blob: 7b0e44f7d66fb839c7c8002fb6e3928b5e3e04a8 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/ext3/namei.c
3 *
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
Dave Kleikampe9ad5622006-09-27 01:49:35 -070018 * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
Linus Torvalds1da177e2005-04-16 15:20:36 -070019 * Hash Tree Directory indexing (c)
Dave Kleikampe9ad5622006-09-27 01:49:35 -070020 * Daniel Phillips, 2001
Linus Torvalds1da177e2005-04-16 15:20:36 -070021 * Hash Tree Directory indexing porting
Dave Kleikampe9ad5622006-09-27 01:49:35 -070022 * Christopher Li, 2002
Linus Torvalds1da177e2005-04-16 15:20:36 -070023 * Hash Tree Directory indexing cleanup
Dave Kleikampe9ad5622006-09-27 01:49:35 -070024 * Theodore Ts'o, 2002
Linus Torvalds1da177e2005-04-16 15:20:36 -070025 */
26
27#include <linux/fs.h>
28#include <linux/pagemap.h>
29#include <linux/jbd.h>
30#include <linux/time.h>
31#include <linux/ext3_fs.h>
32#include <linux/ext3_jbd.h>
33#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>
Jens Axboecaa38fb2006-07-23 01:41:26 +020038#include <linux/bio.h>
Ben Dooks381be2542005-10-30 15:02:56 -080039
40#include "namei.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#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
49#define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
50#define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b))
51
52static struct buffer_head *ext3_append(handle_t *handle,
53 struct inode *inode,
54 u32 *block, int *err)
55{
56 struct buffer_head *bh;
57
58 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
59
Akinobu Mita33575f82008-04-28 02:16:08 -070060 bh = ext3_bread(handle, inode, *block, 1, err);
61 if (bh) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 inode->i_size += inode->i_sb->s_blocksize;
63 EXT3_I(inode)->i_disksize = inode->i_size;
Akinobu Mita33575f82008-04-28 02:16:08 -070064 *err = ext3_journal_get_write_access(handle, bh);
65 if (*err) {
66 brelse(bh);
67 bh = NULL;
68 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 }
70 return bh;
71}
72
73#ifndef assert
74#define assert(test) J_ASSERT(test)
75#endif
76
Linus Torvalds1da177e2005-04-16 15:20:36 -070077#ifdef DX_DEBUG
78#define dxtrace(command) command
79#else
Mingming Caoae6ddcc2006-09-27 01:49:27 -070080#define dxtrace(command)
Linus Torvalds1da177e2005-04-16 15:20:36 -070081#endif
82
83struct fake_dirent
84{
85 __le32 inode;
86 __le16 rec_len;
87 u8 name_len;
88 u8 file_type;
89};
90
91struct dx_countlimit
92{
93 __le16 limit;
94 __le16 count;
95};
96
97struct dx_entry
98{
99 __le32 hash;
100 __le32 block;
101};
102
103/*
104 * dx_root_info is laid out so that if it should somehow get overlaid by a
105 * dirent the two low bits of the hash version will be zero. Therefore, the
106 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
107 */
108
109struct dx_root
110{
111 struct fake_dirent dot;
112 char dot_name[4];
113 struct fake_dirent dotdot;
114 char dotdot_name[4];
115 struct dx_root_info
116 {
117 __le32 reserved_zero;
118 u8 hash_version;
119 u8 info_length; /* 8 */
120 u8 indirect_levels;
121 u8 unused_flags;
122 }
123 info;
124 struct dx_entry entries[0];
125};
126
127struct dx_node
128{
129 struct fake_dirent fake;
130 struct dx_entry entries[0];
131};
132
133
134struct dx_frame
135{
136 struct buffer_head *bh;
137 struct dx_entry *entries;
138 struct dx_entry *at;
139};
140
141struct dx_map_entry
142{
143 u32 hash;
Eric Sandeenef2b02d2007-09-18 22:46:42 -0700144 u16 offs;
145 u16 size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146};
147
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148static inline unsigned dx_get_block (struct dx_entry *entry);
149static void dx_set_block (struct dx_entry *entry, unsigned value);
150static inline unsigned dx_get_hash (struct dx_entry *entry);
151static void dx_set_hash (struct dx_entry *entry, unsigned value);
152static unsigned dx_get_count (struct dx_entry *entries);
153static unsigned dx_get_limit (struct dx_entry *entries);
154static void dx_set_count (struct dx_entry *entries, unsigned value);
155static void dx_set_limit (struct dx_entry *entries, unsigned value);
156static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
157static unsigned dx_node_limit (struct inode *dir);
Al Viro734711a2008-08-24 07:26:48 -0400158static struct dx_frame *dx_probe(struct qstr *entry,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159 struct inode *dir,
160 struct dx_hash_info *hinfo,
161 struct dx_frame *frame,
162 int *err);
163static void dx_release (struct dx_frame *frames);
Wei Yongjun45f90212009-04-02 16:57:14 -0700164static int dx_make_map(struct ext3_dir_entry_2 *de, unsigned blocksize,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 struct dx_hash_info *hinfo, struct dx_map_entry map[]);
166static void dx_sort_map(struct dx_map_entry *map, unsigned count);
167static struct ext3_dir_entry_2 *dx_move_dirents (char *from, char *to,
