
480 * controlled by the hash parameter. If the hash value is even, then
481 * the search is only continued if the next block starts with that
482 * hash value. This is used if we are searching for a specific file.
483 *
484 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
485 *
486 * This function returns 1 if the caller should continue to search,
487 * or 0 if it should not. If there is an error reading one of the
488 * index blocks, it will a negative error code.
489 *
490 * If start_hash is non-null, it will be filled in with the starting
491 * hash of the next page.
492 */
Mingming Cao617ba132006-10-11 01:20:53 -0700493static int ext4_htree_next_block(struct inode *dir, __u32 hash,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700494 struct dx_frame *frame,
495 struct dx_frame *frames,
496 __u32 *start_hash)
497{
498 struct dx_frame *p;
499 struct buffer_head *bh;
500 int err, num_frames = 0;
501 __u32 bhash;
502
503 p = frame;
504 /*
505 * Find the next leaf page by incrementing the frame pointer.
506 * If we run out of entries in the interior node, loop around and
507 * increment pointer in the parent node. When we break out of
508 * this loop, num_frames indicates the number of interior
509 * nodes need to be read.
510 */
511 while (1) {
512 if (++(p->at) < p->entries + dx_get_count(p->entries))
513 break;
514 if (p == frames)
515 return 0;
516 num_frames++;
517 p--;
518 }
519
520 /*
521 * If the hash is 1, then continue only if the next page has a
522 * continuation hash of any value. This is used for readdir
523 * handling. Otherwise, check to see if the hash matches the
524 * desired contiuation hash. If it doesn't, return since
525 * there's no point to read in the successive index pages.
526 */
527 bhash = dx_get_hash(p->at);
528 if (start_hash)
529 *start_hash = bhash;
530 if ((hash & 1) == 0) {
531 if ((bhash & ~1) != hash)
532 return 0;
533 }
534 /*
535 * If the hash is HASH_NB_ALWAYS, we always go to the next
536 * block so no check is necessary
537 */
538 while (num_frames--) {
Mingming Cao617ba132006-10-11 01:20:53 -0700539 if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at),
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700540 0, &err)))
541 return err; /* Failure */
542 p++;
543 brelse (p->bh);
544 p->bh = bh;
545 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
546 }
547 return 1;
548}
549
550
551/*
552 * p is at least 6 bytes before the end of page
553 */
Mingming Cao617ba132006-10-11 01:20:53 -0700554static inline struct ext4_dir_entry_2 *ext4_next_entry(struct ext4_dir_entry_2 *p)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700555{
Jan Karaa72d7f82008-01-28 23:58:27 -0500556 return (struct ext4_dir_entry_2 *)((char *)p +
557 ext4_rec_len_from_disk(p->rec_len));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700558}
559
560/*
561 * This function fills a red-black tree with information from a
562 * directory block. It returns the number directory entries loaded
563 * into the tree. If there is an error it is returned in err.
564 */
565static int htree_dirblock_to_tree(struct file *dir_file,
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500566 struct inode *dir, ext4_lblk_t block,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700567 struct dx_hash_info *hinfo,
568 __u32 start_hash, __u32 start_minor_hash)
569{
570 struct buffer_head *bh;
Mingming Cao617ba132006-10-11 01:20:53 -0700571 struct ext4_dir_entry_2 *de, *top;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700572 int err, count = 0;
573
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500574 dxtrace(printk(KERN_INFO "In htree dirblock_to_tree: block %lu\n",
575 (unsigned long)block));
Mingming Cao617ba132006-10-11 01:20:53 -0700576 if (!(bh = ext4_bread (NULL, dir, block, 0, &err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700577 return err;
578
Mingming Cao617ba132006-10-11 01:20:53 -0700579 de = (struct ext4_dir_entry_2 *) bh->b_data;
580 top = (struct ext4_dir_entry_2 *) ((char *) de +
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700581 dir->i_sb->s_blocksize -
Mingming Cao617ba132006-10-11 01:20:53 -0700582 EXT4_DIR_REC_LEN(0));
583 for (; de < top; de = ext4_next_entry(de)) {
Eric Sandeene6c40212006-12-06 20:36:28 -0800584 if (!ext4_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
585 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
586 +((char *)de - bh->b_data))) {
587 /* On error, skip the f_pos to the next block. */
588 dir_file->f_pos = (dir_file->f_pos |
589 (dir->i_sb->s_blocksize - 1)) + 1;
590 brelse (bh);
591 return count;
592 }
Mingming Cao617ba132006-10-11 01:20:53 -0700593 ext4fs_dirhash(de->name, de->name_len, hinfo);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700594 if ((hinfo->hash < start_hash) ||
595 ((hinfo->hash == start_hash) &&
596 (hinfo->minor_hash < start_minor_hash)))
597 continue;
598 if (de->inode == 0)
599 continue;
Mingming Cao617ba132006-10-11 01:20:53 -0700600 if ((err = ext4_htree_store_dirent(dir_file,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700601 hinfo->hash, hinfo->minor_hash, de)) != 0) {
602 brelse(bh);
603 return err;
604 }
605 count++;
606 }
607 brelse(bh);
608 return count;
609}
610
611
612/*
613 * This function fills a red-black tree with information from a
614 * directory. We start scanning the directory in hash order, starting
615 * at start_hash and start_minor_hash.
616 *
617 * This function returns the number of entries inserted into the tree,
618 * or a negative error code.
619 */
Mingming Cao617ba132006-10-11 01:20:53 -0700620int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700621 __u32 start_minor_hash, __u32 *next_hash)
622{
623 struct dx_hash_info hinfo;
Mingming Cao617ba132006-10-11 01:20:53 -0700624 struct ext4_dir_entry_2 *de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700625 struct dx_frame frames[2], *frame;
626 struct inode *dir;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500627 ext4_lblk_t block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700628 int count = 0;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500629 int ret, err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700630 __u32 hashval;
631
632 dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
633 start_minor_hash));
Josef "Jeff" Sipek9d549892006-12-08 02:36:38 -0800634 dir = dir_file->f_path.dentry->d_inode;
Mingming Cao617ba132006-10-11 01:20:53 -0700635 if (!(EXT4_I(dir)->i_flags & EXT4_INDEX_FL)) {
636 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
637 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700638 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
639 start_hash, start_minor_hash);
640 *next_hash = ~0;
641 return count;
642 }
643 hinfo.hash = start_hash;
644 hinfo.minor_hash = 0;
Josef "Jeff" Sipek9d549892006-12-08 02:36:38 -0800645 frame = dx_probe(NULL, dir_file->f_path.dentry->d_inode, &hinfo, frames, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700646 if (!frame)
647 return err;
648
649 /* Add '.' and '..' from the htree header */
650 if (!start_hash && !start_minor_hash) {
Mingming Cao617ba132006-10-11 01:20:53 -0700651 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
652 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700653 goto errout;
654 count++;
655 }
656 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
Mingming Cao617ba132006-10-11 01:20:53 -0700657 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
658 de = ext4_next_entry(de);
659 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700660 goto errout;
661 count++;
662 }
663
664 while (1) {
665 block = dx_get_block(frame->at);
666 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
667 start_hash, start_minor_hash);
668 if (ret < 0) {
669 err = ret;
670 goto errout;
671 }
672 count += ret;
673 hashval = ~0;
Mingming Cao617ba132006-10-11 01:20:53 -0700674 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700675 frame, frames, &hashval);
676 *next_hash = hashval;
677 if (ret < 0) {
678 err = ret;
679 goto errout;
680 }
681 /*
682 * Stop if: (a) there are no more entries, or
683 * (b) we have inserted at least one entry and the
684 * next hash value is not a continuation
685 */
686 if ((ret == 0) ||
687 (count && ((hashval & 1) == 0)))
688 break;
689 }
690 dx_release(frames);
691 dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
692 count, *next_hash));
693 return count;
694errout:
695 dx_release(frames);
696 return (err);
697}
698
699
700/*
701 * Directory block splitting, compacting
702 */
703
Eric Sandeenef2b02d2007-09-18 22:46:42 -0700704/*
705 * Create map of hash values, offsets, and sizes, stored at end of block.
706 * Returns number of entries mapped.
707 */
Mingming Cao617ba132006-10-11 01:20:53 -0700708static int dx_make_map (struct ext4_dir_entry_2 *de, int size,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700709 struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
710{
711 int count = 0;
712 char *base = (char *) de;
713 struct dx_hash_info h = *hinfo;
714
715 while ((char *) de < base + size)
716 {
717 if (de->name_len && de->inode) {
Mingming Cao617ba132006-10-11 01:20:53 -0700718 ext4fs_dirhash(de->name, de->name_len, &h);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700719 map_tail--;
720 map_tail->hash = h.hash;
Eric Sandeenef2b02d2007-09-18 22:46:42 -0700721 map_tail->offs = (u16) ((char *) de - base);
722 map_tail->size = le16_to_cpu(de->rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700723 count++;
724 cond_resched();
725 }
726 /* XXX: do we need to check rec_len == 0 case? -Chris */
Jan Karaa72d7f82008-01-28 23:58:27 -0500727 de = ext4_next_entry(de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700728 }
729 return count;
730}
731
Eric Sandeenef2b02d2007-09-18 22:46:42 -0700732/* Sort map by hash value */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700733static void dx_sort_map (struct dx_map_entry *map, unsigned count)
734{
Andrew Morton63f57932006-10-11 01:21:24 -0700735 struct dx_map_entry *p, *q, *top = map + count - 1;
736 int more;
737 /* Combsort until bubble sort doesn't suck */
738 while (count > 2) {
739 count = count*10/13;
740 if (count - 9 < 2) /* 9, 10 -> 11 */
741 count = 11;
742 for (p = top, q = p - count; q >= map; p--, q--)
743 if (p->hash < q->hash)
744 swap(*p, *q);
745 }
746 /* Garden variety bubble sort */
747 do {
748 more = 0;
749 q = top;
750 while (q-- > map) {
751 if (q[1].hash >= q[0].hash)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700752 continue;
Andrew Morton63f57932006-10-11 01:21:24 -0700753 swap(*(q+1), *q);
754 more = 1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700755 }
756 } while(more);
757}
758
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500759static void dx_insert_block(struct dx_frame *frame, u32 hash, ext4_lblk_t block)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700760{
761 struct dx_entry *entries = frame->entries;
762 struct dx_entry *old = frame->at, *new = old + 1;
763 int count = dx_get_count(entries);
764
765 assert(count < dx_get_limit(entries));
766 assert(old < entries + count);
767 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
768 dx_set_hash(new, hash);
769 dx_set_block(new, block);
770 dx_set_count(entries, count + 1);
771}
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700772
Mingming Cao617ba132006-10-11 01:20:53 -0700773static void ext4_update_dx_flag(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700774{
Mingming Cao617ba132006-10-11 01:20:53 -0700775 if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
776 EXT4_FEATURE_COMPAT_DIR_INDEX))
777 EXT4_I(inode)->i_flags &= ~EXT4_INDEX_FL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700778}
779
780/*
Mingming Cao617ba132006-10-11 01:20:53 -0700781 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700782 *
Mingming Cao617ba132006-10-11 01:20:53 -0700783 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700784 * `de != NULL' is guaranteed by caller.
