blob: 68611945687d70cb63d461c77908e1217457fe1f [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;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700864
865 *res_dir = NULL;
866 sb = dir->i_sb;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700867 namelen = dentry->d_name.len;
Mingming Cao617ba132006-10-11 01:20:53 -0700868 if (namelen > EXT4_NAME_LEN)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700869 return NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700870 if (is_dx(dir)) {
Mingming Cao617ba132006-10-11 01:20:53 -0700871 bh = ext4_dx_find_entry(dentry, res_dir, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700872 /*
873 * On success, or if the error was file not found,
874 * return. Otherwise, fall back to doing a search the
875 * old fashioned way.
876 */
877 if (bh || (err != ERR_BAD_DX_DIR))
878 return bh;
Mingming Cao617ba132006-10-11 01:20:53 -0700879 dxtrace(printk("ext4_find_entry: dx failed, falling back\n"));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700880 }
Mingming Cao617ba132006-10-11 01:20:53 -0700881 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
882 start = EXT4_I(dir)->i_dir_start_lookup;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700883 if (start >= nblocks)
884 start = 0;
885 block = start;
886restart:
887 do {
888 /*
889 * We deal with the read-ahead logic here.
890 */
891 if (ra_ptr >= ra_max) {
892 /* Refill the readahead buffer */
893 ra_ptr = 0;
894 b = block;
895 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
896 /*
897 * Terminate if we reach the end of the
898 * directory and must wrap, or if our
899 * search has finished at this block.
900 */
901 if (b >= nblocks || (num && block == start)) {
902 bh_use[ra_max] = NULL;
903 break;
904 }
905 num++;
Mingming Cao617ba132006-10-11 01:20:53 -0700906 bh = ext4_getblk(NULL, dir, b++, 0, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700907 bh_use[ra_max] = bh;
908 if (bh)
909 ll_rw_block(READ_META, 1, &bh);
910 }
911 }
912 if ((bh = bh_use[ra_ptr++]) == NULL)
913 goto next;
914 wait_on_buffer(bh);
915 if (!buffer_uptodate(bh)) {
916 /* read error, skip block & hope for the best */
Mingming Cao617ba132006-10-11 01:20:53 -0700917 ext4_error(sb, __FUNCTION__, "reading directory #%lu "
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500918 "offset %lu", dir->i_ino,
919 (unsigned long)block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700920 brelse(bh);
921 goto next;
922 }
923 i = search_dirblock(bh, dir, dentry,
Mingming Cao617ba132006-10-11 01:20:53 -0700924 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700925 if (i == 1) {
Mingming Cao617ba132006-10-11 01:20:53 -0700926 EXT4_I(dir)->i_dir_start_lookup = block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700927 ret = bh;
928 goto cleanup_and_exit;
929 } else {
930 brelse(bh);
931 if (i < 0)
932 goto cleanup_and_exit;
933 }
934 next:
935 if (++block >= nblocks)
936 block = 0;
937 } while (block != start);
938
939 /*
940 * If the directory has grown while we were searching, then
941 * search the last part of the directory before giving up.
942 */
943 block = nblocks;
Mingming Cao617ba132006-10-11 01:20:53 -0700944 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700945 if (block < nblocks) {
946 start = 0;
947 goto restart;
948 }
949
950cleanup_and_exit:
951 /* Clean up the read-ahead blocks */
952 for (; ra_ptr < ra_max; ra_ptr++)
953 brelse (bh_use[ra_ptr]);
954 return ret;
955}
956
Mingming Cao617ba132006-10-11 01:20:53 -0700957static struct buffer_head * ext4_dx_find_entry(struct dentry *dentry,
958 struct ext4_dir_entry_2 **res_dir, int *err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700959{
960 struct super_block * sb;
961 struct dx_hash_info hinfo;
962 u32 hash;
963 struct dx_frame frames[2], *frame;
Mingming Cao617ba132006-10-11 01:20:53 -0700964 struct ext4_dir_entry_2 *de, *top;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700965 struct buffer_head *bh;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -0500966 ext4_lblk_t block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700967 int retval;
968 int namelen = dentry->d_name.len;
969 const u8 *name = dentry->d_name.name;
970 struct inode *dir = dentry->d_parent->d_inode;
971
972 sb = dir->i_sb;
973 /* NFS may look up ".." - look at dx_root directory block */
974 if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
975 if (!(frame = dx_probe(dentry, NULL, &hinfo, frames, err)))
976 return NULL;
977 } else {
978 frame = frames;
979 frame->bh = NULL; /* for dx_release() */
980 frame->at = (struct dx_entry *)frames; /* hack for zero entry*/
981 dx_set_block(frame->at, 0); /* dx_root block is 0 */
982 }
983 hash = hinfo.hash;
984 do {
985 block = dx_get_block(frame->at);
Mingming Cao617ba132006-10-11 01:20:53 -0700986 if (!(bh = ext4_bread (NULL,dir, block, 0, err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700987 goto errout;
Mingming Cao617ba132006-10-11 01:20:53 -0700988 de = (struct ext4_dir_entry_2 *) bh->b_data;
989 top = (struct ext4_dir_entry_2 *) ((char *) de + sb->s_blocksize -
990 EXT4_DIR_REC_LEN(0));
991 for (; de < top; de = ext4_next_entry(de))
992 if (ext4_match (namelen, name, de)) {
993 if (!ext4_check_dir_entry("ext4_find_entry",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700994 dir, de, bh,
Mingming Cao617ba132006-10-11 01:20:53 -0700995 (block<<EXT4_BLOCK_SIZE_BITS(sb))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700996 +((char *)de - bh->b_data))) {
997 brelse (bh);
Dmitriy Monakhovfedee542007-05-08 00:25:34 -0700998 *err = ERR_BAD_DX_DIR;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700999 goto errout;
1000 }
1001 *res_dir = de;
1002 dx_release (frames);
1003 return bh;
1004 }
1005 brelse (bh);
1006 /* Check to see if we should continue to search */
Mingming Cao617ba132006-10-11 01:20:53 -07001007 retval = ext4_htree_next_block(dir, hash, frame,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001008 frames, NULL);
1009 if (retval < 0) {
Mingming Cao617ba132006-10-11 01:20:53 -07001010 ext4_warning(sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001011 "error reading index page in directory #%lu",
1012 dir->i_ino);
1013 *err = retval;
1014 goto errout;
1015 }
1016 } while (retval == 1);
1017
1018 *err = -ENOENT;
1019errout:
1020 dxtrace(printk("%s not found\n", name));
1021 dx_release (frames);
1022 return NULL;
1023}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001024
Mingming Cao617ba132006-10-11 01:20:53 -07001025static struct dentry *ext4_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001026{
1027 struct inode * inode;
Mingming Cao617ba132006-10-11 01:20:53 -07001028 struct ext4_dir_entry_2 * de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001029 struct buffer_head * bh;
1030
Mingming Cao617ba132006-10-11 01:20:53 -07001031 if (dentry->d_name.len > EXT4_NAME_LEN)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001032 return ERR_PTR(-ENAMETOOLONG);
1033
Mingming Cao617ba132006-10-11 01:20:53 -07001034 bh = ext4_find_entry(dentry, &de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001035 inode = NULL;
1036 if (bh) {
1037 unsigned long ino = le32_to_cpu(de->inode);
1038 brelse (bh);
Mingming Cao617ba132006-10-11 01:20:53 -07001039 if (!ext4_valid_inum(dir->i_sb, ino)) {
1040 ext4_error(dir->i_sb, "ext4_lookup",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001041 "bad inode number: %lu", ino);
David Howells1d1fe1e2008-02-07 00:15:37 -08001042 return ERR_PTR(-EIO);
Vasily Averina6c15c22007-07-15 23:40:46 -07001043 }
David Howells1d1fe1e2008-02-07 00:15:37 -08001044 inode = ext4_iget(dir->i_sb, ino);
1045 if (IS_ERR(inode))
1046 return ERR_CAST(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001047 }
1048 return d_splice_alias(inode, dentry);
1049}
1050
1051
Mingming Cao617ba132006-10-11 01:20:53 -07001052struct dentry *ext4_get_parent(struct dentry *child)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001053{
1054 unsigned long ino;
1055 struct dentry *parent;
1056 struct inode *inode;
1057 struct dentry dotdot;
Mingming Cao617ba132006-10-11 01:20:53 -07001058 struct ext4_dir_entry_2 * de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001059 struct buffer_head *bh;
1060
1061 dotdot.d_name.name = "..";
1062 dotdot.d_name.len = 2;
1063 dotdot.d_parent = child; /* confusing, isn't it! */
1064
Mingming Cao617ba132006-10-11 01:20:53 -07001065 bh = ext4_find_entry(&dotdot, &de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001066 inode = NULL;
1067 if (!bh)
1068 return ERR_PTR(-ENOENT);
1069 ino = le32_to_cpu(de->inode);
1070 brelse(bh);
1071
Mingming Cao617ba132006-10-11 01:20:53 -07001072 if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1073 ext4_error(child->d_inode->i_sb, "ext4_get_parent",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001074 "bad inode number: %lu", ino);
David Howells1d1fe1e2008-02-07 00:15:37 -08001075 return ERR_PTR(-EIO);
Vasily Averina6c15c22007-07-15 23:40:46 -07001076 }
1077
David Howells1d1fe1e2008-02-07 00:15:37 -08001078 inode = ext4_iget(child->d_inode->i_sb, ino);
1079 if (IS_ERR(inode))
1080 return ERR_CAST(inode);
1081
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001082 parent = d_alloc_anon(inode);
1083 if (!parent) {
1084 iput(inode);
1085 parent = ERR_PTR(-ENOMEM);
1086 }
1087 return parent;
1088}
1089
1090#define S_SHIFT 12
Mingming Cao617ba132006-10-11 01:20:53 -07001091static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1092 [S_IFREG >> S_SHIFT] = EXT4_FT_REG_FILE,
1093 [S_IFDIR >> S_SHIFT] = EXT4_FT_DIR,
1094 [S_IFCHR >> S_SHIFT] = EXT4_FT_CHRDEV,
1095 [S_IFBLK >> S_SHIFT] = EXT4_FT_BLKDEV,
1096 [S_IFIFO >> S_SHIFT] = EXT4_FT_FIFO,
1097 [S_IFSOCK >> S_SHIFT] = EXT4_FT_SOCK,
1098 [S_IFLNK >> S_SHIFT] = EXT4_FT_SYMLINK,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001099};
1100
Mingming Cao617ba132006-10-11 01:20:53 -07001101static inline void ext4_set_de_type(struct super_block *sb,
1102 struct ext4_dir_entry_2 *de,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001103 umode_t mode) {
Mingming Cao617ba132006-10-11 01:20:53 -07001104 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1105 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001106}
1107
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001108/*
1109 * Move count entries from end of map between two memory locations.