168 struct dx_map_entry *offsets, int count);
Wei Yongjun45f90212009-04-02 16:57:14 -0700169static struct ext3_dir_entry_2 *dx_pack_dirents(char *base, unsigned blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
171static int ext3_htree_next_block(struct inode *dir, __u32 hash,
172 struct dx_frame *frame,
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700173 struct dx_frame *frames,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 __u32 *start_hash);
Al Viro734711a2008-08-24 07:26:48 -0400175static struct buffer_head * ext3_dx_find_entry(struct inode *dir,
176 struct qstr *entry, struct ext3_dir_entry_2 **res_dir,
177 int *err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
179 struct inode *inode);
180
181/*
Jan Kara7c06a8d2007-11-14 17:00:19 -0800182 * p is at least 6 bytes before the end of page
183 */
184static inline struct ext3_dir_entry_2 *
185ext3_next_entry(struct ext3_dir_entry_2 *p)
186{
187 return (struct ext3_dir_entry_2 *)((char *)p +
188 ext3_rec_len_from_disk(p->rec_len));
189}
190
191/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 * Future: use high four bits of block for coalesce-on-delete flags
193 * Mask them off for now.
194 */
195
196static inline unsigned dx_get_block (struct dx_entry *entry)
197{
198 return le32_to_cpu(entry->block) & 0x00ffffff;
199}
200
201static inline void dx_set_block (struct dx_entry *entry, unsigned value)
202{
203 entry->block = cpu_to_le32(value);
204}
205
206static inline unsigned dx_get_hash (struct dx_entry *entry)
207{
208 return le32_to_cpu(entry->hash);
209}
210
211static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
212{
213 entry->hash = cpu_to_le32(value);
214}
215
216static inline unsigned dx_get_count (struct dx_entry *entries)
217{
218 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
219}
220
221static inline unsigned dx_get_limit (struct dx_entry *entries)
222{
223 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
224}
225
226static inline void dx_set_count (struct dx_entry *entries, unsigned value)
227{
228 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
229}
230
231static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
232{
233 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
234}
235
236static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
237{
238 unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(1) -
239 EXT3_DIR_REC_LEN(2) - infosize;
Li Zefan8ef27202008-07-25 01:46:29 -0700240 return entry_space / sizeof(struct dx_entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241}
242
243static inline unsigned dx_node_limit (struct inode *dir)
244{
245 unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(0);
Li Zefan8ef27202008-07-25 01:46:29 -0700246 return entry_space / sizeof(struct dx_entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247}
248
249/*
250 * Debug
251 */
252#ifdef DX_DEBUG
253static void dx_show_index (char * label, struct dx_entry *entries)
254{
255 int i, n = dx_get_count (entries);
256 printk("%s index ", label);
257 for (i = 0; i < n; i++)
258 {
259 printk("%x->%u ", i? dx_get_hash(entries + i): 0, dx_get_block(entries + i));
260 }
261 printk("\n");
262}
263
264struct stats
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700265{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 unsigned names;
267 unsigned space;
268 unsigned bcount;
269};
270
271static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext3_dir_entry_2 *de,
272 int size, int show_names)
273{
274 unsigned names = 0, space = 0;
275 char *base = (char *) de;
276 struct dx_hash_info h = *hinfo;
277
278 printk("names: ");
279 while ((char *) de < base + size)
280 {
281 if (de->inode)
282 {
283 if (show_names)
284 {
285 int len = de->name_len;
286 char *name = de->name;
287 while (len--) printk("%c", *name++);
288 ext3fs_dirhash(de->name, de->name_len, &h);
289 printk(":%x.%u ", h.hash,
290 ((char *) de - base));
291 }
292 space += EXT3_DIR_REC_LEN(de->name_len);
Dave Kleikampe9ad5622006-09-27 01:49:35 -0700293 names++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 }
Jan Kara7c06a8d2007-11-14 17:00:19 -0800295 de = ext3_next_entry(de);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 }
297 printk("(%i)\n", names);
298 return (struct stats) { names, space, 1 };
299}
300
301struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
302 struct dx_entry *entries, int levels)
303{
304 unsigned blocksize = dir->i_sb->s_blocksize;
305 unsigned count = dx_get_count (entries), names = 0, space = 0, i;
306 unsigned bcount = 0;
307 struct buffer_head *bh;
308 int err;
309 printk("%i indexed blocks...\n", count);
310 for (i = 0; i < count; i++, entries++)
311 {
312 u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0;
313 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
314 struct stats stats;
315 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
316 if (!(bh = ext3_bread (NULL,dir, block, 0,&err))) continue;
317 stats = levels?