785 */
Mingming Cao617ba132006-10-11 01:20:53 -0700786static inline int ext4_match (int len, const char * const name,
787 struct ext4_dir_entry_2 * de)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700788{
789 if (len != de->name_len)
790 return 0;
791 if (!de->inode)
792 return 0;
793 return !memcmp(name, de->name, len);
794}
795
796/*
797 * Returns 0 if not found, -1 on failure, and 1 on success
798 */
799static inline int search_dirblock(struct buffer_head * bh,
800 struct inode *dir,
801 struct dentry *dentry,
802 unsigned long offset,
Mingming Cao617ba132006-10-11 01:20:53 -0700803 struct ext4_dir_entry_2 ** res_dir)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700804{
Mingming Cao617ba132006-10-11 01:20:53 -0700805 struct ext4_dir_entry_2 * de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700806 char * dlimit;
807 int de_len;
808 const char *name = dentry->d_name.name;
809 int namelen = dentry->d_name.len;
810
Mingming Cao617ba132006-10-11 01:20:53 -0700811 de = (struct ext4_dir_entry_2 *) bh->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700812 dlimit = bh->b_data + dir->i_sb->s_blocksize;
813 while ((char *) de < dlimit) {
814 /* this code is executed quadratically often */
815 /* do minimal checking `by hand' */
816
817 if ((char *) de + namelen <= dlimit &&
Mingming Cao617ba132006-10-11 01:20:53 -0700818 ext4_match (namelen, name, de)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700819 /* found a match - just to be sure, do a full check */
Mingming Cao617ba132006-10-11 01:20:53 -0700820 if (!ext4_check_dir_entry("ext4_find_entry",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700821 dir, de, bh, offset))
822 return -1;
823 *res_dir = de;
824 return 1;
825 }
826 /* prevent looping on a bad block */
Jan Karaa72d7f82008-01-28 23:58:27 -0500827 de_len = ext4_rec_len_from_disk(de->rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700828 if (de_len <= 0)
829 return -1;
830 offset += de_len;
Mingming Cao617ba132006-10-11 01:20:53 -0700831 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700832 }
833 return 0;
834}
835
836
837/*
Mingming Cao617ba132006-10-11 01:20:53 -0700838 * ext4_find_entry()
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700839 *
840 * finds an entry in the specified directory with the wanted name. It
841 * returns the cache buffer in which the entry was found, and the entry
842 * itself (as a parameter - res_dir). It does NOT read the inode of the
843 * entry - you'll have to do that yourself if you want to.
844 *
845 * The returned buffer_head has ->b_count elevated. The caller is expected
846 * to brelse() it when appropriate.
847 */
Mingming Cao617ba132006-10-11 01:20:53 -0700848static struct buffer_head * ext4_find_entry (struct dentry *dentry,
849 struct ext4_dir_entry_2 ** res_dir)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700850{
851 struct super_block * sb;
852 struct buffer_head * bh_use[NAMEI_RA_SIZE];
853 struct buffer_head * bh, *ret = NULL;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500854 ext4_lblk_t start, block, b;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700855 int ra_max = 0; /* Number of bh's in the readahead
856 buffer, bh_use[] */
857 int ra_ptr = 0; /* Current index into readahead
858 buffer */
859 int num = 0;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500860 ext4_lblk_t nblocks;
861 int i, err;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700862 struct inode *dir = dentry->d_parent->d_inode;
863 int namelen;
864 const u8 *name;
865 unsigned blocksize;
866
867 *res_dir = NULL;
868 sb = dir->i_sb;
869 blocksize = sb->s_blocksize;
870 namelen = dentry->d_name.len;
871 name = dentry->d_name.name;
Mingming Cao617ba132006-10-11 01:20:53 -0700872 if (namelen > EXT4_NAME_LEN)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700873 return NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700874 if (is_dx(dir)) {
Mingming Cao617ba132006-10-11 01:20:53 -0700875 bh = ext4_dx_find_entry(dentry, res_dir, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700876 /*
877 * On success, or if the error was file not found,
878 * return. Otherwise, fall back to doing a search the
879 * old fashioned way.
880 */
881 if (bh || (err != ERR_BAD_DX_DIR))
882 return bh;
Mingming Cao617ba132006-10-11 01:20:53 -0700883 dxtrace(printk("ext4_find_entry: dx failed, falling back\n"));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700884 }
Mingming Cao617ba132006-10-11 01:20:53 -0700885 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
886 start = EXT4_I(dir)->i_dir_start_lookup;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700887 if (start >= nblocks)
888 start = 0;
889 block = start;
890restart:
891 do {
892 /*
893 * We deal with the read-ahead logic here.
894 */
895 if (ra_ptr >= ra_max) {
896 /* Refill the readahead buffer */
897 ra_ptr = 0;
898 b = block;
899 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
900 /*
901 * Terminate if we reach the end of the
902 * directory and must wrap, or if our
903 * search has finished at this block.
904 */
905 if (b >= nblocks || (num && block == start)) {
906 bh_use[ra_max] = NULL;
907 break;
908 }
909 num++;
Mingming Cao617ba132006-10-11 01:20:53 -0700910 bh = ext4_getblk(NULL, dir, b++, 0, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700911 bh_use[ra_max] = bh;
912 if (bh)
913 ll_rw_block(READ_META, 1, &bh);
914 }
915 }
916 if ((bh = bh_use[ra_ptr++]) == NULL)
917 goto next;
918 wait_on_buffer(bh);
919 if (!buffer_uptodate(bh)) {
920 /* read error, skip block & hope for the best */
Mingming Cao617ba132006-10-11 01:20:53 -0700921 ext4_error(sb, __FUNCTION__, "reading directory #%lu "
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500922 "offset %lu", dir->i_ino,
923 (unsigned long)block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700924 brelse(bh);
925 goto next;
926 }
927 i = search_dirblock(bh, dir, dentry,
Mingming Cao617ba132006-10-11 01:20:53 -0700928 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700929 if (i == 1) {
Mingming Cao617ba132006-10-11 01:20:53 -0700930 EXT4_I(dir)->i_dir_start_lookup = block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700931 ret = bh;
932 goto cleanup_and_exit;
933 } else {
934 brelse(bh);
935 if (i < 0)
936 goto cleanup_and_exit;
937 }
938 next:
939 if (++block >= nblocks)
940 block = 0;
941 } while (block != start);
942
943 /*
944 * If the directory has grown while we were searching, then
945 * search the last part of the directory before giving up.
946 */
947 block = nblocks;
Mingming Cao617ba132006-10-11 01:20:53 -0700948 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700949 if (block < nblocks) {
950 start = 0;
951 goto restart;
952 }
953
954cleanup_and_exit:
955 /* Clean up the read-ahead blocks */
956 for (; ra_ptr < ra_max; ra_ptr++)
957 brelse (bh_use[ra_ptr]);
958 return ret;
959}
960
Mingming Cao617ba132006-10-11 01:20:53 -0700961static struct buffer_head * ext4_dx_find_entry(struct dentry *dentry,
962 struct ext4_dir_entry_2 **res_dir, int *err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700963{
964 struct super_block * sb;
965 struct dx_hash_info hinfo;
966 u32 hash;
967 struct dx_frame frames[2], *frame;
Mingming Cao617ba132006-10-11 01:20:53 -0700968 struct ext4_dir_entry_2 *de, *top;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700969 struct buffer_head *bh;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500970 ext4_lblk_t block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700971 int retval;
972 int namelen = dentry->d_name.len;
973 const u8 *name = dentry->d_name.name;
974 struct inode *dir = dentry->d_parent->d_inode;
975
976 sb = dir->i_sb;
977 /* NFS may look up ".." - look at dx_root directory block */
978 if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
979 if (!(frame = dx_probe(dentry, NULL, &hinfo, frames, err)))
980 return NULL;
981 } else {
982 frame = frames;
983 frame->bh = NULL; /* for dx_release() */
984 frame->at = (struct dx_entry *)frames; /* hack for zero entry*/
985 dx_set_block(frame->at, 0); /* dx_root block is 0 */
986 }
987 hash = hinfo.hash;
988 do {
989 block = dx_get_block(frame->at);
Mingming Cao617ba132006-10-11 01:20:53 -0700990 if (!(bh = ext4_bread (NULL,dir, block, 0, err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700991 goto errout;
Mingming Cao617ba132006-10-11 01:20:53 -0700992 de = (struct ext4_dir_entry_2 *) bh->b_data;
993 top = (struct ext4_dir_entry_2 *) ((char *) de + sb->s_blocksize -
994 EXT4_DIR_REC_LEN(0));
995 for (; de < top; de = ext4_next_entry(de))
996 if (ext4_match (namelen, name, de)) {
997 if (!ext4_check_dir_entry("ext4_find_entry",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700998 dir, de, bh,
Mingming Cao617ba132006-10-11 01:20:53 -0700999 (block<<EXT4_BLOCK_SIZE_BITS(sb))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001000 +((char *)de - bh->b_data))) {
1001 brelse (bh);
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001002 *err = ERR_BAD_DX_DIR;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001003 goto errout;
1004 }
1005 *res_dir = de;
1006 dx_release (frames);
1007 return bh;
1008 }
1009 brelse (bh);
1010 /* Check to see if we should continue to search */
Mingming Cao617ba132006-10-11 01:20:53 -07001011 retval = ext4_htree_next_block(dir, hash, frame,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001012 frames, NULL);
1013 if (retval < 0) {
Mingming Cao617ba132006-10-11 01:20:53 -07001014 ext4_warning(sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001015 "error reading index page in directory #%lu",
1016 dir->i_ino);
1017 *err = retval;
1018 goto errout;
1019 }
1020 } while (retval == 1);
1021
1022 *err = -ENOENT;
1023errout:
1024 dxtrace(printk("%s not found\n", name));
1025 dx_release (frames);
1026 return NULL;
1027}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001028
Mingming Cao617ba132006-10-11 01:20:53 -07001029static struct dentry *ext4_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001030{
1031 struct inode * inode;
Mingming Cao617ba132006-10-11 01:20:53 -07001032 struct ext4_dir_entry_2 * de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001033 struct buffer_head * bh;
1034
Mingming Cao617ba132006-10-11 01:20:53 -07001035 if (dentry->d_name.len > EXT4_NAME_LEN)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001036 return ERR_PTR(-ENAMETOOLONG);
1037
Mingming Cao617ba132006-10-11 01:20:53 -07001038 bh = ext4_find_entry(dentry, &de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001039 inode = NULL;