1110 * Returns pointer to last entry moved.
1111 */
Mingming Cao617ba132006-10-11 01:20:53 -07001112static struct ext4_dir_entry_2 *
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001113dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1114{
1115 unsigned rec_len = 0;
1116
1117 while (count--) {
Mingming Cao617ba132006-10-11 01:20:53 -07001118 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *) (from + map->offs);
1119 rec_len = EXT4_DIR_REC_LEN(de->name_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001120 memcpy (to, de, rec_len);
Mingming Cao617ba132006-10-11 01:20:53 -07001121 ((struct ext4_dir_entry_2 *) to)->rec_len =
Jan Karaa72d7f82008-01-28 23:58:27 -05001122 ext4_rec_len_to_disk(rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001123 de->inode = 0;
1124 map++;
1125 to += rec_len;
1126 }
Mingming Cao617ba132006-10-11 01:20:53 -07001127 return (struct ext4_dir_entry_2 *) (to - rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001128}
1129
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001130/*
1131 * Compact each dir entry in the range to the minimal rec_len.
1132 * Returns pointer to last entry in range.
1133 */
Mingming Cao617ba132006-10-11 01:20:53 -07001134static struct ext4_dir_entry_2* dx_pack_dirents(char *base, int size)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001135{
Mingming Cao617ba132006-10-11 01:20:53 -07001136 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001137 unsigned rec_len = 0;
1138
1139 prev = to = de;
1140 while ((char*)de < base + size) {
Jan Karaa72d7f82008-01-28 23:58:27 -05001141 next = ext4_next_entry(de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001142 if (de->inode && de->name_len) {
Mingming Cao617ba132006-10-11 01:20:53 -07001143 rec_len = EXT4_DIR_REC_LEN(de->name_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001144 if (de > to)
1145 memmove(to, de, rec_len);
Jan Karaa72d7f82008-01-28 23:58:27 -05001146 to->rec_len = ext4_rec_len_to_disk(rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001147 prev = to;
Mingming Cao617ba132006-10-11 01:20:53 -07001148 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001149 }
1150 de = next;
1151 }
1152 return prev;
1153}
1154
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001155/*
1156 * Split a full leaf block to make room for a new dir entry.
1157 * Allocate a new block, and move entries so that they are approx. equally full.
1158 * Returns pointer to de in block into which the new entry will be inserted.
1159 */
Mingming Cao617ba132006-10-11 01:20:53 -07001160static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001161 struct buffer_head **bh,struct dx_frame *frame,
1162 struct dx_hash_info *hinfo, int *error)
1163{
1164 unsigned blocksize = dir->i_sb->s_blocksize;
1165 unsigned count, continued;
1166 struct buffer_head *bh2;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001167 ext4_lblk_t newblock;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001168 u32 hash2;
1169 struct dx_map_entry *map;
1170 char *data1 = (*bh)->b_data, *data2;
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001171 unsigned split, move, size, i;
Mingming Cao617ba132006-10-11 01:20:53 -07001172 struct ext4_dir_entry_2 *de = NULL, *de2;
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001173 int err = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001174
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001175 bh2 = ext4_append (handle, dir, &newblock, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001176 if (!(bh2)) {
1177 brelse(*bh);
1178 *bh = NULL;
1179 goto errout;
1180 }
1181
1182 BUFFER_TRACE(*bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001183 err = ext4_journal_get_write_access(handle, *bh);
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001184 if (err)
1185 goto journal_error;
1186
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001187 BUFFER_TRACE(frame->bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001188 err = ext4_journal_get_write_access(handle, frame->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001189 if (err)
1190 goto journal_error;
1191
1192 data2 = bh2->b_data;
1193
1194 /* create map in the end of data2 block */
1195 map = (struct dx_map_entry *) (data2 + blocksize);
Mingming Cao617ba132006-10-11 01:20:53 -07001196 count = dx_make_map ((struct ext4_dir_entry_2 *) data1,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001197 blocksize, hinfo, map);
1198 map -= count;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001199 dx_sort_map (map, count);
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001200 /* Split the existing block in the middle, size-wise */
1201 size = 0;
1202 move = 0;
1203 for (i = count-1; i >= 0; i--) {
1204 /* is more than half of this entry in 2nd half of the block? */
1205 if (size + map[i].size/2 > blocksize/2)
1206 break;
1207 size += map[i].size;
1208 move++;
1209 }
1210 /* map index at which we will split */
1211 split = count - move;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001212 hash2 = map[split].hash;
1213 continued = hash2 == map[split - 1].hash;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001214 dxtrace(printk(KERN_INFO "Split block %lu at %x, %i/%i\n",
1215 (unsigned long)dx_get_block(frame->at),
1216 hash2, split, count-split));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001217
1218 /* Fancy dance to stay within two buffers */
1219 de2 = dx_move_dirents(data1, data2, map + split, count - split);
1220 de = dx_pack_dirents(data1,blocksize);
Jan Karaa72d7f82008-01-28 23:58:27 -05001221 de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de);
1222 de2->rec_len = ext4_rec_len_to_disk(data2 + blocksize - (char *) de2);
Mingming Cao617ba132006-10-11 01:20:53 -07001223 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1224 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001225
1226 /* Which block gets the new entry? */
1227 if (hinfo->hash >= hash2)
1228 {
1229 swap(*bh, bh2);
1230 de = de2;
1231 }
1232 dx_insert_block (frame, hash2 + continued, newblock);
Mingming Cao617ba132006-10-11 01:20:53 -07001233 err = ext4_journal_dirty_metadata (handle, bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001234 if (err)
1235 goto journal_error;
Mingming Cao617ba132006-10-11 01:20:53 -07001236 err = ext4_journal_dirty_metadata (handle, frame->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001237 if (err)
1238 goto journal_error;
1239 brelse (bh2);
1240 dxtrace(dx_show_index ("frame", frame->entries));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001241 return de;
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001242
1243journal_error:
1244 brelse(*bh);
1245 brelse(bh2);
1246 *bh = NULL;
1247 ext4_std_error(dir->i_sb, err);
1248errout:
1249 *error = err;
1250 return NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001251}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001252
1253/*
1254 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1255 * it points to a directory entry which is guaranteed to be large
1256 * enough for new directory entry. If de is NULL, then
1257 * add_dirent_to_buf will attempt search the directory block for
1258 * space. It will return -ENOSPC if no space is available, and -EIO
1259 * and -EEXIST if directory entry already exists.
1260 *
1261 * NOTE! bh is NOT released in the case where ENOSPC is returned. In
1262 * all other cases bh is released.