318 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
319 dx_show_leaf(hinfo, (struct ext3_dir_entry_2 *) bh->b_data, blocksize, 0);
320 names += stats.names;
321 space += stats.space;
322 bcount += stats.bcount;
323 brelse (bh);
324 }
325 if (bcount)
326 printk("%snames %u, fullness %u (%u%%)\n", levels?"":" ",
327 names, space/bcount,(space/bcount)*100/blocksize);
328 return (struct stats) { names, space, bcount};
329}
330#endif /* DX_DEBUG */
331
332/*
333 * Probe for a directory leaf block to search.
334 *
335 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
336 * error in the directory index, and the caller should fall back to
337 * searching the directory normally. The callers of dx_probe **MUST**
338 * check for this error code, and make sure it never gets reflected
339 * back to userspace.
340 */
341static struct dx_frame *
Al Viro734711a2008-08-24 07:26:48 -0400342dx_probe(struct qstr *entry, struct inode *dir,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
344{
345 unsigned count, indirect;
346 struct dx_entry *at, *entries, *p, *q, *m;
347 struct dx_root *root;
348 struct buffer_head *bh;
349 struct dx_frame *frame = frame_in;
350 u32 hash;
351
352 frame->bh = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 if (!(bh = ext3_bread (NULL,dir, 0, 0, err)))
354 goto fail;
355 root = (struct dx_root *) bh->b_data;
356 if (root->info.hash_version != DX_HASH_TEA &&
357 root->info.hash_version != DX_HASH_HALF_MD4 &&
358 root->info.hash_version != DX_HASH_LEGACY) {
Harvey Harrisone05b6b52008-04-28 02:16:15 -0700359 ext3_warning(dir->i_sb, __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 "Unrecognised inode hash code %d",
361 root->info.hash_version);
362 brelse(bh);
363 *err = ERR_BAD_DX_DIR;
364 goto fail;
365 }
366 hinfo->hash_version = root->info.hash_version;
Theodore Ts'o5e1f8c92008-10-28 13:21:55 -0400367 if (hinfo->hash_version <= DX_HASH_TEA)
368 hinfo->hash_version += EXT3_SB(dir->i_sb)->s_hash_unsigned;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 hinfo->seed = EXT3_SB(dir->i_sb)->s_hash_seed;
Al Viro734711a2008-08-24 07:26:48 -0400370 if (entry)
371 ext3fs_dirhash(entry->name, entry->len, hinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 hash = hinfo->hash;
373
374 if (root->info.unused_flags & 1) {
Harvey Harrisone05b6b52008-04-28 02:16:15 -0700375 ext3_warning(dir->i_sb, __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 "Unimplemented inode hash flags: %#06x",
377 root->info.unused_flags);
378 brelse(bh);
379 *err = ERR_BAD_DX_DIR;
380 goto fail;
381 }
382
383 if ((indirect = root->info.indirect_levels) > 1) {
Harvey Harrisone05b6b52008-04-28 02:16:15 -0700384 ext3_warning(dir->i_sb, __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 "Unimplemented inode hash depth: %#06x",
386 root->info.indirect_levels);
387 brelse(bh);
388 *err = ERR_BAD_DX_DIR;
389 goto fail;
390 }
391
392 entries = (struct dx_entry *) (((char *)&root->info) +
393 root->info.info_length);
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700394
395 if (dx_get_limit(entries) != dx_root_limit(dir,
396 root->info.info_length)) {
Harvey Harrisone05b6b52008-04-28 02:16:15 -0700397 ext3_warning(dir->i_sb, __func__,
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700398 "dx entry: limit != root limit");
399 brelse(bh);
400 *err = ERR_BAD_DX_DIR;
401 goto fail;
402 }
403
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 dxtrace (printk("Look up %x", hash));
405 while (1)
406 {
407 count = dx_get_count(entries);
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700408 if (!count || count > dx_get_limit(entries)) {
Harvey Harrisone05b6b52008-04-28 02:16:15 -0700409 ext3_warning(dir->i_sb, __func__,
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700410 "dx entry: no count or count > limit");
411 brelse(bh);
412 *err = ERR_BAD_DX_DIR;
413 goto fail2;
414 }
415
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 p = entries + 1;