1040 if (bh) {
1041 unsigned long ino = le32_to_cpu(de->inode);
1042 brelse (bh);
Mingming Cao617ba132006-10-11 01:20:53 -07001043 if (!ext4_valid_inum(dir->i_sb, ino)) {
1044 ext4_error(dir->i_sb, "ext4_lookup",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001045 "bad inode number: %lu", ino);
1046 inode = NULL;
1047 } else
1048 inode = iget(dir->i_sb, ino);
1049
1050 if (!inode)
1051 return ERR_PTR(-EACCES);
Vasily Averina6c15c22007-07-15 23:40:46 -07001052
1053 if (is_bad_inode(inode)) {
1054 iput(inode);
1055 return ERR_PTR(-ENOENT);
1056 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001057 }
1058 return d_splice_alias(inode, dentry);
1059}
1060
1061
Mingming Cao617ba132006-10-11 01:20:53 -07001062struct dentry *ext4_get_parent(struct dentry *child)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001063{
1064 unsigned long ino;
1065 struct dentry *parent;
1066 struct inode *inode;
1067 struct dentry dotdot;
Mingming Cao617ba132006-10-11 01:20:53 -07001068 struct ext4_dir_entry_2 * de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001069 struct buffer_head *bh;
1070
1071 dotdot.d_name.name = "..";
1072 dotdot.d_name.len = 2;
1073 dotdot.d_parent = child; /* confusing, isn't it! */
1074
Mingming Cao617ba132006-10-11 01:20:53 -07001075 bh = ext4_find_entry(&dotdot, &de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001076 inode = NULL;
1077 if (!bh)
1078 return ERR_PTR(-ENOENT);
1079 ino = le32_to_cpu(de->inode);
1080 brelse(bh);
1081
Mingming Cao617ba132006-10-11 01:20:53 -07001082 if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1083 ext4_error(child->d_inode->i_sb, "ext4_get_parent",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001084 "bad inode number: %lu", ino);
1085 inode = NULL;
1086 } else
1087 inode = iget(child->d_inode->i_sb, ino);
1088
1089 if (!inode)
1090 return ERR_PTR(-EACCES);
1091
Vasily Averina6c15c22007-07-15 23:40:46 -07001092 if (is_bad_inode(inode)) {
1093 iput(inode);
1094 return ERR_PTR(-ENOENT);
1095 }
1096
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001097 parent = d_alloc_anon(inode);
1098 if (!parent) {
1099 iput(inode);
1100 parent = ERR_PTR(-ENOMEM);
1101 }
1102 return parent;
1103}
1104
1105#define S_SHIFT 12
Mingming Cao617ba132006-10-11 01:20:53 -07001106static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1107 [S_IFREG >> S_SHIFT] = EXT4_FT_REG_FILE,
1108 [S_IFDIR >> S_SHIFT] = EXT4_FT_DIR,
1109 [S_IFCHR >> S_SHIFT] = EXT4_FT_CHRDEV,
1110 [S_IFBLK >> S_SHIFT] = EXT4_FT_BLKDEV,
1111 [S_IFIFO >> S_SHIFT] = EXT4_FT_FIFO,
1112 [S_IFSOCK >> S_SHIFT] = EXT4_FT_SOCK,
1113 [S_IFLNK >> S_SHIFT] = EXT4_FT_SYMLINK,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001114};
1115
Mingming Cao617ba132006-10-11 01:20:53 -07001116static inline void ext4_set_de_type(struct super_block *sb,
1117 struct ext4_dir_entry_2 *de,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001118 umode_t mode) {
Mingming Cao617ba132006-10-11 01:20:53 -07001119 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1120 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001121}
1122
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001123/*
1124 * Move count entries from end of map between two memory locations.
1125 * Returns pointer to last entry moved.
1126 */
Mingming Cao617ba132006-10-11 01:20:53 -07001127static struct ext4_dir_entry_2 *
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001128dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1129{
1130 unsigned rec_len = 0;
1131
1132 while (count--) {
Mingming Cao617ba132006-10-11 01:20:53 -07001133 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *) (from + map->offs);
1134 rec_len = EXT4_DIR_REC_LEN(de->name_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001135 memcpy (to, de, rec_len);
Mingming Cao617ba132006-10-11 01:20:53 -07001136 ((struct ext4_dir_entry_2 *) to)->rec_len =
Jan Karaa72d7f82008-01-28 23:58:27 -05001137 ext4_rec_len_to_disk(rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001138 de->inode = 0;
1139 map++;
1140 to += rec_len;
1141 }
Mingming Cao617ba132006-10-11 01:20:53 -07001142 return (struct ext4_dir_entry_2 *) (to - rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001143}
1144
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001145/*
1146 * Compact each dir entry in the range to the minimal rec_len.
1147 * Returns pointer to last entry in range.
1148 */
Mingming Cao617ba132006-10-11 01:20:53 -07001149static struct ext4_dir_entry_2* dx_pack_dirents(char *base, int size)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001150{
Mingming Cao617ba132006-10-11 01:20:53 -07001151 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001152 unsigned rec_len = 0;
1153
1154 prev = to = de;
1155 while ((char*)de < base + size) {
Jan Karaa72d7f82008-01-28 23:58:27 -05001156 next = ext4_next_entry(de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001157 if (de->inode && de->name_len) {
Mingming Cao617ba132006-10-11 01:20:53 -07001158 rec_len = EXT4_DIR_REC_LEN(de->name_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001159 if (de > to)
1160 memmove(to, de, rec_len);
Jan Karaa72d7f82008-01-28 23:58:27 -05001161 to->rec_len = ext4_rec_len_to_disk(rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001162 prev = to;
Mingming Cao617ba132006-10-11 01:20:53 -07001163 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001164 }
1165 de = next;
1166 }
1167 return prev;
1168}
1169
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001170/*
1171 * Split a full leaf block to make room for a new dir entry.
1172 * Allocate a new block, and move entries so that they are approx. equally full.
1173 * Returns pointer to de in block into which the new entry will be inserted.
1174 */
Mingming Cao617ba132006-10-11 01:20:53 -07001175static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001176 struct buffer_head **bh,struct dx_frame *frame,
1177 struct dx_hash_info *hinfo, int *error)
1178{
1179 unsigned blocksize = dir->i_sb->s_blocksize;
1180 unsigned count, continued;
1181 struct buffer_head *bh2;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001182 ext4_lblk_t newblock;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001183 u32 hash2;
1184 struct dx_map_entry *map;
1185 char *data1 = (*bh)->b_data, *data2;
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001186 unsigned split, move, size, i;
Mingming Cao617ba132006-10-11 01:20:53 -07001187 struct ext4_dir_entry_2 *de = NULL, *de2;
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001188 int err = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001189
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001190 bh2 = ext4_append (handle, dir, &newblock, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001191 if (!(bh2)) {
1192 brelse(*bh);
1193 *bh = NULL;
1194 goto errout;
1195 }
1196
1197 BUFFER_TRACE(*bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001198 err = ext4_journal_get_write_access(handle, *bh);
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001199 if (err)
1200 goto journal_error;
1201
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001202 BUFFER_TRACE(frame->bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001203 err = ext4_journal_get_write_access(handle, frame->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001204 if (err)
1205 goto journal_error;
1206
1207 data2 = bh2->b_data;
1208
1209 /* create map in the end of data2 block */
1210 map = (struct dx_map_entry *) (data2 + blocksize);
Mingming Cao617ba132006-10-11 01:20:53 -07001211 count = dx_make_map ((struct ext4_dir_entry_2 *) data1,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001212 blocksize, hinfo, map);
1213 map -= count;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001214 dx_sort_map (map, count);
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001215 /* Split the existing block in the middle, size-wise */
1216 size = 0;
1217 move = 0;
1218 for (i = count-1; i >= 0; i--) {
1219 /* is more than half of this entry in 2nd half of the block? */
1220 if (size + map[i].size/2 > blocksize/2)
1221 break;
1222 size += map[i].size;
1223 move++;
1224 }
1225 /* map index at which we will split */
1226 split = count - move;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001227 hash2 = map[split].hash;
1228 continued = hash2 == map[split - 1].hash;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001229 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1230 (unsigned long)dx_get_block(frame->at),
1231 hash2, split, count-split));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001232
1233 /* Fancy dance to stay within two buffers */
1234 de2 = dx_move_dirents(data1, data2, map + split, count - split);
1235 de = dx_pack_dirents(data1,blocksize);
Jan Karaa72d7f82008-01-28 23:58:27 -05001236 de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de);
1237 de2->rec_len = ext4_rec_len_to_disk(data2 + blocksize - (char *) de2);
Mingming Cao617ba132006-10-11 01:20:53 -07001238 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1239 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001240
1241 /* Which block gets the new entry? */
1242 if (hinfo->hash >= hash2)
1243 {
1244 swap(*bh, bh2);
1245 de = de2;
1246 }
1247 dx_insert_block (frame, hash2 + continued, newblock);
Mingming Cao617ba132006-10-11 01:20:53 -07001248 err = ext4_journal_dirty_metadata (handle, bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001249 if (err)
1250 goto journal_error;
Mingming Cao617ba132006-10-11 01:20:53 -07001251 err = ext4_journal_dirty_metadata (handle, frame->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001252 if (err)
1253 goto journal_error;
1254 brelse (bh2);
1255 dxtrace(dx_show_index ("frame", frame->entries));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001256 return de;
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001257
1258journal_error:
1259 brelse(*bh);
1260 brelse(bh2);
1261 *bh = NULL;
1262 ext4_std_error(dir->i_sb, err);
1263errout:
1264 *error = err;
1265 return NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001266}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001267
1268/*
1269 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1270 * it points to a directory entry which is guaranteed to be large
1271 * enough for new directory entry. If de is NULL, then
1272 * add_dirent_to_buf will attempt search the directory block for
1273 * space. It will return -ENOSPC if no space is available, and -EIO
1274 * and -EEXIST if directory entry already exists.