1263 */
1264static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
Mingming Cao617ba132006-10-11 01:20:53 -07001265 struct inode *inode, struct ext4_dir_entry_2 *de,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001266 struct buffer_head * bh)
1267{
1268 struct inode *dir = dentry->d_parent->d_inode;
1269 const char *name = dentry->d_name.name;
1270 int namelen = dentry->d_name.len;
1271 unsigned long offset = 0;
1272 unsigned short reclen;
1273 int nlen, rlen, err;
1274 char *top;
1275
Mingming Cao617ba132006-10-11 01:20:53 -07001276 reclen = EXT4_DIR_REC_LEN(namelen);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001277 if (!de) {
Mingming Cao617ba132006-10-11 01:20:53 -07001278 de = (struct ext4_dir_entry_2 *)bh->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001279 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1280 while ((char *) de <= top) {
Mingming Cao617ba132006-10-11 01:20:53 -07001281 if (!ext4_check_dir_entry("ext4_add_entry", dir, de,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001282 bh, offset)) {
1283 brelse (bh);
1284 return -EIO;
1285 }
Mingming Cao617ba132006-10-11 01:20:53 -07001286 if (ext4_match (namelen, name, de)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001287 brelse (bh);
1288 return -EEXIST;
1289 }
Mingming Cao617ba132006-10-11 01:20:53 -07001290 nlen = EXT4_DIR_REC_LEN(de->name_len);
Jan Karaa72d7f82008-01-28 23:58:27 -05001291 rlen = ext4_rec_len_from_disk(de->rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001292 if ((de->inode? rlen - nlen: rlen) >= reclen)
1293 break;
Mingming Cao617ba132006-10-11 01:20:53 -07001294 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001295 offset += rlen;
1296 }
1297 if ((char *) de > top)
1298 return -ENOSPC;
1299 }
1300 BUFFER_TRACE(bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001301 err = ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001302 if (err) {
Mingming Cao617ba132006-10-11 01:20:53 -07001303 ext4_std_error(dir->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001304 brelse(bh);
1305 return err;
1306 }
1307
1308 /* By now the buffer is marked for journaling */
Mingming Cao617ba132006-10-11 01:20:53 -07001309 nlen = EXT4_DIR_REC_LEN(de->name_len);
Jan Karaa72d7f82008-01-28 23:58:27 -05001310 rlen = ext4_rec_len_from_disk(de->rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001311 if (de->inode) {
Mingming Cao617ba132006-10-11 01:20:53 -07001312 struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
Jan Karaa72d7f82008-01-28 23:58:27 -05001313 de1->rec_len = ext4_rec_len_to_disk(rlen - nlen);
1314 de->rec_len = ext4_rec_len_to_disk(nlen);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001315 de = de1;
1316 }
Mingming Cao617ba132006-10-11 01:20:53 -07001317 de->file_type = EXT4_FT_UNKNOWN;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001318 if (inode) {
1319 de->inode = cpu_to_le32(inode->i_ino);
Mingming Cao617ba132006-10-11 01:20:53 -07001320 ext4_set_de_type(dir->i_sb, de, inode->i_mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001321 } else
1322 de->inode = 0;
1323 de->name_len = namelen;
1324 memcpy (de->name, name, namelen);
1325 /*
1326 * XXX shouldn't update any times until successful
1327 * completion of syscall, but too many callers depend
1328 * on this.
1329 *
1330 * XXX similarly, too many callers depend on
Mingming Cao617ba132006-10-11 01:20:53 -07001331 * ext4_new_inode() setting the times, but error
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001332 * recovery deletes the inode, so the worst that can
1333 * happen is that the times are slightly out of date
1334 * and/or different from the directory change time.
1335 */
Kalpak Shahef7f3832007-07-18 09:15:20 -04001336 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
Mingming Cao617ba132006-10-11 01:20:53 -07001337 ext4_update_dx_flag(dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001338 dir->i_version++;
Mingming Cao617ba132006-10-11 01:20:53 -07001339 ext4_mark_inode_dirty(handle, dir);
1340 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1341 err = ext4_journal_dirty_metadata(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001342 if (err)
Mingming Cao617ba132006-10-11 01:20:53 -07001343 ext4_std_error(dir->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001344 brelse(bh);
1345 return 0;
1346}
1347
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001348/*
1349 * This converts a one block unindexed directory to a 3 block indexed
1350 * directory, and adds the dentry to the indexed directory.
1351 */
1352static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1353 struct inode *inode, struct buffer_head *bh)
1354{
1355 struct inode *dir = dentry->d_parent->d_inode;
1356 const char *name = dentry->d_name.name;
1357 int namelen = dentry->d_name.len;
1358 struct buffer_head *bh2;
1359 struct dx_root *root;
1360 struct dx_frame frames[2], *frame;
1361 struct dx_entry *entries;
Mingming Cao617ba132006-10-11 01:20:53 -07001362 struct ext4_dir_entry_2 *de, *de2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001363 char *data1, *top;
1364 unsigned len;
1365 int retval;
1366 unsigned blocksize;
1367 struct dx_hash_info hinfo;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001368 ext4_lblk_t block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001369 struct fake_dirent *fde;
1370
1371 blocksize = dir->i_sb->s_blocksize;
1372 dxtrace(printk("Creating index\n"));
Mingming Cao617ba132006-10-11 01:20:53 -07001373 retval = ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001374 if (retval) {
Mingming Cao617ba132006-10-11 01:20:53 -07001375 ext4_std_error(dir->i_sb, retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001376 brelse(bh);
1377 return retval;
1378 }
1379 root = (struct dx_root *) bh->b_data;
1380
Mingming Cao617ba132006-10-11 01:20:53 -07001381 bh2 = ext4_append (handle, dir, &block, &retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001382 if (!(bh2)) {
1383 brelse(bh);
1384 return retval;
1385 }
Mingming Cao617ba132006-10-11 01:20:53 -07001386 EXT4_I(dir)->i_flags |= EXT4_INDEX_FL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001387 data1 = bh2->b_data;
1388
1389 /* The 0th block becomes the root, move the dirents out */
1390 fde = &root->dotdot;
Jan Karaa72d7f82008-01-28 23:58:27 -05001391 de = (struct ext4_dir_entry_2 *)((char *)fde +
1392 ext4_rec_len_from_disk(fde->rec_len));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001393 len = ((char *) root) + blocksize - (char *) de;
1394 memcpy (data1, de, len);
Mingming Cao617ba132006-10-11 01:20:53 -07001395 de = (struct ext4_dir_entry_2 *) data1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001396 top = data1 + len;
Jan Karaa72d7f82008-01-28 23:58:27 -05001397 while ((char *)(de2 = ext4_next_entry(de)) < top)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001398 de = de2;
Jan Karaa72d7f82008-01-28 23:58:27 -05001399 de->rec_len = ext4_rec_len_to_disk(data1 + blocksize - (char *) de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001400 /* Initialize the root; the dot dirents already exist */
Mingming Cao617ba132006-10-11 01:20:53 -07001401 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
Jan Karaa72d7f82008-01-28 23:58:27 -05001402 de->rec_len = ext4_rec_len_to_disk(blocksize - EXT4_DIR_REC_LEN(2));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001403 memset (&root->info, 0, sizeof(root->info));
1404 root->info.info_length = sizeof(root->info);
Mingming Cao617ba132006-10-11 01:20:53 -07001405 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001406 entries = root->entries;
1407 dx_set_block (entries, 1);
1408 dx_set_count (entries, 1);
1409 dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1410
1411 /* Initialize as for dx_probe */
1412 hinfo.hash_version = root->info.hash_version;
Mingming Cao617ba132006-10-11 01:20:53 -07001413 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1414 ext4fs_dirhash(name, namelen, &hinfo);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001415 frame = frames;
1416 frame->entries = entries;
1417 frame->at = entries;
1418 frame->bh = bh;
1419 bh = bh2;
1420 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1421 dx_release (frames);
1422 if (!(de))
1423 return retval;
1424
1425 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1426}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001427
1428/*
Mingming Cao617ba132006-10-11 01:20:53 -07001429 * ext4_add_entry()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001430 *
1431 * adds a file entry to the specified directory, using the same
Mingming Cao617ba132006-10-11 01:20:53 -07001432 * semantics as ext4_find_entry(). It returns NULL if it failed.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001433 *
1434 * NOTE!! The inode part of 'de' is left at 0 - which means you
1435 * may not sleep between calling this and putting something into
1436 * the entry, as someone else might have used it while you slept.