417 q = entries + count - 1;
418 while (p <= q)
419 {
420 m = p + (q - p)/2;
421 dxtrace(printk("."));
422 if (dx_get_hash(m) > hash)
423 q = m - 1;
424 else
425 p = m + 1;
426 }
427
428 if (0) // linear search cross check
429 {
430 unsigned n = count - 1;
431 at = entries;
432 while (n--)
433 {
434 dxtrace(printk(","));
435 if (dx_get_hash(++at) > hash)
436 {
437 at--;
438 break;
439 }
440 }
441 assert (at == p - 1);
442 }
443
444 at = p - 1;
445 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
446 frame->bh = bh;
447 frame->entries = entries;
448 frame->at = at;
449 if (!indirect--) return frame;
450 if (!(bh = ext3_bread (NULL,dir, dx_get_block(at), 0, err)))
451 goto fail2;
452 at = entries = ((struct dx_node *) bh->b_data)->entries;
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700453 if (dx_get_limit(entries) != dx_node_limit (dir)) {
Harvey Harrisone05b6b52008-04-28 02:16:15 -0700454 ext3_warning(dir->i_sb, __func__,
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700455 "dx entry: limit != node limit");
456 brelse(bh);
457 *err = ERR_BAD_DX_DIR;
458 goto fail2;
459 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 frame++;
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700461 frame->bh = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 }
463fail2:
464 while (frame >= frame_in) {
465 brelse(frame->bh);
466 frame--;
467 }
468fail:
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700469 if (*err == ERR_BAD_DX_DIR)
Harvey Harrisone05b6b52008-04-28 02:16:15 -0700470 ext3_warning(dir->i_sb, __func__,
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700471 "Corrupt dir inode %ld, running e2fsck is "
472 "recommended.", dir->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 return NULL;
474}
475
476static void dx_release (struct dx_frame *frames)
477{
478 if (frames[0].bh == NULL)
479 return;
480
481 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
482 brelse(frames[1].bh);
483 brelse(frames[0].bh);
484}
485
486/*
487 * This function increments the frame pointer to search the next leaf
488 * block, and reads in the necessary intervening nodes if the search
489 * should be necessary. Whether or not the search is necessary is
490 * controlled by the hash parameter. If the hash value is even, then
491 * the search is only continued if the next block starts with that
492 * hash value. This is used if we are searching for a specific file.
493 *
494 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
495 *
496 * This function returns 1 if the caller should continue to search,
497 * or 0 if it should not. If there is an error reading one of the
498 * index blocks, it will a negative error code.
499 *
500 * If start_hash is non-null, it will be filled in with the starting
501 * hash of the next page.
502 */
503static int ext3_htree_next_block(struct inode *dir, __u32 hash,
504 struct dx_frame *frame,
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700505 struct dx_frame *frames,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 __u32 *start_hash)
507{
508 struct dx_frame *p;
509 struct buffer_head *bh;
510 int err, num_frames = 0;
511 __u32 bhash;
512
513 p = frame;
514 /*
515 * Find the next leaf page by incrementing the frame pointer.
516 * If we run out of entries in the interior node, loop around and
517 * increment pointer in the parent node. When we break out of
518 * this loop, num_frames indicates the number of interior
519 * nodes need to be read.
520 */
521 while (1) {
522 if (++(p->at) < p->entries + dx_get_count(p->entries))
523 break;
524 if (p == frames)
525 return 0;
526 num_frames++;
527 p--;
528 }
529
530 /*
531 * If the hash is 1, then continue only if the next page has a
532 * continuation hash of any value. This is used for readdir
533 * handling. Otherwise, check to see if the hash matches the
534 * desired contiuation hash. If it doesn't, return since
535 * there's no point to read in the successive index pages.