1275 *
1276 * NOTE! bh is NOT released in the case where ENOSPC is returned. In
1277 * all other cases bh is released.
1278 */
1279static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
Mingming Cao617ba132006-10-11 01:20:53 -07001280 struct inode *inode, struct ext4_dir_entry_2 *de,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001281 struct buffer_head * bh)
1282{
1283 struct inode *dir = dentry->d_parent->d_inode;
1284 const char *name = dentry->d_name.name;
1285 int namelen = dentry->d_name.len;
1286 unsigned long offset = 0;
1287 unsigned short reclen;
1288 int nlen, rlen, err;
1289 char *top;
1290
Mingming Cao617ba132006-10-11 01:20:53 -07001291 reclen = EXT4_DIR_REC_LEN(namelen);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001292 if (!de) {
Mingming Cao617ba132006-10-11 01:20:53 -07001293 de = (struct ext4_dir_entry_2 *)bh->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001294 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1295 while ((char *) de <= top) {
Mingming Cao617ba132006-10-11 01:20:53 -07001296 if (!ext4_check_dir_entry("ext4_add_entry", dir, de,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001297 bh, offset)) {
1298 brelse (bh);
1299 return -EIO;
1300 }
Mingming Cao617ba132006-10-11 01:20:53 -07001301 if (ext4_match (namelen, name, de)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001302 brelse (bh);
1303 return -EEXIST;
1304 }
Mingming Cao617ba132006-10-11 01:20:53 -07001305 nlen = EXT4_DIR_REC_LEN(de->name_len);
Jan Karaa72d7f82008-01-28 23:58:27 -05001306 rlen = ext4_rec_len_from_disk(de->rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001307 if ((de->inode? rlen - nlen: rlen) >= reclen)
1308 break;
Mingming Cao617ba132006-10-11 01:20:53 -07001309 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001310 offset += rlen;
1311 }
1312 if ((char *) de > top)
1313 return -ENOSPC;
1314 }
1315 BUFFER_TRACE(bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001316 err = ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001317 if (err) {
Mingming Cao617ba132006-10-11 01:20:53 -07001318 ext4_std_error(dir->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001319 brelse(bh);
1320 return err;
1321 }
1322
1323 /* By now the buffer is marked for journaling */
Mingming Cao617ba132006-10-11 01:20:53 -07001324 nlen = EXT4_DIR_REC_LEN(de->name_len);
Jan Karaa72d7f82008-01-28 23:58:27 -05001325 rlen = ext4_rec_len_from_disk(de->rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001326 if (de->inode) {
Mingming Cao617ba132006-10-11 01:20:53 -07001327 struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
Jan Karaa72d7f82008-01-28 23:58:27 -05001328 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen);
1329 de->rec_len = ext4_rec_len_to_disk(nlen);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001330 de = de1;
1331 }
Mingming Cao617ba132006-10-11 01:20:53 -07001332 de->file_type = EXT4_FT_UNKNOWN;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001333 if (inode) {
1334 de->inode = cpu_to_le32(inode->i_ino);
Mingming Cao617ba132006-10-11 01:20:53 -07001335 ext4_set_de_type(dir->i_sb, de, inode->i_mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001336 } else
1337 de->inode = 0;
1338 de->name_len = namelen;
1339 memcpy (de->name, name, namelen);
1340 /*
1341 * XXX shouldn't update any times until successful
1342 * completion of syscall, but too many callers depend
1343 * on this.
1344 *
1345 * XXX similarly, too many callers depend on
Mingming Cao617ba132006-10-11 01:20:53 -07001346 * ext4_new_inode() setting the times, but error
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001347 * recovery deletes the inode, so the worst that can
1348 * happen is that the times are slightly out of date
1349 * and/or different from the directory change time.
1350 */
Kalpak Shahef7f3832007-07-18 09:15:20 -04001351 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
Mingming Cao617ba132006-10-11 01:20:53 -07001352 ext4_update_dx_flag(dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001353 dir->i_version++;
Mingming Cao617ba132006-10-11 01:20:53 -07001354 ext4_mark_inode_dirty(handle, dir);
1355 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1356 err = ext4_journal_dirty_metadata(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001357 if (err)
Mingming Cao617ba132006-10-11 01:20:53 -07001358 ext4_std_error(dir->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001359 brelse(bh);
1360 return 0;
1361}
1362
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001363/*
1364 * This converts a one block unindexed directory to a 3 block indexed
1365 * directory, and adds the dentry to the indexed directory.
1366 */
1367static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1368 struct inode *inode, struct buffer_head *bh)
1369{
1370 struct inode *dir = dentry->d_parent->d_inode;
1371 const char *name = dentry->d_name.name;
1372 int namelen = dentry->d_name.len;
1373 struct buffer_head *bh2;
1374 struct dx_root *root;
1375 struct dx_frame frames[2], *frame;
1376 struct dx_entry *entries;
Mingming Cao617ba132006-10-11 01:20:53 -07001377 struct ext4_dir_entry_2 *de, *de2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001378 char *data1, *top;
1379 unsigned len;
1380 int retval;
1381 unsigned blocksize;
1382 struct dx_hash_info hinfo;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001383 ext4_lblk_t block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001384 struct fake_dirent *fde;
1385
1386 blocksize = dir->i_sb->s_blocksize;
1387 dxtrace(printk("Creating index\n"));
Mingming Cao617ba132006-10-11 01:20:53 -07001388 retval = ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001389 if (retval) {
Mingming Cao617ba132006-10-11 01:20:53 -07001390 ext4_std_error(dir->i_sb, retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001391 brelse(bh);
1392 return retval;
1393 }
1394 root = (struct dx_root *) bh->b_data;
1395
Mingming Cao617ba132006-10-11 01:20:53 -07001396 bh2 = ext4_append (handle, dir, &block, &retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001397 if (!(bh2)) {
1398 brelse(bh);
1399 return retval;
1400 }
Mingming Cao617ba132006-10-11 01:20:53 -07001401 EXT4_I(dir)->i_flags |= EXT4_INDEX_FL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001402 data1 = bh2->b_data;
1403
1404 /* The 0th block becomes the root, move the dirents out */
1405 fde = &root->dotdot;
Jan Karaa72d7f82008-01-28 23:58:27 -05001406 de = (struct ext4_dir_entry_2 *)((char *)fde +
1407 ext4_rec_len_from_disk(fde->rec_len));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001408 len = ((char *) root) + blocksize - (char *) de;
1409 memcpy (data1, de, len);
Mingming Cao617ba132006-10-11 01:20:53 -07001410 de = (struct ext4_dir_entry_2 *) data1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001411 top = data1 + len;
Jan Karaa72d7f82008-01-28 23:58:27 -05001412 while ((char *)(de2 = ext4_next_entry(de)) < top)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001413 de = de2;
Jan Karaa72d7f82008-01-28 23:58:27 -05001414 de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001415 /* Initialize the root; the dot dirents already exist */
Mingming Cao617ba132006-10-11 01:20:53 -07001416 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
Jan Karaa72d7f82008-01-28 23:58:27 -05001417 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001418 memset (&root->info, 0, sizeof(root->info));
1419 root->info.info_length = sizeof(root->info);
Mingming Cao617ba132006-10-11 01:20:53 -07001420 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001421 entries = root->entries;
1422 dx_set_block (entries, 1);
1423 dx_set_count (entries, 1);
1424 dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1425
1426 /* Initialize as for dx_probe */
1427 hinfo.hash_version = root->info.hash_version;
Mingming Cao617ba132006-10-11 01:20:53 -07001428 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1429 ext4fs_dirhash(name, namelen, &hinfo);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001430 frame = frames;
1431 frame->entries = entries;
1432 frame->at = entries;
1433 frame->bh = bh;
1434 bh = bh2;
1435 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1436 dx_release (frames);
1437 if (!(de))
1438 return retval;
1439
1440 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1441}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001442
1443/*
Mingming Cao617ba132006-10-11 01:20:53 -07001444 * ext4_add_entry()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001445 *
1446 * adds a file entry to the specified directory, using the same
Mingming Cao617ba132006-10-11 01:20:53 -07001447 * semantics as ext4_find_entry(). It returns NULL if it failed.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001448 *
1449 * NOTE!! The inode part of 'de' is left at 0 - which means you
1450 * may not sleep between calling this and putting something into
1451 * the entry, as someone else might have used it while you slept.