1437 */
Mingming Cao617ba132006-10-11 01:20:53 -07001438static int ext4_add_entry (handle_t *handle, struct dentry *dentry,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001439 struct inode *inode)
1440{
1441 struct inode *dir = dentry->d_parent->d_inode;
1442 unsigned long offset;
1443 struct buffer_head * bh;
Mingming Cao617ba132006-10-11 01:20:53 -07001444 struct ext4_dir_entry_2 *de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001445 struct super_block * sb;
1446 int retval;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001447 int dx_fallback=0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001448 unsigned blocksize;
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001449 ext4_lblk_t block, blocks;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001450
1451 sb = dir->i_sb;
1452 blocksize = sb->s_blocksize;
1453 if (!dentry->d_name.len)
1454 return -EINVAL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001455 if (is_dx(dir)) {
Mingming Cao617ba132006-10-11 01:20:53 -07001456 retval = ext4_dx_add_entry(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001457 if (!retval || (retval != ERR_BAD_DX_DIR))
1458 return retval;
Mingming Cao617ba132006-10-11 01:20:53 -07001459 EXT4_I(dir)->i_flags &= ~EXT4_INDEX_FL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001460 dx_fallback++;
Mingming Cao617ba132006-10-11 01:20:53 -07001461 ext4_mark_inode_dirty(handle, dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001462 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001463 blocks = dir->i_size >> sb->s_blocksize_bits;
1464 for (block = 0, offset = 0; block < blocks; block++) {
Mingming Cao617ba132006-10-11 01:20:53 -07001465 bh = ext4_bread(handle, dir, block, 0, &retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001466 if(!bh)
1467 return retval;
1468 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1469 if (retval != -ENOSPC)
1470 return retval;
1471
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001472 if (blocks == 1 && !dx_fallback &&
Mingming Cao617ba132006-10-11 01:20:53 -07001473 EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001474 return make_indexed_dir(handle, dentry, inode, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001475 brelse(bh);
1476 }
Mingming Cao617ba132006-10-11 01:20:53 -07001477 bh = ext4_append(handle, dir, &block, &retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001478 if (!bh)
1479 return retval;
Mingming Cao617ba132006-10-11 01:20:53 -07001480 de = (struct ext4_dir_entry_2 *) bh->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001481 de->inode = 0;
Jan Karaa72d7f82008-01-28 23:58:27 -05001482 de->rec_len = ext4_rec_len_to_disk(blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001483 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1484}
1485
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001486/*
1487 * Returns 0 for success, or a negative error value
1488 */
Mingming Cao617ba132006-10-11 01:20:53 -07001489static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001490 struct inode *inode)
1491{
1492 struct dx_frame frames[2], *frame;
1493 struct dx_entry *entries, *at;
1494 struct dx_hash_info hinfo;
1495 struct buffer_head * bh;
1496 struct inode *dir = dentry->d_parent->d_inode;
1497 struct super_block * sb = dir->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -07001498 struct ext4_dir_entry_2 *de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001499 int err;
1500
1501 frame = dx_probe(dentry, NULL, &hinfo, frames, &err);
1502 if (!frame)
1503 return err;
1504 entries = frame->entries;
1505 at = frame->at;
1506
Mingming Cao617ba132006-10-11 01:20:53 -07001507 if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001508 goto cleanup;
1509
1510 BUFFER_TRACE(bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001511 err = ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001512 if (err)
1513 goto journal_error;
1514
1515 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1516 if (err != -ENOSPC) {
1517 bh = NULL;
1518 goto cleanup;
1519 }
1520
1521 /* Block full, should compress but for now just split */
1522 dxtrace(printk("using %u of %u node entries\n",
1523 dx_get_count(entries), dx_get_limit(entries)));
1524 /* Need to split index? */
1525 if (dx_get_count(entries) == dx_get_limit(entries)) {
Aneesh Kumar K.V725d26d2008-01-28 23:58:27 -05001526 ext4_lblk_t newblock;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001527 unsigned icount = dx_get_count(entries);
1528 int levels = frame - frames;
1529 struct dx_entry *entries2;
1530 struct dx_node *node2;
1531 struct buffer_head *bh2;
1532
1533 if (levels && (dx_get_count(frames->entries) ==
1534 dx_get_limit(frames->entries))) {
Mingming Cao617ba132006-10-11 01:20:53 -07001535 ext4_warning(sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001536 "Directory index full!");
1537 err = -ENOSPC;
1538 goto cleanup;
1539 }
Mingming Cao617ba132006-10-11 01:20:53 -07001540 bh2 = ext4_append (handle, dir, &newblock, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001541 if (!(bh2))
1542 goto cleanup;
1543 node2 = (struct dx_node *)(bh2->b_data);
1544 entries2 = node2->entries;
Jan Karaa72d7f82008-01-28 23:58:27 -05001545 node2->fake.rec_len = ext4_rec_len_to_disk(sb->s_blocksize);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001546 node2->fake.inode = 0;
1547 BUFFER_TRACE(frame->bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001548 err = ext4_journal_get_write_access(handle, frame->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001549 if (err)
1550 goto journal_error;
1551 if (levels) {
1552 unsigned icount1 = icount/2, icount2 = icount - icount1;
1553 unsigned hash2 = dx_get_hash(entries + icount1);
1554 dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1555
1556 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
Mingming Cao617ba132006-10-11 01:20:53 -07001557 err = ext4_journal_get_write_access(handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001558 frames[0].bh);
1559 if (err)
1560 goto journal_error;
1561
1562 memcpy ((char *) entries2, (char *) (entries + icount1),
1563 icount2 * sizeof(struct dx_entry));
1564 dx_set_count (entries, icount1);
1565 dx_set_count (entries2, icount2);
1566 dx_set_limit (entries2, dx_node_limit(dir));
1567
1568 /* Which index block gets the new entry? */
1569 if (at - entries >= icount1) {
1570 frame->at = at = at - entries - icount1 + entries2;
1571 frame->entries = entries = entries2;
1572 swap(frame->bh, bh2);
1573 }
1574 dx_insert_block (frames + 0, hash2, newblock);
1575 dxtrace(dx_show_index ("node", frames[1].entries));
1576 dxtrace(dx_show_index ("node",
1577 ((struct dx_node *) bh2->b_data)->entries));
Mingming Cao617ba132006-10-11 01:20:53 -07001578 err = ext4_journal_dirty_metadata(handle, bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001579 if (err)
1580 goto journal_error;
1581 brelse (bh2);
1582 } else {
1583 dxtrace(printk("Creating second level index...\n"));
1584 memcpy((char *) entries2, (char *) entries,
1585 icount * sizeof(struct dx_entry));
1586 dx_set_limit(entries2, dx_node_limit(dir));
1587
1588 /* Set up root */
1589 dx_set_count(entries, 1);
1590 dx_set_block(entries + 0, newblock);
1591 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1592
1593 /* Add new access path frame */
1594 frame = frames + 1;
1595 frame->at = at = at - entries + entries2;
1596 frame->entries = entries = entries2;
1597 frame->bh = bh2;
Mingming Cao617ba132006-10-11 01:20:53 -07001598 err = ext4_journal_get_write_access(handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001599 frame->bh);
1600 if (err)
1601 goto journal_error;
1602 }
Mingming Cao617ba132006-10-11 01:20:53 -07001603 ext4_journal_dirty_metadata(handle, frames[0].bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001604 }
1605 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1606 if (!de)
1607 goto cleanup;
1608 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1609 bh = NULL;
1610 goto cleanup;
1611
1612journal_error:
Mingming Cao617ba132006-10-11 01:20:53 -07001613 ext4_std_error(dir->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001614cleanup:
1615 if (bh)
1616 brelse(bh);
1617 dx_release(frames);
1618 return err;
1619}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001620
1621/*
Mingming Cao617ba132006-10-11 01:20:53 -07001622 * ext4_delete_entry deletes a directory entry by merging it with the
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001623 * previous entry
1624 */
Mingming Cao617ba132006-10-11 01:20:53 -07001625static int ext4_delete_entry (handle_t *handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001626 struct inode * dir,
Mingming Cao617ba132006-10-11 01:20:53 -07001627 struct ext4_dir_entry_2 * de_del,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001628 struct buffer_head * bh)
1629{
Mingming Cao617ba132006-10-11 01:20:53 -07001630 struct ext4_dir_entry_2 * de, * pde;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001631 int i;
1632
1633 i = 0;
1634 pde = NULL;
Mingming Cao617ba132006-10-11 01:20:53 -07001635 de = (struct ext4_dir_entry_2 *) bh->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001636 while (i < bh->b_size) {
Mingming Cao617ba132006-10-11 01:20:53 -07001637 if (!ext4_check_dir_entry("ext4_delete_entry", dir, de, bh, i))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001638 return -EIO;
1639 if (de == de_del) {
1640 BUFFER_TRACE(bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001641 ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001642 if (pde)
Jan Karaa72d7f82008-01-28 23:58:27 -05001643 pde->rec_len = ext4_rec_len_to_disk(
1644 ext4_rec_len_from_disk(pde->rec_len) +
1645 ext4_rec_len_from_disk(de->rec_len));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001646 else
1647 de->inode = 0;
1648 dir->i_version++;
Mingming Cao617ba132006-10-11 01:20:53 -07001649 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1650 ext4_journal_dirty_metadata(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001651 return 0;
1652 }
Jan Karaa72d7f82008-01-28 23:58:27 -05001653 i += ext4_rec_len_from_disk(de->rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001654 pde = de;
Jan Karaa72d7f82008-01-28 23:58:27 -05001655 de = ext4_next_entry(de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001656 }
1657 return -ENOENT;
1658}
1659
Andreas Dilgerf8628a12007-07-18 08:38:01 -04001660/*
1661 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
1662 * since this indicates that nlinks count was previously 1.