536 */
537 bhash = dx_get_hash(p->at);
538 if (start_hash)
539 *start_hash = bhash;
540 if ((hash & 1) == 0) {
541 if ((bhash & ~1) != hash)
542 return 0;
543 }
544 /*
545 * If the hash is HASH_NB_ALWAYS, we always go to the next
546 * block so no check is necessary
547 */
548 while (num_frames--) {
549 if (!(bh = ext3_bread(NULL, dir, dx_get_block(p->at),
550 0, &err)))
551 return err; /* Failure */
552 p++;
553 brelse (p->bh);
554 p->bh = bh;
555 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
556 }
557 return 1;
558}
559
560
561/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 * This function fills a red-black tree with information from a
563 * directory block. It returns the number directory entries loaded
564 * into the tree. If there is an error it is returned in err.
565 */
566static int htree_dirblock_to_tree(struct file *dir_file,
567 struct inode *dir, int block,
568 struct dx_hash_info *hinfo,
569 __u32 start_hash, __u32 start_minor_hash)
570{
571 struct buffer_head *bh;
572 struct ext3_dir_entry_2 *de, *top;
573 int err, count = 0;
574
575 dxtrace(printk("In htree dirblock_to_tree: block %d\n", block));
576 if (!(bh = ext3_bread (NULL, dir, block, 0, &err)))
577 return err;
578
579 de = (struct ext3_dir_entry_2 *) bh->b_data;
580 top = (struct ext3_dir_entry_2 *) ((char *) de +
581 dir->i_sb->s_blocksize -
582 EXT3_DIR_REC_LEN(0));
583 for (; de < top; de = ext3_next_entry(de)) {
Eric Sandeen40b85132006-12-06 20:36:26 -0800584 if (!ext3_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
585 (block<<EXT3_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 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 ext3fs_dirhash(de->name, de->name_len, hinfo);
594 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;
600 if ((err = ext3_htree_store_dirent(dir_file,
601 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 */
620int ext3_htree_fill_tree(struct file *dir_file, __u32 start_hash,
621 __u32 start_minor_hash, __u32 *next_hash)
622{
623 struct dx_hash_info hinfo;
624 struct ext3_dir_entry_2 *de;
625 struct dx_frame frames[2], *frame;
626 struct inode *dir;
627 int block, err;
628 int count = 0;
629 int ret;
630 __u32 hashval;
631
632 dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
633 start_minor_hash));
Josef "Jeff" Sipekfe21a692006-12-08 02:36:38 -0800634 dir = dir_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 if (!(EXT3_I(dir)->i_flags & EXT3_INDEX_FL)) {
636 hinfo.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
Theodore Ts'o5e1f8c92008-10-28 13:21:55 -0400637 if (hinfo.hash_version <= DX_HASH_TEA)
638 hinfo.hash_version +=
639 EXT3_SB(dir->i_sb)->s_hash_unsigned;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
641 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
642 start_hash, start_minor_hash);
643 *next_hash = ~0;
644 return count;
645 }
646 hinfo.hash = start_hash;
647 hinfo.minor_hash = 0;
Josef "Jeff" Sipekfe21a692006-12-08 02:36:38 -0800648 frame = dx_probe(NULL, dir_file->f_path.dentry->d_inode, &hinfo, frames, &err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 if (!frame)
650 return err;
651
652 /* Add '.' and '..' from the htree header */
653 if (!start_hash && !start_minor_hash) {
654 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
655 if ((err = ext3_htree_store_dirent(dir_file, 0, 0, de)) != 0)
656 goto errout;
657 count++;
658 }
659 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
660 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
661 de = ext3_next_entry(de);
662 if ((err = ext3_htree_store_dirent(dir_file, 2, 0, de)) != 0)
663 goto errout;
664 count++;
665 }
666
667 while (1) {
668 block = dx_get_block(frame->at);
669 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
670 start_hash, start_minor_hash);
671 if (ret < 0) {
672 err = ret;
673 goto errout;
674 }
675 count += ret;
676 hashval = ~0;
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700677 ret = ext3_htree_next_block(dir, HASH_NB_ALWAYS,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 frame, frames, &hashval);
679 *next_hash = hashval;
680 if (ret < 0) {
681 err = ret;
682 goto errout;
683 }
684 /*
685 * Stop if: (a) there are no more entries, or
686 * (b) we have inserted at least one entry and the
687 * next hash value is not a continuation
688 */
689 if ((ret == 0) ||
690 (count && ((hashval & 1) == 0)))
691 break;
692 }
693 dx_release(frames);
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700694 dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 count, *next_hash));
696 return count;
697errout:
698 dx_release(frames);
699 return (err);
700}
701
702
703/*
704 * Directory block splitting, compacting
705 */
706
Eric Sandeenef2b02d2007-09-18 22:46:42 -0700707/*
708 * Create map of hash values, offsets, and sizes, stored at end of block.