1452 */
Mingming Cao617ba132006-10-11 01:20:53 -07001453static int ext4_add_entry (handle_t *handle, struct dentry *dentry,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001454 struct inode *inode)
1455{
1456 struct inode *dir = dentry->d_parent->d_inode;
1457 unsigned long offset;
1458 struct buffer_head * bh;
Mingming Cao617ba132006-10-11 01:20:53 -07001459 struct ext4_dir_entry_2 *de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001460 struct super_block * sb;
1461 int retval;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001462 int dx_fallback=0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001463 unsigned blocksize;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001464 ext4_lblk_t block, blocks;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001465
1466 sb = dir->i_sb;
1467 blocksize = sb->s_blocksize;
1468 if (!dentry->d_name.len)
1469 return -EINVAL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001470 if (is_dx(dir)) {
Mingming Cao617ba132006-10-11 01:20:53 -07001471 retval = ext4_dx_add_entry(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001472 if (!retval || (retval != ERR_BAD_DX_DIR))
1473 return retval;
Mingming Cao617ba132006-10-11 01:20:53 -07001474 EXT4_I(dir)->i_flags &= ~EXT4_INDEX_FL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001475 dx_fallback++;
Mingming Cao617ba132006-10-11 01:20:53 -07001476 ext4_mark_inode_dirty(handle, dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001477 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001478 blocks = dir->i_size >> sb->s_blocksize_bits;
1479 for (block = 0, offset = 0; block < blocks; block++) {
Mingming Cao617ba132006-10-11 01:20:53 -07001480 bh = ext4_bread(handle, dir, block, 0, &retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001481 if(!bh)
1482 return retval;
1483 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1484 if (retval != -ENOSPC)
1485 return retval;
1486
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001487 if (blocks == 1 && !dx_fallback &&
Mingming Cao617ba132006-10-11 01:20:53 -07001488 EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001489 return make_indexed_dir(handle, dentry, inode, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001490 brelse(bh);
1491 }
Mingming Cao617ba132006-10-11 01:20:53 -07001492 bh = ext4_append(handle, dir, &block, &retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001493 if (!bh)
1494 return retval;
Mingming Cao617ba132006-10-11 01:20:53 -07001495 de = (struct ext4_dir_entry_2 *) bh->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001496 de->inode = 0;
Jan Karaa72d7f82008-01-28 23:58:27 -05001497 de->rec_len = ext4_rec_len_to_disk(blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001498 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1499}
1500
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001501/*
1502 * Returns 0 for success, or a negative error value
1503 */
Mingming Cao617ba132006-10-11 01:20:53 -07001504static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001505 struct inode *inode)
1506{
1507 struct dx_frame frames[2], *frame;
1508 struct dx_entry *entries, *at;
1509 struct dx_hash_info hinfo;
1510 struct buffer_head * bh;
1511 struct inode *dir = dentry->d_parent->d_inode;
1512 struct super_block * sb = dir->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -07001513 struct ext4_dir_entry_2 *de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001514 int err;
1515
1516 frame = dx_probe(dentry, NULL, &hinfo, frames, &err);
1517 if (!frame)
1518 return err;
1519 entries = frame->entries;
1520 at = frame->at;
1521
Mingming Cao617ba132006-10-11 01:20:53 -07001522 if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001523 goto cleanup;
1524
1525 BUFFER_TRACE(bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001526 err = ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001527 if (err)
1528 goto journal_error;
1529
1530 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1531 if (err != -ENOSPC) {
1532 bh = NULL;
1533 goto cleanup;
1534 }
1535
1536 /* Block full, should compress but for now just split */
1537 dxtrace(printk("using %u of %u node entries\n",
1538 dx_get_count(entries), dx_get_limit(entries)));
1539 /* Need to split index? */
1540 if (dx_get_count(entries) == dx_get_limit(entries)) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001541 ext4_lblk_t newblock;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001542 unsigned icount = dx_get_count(entries);
1543 int levels = frame - frames;
1544 struct dx_entry *entries2;
1545 struct dx_node *node2;
1546 struct buffer_head *bh2;
1547
1548 if (levels && (dx_get_count(frames->entries) ==
1549 dx_get_limit(frames->entries))) {
Mingming Cao617ba132006-10-11 01:20:53 -07001550 ext4_warning(sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001551 "Directory index full!");
1552 err = -ENOSPC;
1553 goto cleanup;
1554 }
Mingming Cao617ba132006-10-11 01:20:53 -07001555 bh2 = ext4_append (handle, dir, &newblock, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001556 if (!(bh2))
1557 goto cleanup;
1558 node2 = (struct dx_node *)(bh2->b_data);
1559 entries2 = node2->entries;
Jan Karaa72d7f82008-01-28 23:58:27 -05001560 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001561 node2->fake.inode = 0;
1562 BUFFER_TRACE(frame->bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001563 err = ext4_journal_get_write_access(handle, frame->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001564 if (err)
1565 goto journal_error;
1566 if (levels) {
1567 unsigned icount1 = icount/2, icount2 = icount - icount1;
1568 unsigned hash2 = dx_get_hash(entries + icount1);
1569 dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1570
1571 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
Mingming Cao617ba132006-10-11 01:20:53 -07001572 err = ext4_journal_get_write_access(handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001573 frames[0].bh);
1574 if (err)
1575 goto journal_error;
1576
1577 memcpy ((char *) entries2, (char *) (entries + icount1),
1578 icount2 * sizeof(struct dx_entry));
1579 dx_set_count (entries, icount1);
1580 dx_set_count (entries2, icount2);
1581 dx_set_limit (entries2, dx_node_limit(dir));
1582
1583 /* Which index block gets the new entry? */
1584 if (at - entries >= icount1) {
1585 frame->at = at = at - entries - icount1 + entries2;
1586 frame->entries = entries = entries2;
1587 swap(frame->bh, bh2);
1588 }
1589 dx_insert_block (frames + 0, hash2, newblock);
1590 dxtrace(dx_show_index ("node", frames[1].entries));
1591 dxtrace(dx_show_index ("node",
1592 ((struct dx_node *) bh2->b_data)->entries));
Mingming Cao617ba132006-10-11 01:20:53 -07001593 err = ext4_journal_dirty_metadata(handle, bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001594 if (err)
1595 goto journal_error;
1596 brelse (bh2);
1597 } else {
1598 dxtrace(printk("Creating second level index...\n"));
1599 memcpy((char *) entries2, (char *) entries,
1600 icount * sizeof(struct dx_entry));
1601 dx_set_limit(entries2, dx_node_limit(dir));
1602
1603 /* Set up root */
1604 dx_set_count(entries, 1);
1605 dx_set_block(entries + 0, newblock);
1606 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1607
1608 /* Add new access path frame */
1609 frame = frames + 1;
1610 frame->at = at = at - entries + entries2;
1611 frame->entries = entries = entries2;
1612 frame->bh = bh2;
Mingming Cao617ba132006-10-11 01:20:53 -07001613 err = ext4_journal_get_write_access(handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001614 frame->bh);
1615 if (err)
1616 goto journal_error;
1617 }
Mingming Cao617ba132006-10-11 01:20:53 -07001618 ext4_journal_dirty_metadata(handle, frames[0].bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001619 }
1620 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1621 if (!de)
1622 goto cleanup;
1623 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1624 bh = NULL;
1625 goto cleanup;
1626
1627journal_error:
Mingming Cao617ba132006-10-11 01:20:53 -07001628 ext4_std_error(dir->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001629cleanup:
1630 if (bh)
1631 brelse(bh);
1632 dx_release(frames);
1633 return err;
1634}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001635
1636/*
Mingming Cao617ba132006-10-11 01:20:53 -07001637 * ext4_delete_entry deletes a directory entry by merging it with the
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001638 * previous entry
1639 */
Mingming Cao617ba132006-10-11 01:20:53 -07001640static int ext4_delete_entry (handle_t *handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001641 struct inode * dir,
Mingming Cao617ba132006-10-11 01:20:53 -07001642 struct ext4_dir_entry_2 * de_del,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001643 struct buffer_head * bh)
1644{
Mingming Cao617ba132006-10-11 01:20:53 -07001645 struct ext4_dir_entry_2 * de, * pde;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001646 int i;
1647
1648 i = 0;
1649 pde = NULL;
Mingming Cao617ba132006-10-11 01:20:53 -07001650 de = (struct ext4_dir_entry_2 *) bh->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001651 while (i < bh->b_size) {
Mingming Cao617ba132006-10-11 01:20:53 -07001652 if (!ext4_check_dir_entry("ext4_delete_entry", dir, de, bh, i))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001653 return -EIO;
1654 if (de == de_del) {
1655 BUFFER_TRACE(bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001656 ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001657 if (pde)
Jan Karaa72d7f82008-01-28 23:58:27 -05001658 pde->rec_len = ext4_rec_len_to_disk(
1659 ext4_rec_len_from_disk(pde->rec_len) +
1660 ext4_rec_len_from_disk(de->rec_len));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001661 else
1662 de->inode = 0;
1663 dir->i_version++;
Mingming Cao617ba132006-10-11 01:20:53 -07001664 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1665 ext4_journal_dirty_metadata(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001666 return 0;
1667 }
Jan Karaa72d7f82008-01-28 23:58:27 -05001668 i += ext4_rec_len_from_disk(de->rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001669 pde = de;
Jan Karaa72d7f82008-01-28 23:58:27 -05001670 de = ext4_next_entry(de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001671 }
1672 return -ENOENT;
1673}
1674
Andreas Dilgerf8628a12007-07-18 08:38:01 -04001675/*
1676 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
1677 * since this indicates that nlinks count was previously 1.
1678 */
1679static void ext4_inc_count(handle_t *handle, struct inode *inode)
1680{
1681 inc_nlink(inode);
1682 if (is_dx(inode) && inode->i_nlink > 1) {
1683 /* limit is 16-bit i_links_count */
1684 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
1685 inode->i_nlink = 1;
1686 EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
1687 EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
1688 }
1689 }
1690}
1691
1692/*
1693 * If a directory had nlink == 1, then we should let it be 1. This indicates
1694 * directory has >EXT4_LINK_MAX subdirs.
1695 */
1696static void ext4_dec_count(handle_t *handle, struct inode *inode)
1697{
1698 drop_nlink(inode);
1699 if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
1700 inc_nlink(inode);
1701}
1702
1703
Mingming Cao617ba132006-10-11 01:20:53 -07001704static int ext4_add_nondir(handle_t *handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001705 struct dentry *dentry, struct inode *inode)
1706{
Mingming Cao617ba132006-10-11 01:20:53 -07001707 int err = ext4_add_entry(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001708 if (!err) {
Mingming Cao617ba132006-10-11 01:20:53 -07001709 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001710 d_instantiate(dentry, inode);
1711 return 0;
1712 }
Eric Sandeen731b9a52007-02-10 01:46:16 -08001713 drop_nlink(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001714 iput(inode);
1715 return err;
1716}
1717
1718/*
1719 * By the time this is called, we already have created
1720 * the directory cache entry for the new file, but it
1721 * is so far negative - it has no inode.
1722 *
1723 * If the create succeeds, we fill in the inode information
1724 * with d_instantiate().