1663 */
1664static void ext4_inc_count(handle_t *handle, struct inode *inode)
1665{
1666 inc_nlink(inode);
1667 if (is_dx(inode) && inode->i_nlink > 1) {
1668 /* limit is 16-bit i_links_count */
1669 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
1670 inode->i_nlink = 1;
1671 EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
1672 EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
1673 }
1674 }
1675}
1676
1677/*
1678 * If a directory had nlink == 1, then we should let it be 1. This indicates
1679 * directory has >EXT4_LINK_MAX subdirs.
1680 */
1681static void ext4_dec_count(handle_t *handle, struct inode *inode)
1682{
1683 drop_nlink(inode);
1684 if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
1685 inc_nlink(inode);
1686}
1687
1688
Mingming Cao617ba132006-10-11 01:20:53 -07001689static int ext4_add_nondir(handle_t *handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001690 struct dentry *dentry, struct inode *inode)
1691{
Mingming Cao617ba132006-10-11 01:20:53 -07001692 int err = ext4_add_entry(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001693 if (!err) {
Mingming Cao617ba132006-10-11 01:20:53 -07001694 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001695 d_instantiate(dentry, inode);
1696 return 0;
1697 }
Eric Sandeen731b9a52007-02-10 01:46:16 -08001698 drop_nlink(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001699 iput(inode);
1700 return err;
1701}
1702
1703/*
1704 * By the time this is called, we already have created
1705 * the directory cache entry for the new file, but it
1706 * is so far negative - it has no inode.
1707 *
1708 * If the create succeeds, we fill in the inode information
1709 * with d_instantiate().
1710 */
Mingming Cao617ba132006-10-11 01:20:53 -07001711static int ext4_create (struct inode * dir, struct dentry * dentry, int mode,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001712 struct nameidata *nd)
1713{
1714 handle_t *handle;
1715 struct inode * inode;
1716 int err, retries = 0;
1717
1718retry:
Mingming Cao617ba132006-10-11 01:20:53 -07001719 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1720 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1721 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001722 if (IS_ERR(handle))
1723 return PTR_ERR(handle);
1724
1725 if (IS_DIRSYNC(dir))
1726 handle->h_sync = 1;
1727
Mingming Cao617ba132006-10-11 01:20:53 -07001728 inode = ext4_new_inode (handle, dir, mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001729 err = PTR_ERR(inode);
1730 if (!IS_ERR(inode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07001731 inode->i_op = &ext4_file_inode_operations;
1732 inode->i_fop = &ext4_file_operations;
1733 ext4_set_aops(inode);
1734 err = ext4_add_nondir(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001735 }
Mingming Cao617ba132006-10-11 01:20:53 -07001736 ext4_journal_stop(handle);
1737 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001738 goto retry;
1739 return err;
1740}
1741
Mingming Cao617ba132006-10-11 01:20:53 -07001742static int ext4_mknod (struct inode * dir, struct dentry *dentry,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001743 int mode, dev_t rdev)
1744{
1745 handle_t *handle;
1746 struct inode *inode;
1747 int err, retries = 0;
1748
1749 if (!new_valid_dev(rdev))
1750 return -EINVAL;
1751
1752retry:
Mingming Cao617ba132006-10-11 01:20:53 -07001753 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1754 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1755 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001756 if (IS_ERR(handle))
1757 return PTR_ERR(handle);
1758
1759 if (IS_DIRSYNC(dir))
1760 handle->h_sync = 1;
1761
Mingming Cao617ba132006-10-11 01:20:53 -07001762 inode = ext4_new_inode (handle, dir, mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001763 err = PTR_ERR(inode);
1764 if (!IS_ERR(inode)) {
1765 init_special_inode(inode, inode->i_mode, rdev);
Mingming Cao617ba132006-10-11 01:20:53 -07001766#ifdef CONFIG_EXT4DEV_FS_XATTR
1767 inode->i_op = &ext4_special_inode_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001768#endif
Mingming Cao617ba132006-10-11 01:20:53 -07001769 err = ext4_add_nondir(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001770 }
Mingming Cao617ba132006-10-11 01:20:53 -07001771 ext4_journal_stop(handle);
1772 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001773 goto retry;
1774 return err;
1775}
1776
Mingming Cao617ba132006-10-11 01:20:53 -07001777static int ext4_mkdir(struct inode * dir, struct dentry * dentry, int mode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001778{
1779 handle_t *handle;
1780 struct inode * inode;
1781 struct buffer_head * dir_block;
Mingming Cao617ba132006-10-11 01:20:53 -07001782 struct ext4_dir_entry_2 * de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001783 int err, retries = 0;
1784
Andreas Dilgerf8628a12007-07-18 08:38:01 -04001785 if (EXT4_DIR_LINK_MAX(dir))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001786 return -EMLINK;
1787
1788retry:
Mingming Cao617ba132006-10-11 01:20:53 -07001789 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1790 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1791 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001792 if (IS_ERR(handle))
1793 return PTR_ERR(handle);
1794
1795 if (IS_DIRSYNC(dir))
1796 handle->h_sync = 1;
1797
Mingming Cao617ba132006-10-11 01:20:53 -07001798 inode = ext4_new_inode (handle, dir, S_IFDIR | mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001799 err = PTR_ERR(inode);
1800 if (IS_ERR(inode))
1801 goto out_stop;
1802
Mingming Cao617ba132006-10-11 01:20:53 -07001803 inode->i_op = &ext4_dir_inode_operations;
1804 inode->i_fop = &ext4_dir_operations;
1805 inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1806 dir_block = ext4_bread (handle, inode, 0, 1, &err);
Aneesh Kumar K.V4cdeed82008-02-22 06:17:31 -05001807 if (!dir_block)
1808 goto out_clear_inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001809 BUFFER_TRACE(dir_block, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001810 ext4_journal_get_write_access(handle, dir_block);
1811 de = (struct ext4_dir_entry_2 *) dir_block->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001812 de->inode = cpu_to_le32(inode->i_ino);
1813 de->name_len = 1;
Jan Karaa72d7f82008-01-28 23:58:27 -05001814 de->rec_len = ext4_rec_len_to_disk(EXT4_DIR_REC_LEN(de->name_len));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001815 strcpy (de->name, ".");
Mingming Cao617ba132006-10-11 01:20:53 -07001816 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
Jan Karaa72d7f82008-01-28 23:58:27 -05001817 de = ext4_next_entry(de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001818 de->inode = cpu_to_le32(dir->i_ino);
Jan Karaa72d7f82008-01-28 23:58:27 -05001819 de->rec_len = ext4_rec_len_to_disk(inode->i_sb->s_blocksize -
1820 EXT4_DIR_REC_LEN(1));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001821 de->name_len = 2;
1822 strcpy (de->name, "..");
Mingming Cao617ba132006-10-11 01:20:53 -07001823 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001824 inode->i_nlink = 2;
Mingming Cao617ba132006-10-11 01:20:53 -07001825 BUFFER_TRACE(dir_block, "call ext4_journal_dirty_metadata");
1826 ext4_journal_dirty_metadata(handle, dir_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001827 brelse (dir_block);
Mingming Cao617ba132006-10-11 01:20:53 -07001828 ext4_mark_inode_dirty(handle, inode);
1829 err = ext4_add_entry (handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001830 if (err) {
Aneesh Kumar K.V4cdeed82008-02-22 06:17:31 -05001831out_clear_inode:
1832 clear_nlink(inode);
Mingming Cao617ba132006-10-11 01:20:53 -07001833 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001834 iput (inode);
1835 goto out_stop;
1836 }
Andreas Dilgerf8628a12007-07-18 08:38:01 -04001837 ext4_inc_count(handle, dir);
Mingming Cao617ba132006-10-11 01:20:53 -07001838 ext4_update_dx_flag(dir);
1839 ext4_mark_inode_dirty(handle, dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001840 d_instantiate(dentry, inode);
1841out_stop:
Mingming Cao617ba132006-10-11 01:20:53 -07001842 ext4_journal_stop(handle);
1843 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001844 goto retry;
1845 return err;
1846}
1847
1848/*
1849 * routine to check that the specified directory is empty (for rmdir)
1850 */
1851static int empty_dir (struct inode * inode)
1852{
1853 unsigned long offset;
1854 struct buffer_head * bh;
Mingming Cao617ba132006-10-11 01:20:53 -07001855 struct ext4_dir_entry_2 * de, * de1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001856 struct super_block * sb;
1857 int err = 0;
1858
1859 sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -07001860 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
1861 !(bh = ext4_bread (NULL, inode, 0, 0, &err))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001862 if (err)
Mingming Cao617ba132006-10-11 01:20:53 -07001863 ext4_error(inode->i_sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001864 "error %d reading directory #%lu offset 0",
1865 err, inode->i_ino);
1866 else
Mingming Cao617ba132006-10-11 01:20:53 -07001867 ext4_warning(inode->i_sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001868 "bad directory (dir #%lu) - no data block",