709 * Returns number of entries mapped.
710 */
Wei Yongjun45f90212009-04-02 16:57:14 -0700711static int dx_make_map(struct ext3_dir_entry_2 *de, unsigned blocksize,
712 struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713{
714 int count = 0;
715 char *base = (char *) de;
716 struct dx_hash_info h = *hinfo;
717
Wei Yongjun45f90212009-04-02 16:57:14 -0700718 while ((char *) de < base + blocksize)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 {
720 if (de->name_len && de->inode) {
721 ext3fs_dirhash(de->name, de->name_len, &h);
722 map_tail--;
723 map_tail->hash = h.hash;
Eric Sandeenef2b02d2007-09-18 22:46:42 -0700724 map_tail->offs = (u16) ((char *) de - base);
725 map_tail->size = le16_to_cpu(de->rec_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 count++;
727 cond_resched();
728 }
729 /* XXX: do we need to check rec_len == 0 case? -Chris */
Jan Kara7c06a8d2007-11-14 17:00:19 -0800730 de = ext3_next_entry(de);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 }
732 return count;
733}
734
Eric Sandeenef2b02d2007-09-18 22:46:42 -0700735/* Sort map by hash value */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736static void dx_sort_map (struct dx_map_entry *map, unsigned count)
737{
738 struct dx_map_entry *p, *q, *top = map + count - 1;
739 int more;
740 /* Combsort until bubble sort doesn't suck */
741 while (count > 2)
742 {
743 count = count*10/13;
744 if (count - 9 < 2) /* 9, 10 -> 11 */
745 count = 11;
746 for (p = top, q = p - count; q >= map; p--, q--)
747 if (p->hash < q->hash)
748 swap(*p, *q);
749 }
750 /* Garden variety bubble sort */
751 do {
752 more = 0;
753 q = top;
754 while (q-- > map)
755 {
756 if (q[1].hash >= q[0].hash)
757 continue;
758 swap(*(q+1), *q);
759 more = 1;
760 }
761 } while(more);
762}
763
764static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
765{
766 struct dx_entry *entries = frame->entries;
767 struct dx_entry *old = frame->at, *new = old + 1;
768 int count = dx_get_count(entries);
769
770 assert(count < dx_get_limit(entries));
771 assert(old < entries + count);
772 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
773 dx_set_hash(new, hash);
774 dx_set_block(new, block);
775 dx_set_count(entries, count + 1);
776}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777
778static void ext3_update_dx_flag(struct inode *inode)
779{
780 if (!EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
781 EXT3_FEATURE_COMPAT_DIR_INDEX))
782 EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL;
783}
784
785/*
786 * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
787 *
788 * `len <= EXT3_NAME_LEN' is guaranteed by caller.
789 * `de != NULL' is guaranteed by caller.
790 */
791static inline int ext3_match (int len, const char * const name,
792 struct ext3_dir_entry_2 * de)
793{
794 if (len != de->name_len)
795 return 0;
796 if (!de->inode)
797 return 0;
798 return !memcmp(name, de->name, len);
799}
800
801/*
802 * Returns 0 if not found, -1 on failure, and 1 on success
803 */
804static inline int search_dirblock(struct buffer_head * bh,
805 struct inode *dir,
Al Viro734711a2008-08-24 07:26:48 -0400806 struct qstr *child,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 unsigned long offset,
808 struct ext3_dir_entry_2 ** res_dir)
809{
810 struct ext3_dir_entry_2 * de;
811 char * dlimit;
812 int de_len;
Al Viro734711a2008-08-24 07:26:48 -0400813 const char *name = child->name;
814 int namelen = child->len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815
816 de = (struct ext3_dir_entry_2 *) bh->b_data;
817 dlimit = bh->b_data + dir->i_sb->s_blocksize;
818 while ((char *) de < dlimit) {
819 /* this code is executed quadratically often */
820 /* do minimal checking `by hand' */
821
822 if ((char *) de + namelen <= dlimit &&
823 ext3_match (namelen, name, de)) {
824 /* found a match - just to be sure, do a full check */
825 if (!ext3_check_dir_entry("ext3_find_entry",
826 dir, de, bh, offset))
827 return -1;
828 *res_dir = de;
829 return 1;
830 }
831 /* prevent looping on a bad block */
Jan Kara7c06a8d2007-11-14 17:00:19 -0800832 de_len = ext3_rec_len_from_disk(de->rec_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 if (de_len <= 0)
834 return -1;
835 offset += de_len;
836 de = (struct ext3_dir_entry_2 *) ((char *) de + de_len);
837 }
838 return 0;
839}
840
841
842/*
843 * ext3_find_entry()
844 *
845 * finds an entry in the specified directory with the wanted name. It
846 * returns the cache buffer in which the entry was found, and the entry
847 * itself (as a parameter - res_dir). It does NOT read the inode of the
848 * entry - you'll have to do that yourself if you want to.