1725 */
Mingming Cao617ba132006-10-11 01:20:53 -07001726static int ext4_create (struct inode * dir, struct dentry * dentry, int mode,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001727 struct nameidata *nd)
1728{
1729 handle_t *handle;
1730 struct inode * inode;
1731 int err, retries = 0;
1732
1733retry:
Mingming Cao617ba132006-10-11 01:20:53 -07001734 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1735 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1736 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001737 if (IS_ERR(handle))
1738 return PTR_ERR(handle);
1739
1740 if (IS_DIRSYNC(dir))
1741 handle->h_sync = 1;
1742
Mingming Cao617ba132006-10-11 01:20:53 -07001743 inode = ext4_new_inode (handle, dir, mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001744 err = PTR_ERR(inode);
1745 if (!IS_ERR(inode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07001746 inode->i_op = &ext4_file_inode_operations;
1747 inode->i_fop = &ext4_file_operations;
1748 ext4_set_aops(inode);
1749 err = ext4_add_nondir(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001750 }
Mingming Cao617ba132006-10-11 01:20:53 -07001751 ext4_journal_stop(handle);
1752 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001753 goto retry;
1754 return err;
1755}
1756
Mingming Cao617ba132006-10-11 01:20:53 -07001757static int ext4_mknod (struct inode * dir, struct dentry *dentry,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001758 int mode, dev_t rdev)
1759{
1760 handle_t *handle;
1761 struct inode *inode;
1762 int err, retries = 0;
1763
1764 if (!new_valid_dev(rdev))
1765 return -EINVAL;
1766
1767retry:
Mingming Cao617ba132006-10-11 01:20:53 -07001768 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1769 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1770 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001771 if (IS_ERR(handle))
1772 return PTR_ERR(handle);
1773
1774 if (IS_DIRSYNC(dir))
1775 handle->h_sync = 1;
1776
Mingming Cao617ba132006-10-11 01:20:53 -07001777 inode = ext4_new_inode (handle, dir, mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001778 err = PTR_ERR(inode);
1779 if (!IS_ERR(inode)) {
1780 init_special_inode(inode, inode->i_mode, rdev);
Mingming Cao617ba132006-10-11 01:20:53 -07001781#ifdef CONFIG_EXT4DEV_FS_XATTR
1782 inode->i_op = &ext4_special_inode_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001783#endif
Mingming Cao617ba132006-10-11 01:20:53 -07001784 err = ext4_add_nondir(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001785 }
Mingming Cao617ba132006-10-11 01:20:53 -07001786 ext4_journal_stop(handle);
1787 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001788 goto retry;
1789 return err;
1790}
1791
Mingming Cao617ba132006-10-11 01:20:53 -07001792static int ext4_mkdir(struct inode * dir, struct dentry * dentry, int mode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001793{
1794 handle_t *handle;
1795 struct inode * inode;
1796 struct buffer_head * dir_block;
Mingming Cao617ba132006-10-11 01:20:53 -07001797 struct ext4_dir_entry_2 * de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001798 int err, retries = 0;
1799
Andreas Dilgerf8628a12007-07-18 08:38:01 -04001800 if (EXT4_DIR_LINK_MAX(dir))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001801 return -EMLINK;
1802
1803retry:
Mingming Cao617ba132006-10-11 01:20:53 -07001804 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1805 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1806 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001807 if (IS_ERR(handle))
1808 return PTR_ERR(handle);
1809
1810 if (IS_DIRSYNC(dir))
1811 handle->h_sync = 1;
1812
Mingming Cao617ba132006-10-11 01:20:53 -07001813 inode = ext4_new_inode (handle, dir, S_IFDIR | mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001814 err = PTR_ERR(inode);
1815 if (IS_ERR(inode))
1816 goto out_stop;
1817
Mingming Cao617ba132006-10-11 01:20:53 -07001818 inode->i_op = &ext4_dir_inode_operations;
1819 inode->i_fop = &ext4_dir_operations;
1820 inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1821 dir_block = ext4_bread (handle, inode, 0, 1, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001822 if (!dir_block) {
Andreas Dilgerf8628a12007-07-18 08:38:01 -04001823 ext4_dec_count(handle, inode); /* is this nlink == 0? */
Mingming Cao617ba132006-10-11 01:20:53 -07001824 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001825 iput (inode);
1826 goto out_stop;
1827 }
1828 BUFFER_TRACE(dir_block, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001829 ext4_journal_get_write_access(handle, dir_block);
1830 de = (struct ext4_dir_entry_2 *) dir_block->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001831 de->inode = cpu_to_le32(inode->i_ino);
1832 de->name_len = 1;
Jan Karaa72d7f82008-01-28 23:58:27 -05001833 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001834 strcpy (de->name, ".");
Mingming Cao617ba132006-10-11 01:20:53 -07001835 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
Jan Karaa72d7f82008-01-28 23:58:27 -05001836 de = ext4_next_entry(de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001837 de->inode = cpu_to_le32(dir->i_ino);
Jan Karaa72d7f82008-01-28 23:58:27 -05001838 de->rec_len = ext4_rec_len_to_disk(inode->i_sb->s_blocksize -
1839 EXT4_DIR_REC_LEN(1));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001840 de->name_len = 2;
1841 strcpy (de->name, "..");
Mingming Cao617ba132006-10-11 01:20:53 -07001842 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001843 inode->i_nlink = 2;
Mingming Cao617ba132006-10-11 01:20:53 -07001844 BUFFER_TRACE(dir_block, "call ext4_journal_dirty_metadata");
1845 ext4_journal_dirty_metadata(handle, dir_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001846 brelse (dir_block);
Mingming Cao617ba132006-10-11 01:20:53 -07001847 ext4_mark_inode_dirty(handle, inode);
1848 err = ext4_add_entry (handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001849 if (err) {
1850 inode->i_nlink = 0;
Mingming Cao617ba132006-10-11 01:20:53 -07001851 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001852 iput (inode);
1853 goto out_stop;
1854 }
Andreas Dilgerf8628a12007-07-18 08:38:01 -04001855 ext4_inc_count(handle, dir);
Mingming Cao617ba132006-10-11 01:20:53 -07001856 ext4_update_dx_flag(dir);
1857 ext4_mark_inode_dirty(handle, dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001858 d_instantiate(dentry, inode);
1859out_stop:
Mingming Cao617ba132006-10-11 01:20:53 -07001860 ext4_journal_stop(handle);
1861 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001862 goto retry;
1863 return err;
1864}
1865
1866/*
1867 * routine to check that the specified directory is empty (for rmdir)
1868 */
1869static int empty_dir (struct inode * inode)
1870{
1871 unsigned long offset;
1872 struct buffer_head * bh;
Mingming Cao617ba132006-10-11 01:20:53 -07001873 struct ext4_dir_entry_2 * de, * de1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001874 struct super_block * sb;
1875 int err = 0;
1876
1877 sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -07001878 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
1879 !(bh = ext4_bread (NULL, inode, 0, 0, &err))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001880 if (err)
Mingming Cao617ba132006-10-11 01:20:53 -07001881 ext4_error(inode->i_sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001882 "error %d reading directory #%lu offset 0",
1883 err, inode->i_ino);
1884 else
Mingming Cao617ba132006-10-11 01:20:53 -07001885 ext4_warning(inode->i_sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001886 "bad directory (dir #%lu) - no data block",
1887 inode->i_ino);
1888 return 1;
1889 }
Mingming Cao617ba132006-10-11 01:20:53 -07001890 de = (struct ext4_dir_entry_2 *) bh->b_data;
Jan Karaa72d7f82008-01-28 23:58:27 -05001891 de1 = ext4_next_entry(de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001892 if (le32_to_cpu(de->inode) != inode->i_ino ||
1893 !le32_to_cpu(de1->inode) ||
1894 strcmp (".", de->name) ||
1895 strcmp ("..", de1->name)) {
Mingming Cao617ba132006-10-11 01:20:53 -07001896 ext4_warning (inode->i_sb, "empty_dir",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001897 "bad directory (dir #%lu) - no `.' or `..'",
1898 inode->i_ino);
1899 brelse (bh);
1900 return 1;
1901 }
Jan Karaa72d7f82008-01-28 23:58:27 -05001902 offset = ext4_rec_len_from_disk(de->rec_len) +
1903 ext4_rec_len_from_disk(de1->rec_len);
1904 de = ext4_next_entry(de1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001905 while (offset < inode->i_size ) {
1906 if (!bh ||
1907 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1908 err = 0;
1909 brelse (bh);
Mingming Cao617ba132006-10-11 01:20:53 -07001910 bh = ext4_bread (NULL, inode,
1911 offset >> EXT4_BLOCK_SIZE_BITS(sb), 0, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001912 if (!bh) {
1913 if (err)
Mingming Cao617ba132006-10-11 01:20:53 -07001914 ext4_error(sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001915 "error %d reading directory"
1916 " #%lu offset %lu",
1917 err, inode->i_ino, offset);
1918 offset += sb->s_blocksize;
1919 continue;
1920 }
Mingming Cao617ba132006-10-11 01:20:53 -07001921 de = (struct ext4_dir_entry_2 *) bh->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001922 }
Mingming Cao617ba132006-10-11 01:20:53 -07001923 if (!ext4_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1924 de = (struct ext4_dir_entry_2 *)(bh->b_data +
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001925 sb->s_blocksize);
1926 offset = (offset | (sb->s_blocksize - 1)) + 1;
1927 continue;
1928 }
1929 if (le32_to_cpu(de->inode)) {
1930 brelse (bh);
1931 return 0;
1932 }
Jan Karaa72d7f82008-01-28 23:58:27 -05001933 offset += ext4_rec_len_from_disk(de->rec_len);
1934 de = ext4_next_entry(de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001935 }
1936 brelse (bh);
1937 return 1;
1938}
1939
Mingming Cao617ba132006-10-11 01:20:53 -07001940/* ext4_orphan_add() links an unlinked or truncated inode into a list of
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001941 * such inodes, starting at the superblock, in case we crash before the
1942 * file is closed/deleted, or in case the inode truncate spans multiple
1943 * transactions and the last transaction is not recovered after a crash.
1944 *
1945 * At filesystem recovery time, we walk this list deleting unlinked
Mingming Cao617ba132006-10-11 01:20:53 -07001946 * inodes and truncating linked inodes in ext4_orphan_cleanup().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001947 */
Mingming Cao617ba132006-10-11 01:20:53 -07001948int ext4_orphan_add(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001949{
1950 struct super_block *sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -07001951 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001952 int err = 0, rc;
1953
1954 lock_super(sb);
Mingming Cao617ba132006-10-11 01:20:53 -07001955 if (!list_empty(&EXT4_I(inode)->i_orphan))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001956 goto out_unlock;
1957
1958 /* Orphan handling is only valid for files with data blocks
1959 * being truncated, or files being unlinked. */
1960
1961 /* @@@ FIXME: Observation from aviro:
Mingming Cao617ba132006-10-11 01:20:53 -07001962 * I think I can trigger J_ASSERT in ext4_orphan_add(). We block
1963 * here (on lock_super()), so race with ext4_link() which might bump
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001964 * ->i_nlink. For, say it, character device. Not a regular file,
1965 * not a directory, not a symlink and ->i_nlink > 0.
1966 */
1967 J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1968 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1969
Mingming Cao617ba132006-10-11 01:20:53 -07001970 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1971 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001972 if (err)
1973 goto out_unlock;
1974
Mingming Cao617ba132006-10-11 01:20:53 -07001975 err = ext4_reserve_inode_write(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001976 if (err)
1977 goto out_unlock;
1978
1979 /* Insert this inode at the head of the on-disk orphan list... */
Mingming Cao617ba132006-10-11 01:20:53 -07001980 NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
1981 EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1982 err = ext4_journal_dirty_metadata(handle, EXT4_SB(sb)->s_sbh);
1983 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001984 if (!err)
1985 err = rc;
1986
1987 /* Only add to the head of the in-memory list if all the
1988 * previous operations succeeded. If the orphan_add is going to
1989 * fail (possibly taking the journal offline), we can't risk
1990 * leaving the inode on the orphan list: stray orphan-list
1991 * entries can cause panics at unmount time.