1869 inode->i_ino);
1870 return 1;
1871 }
Mingming Cao617ba132006-10-11 01:20:53 -07001872 de = (struct ext4_dir_entry_2 *) bh->b_data;
Jan Karaa72d7f82008-01-28 23:58:27 -05001873 de1 = ext4_next_entry(de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001874 if (le32_to_cpu(de->inode) != inode->i_ino ||
1875 !le32_to_cpu(de1->inode) ||
1876 strcmp (".", de->name) ||
1877 strcmp ("..", de1->name)) {
Mingming Cao617ba132006-10-11 01:20:53 -07001878 ext4_warning (inode->i_sb, "empty_dir",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001879 "bad directory (dir #%lu) - no `.' or `..'",
1880 inode->i_ino);
1881 brelse (bh);
1882 return 1;
1883 }
Jan Karaa72d7f82008-01-28 23:58:27 -05001884 offset = ext4_rec_len_from_disk(de->rec_len) +
1885 ext4_rec_len_from_disk(de1->rec_len);
1886 de = ext4_next_entry(de1);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001887 while (offset < inode->i_size ) {
1888 if (!bh ||
1889 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1890 err = 0;
1891 brelse (bh);
Mingming Cao617ba132006-10-11 01:20:53 -07001892 bh = ext4_bread (NULL, inode,
1893 offset >> EXT4_BLOCK_SIZE_BITS(sb), 0, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001894 if (!bh) {
1895 if (err)
Mingming Cao617ba132006-10-11 01:20:53 -07001896 ext4_error(sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001897 "error %d reading directory"
1898 " #%lu offset %lu",
1899 err, inode->i_ino, offset);
1900 offset += sb->s_blocksize;
1901 continue;
1902 }
Mingming Cao617ba132006-10-11 01:20:53 -07001903 de = (struct ext4_dir_entry_2 *) bh->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001904 }
Mingming Cao617ba132006-10-11 01:20:53 -07001905 if (!ext4_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1906 de = (struct ext4_dir_entry_2 *)(bh->b_data +
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001907 sb->s_blocksize);
1908 offset = (offset | (sb->s_blocksize - 1)) + 1;
1909 continue;
1910 }
1911 if (le32_to_cpu(de->inode)) {
1912 brelse (bh);
1913 return 0;
1914 }
Jan Karaa72d7f82008-01-28 23:58:27 -05001915 offset += ext4_rec_len_from_disk(de->rec_len);
1916 de = ext4_next_entry(de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001917 }
1918 brelse (bh);
1919 return 1;
1920}
1921
Mingming Cao617ba132006-10-11 01:20:53 -07001922/* ext4_orphan_add() links an unlinked or truncated inode into a list of
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001923 * such inodes, starting at the superblock, in case we crash before the
1924 * file is closed/deleted, or in case the inode truncate spans multiple
1925 * transactions and the last transaction is not recovered after a crash.
1926 *
1927 * At filesystem recovery time, we walk this list deleting unlinked
Mingming Cao617ba132006-10-11 01:20:53 -07001928 * inodes and truncating linked inodes in ext4_orphan_cleanup().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001929 */
Mingming Cao617ba132006-10-11 01:20:53 -07001930int ext4_orphan_add(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001931{
1932 struct super_block *sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -07001933 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001934 int err = 0, rc;
1935
1936 lock_super(sb);
Mingming Cao617ba132006-10-11 01:20:53 -07001937 if (!list_empty(&EXT4_I(inode)->i_orphan))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001938 goto out_unlock;
1939
1940 /* Orphan handling is only valid for files with data blocks
1941 * being truncated, or files being unlinked. */
1942
1943 /* @@@ FIXME: Observation from aviro:
Mingming Cao617ba132006-10-11 01:20:53 -07001944 * I think I can trigger J_ASSERT in ext4_orphan_add(). We block
1945 * here (on lock_super()), so race with ext4_link() which might bump
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001946 * ->i_nlink. For, say it, character device. Not a regular file,
1947 * not a directory, not a symlink and ->i_nlink > 0.
1948 */
1949 J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1950 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1951
Mingming Cao617ba132006-10-11 01:20:53 -07001952 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1953 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001954 if (err)
1955 goto out_unlock;
1956
Mingming Cao617ba132006-10-11 01:20:53 -07001957 err = ext4_reserve_inode_write(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001958 if (err)
1959 goto out_unlock;
1960
1961 /* Insert this inode at the head of the on-disk orphan list... */
Mingming Cao617ba132006-10-11 01:20:53 -07001962 NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
1963 EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1964 err = ext4_journal_dirty_metadata(handle, EXT4_SB(sb)->s_sbh);
1965 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001966 if (!err)
1967 err = rc;
1968
1969 /* Only add to the head of the in-memory list if all the
1970 * previous operations succeeded. If the orphan_add is going to
1971 * fail (possibly taking the journal offline), we can't risk
1972 * leaving the inode on the orphan list: stray orphan-list
1973 * entries can cause panics at unmount time.
1974 *
1975 * This is safe: on error we're going to ignore the orphan list
1976 * anyway on the next recovery. */
1977 if (!err)
Mingming Cao617ba132006-10-11 01:20:53 -07001978 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001979
1980 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1981 jbd_debug(4, "orphan inode %lu will point to %d\n",
1982 inode->i_ino, NEXT_ORPHAN(inode));
1983out_unlock:
1984 unlock_super(sb);
Mingming Cao617ba132006-10-11 01:20:53 -07001985 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001986 return err;
1987}
1988
1989/*
Mingming Cao617ba132006-10-11 01:20:53 -07001990 * ext4_orphan_del() removes an unlinked or truncated inode from the list
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001991 * of such inodes stored on disk, because it is finally being cleaned up.
1992 */
Mingming Cao617ba132006-10-11 01:20:53 -07001993int ext4_orphan_del(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001994{
1995 struct list_head *prev;
Mingming Cao617ba132006-10-11 01:20:53 -07001996 struct ext4_inode_info *ei = EXT4_I(inode);
1997 struct ext4_sb_info *sbi;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001998 unsigned long ino_next;
Mingming Cao617ba132006-10-11 01:20:53 -07001999 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002000 int err = 0;
2001
2002 lock_super(inode->i_sb);
2003 if (list_empty(&ei->i_orphan)) {
2004 unlock_super(inode->i_sb);
2005 return 0;
2006 }
2007
2008 ino_next = NEXT_ORPHAN(inode);
2009 prev = ei->i_orphan.prev;
Mingming Cao617ba132006-10-11 01:20:53 -07002010 sbi = EXT4_SB(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002011
2012 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2013
2014 list_del_init(&ei->i_orphan);
2015
2016 /* If we're on an error path, we may not have a valid
2017 * transaction handle with which to update the orphan list on
2018 * disk, but we still need to remove the inode from the linked
2019 * list in memory. */
2020 if (!handle)
2021 goto out;
2022
Mingming Cao617ba132006-10-11 01:20:53 -07002023 err = ext4_reserve_inode_write(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002024 if (err)
2025 goto out_err;
2026
2027 if (prev == &sbi->s_orphan) {
2028 jbd_debug(4, "superblock will point to %lu\n", ino_next);
2029 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07002030 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002031 if (err)
2032 goto out_brelse;
2033 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
Mingming Cao617ba132006-10-11 01:20:53 -07002034 err = ext4_journal_dirty_metadata(handle, sbi->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002035 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07002036 struct ext4_iloc iloc2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002037 struct inode *i_prev =
Mingming Cao617ba132006-10-11 01:20:53 -07002038 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002039
2040 jbd_debug(4, "orphan inode %lu will point to %lu\n",
2041 i_prev->i_ino, ino_next);
Mingming Cao617ba132006-10-11 01:20:53 -07002042 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002043 if (err)
2044 goto out_brelse;
2045 NEXT_ORPHAN(i_prev) = ino_next;
Mingming Cao617ba132006-10-11 01:20:53 -07002046 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002047 }
2048 if (err)
2049 goto out_brelse;
2050 NEXT_ORPHAN(inode) = 0;
Mingming Cao617ba132006-10-11 01:20:53 -07002051 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002052
2053out_err:
Mingming Cao617ba132006-10-11 01:20:53 -07002054 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002055out:
2056 unlock_super(inode->i_sb);
2057 return err;
2058
2059out_brelse:
2060 brelse(iloc.bh);
2061 goto out_err;
2062}
2063
Mingming Cao617ba132006-10-11 01:20:53 -07002064static int ext4_rmdir (struct inode * dir, struct dentry *dentry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002065{
2066 int retval;
2067 struct inode * inode;
2068 struct buffer_head * bh;