849 *
850 * The returned buffer_head has ->b_count elevated. The caller is expected
851 * to brelse() it when appropriate.
852 */
Al Viro734711a2008-08-24 07:26:48 -0400853static struct buffer_head *ext3_find_entry(struct inode *dir,
854 struct qstr *entry,
855 struct ext3_dir_entry_2 **res_dir)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856{
857 struct super_block * sb;
858 struct buffer_head * bh_use[NAMEI_RA_SIZE];
859 struct buffer_head * bh, *ret = NULL;
860 unsigned long start, block, b;
861 int ra_max = 0; /* Number of bh's in the readahead
862 buffer, bh_use[] */
863 int ra_ptr = 0; /* Current index into readahead
864 buffer */
865 int num = 0;
866 int nblocks, i, err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 int namelen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868
869 *res_dir = NULL;
870 sb = dir->i_sb;
Al Viro734711a2008-08-24 07:26:48 -0400871 namelen = entry->len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 if (namelen > EXT3_NAME_LEN)
873 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 if (is_dx(dir)) {
Al Viro734711a2008-08-24 07:26:48 -0400875 bh = ext3_dx_find_entry(dir, entry, res_dir, &err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 /*
877 * On success, or if the error was file not found,
878 * return. Otherwise, fall back to doing a search the
879 * old fashioned way.
880 */
881 if (bh || (err != ERR_BAD_DX_DIR))
882 return bh;
883 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
884 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
886 start = EXT3_I(dir)->i_dir_start_lookup;
887 if (start >= nblocks)
888 start = 0;
889 block = start;
890restart:
891 do {
892 /*
893 * We deal with the read-ahead logic here.
894 */
895 if (ra_ptr >= ra_max) {
896 /* Refill the readahead buffer */
897 ra_ptr = 0;
898 b = block;
899 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
900 /*
901 * Terminate if we reach the end of the
902 * directory and must wrap, or if our
903 * search has finished at this block.
904 */
905 if (b >= nblocks || (num && block == start)) {
906 bh_use[ra_max] = NULL;
907 break;
908 }
909 num++;
910 bh = ext3_getblk(NULL, dir, b++, 0, &err);
911 bh_use[ra_max] = bh;
912 if (bh)
Jens Axboecaa38fb2006-07-23 01:41:26 +0200913 ll_rw_block(READ_META, 1, &bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 }
915 }
916 if ((bh = bh_use[ra_ptr++]) == NULL)
917 goto next;
918 wait_on_buffer(bh);
919 if (!buffer_uptodate(bh)) {
920 /* read error, skip block & hope for the best */
Harvey Harrisone05b6b52008-04-28 02:16:15 -0700921 ext3_error(sb, __func__, "reading directory #%lu "
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 "offset %lu", dir->i_ino, block);
923 brelse(bh);
924 goto next;
925 }
Al Viro734711a2008-08-24 07:26:48 -0400926 i = search_dirblock(bh, dir, entry,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 block << EXT3_BLOCK_SIZE_BITS(sb), res_dir);
928 if (i == 1) {
929 EXT3_I(dir)->i_dir_start_lookup = block;
930 ret = bh;
931 goto cleanup_and_exit;
932 } else {
933 brelse(bh);
934 if (i < 0)
935 goto cleanup_and_exit;
936 }
937 next:
938 if (++block >= nblocks)
939 block = 0;
940 } while (block != start);
941
942 /*
943 * If the directory has grown while we were searching, then
944 * search the last part of the directory before giving up.