1992 *
1993 * This is safe: on error we're going to ignore the orphan list
1994 * anyway on the next recovery. */
1995 if (!err)
Mingming Cao617ba132006-10-11 01:20:53 -07001996 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001997
1998 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1999 jbd_debug(4, "orphan inode %lu will point to %d\n",
2000 inode->i_ino, NEXT_ORPHAN(inode));
2001out_unlock:
2002 unlock_super(sb);
Mingming Cao617ba132006-10-11 01:20:53 -07002003 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002004 return err;
2005}
2006
2007/*
Mingming Cao617ba132006-10-11 01:20:53 -07002008 * ext4_orphan_del() removes an unlinked or truncated inode from the list
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002009 * of such inodes stored on disk, because it is finally being cleaned up.
2010 */
Mingming Cao617ba132006-10-11 01:20:53 -07002011int ext4_orphan_del(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002012{
2013 struct list_head *prev;
Mingming Cao617ba132006-10-11 01:20:53 -07002014 struct ext4_inode_info *ei = EXT4_I(inode);
2015 struct ext4_sb_info *sbi;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002016 unsigned long ino_next;
Mingming Cao617ba132006-10-11 01:20:53 -07002017 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002018 int err = 0;
2019
2020 lock_super(inode->i_sb);
2021 if (list_empty(&ei->i_orphan)) {
2022 unlock_super(inode->i_sb);
2023 return 0;
2024 }
2025
2026 ino_next = NEXT_ORPHAN(inode);
2027 prev = ei->i_orphan.prev;
Mingming Cao617ba132006-10-11 01:20:53 -07002028 sbi = EXT4_SB(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002029
2030 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2031
2032 list_del_init(&ei->i_orphan);
2033
2034 /* If we're on an error path, we may not have a valid
2035 * transaction handle with which to update the orphan list on
2036 * disk, but we still need to remove the inode from the linked
2037 * list in memory. */
2038 if (!handle)
2039 goto out;
2040
Mingming Cao617ba132006-10-11 01:20:53 -07002041 err = ext4_reserve_inode_write(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002042 if (err)
2043 goto out_err;
2044
2045 if (prev == &sbi->s_orphan) {
2046 jbd_debug(4, "superblock will point to %lu\n", ino_next);
2047 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07002048 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002049 if (err)
2050 goto out_brelse;
2051 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
Mingming Cao617ba132006-10-11 01:20:53 -07002052 err = ext4_journal_dirty_metadata(handle, sbi->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002053 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07002054 struct ext4_iloc iloc2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002055 struct inode *i_prev =
Mingming Cao617ba132006-10-11 01:20:53 -07002056 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002057
2058 jbd_debug(4, "orphan inode %lu will point to %lu\n",
2059 i_prev->i_ino, ino_next);
Mingming Cao617ba132006-10-11 01:20:53 -07002060 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002061 if (err)
2062 goto out_brelse;
2063 NEXT_ORPHAN(i_prev) = ino_next;
Mingming Cao617ba132006-10-11 01:20:53 -07002064 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002065 }
2066 if (err)
2067 goto out_brelse;
2068 NEXT_ORPHAN(inode) = 0;
Mingming Cao617ba132006-10-11 01:20:53 -07002069 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002070
2071out_err:
Mingming Cao617ba132006-10-11 01:20:53 -07002072 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002073out:
2074 unlock_super(inode->i_sb);
2075 return err;
2076
2077out_brelse:
2078 brelse(iloc.bh);
2079 goto out_err;
2080}
2081
Mingming Cao617ba132006-10-11 01:20:53 -07002082static int ext4_rmdir (struct inode * dir, struct dentry *dentry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002083{
2084 int retval;
2085 struct inode * inode;
2086 struct buffer_head * bh;
Mingming Cao617ba132006-10-11 01:20:53 -07002087 struct ext4_dir_entry_2 * de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002088 handle_t *handle;
2089
2090 /* Initialize quotas before so that eventual writes go in
2091 * separate transaction */
2092 DQUOT_INIT(dentry->d_inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002093 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002094 if (IS_ERR(handle))
2095 return PTR_ERR(handle);
2096
2097 retval = -ENOENT;
Mingming Cao617ba132006-10-11 01:20:53 -07002098 bh = ext4_find_entry (dentry, &de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002099 if (!bh)
2100 goto end_rmdir;
2101
2102 if (IS_DIRSYNC(dir))
2103 handle->h_sync = 1;
2104
2105 inode = dentry->d_inode;
2106
2107 retval = -EIO;
2108 if (le32_to_cpu(de->inode) != inode->i_ino)
2109 goto end_rmdir;
2110
2111 retval = -ENOTEMPTY;
2112 if (!empty_dir (inode))
2113 goto end_rmdir;
2114
Mingming Cao617ba132006-10-11 01:20:53 -07002115 retval = ext4_delete_entry(handle, dir, de, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002116 if (retval)
2117 goto end_rmdir;
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002118 if (!EXT4_DIR_LINK_EMPTY(inode))
Mingming Cao617ba132006-10-11 01:20:53 -07002119 ext4_warning (inode->i_sb, "ext4_rmdir",
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002120 "empty directory has too many links (%d)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002121 inode->i_nlink);
2122 inode->i_version++;
2123 clear_nlink(inode);
2124 /* There's no need to set i_disksize: the fact that i_nlink is
2125 * zero will ensure that the right thing happens during any
2126 * recovery. */
2127 inode->i_size = 0;
Mingming Cao617ba132006-10-11 01:20:53 -07002128 ext4_orphan_add(handle, inode);
Kalpak Shahef7f3832007-07-18 09:15:20 -04002129 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002130 ext4_mark_inode_dirty(handle, inode);
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002131 ext4_dec_count(handle, dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002132 ext4_update_dx_flag(dir);
2133 ext4_mark_inode_dirty(handle, dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002134
2135end_rmdir:
Mingming Cao617ba132006-10-11 01:20:53 -07002136 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002137 brelse (bh);
2138 return retval;
2139}
2140
Mingming Cao617ba132006-10-11 01:20:53 -07002141static int ext4_unlink(struct inode * dir, struct dentry *dentry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002142{
2143 int retval;
2144 struct inode * inode;
2145 struct buffer_head * bh;
Mingming Cao617ba132006-10-11 01:20:53 -07002146 struct ext4_dir_entry_2 * de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002147 handle_t *handle;
2148
2149 /* Initialize quotas before so that eventual writes go
2150 * in separate transaction */
2151 DQUOT_INIT(dentry->d_inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002152 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002153 if (IS_ERR(handle))
2154 return PTR_ERR(handle);
2155
2156 if (IS_DIRSYNC(dir))
2157 handle->h_sync = 1;
2158
2159 retval = -ENOENT;
Mingming Cao617ba132006-10-11 01:20:53 -07002160 bh = ext4_find_entry (dentry, &de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002161 if (!bh)
2162 goto end_unlink;
2163
2164 inode = dentry->d_inode;
2165
2166 retval = -EIO;
2167 if (le32_to_cpu(de->inode) != inode->i_ino)
2168 goto end_unlink;
2169
2170 if (!inode->i_nlink) {
Mingming Cao617ba132006-10-11 01:20:53 -07002171 ext4_warning (inode->i_sb, "ext4_unlink",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002172 "Deleting nonexistent file (%lu), %d",
2173 inode->i_ino, inode->i_nlink);
2174 inode->i_nlink = 1;
2175 }
Mingming Cao617ba132006-10-11 01:20:53 -07002176 retval = ext4_delete_entry(handle, dir, de, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002177 if (retval)
2178 goto end_unlink;
Kalpak Shahef7f3832007-07-18 09:15:20 -04002179 dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002180 ext4_update_dx_flag(dir);
2181 ext4_mark_inode_dirty(handle, dir);
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002182 ext4_dec_count(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002183 if (!inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07002184 ext4_orphan_add(handle, inode);
Kalpak Shahef7f3832007-07-18 09:15:20 -04002185 inode->i_ctime = ext4_current_time(inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002186 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002187 retval = 0;
2188
2189end_unlink:
Mingming Cao617ba132006-10-11 01:20:53 -07002190 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002191 brelse (bh);
2192 return retval;
2193}
2194
Mingming Cao617ba132006-10-11 01:20:53 -07002195static int ext4_symlink (struct inode * dir,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002196 struct dentry *dentry, const char * symname)
2197{
2198 handle_t *handle;
2199 struct inode * inode;
2200 int l, err, retries = 0;
2201
2202 l = strlen(symname)+1;
2203 if (l > dir->i_sb->s_blocksize)
2204 return -ENAMETOOLONG;
2205
2206retry:
Mingming Cao617ba132006-10-11 01:20:53 -07002207 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2208 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2209 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002210 if (IS_ERR(handle))
2211 return PTR_ERR(handle);
2212
2213 if (IS_DIRSYNC(dir))
2214 handle->h_sync = 1;
2215
Mingming Cao617ba132006-10-11 01:20:53 -07002216 inode = ext4_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002217 err = PTR_ERR(inode);
2218 if (IS_ERR(inode))
2219 goto out_stop;
2220
Mingming Cao617ba132006-10-11 01:20:53 -07002221 if (l > sizeof (EXT4_I(inode)->i_data)) {
2222 inode->i_op = &ext4_symlink_inode_operations;
2223 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002224 /*
Mingming Cao617ba132006-10-11 01:20:53 -07002225 * page_symlink() calls into ext4_prepare/commit_write.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002226 * We have a transaction open. All is sweetness. It also sets
2227 * i_size in generic_commit_write().
2228 */
2229 err = __page_symlink(inode, symname, l,
2230 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
2231 if (err) {
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002232 ext4_dec_count(handle, inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002233 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002234 iput (inode);
2235 goto out_stop;
2236 }
2237 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07002238 inode->i_op = &ext4_fast_symlink_inode_operations;
2239 memcpy((char*)&EXT4_I(inode)->i_data,symname,l);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002240 inode->i_size = l-1;
2241 }
Mingming Cao617ba132006-10-11 01:20:53 -07002242 EXT4_I(inode)->i_disksize = inode->i_size;
2243 err = ext4_add_nondir(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002244out_stop:
Mingming Cao617ba132006-10-11 01:20:53 -07002245 ext4_journal_stop(handle);
2246 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002247 goto retry;
2248 return err;
2249}
2250
Mingming Cao617ba132006-10-11 01:20:53 -07002251static int ext4_link (struct dentry * old_dentry,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002252 struct inode * dir, struct dentry *dentry)
2253{
2254 handle_t *handle;
2255 struct inode *inode = old_dentry->d_inode;
2256 int err, retries = 0;
2257
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002258 if (EXT4_DIR_LINK_MAX(inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002259 return -EMLINK;
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002260
Eric Sandeen2988a772007-02-10 01:46:16 -08002261 /*
2262 * Return -ENOENT if we've raced with unlink and i_nlink is 0. Doing
2263 * otherwise has the potential to corrupt the orphan inode list.