Mingming Cao617ba132006-10-11 01:20:53 -07002069 struct ext4_dir_entry_2 * de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002070 handle_t *handle;
2071
2072 /* Initialize quotas before so that eventual writes go in
2073 * separate transaction */
2074 DQUOT_INIT(dentry->d_inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002075 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002076 if (IS_ERR(handle))
2077 return PTR_ERR(handle);
2078
2079 retval = -ENOENT;
Mingming Cao617ba132006-10-11 01:20:53 -07002080 bh = ext4_find_entry (dentry, &de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002081 if (!bh)
2082 goto end_rmdir;
2083
2084 if (IS_DIRSYNC(dir))
2085 handle->h_sync = 1;
2086
2087 inode = dentry->d_inode;
2088
2089 retval = -EIO;
2090 if (le32_to_cpu(de->inode) != inode->i_ino)
2091 goto end_rmdir;
2092
2093 retval = -ENOTEMPTY;
2094 if (!empty_dir (inode))
2095 goto end_rmdir;
2096
Mingming Cao617ba132006-10-11 01:20:53 -07002097 retval = ext4_delete_entry(handle, dir, de, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002098 if (retval)
2099 goto end_rmdir;
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002100 if (!EXT4_DIR_LINK_EMPTY(inode))
Mingming Cao617ba132006-10-11 01:20:53 -07002101 ext4_warning (inode->i_sb, "ext4_rmdir",
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002102 "empty directory has too many links (%d)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002103 inode->i_nlink);
2104 inode->i_version++;
2105 clear_nlink(inode);
2106 /* There's no need to set i_disksize: the fact that i_nlink is
2107 * zero will ensure that the right thing happens during any
2108 * recovery. */
2109 inode->i_size = 0;
Mingming Cao617ba132006-10-11 01:20:53 -07002110 ext4_orphan_add(handle, inode);
Kalpak Shahef7f3832007-07-18 09:15:20 -04002111 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002112 ext4_mark_inode_dirty(handle, inode);
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002113 ext4_dec_count(handle, dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002114 ext4_update_dx_flag(dir);
2115 ext4_mark_inode_dirty(handle, dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002116
2117end_rmdir:
Mingming Cao617ba132006-10-11 01:20:53 -07002118 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002119 brelse (bh);
2120 return retval;
2121}
2122
Mingming Cao617ba132006-10-11 01:20:53 -07002123static int ext4_unlink(struct inode * dir, struct dentry *dentry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002124{
2125 int retval;
2126 struct inode * inode;
2127 struct buffer_head * bh;
Mingming Cao617ba132006-10-11 01:20:53 -07002128 struct ext4_dir_entry_2 * de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002129 handle_t *handle;
2130
2131 /* Initialize quotas before so that eventual writes go
2132 * in separate transaction */
2133 DQUOT_INIT(dentry->d_inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002134 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002135 if (IS_ERR(handle))
2136 return PTR_ERR(handle);
2137
2138 if (IS_DIRSYNC(dir))
2139 handle->h_sync = 1;
2140
2141 retval = -ENOENT;
Mingming Cao617ba132006-10-11 01:20:53 -07002142 bh = ext4_find_entry (dentry, &de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002143 if (!bh)
2144 goto end_unlink;
2145
2146 inode = dentry->d_inode;
2147
2148 retval = -EIO;
2149 if (le32_to_cpu(de->inode) != inode->i_ino)
2150 goto end_unlink;
2151
2152 if (!inode->i_nlink) {
Mingming Cao617ba132006-10-11 01:20:53 -07002153 ext4_warning (inode->i_sb, "ext4_unlink",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002154 "Deleting nonexistent file (%lu), %d",
2155 inode->i_ino, inode->i_nlink);
2156 inode->i_nlink = 1;
2157 }
Mingming Cao617ba132006-10-11 01:20:53 -07002158 retval = ext4_delete_entry(handle, dir, de, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002159 if (retval)
2160 goto end_unlink;
Kalpak Shahef7f3832007-07-18 09:15:20 -04002161 dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002162 ext4_update_dx_flag(dir);
2163 ext4_mark_inode_dirty(handle, dir);
Theodore Ts'o825f1482008-02-15 15:00:38 -05002164 drop_nlink(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002165 if (!inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07002166 ext4_orphan_add(handle, inode);
Kalpak Shahef7f3832007-07-18 09:15:20 -04002167 inode->i_ctime = ext4_current_time(inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002168 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002169 retval = 0;
2170
2171end_unlink:
Mingming Cao617ba132006-10-11 01:20:53 -07002172 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002173 brelse (bh);
2174 return retval;
2175}
2176
Mingming Cao617ba132006-10-11 01:20:53 -07002177static int ext4_symlink (struct inode * dir,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002178 struct dentry *dentry, const char * symname)
2179{
2180 handle_t *handle;
2181 struct inode * inode;
2182 int l, err, retries = 0;
2183
2184 l = strlen(symname)+1;
2185 if (l > dir->i_sb->s_blocksize)
2186 return -ENAMETOOLONG;
2187
2188retry:
Mingming Cao617ba132006-10-11 01:20:53 -07002189 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2190 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2191 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002192 if (IS_ERR(handle))
2193 return PTR_ERR(handle);
2194
2195 if (IS_DIRSYNC(dir))
2196 handle->h_sync = 1;
2197
Mingming Cao617ba132006-10-11 01:20:53 -07002198 inode = ext4_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002199 err = PTR_ERR(inode);
2200 if (IS_ERR(inode))
2201 goto out_stop;
2202
Mingming Cao617ba132006-10-11 01:20:53 -07002203 if (l > sizeof (EXT4_I(inode)->i_data)) {
2204 inode->i_op = &ext4_symlink_inode_operations;
2205 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002206 /*
Mingming Cao617ba132006-10-11 01:20:53 -07002207 * page_symlink() calls into ext4_prepare/commit_write.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002208 * We have a transaction open. All is sweetness. It also sets
2209 * i_size in generic_commit_write().
2210 */
2211 err = __page_symlink(inode, symname, l,
2212 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
2213 if (err) {
Theodore Ts'o825f1482008-02-15 15:00:38 -05002214 clear_nlink(inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002215 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002216 iput (inode);
2217 goto out_stop;
2218 }
2219 } else {
Aneesh Kumar K.Ve65187e2008-04-29 08:11:12 -04002220 /* clear the extent format for fast symlink */
2221 EXT4_I(inode)->i_flags &= ~EXT4_EXTENTS_FL;
Mingming Cao617ba132006-10-11 01:20:53 -07002222 inode->i_op = &ext4_fast_symlink_inode_operations;
2223 memcpy((char*)&EXT4_I(inode)->i_data,symname,l);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002224 inode->i_size = l-1;
2225 }
Mingming Cao617ba132006-10-11 01:20:53 -07002226 EXT4_I(inode)->i_disksize = inode->i_size;
2227 err = ext4_add_nondir(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002228out_stop:
Mingming Cao617ba132006-10-11 01:20:53 -07002229 ext4_journal_stop(handle);
2230 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002231 goto retry;
2232 return err;
2233}
2234
Mingming Cao617ba132006-10-11 01:20:53 -07002235static int ext4_link (struct dentry * old_dentry,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002236 struct inode * dir, struct dentry *dentry)
2237{
2238 handle_t *handle;
2239 struct inode *inode = old_dentry->d_inode;
2240 int err, retries = 0;
2241
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002242 if (EXT4_DIR_LINK_MAX(inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002243 return -EMLINK;
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002244
Eric Sandeen2988a772007-02-10 01:46:16 -08002245 /*
2246 * Return -ENOENT if we've raced with unlink and i_nlink is 0. Doing
2247 * otherwise has the potential to corrupt the orphan inode list.
2248 */
2249 if (inode->i_nlink == 0)
2250 return -ENOENT;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002251
2252retry:
Mingming Cao617ba132006-10-11 01:20:53 -07002253 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2254 EXT4_INDEX_EXTRA_TRANS_BLOCKS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002255 if (IS_ERR(handle))
2256 return PTR_ERR(handle);
2257
2258 if (IS_DIRSYNC(dir))
2259 handle->h_sync = 1;
2260
Kalpak Shahef7f3832007-07-18 09:15:20 -04002261 inode->i_ctime = ext4_current_time(inode);
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002262 ext4_inc_count(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002263 atomic_inc(&inode->i_count);
2264
Mingming Cao617ba132006-10-11 01:20:53 -07002265 err = ext4_add_nondir(handle, dentry, inode);
2266 ext4_journal_stop(handle);
2267 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002268 goto retry;
2269 return err;
2270}
2271
2272#define PARENT_INO(buffer) \
Jan Karaa72d7f82008-01-28 23:58:27 -05002273 (ext4_next_entry((struct ext4_dir_entry_2 *)(buffer))->inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002274
2275/*
2276 * Anybody can rename anything with this: the permission checks are left to the
2277 * higher-level routines.