2264 */
2265 if (inode->i_nlink == 0)
2266 return -ENOENT;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002267
2268retry:
Mingming Cao617ba132006-10-11 01:20:53 -07002269 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2270 EXT4_INDEX_EXTRA_TRANS_BLOCKS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002271 if (IS_ERR(handle))
2272 return PTR_ERR(handle);
2273
2274 if (IS_DIRSYNC(dir))
2275 handle->h_sync = 1;
2276
Kalpak Shahef7f3832007-07-18 09:15:20 -04002277 inode->i_ctime = ext4_current_time(inode);
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002278 ext4_inc_count(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002279 atomic_inc(&inode->i_count);
2280
Mingming Cao617ba132006-10-11 01:20:53 -07002281 err = ext4_add_nondir(handle, dentry, inode);
2282 ext4_journal_stop(handle);
2283 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002284 goto retry;
2285 return err;
2286}
2287
2288#define PARENT_INO(buffer) \
Jan Karaa72d7f82008-01-28 23:58:27 -05002289 (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer))->inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002290
2291/*
2292 * Anybody can rename anything with this: the permission checks are left to the
2293 * higher-level routines.
2294 */
Mingming Cao617ba132006-10-11 01:20:53 -07002295static int ext4_rename (struct inode * old_dir, struct dentry *old_dentry,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002296 struct inode * new_dir,struct dentry *new_dentry)
2297{
2298 handle_t *handle;
2299 struct inode * old_inode, * new_inode;
2300 struct buffer_head * old_bh, * new_bh, * dir_bh;
Mingming Cao617ba132006-10-11 01:20:53 -07002301 struct ext4_dir_entry_2 * old_de, * new_de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002302 int retval;
2303
2304 old_bh = new_bh = dir_bh = NULL;
2305
2306 /* Initialize quotas before so that eventual writes go
2307 * in separate transaction */
2308 if (new_dentry->d_inode)
2309 DQUOT_INIT(new_dentry->d_inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002310 handle = ext4_journal_start(old_dir, 2 *
2311 EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2312 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002313 if (IS_ERR(handle))
2314 return PTR_ERR(handle);
2315
2316 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2317 handle->h_sync = 1;
2318
Mingming Cao617ba132006-10-11 01:20:53 -07002319 old_bh = ext4_find_entry (old_dentry, &old_de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002320 /*
2321 * Check for inode number is _not_ due to possible IO errors.
2322 * We might rmdir the source, keep it as pwd of some process
2323 * and merrily kill the link to whatever was created under the
2324 * same name. Goodbye sticky bit ;-<
2325 */
2326 old_inode = old_dentry->d_inode;
2327 retval = -ENOENT;
2328 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2329 goto end_rename;
2330
2331 new_inode = new_dentry->d_inode;
Mingming Cao617ba132006-10-11 01:20:53 -07002332 new_bh = ext4_find_entry (new_dentry, &new_de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002333 if (new_bh) {
2334 if (!new_inode) {
2335 brelse (new_bh);
2336 new_bh = NULL;
2337 }
2338 }
2339 if (S_ISDIR(old_inode->i_mode)) {
2340 if (new_inode) {
2341 retval = -ENOTEMPTY;
2342 if (!empty_dir (new_inode))
2343 goto end_rename;
2344 }
2345 retval = -EIO;
Mingming Cao617ba132006-10-11 01:20:53 -07002346 dir_bh = ext4_bread (handle, old_inode, 0, 0, &retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002347 if (!dir_bh)
2348 goto end_rename;
2349 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2350 goto end_rename;
2351 retval = -EMLINK;
2352 if (!new_inode && new_dir!=old_dir &&
Mingming Cao617ba132006-10-11 01:20:53 -07002353 new_dir->i_nlink >= EXT4_LINK_MAX)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002354 goto end_rename;
2355 }
2356 if (!new_bh) {
Mingming Cao617ba132006-10-11 01:20:53 -07002357 retval = ext4_add_entry (handle, new_dentry, old_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002358 if (retval)
2359 goto end_rename;
2360 } else {
2361 BUFFER_TRACE(new_bh, "get write access");
Mingming Cao617ba132006-10-11 01:20:53 -07002362 ext4_journal_get_write_access(handle, new_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002363 new_de->inode = cpu_to_le32(old_inode->i_ino);
Mingming Cao617ba132006-10-11 01:20:53 -07002364 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2365 EXT4_FEATURE_INCOMPAT_FILETYPE))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002366 new_de->file_type = old_de->file_type;
2367 new_dir->i_version++;
Mingming Cao617ba132006-10-11 01:20:53 -07002368 BUFFER_TRACE(new_bh, "call ext4_journal_dirty_metadata");
2369 ext4_journal_dirty_metadata(handle, new_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002370 brelse(new_bh);
2371 new_bh = NULL;
2372 }
2373
2374 /*
2375 * Like most other Unix systems, set the ctime for inodes on a
2376 * rename.
2377 */
Kalpak Shahef7f3832007-07-18 09:15:20 -04002378 old_inode->i_ctime = ext4_current_time(old_inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002379 ext4_mark_inode_dirty(handle, old_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002380
2381 /*
2382 * ok, that's it
2383 */
2384 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2385 old_de->name_len != old_dentry->d_name.len ||
2386 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
Mingming Cao617ba132006-10-11 01:20:53 -07002387 (retval = ext4_delete_entry(handle, old_dir,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002388 old_de, old_bh)) == -ENOENT) {
2389 /* old_de could have moved from under us during htree split, so
2390 * make sure that we are deleting the right entry. We might
2391 * also be pointing to a stale entry in the unused part of
2392 * old_bh so just checking inum and the name isn't enough. */
2393 struct buffer_head *old_bh2;
Mingming Cao617ba132006-10-11 01:20:53 -07002394 struct ext4_dir_entry_2 *old_de2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002395
Mingming Cao617ba132006-10-11 01:20:53 -07002396 old_bh2 = ext4_find_entry(old_dentry, &old_de2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002397 if (old_bh2) {
Mingming Cao617ba132006-10-11 01:20:53 -07002398 retval = ext4_delete_entry(handle, old_dir,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002399 old_de2, old_bh2);
2400 brelse(old_bh2);
2401 }
2402 }
2403 if (retval) {
Mingming Cao617ba132006-10-11 01:20:53 -07002404 ext4_warning(old_dir->i_sb, "ext4_rename",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002405 "Deleting old file (%lu), %d, error=%d",
2406 old_dir->i_ino, old_dir->i_nlink, retval);
2407 }
2408
2409 if (new_inode) {
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002410 ext4_dec_count(handle, new_inode);
Kalpak Shahef7f3832007-07-18 09:15:20 -04002411 new_inode->i_ctime = ext4_current_time(new_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002412 }
Kalpak Shahef7f3832007-07-18 09:15:20 -04002413 old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002414 ext4_update_dx_flag(old_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002415 if (dir_bh) {
2416 BUFFER_TRACE(dir_bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07002417 ext4_journal_get_write_access(handle, dir_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002418 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
Mingming Cao617ba132006-10-11 01:20:53 -07002419 BUFFER_TRACE(dir_bh, "call ext4_journal_dirty_metadata");
2420 ext4_journal_dirty_metadata(handle, dir_bh);
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002421 ext4_dec_count(handle, old_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002422 if (new_inode) {
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002423 /* checked empty_dir above, can't have another parent,
2424 * ext3_dec_count() won't work for many-linked dirs */
2425 new_inode->i_nlink = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002426 } else {
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002427 ext4_inc_count(handle, new_dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002428 ext4_update_dx_flag(new_dir);
2429 ext4_mark_inode_dirty(handle, new_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002430 }
2431 }
Mingming Cao617ba132006-10-11 01:20:53 -07002432 ext4_mark_inode_dirty(handle, old_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002433 if (new_inode) {
Mingming Cao617ba132006-10-11 01:20:53 -07002434 ext4_mark_inode_dirty(handle, new_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002435 if (!new_inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07002436 ext4_orphan_add(handle, new_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002437 }
2438 retval = 0;
2439
2440end_rename:
2441 brelse (dir_bh);
2442 brelse (old_bh);
2443 brelse (new_bh);
Mingming Cao617ba132006-10-11 01:20:53 -07002444 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002445 return retval;
2446}
2447
2448/*
2449 * directories can handle most operations...
2450 */
Arjan van de Ven754661f2007-02-12 00:55:38 -08002451const struct inode_operations ext4_dir_inode_operations = {
Mingming Cao617ba132006-10-11 01:20:53 -07002452 .create = ext4_create,
2453 .lookup = ext4_lookup,
2454 .link = ext4_link,
2455 .unlink = ext4_unlink,
2456 .symlink = ext4_symlink,
2457 .mkdir = ext4_mkdir,
2458 .rmdir = ext4_rmdir,
2459 .mknod = ext4_mknod,
2460 .rename = ext4_rename,
2461 .setattr = ext4_setattr,
2462#ifdef CONFIG_EXT4DEV_FS_XATTR
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002463 .setxattr = generic_setxattr,
2464 .getxattr = generic_getxattr,
Mingming Cao617ba132006-10-11 01:20:53 -07002465 .listxattr = ext4_listxattr,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002466 .removexattr = generic_removexattr,
2467#endif
Mingming Cao617ba132006-10-11 01:20:53 -07002468 .permission = ext4_permission,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002469};
2470
Arjan van de Ven754661f2007-02-12 00:55:38 -08002471const struct inode_operations ext4_special_inode_operations = {
Mingming Cao617ba132006-10-11 01:20:53 -07002472 .setattr = ext4_setattr,
2473#ifdef CONFIG_EXT4DEV_FS_XATTR
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002474 .setxattr = generic_setxattr,
2475 .getxattr = generic_getxattr,
Mingming Cao617ba132006-10-11 01:20:53 -07002476 .listxattr = ext4_listxattr,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002477 .removexattr = generic_removexattr,
2478#endif
Mingming Cao617ba132006-10-11 01:20:53 -07002479 .permission = ext4_permission,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002480};