2278 */
Mingming Cao617ba132006-10-11 01:20:53 -07002279static int ext4_rename (struct inode * old_dir, struct dentry *old_dentry,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002280 struct inode * new_dir,struct dentry *new_dentry)
2281{
2282 handle_t *handle;
2283 struct inode * old_inode, * new_inode;
2284 struct buffer_head * old_bh, * new_bh, * dir_bh;
Mingming Cao617ba132006-10-11 01:20:53 -07002285 struct ext4_dir_entry_2 * old_de, * new_de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002286 int retval;
2287
2288 old_bh = new_bh = dir_bh = NULL;
2289
2290 /* Initialize quotas before so that eventual writes go
2291 * in separate transaction */
2292 if (new_dentry->d_inode)
2293 DQUOT_INIT(new_dentry->d_inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002294 handle = ext4_journal_start(old_dir, 2 *
2295 EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2296 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002297 if (IS_ERR(handle))
2298 return PTR_ERR(handle);
2299
2300 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2301 handle->h_sync = 1;
2302
Mingming Cao617ba132006-10-11 01:20:53 -07002303 old_bh = ext4_find_entry (old_dentry, &old_de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002304 /*
2305 * Check for inode number is _not_ due to possible IO errors.
2306 * We might rmdir the source, keep it as pwd of some process
2307 * and merrily kill the link to whatever was created under the
2308 * same name. Goodbye sticky bit ;-<
2309 */
2310 old_inode = old_dentry->d_inode;
2311 retval = -ENOENT;
2312 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2313 goto end_rename;
2314
2315 new_inode = new_dentry->d_inode;
Mingming Cao617ba132006-10-11 01:20:53 -07002316 new_bh = ext4_find_entry (new_dentry, &new_de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002317 if (new_bh) {
2318 if (!new_inode) {
2319 brelse (new_bh);
2320 new_bh = NULL;
2321 }
2322 }
2323 if (S_ISDIR(old_inode->i_mode)) {
2324 if (new_inode) {
2325 retval = -ENOTEMPTY;
2326 if (!empty_dir (new_inode))
2327 goto end_rename;
2328 }
2329 retval = -EIO;
Mingming Cao617ba132006-10-11 01:20:53 -07002330 dir_bh = ext4_bread (handle, old_inode, 0, 0, &retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002331 if (!dir_bh)
2332 goto end_rename;
2333 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2334 goto end_rename;
2335 retval = -EMLINK;
2336 if (!new_inode && new_dir!=old_dir &&
Mingming Cao617ba132006-10-11 01:20:53 -07002337 new_dir->i_nlink >= EXT4_LINK_MAX)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002338 goto end_rename;
2339 }
2340 if (!new_bh) {
Mingming Cao617ba132006-10-11 01:20:53 -07002341 retval = ext4_add_entry (handle, new_dentry, old_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002342 if (retval)
2343 goto end_rename;
2344 } else {
2345 BUFFER_TRACE(new_bh, "get write access");
Mingming Cao617ba132006-10-11 01:20:53 -07002346 ext4_journal_get_write_access(handle, new_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002347 new_de->inode = cpu_to_le32(old_inode->i_ino);
Mingming Cao617ba132006-10-11 01:20:53 -07002348 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2349 EXT4_FEATURE_INCOMPAT_FILETYPE))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002350 new_de->file_type = old_de->file_type;
2351 new_dir->i_version++;
Mingming Cao617ba132006-10-11 01:20:53 -07002352 BUFFER_TRACE(new_bh, "call ext4_journal_dirty_metadata");
2353 ext4_journal_dirty_metadata(handle, new_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002354 brelse(new_bh);
2355 new_bh = NULL;
2356 }
2357
2358 /*
2359 * Like most other Unix systems, set the ctime for inodes on a
2360 * rename.
2361 */
Kalpak Shahef7f3832007-07-18 09:15:20 -04002362 old_inode->i_ctime = ext4_current_time(old_inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002363 ext4_mark_inode_dirty(handle, old_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002364
2365 /*
2366 * ok, that's it
2367 */
2368 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2369 old_de->name_len != old_dentry->d_name.len ||
2370 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
Mingming Cao617ba132006-10-11 01:20:53 -07002371 (retval = ext4_delete_entry(handle, old_dir,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002372 old_de, old_bh)) == -ENOENT) {
2373 /* old_de could have moved from under us during htree split, so
2374 * make sure that we are deleting the right entry. We might
2375 * also be pointing to a stale entry in the unused part of
2376 * old_bh so just checking inum and the name isn't enough. */
2377 struct buffer_head *old_bh2;
Mingming Cao617ba132006-10-11 01:20:53 -07002378 struct ext4_dir_entry_2 *old_de2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002379
Mingming Cao617ba132006-10-11 01:20:53 -07002380 old_bh2 = ext4_find_entry(old_dentry, &old_de2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002381 if (old_bh2) {
Mingming Cao617ba132006-10-11 01:20:53 -07002382 retval = ext4_delete_entry(handle, old_dir,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002383 old_de2, old_bh2);
2384 brelse(old_bh2);
2385 }
2386 }
2387 if (retval) {
Mingming Cao617ba132006-10-11 01:20:53 -07002388 ext4_warning(old_dir->i_sb, "ext4_rename",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002389 "Deleting old file (%lu), %d, error=%d",
2390 old_dir->i_ino, old_dir->i_nlink, retval);
2391 }
2392
2393 if (new_inode) {
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002394 ext4_dec_count(handle, new_inode);
Kalpak Shahef7f3832007-07-18 09:15:20 -04002395 new_inode->i_ctime = ext4_current_time(new_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002396 }
Kalpak Shahef7f3832007-07-18 09:15:20 -04002397 old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002398 ext4_update_dx_flag(old_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002399 if (dir_bh) {
2400 BUFFER_TRACE(dir_bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07002401 ext4_journal_get_write_access(handle, dir_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002402 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
Mingming Cao617ba132006-10-11 01:20:53 -07002403 BUFFER_TRACE(dir_bh, "call ext4_journal_dirty_metadata");
2404 ext4_journal_dirty_metadata(handle, dir_bh);
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002405 ext4_dec_count(handle, old_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002406 if (new_inode) {
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002407 /* checked empty_dir above, can't have another parent,
Theodore Ts'o825f1482008-02-15 15:00:38 -05002408 * ext4_dec_count() won't work for many-linked dirs */
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002409 new_inode->i_nlink = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002410 } else {
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002411 ext4_inc_count(handle, new_dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002412 ext4_update_dx_flag(new_dir);
2413 ext4_mark_inode_dirty(handle, new_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002414 }
2415 }
Mingming Cao617ba132006-10-11 01:20:53 -07002416 ext4_mark_inode_dirty(handle, old_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002417 if (new_inode) {
Mingming Cao617ba132006-10-11 01:20:53 -07002418 ext4_mark_inode_dirty(handle, new_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002419 if (!new_inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07002420 ext4_orphan_add(handle, new_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002421 }
2422 retval = 0;
2423
2424end_rename:
2425 brelse (dir_bh);
2426 brelse (old_bh);
2427 brelse (new_bh);
Mingming Cao617ba132006-10-11 01:20:53 -07002428 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002429 return retval;
2430}
2431
2432/*
2433 * directories can handle most operations...
2434 */
Arjan van de Ven754661f2007-02-12 00:55:38 -08002435const struct inode_operations ext4_dir_inode_operations = {
Mingming Cao617ba132006-10-11 01:20:53 -07002436 .create = ext4_create,
2437 .lookup = ext4_lookup,
2438 .link = ext4_link,
2439 .unlink = ext4_unlink,
2440 .symlink = ext4_symlink,
2441 .mkdir = ext4_mkdir,
2442 .rmdir = ext4_rmdir,
2443 .mknod = ext4_mknod,
2444 .rename = ext4_rename,
2445 .setattr = ext4_setattr,
2446#ifdef CONFIG_EXT4DEV_FS_XATTR
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002447 .setxattr = generic_setxattr,
2448 .getxattr = generic_getxattr,
Mingming Cao617ba132006-10-11 01:20:53 -07002449 .listxattr = ext4_listxattr,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002450 .removexattr = generic_removexattr,
2451#endif
Mingming Cao617ba132006-10-11 01:20:53 -07002452 .permission = ext4_permission,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002453};
2454
Arjan van de Ven754661f2007-02-12 00:55:38 -08002455const struct inode_operations ext4_special_inode_operations = {
Mingming Cao617ba132006-10-11 01:20:53 -07002456 .setattr = ext4_setattr,
2457#ifdef CONFIG_EXT4DEV_FS_XATTR
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002458 .setxattr = generic_setxattr,
2459 .getxattr = generic_getxattr,
Mingming Cao617ba132006-10-11 01:20:53 -07002460 .listxattr = ext4_listxattr,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002461 .removexattr = generic_removexattr,
2462#endif
Mingming Cao617ba132006-10-11 01:20:53 -07002463 .permission = ext4_permission,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002464};