945 */
946 block = nblocks;
947 nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
948 if (block < nblocks) {
949 start = 0;
950 goto restart;
951 }
952
953cleanup_and_exit:
954 /* Clean up the read-ahead blocks */
955 for (; ra_ptr < ra_max; ra_ptr++)
956 brelse (bh_use[ra_ptr]);
957 return ret;
958}
959
Al Viro734711a2008-08-24 07:26:48 -0400960static struct buffer_head * ext3_dx_find_entry(struct inode *dir,
961 struct qstr *entry, struct ext3_dir_entry_2 **res_dir,
962 int *err)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963{
964 struct super_block * sb;
965 struct dx_hash_info hinfo;
966 u32 hash;
967 struct dx_frame frames[2], *frame;
968 struct ext3_dir_entry_2 *de, *top;
969 struct buffer_head *bh;
970 unsigned long block;
971 int retval;
Al Viro734711a2008-08-24 07:26:48 -0400972 int namelen = entry->len;
973 const u8 *name = entry->name;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974
975 sb = dir->i_sb;
Andreas Dilgeracfa1823d2005-06-23 00:09:45 -0700976 /* NFS may look up ".." - look at dx_root directory block */
Al Viro734711a2008-08-24 07:26:48 -0400977 if (namelen > 2 || name[0] != '.'|| (namelen == 2 && name[1] != '.')) {
978 if (!(frame = dx_probe(entry, dir, &hinfo, frames, err)))
Andreas Dilgeracfa1823d2005-06-23 00:09:45 -0700979 return NULL;
980 } else {
981 frame = frames;
982 frame->bh = NULL; /* for dx_release() */
983 frame->at = (struct dx_entry *)frames; /* hack for zero entry*/
984 dx_set_block(frame->at, 0); /* dx_root block is 0 */
985 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 hash = hinfo.hash;
987 do {
988 block = dx_get_block(frame->at);
989 if (!(bh = ext3_bread (NULL,dir, block, 0, err)))
990 goto errout;
991 de = (struct ext3_dir_entry_2 *) bh->b_data;
992 top = (struct ext3_dir_entry_2 *) ((char *) de + sb->s_blocksize -
993 EXT3_DIR_REC_LEN(0));
Duane Griffin275c0a82008-07-25 01:46:31 -0700994 for (; de < top; de = ext3_next_entry(de)) {
995 int off = (block << EXT3_BLOCK_SIZE_BITS(sb))
996 + ((char *) de - bh->b_data);
997
998 if (!ext3_check_dir_entry(__func__, dir, de, bh, off)) {
999 brelse(bh);
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001000 *err = ERR_BAD_DX_DIR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 goto errout;
1002 }
Duane Griffin275c0a82008-07-25 01:46:31 -07001003
1004 if (ext3_match(namelen, name, de)) {
1005 *res_dir = de;
1006 dx_release(frames);
1007 return bh;
1008 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 }
1010 brelse (bh);
1011 /* Check to see if we should continue to search */
1012 retval = ext3_htree_next_block(dir, hash, frame,
1013 frames, NULL);
1014 if (retval < 0) {
Harvey Harrisone05b6b52008-04-28 02:16:15 -07001015 ext3_warning(sb, __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 "error reading index page in directory #%lu",
1017 dir->i_ino);
1018 *err = retval;
1019 goto errout;
1020 }
1021 } while (retval == 1);
1022
1023 *err = -ENOENT;
1024errout:
1025 dxtrace(printk("%s not found\n", name));
1026 dx_release (frames);
1027 return NULL;
1028}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029
1030static struct dentry *ext3_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
1031{
1032 struct inode * inode;
1033 struct ext3_dir_entry_2 * de;
1034 struct buffer_head * bh;
1035
1036 if (dentry->d_name.len > EXT3_NAME_LEN)
1037 return ERR_PTR(-ENAMETOOLONG);
1038
Al Viro734711a2008-08-24 07:26:48 -04001039 bh = ext3_find_entry(dir, &dentry->d_name, &de);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 inode = NULL;
1041 if (bh) {
1042 unsigned long ino = le32_to_cpu(de->inode);
1043 brelse (bh);
Neil Brown2ccb48e2006-07-30 03:03:01 -07001044 if (!ext3_valid_inum(dir->i_sb, ino)) {
1045 ext3_error(dir->i_sb, "ext3_lookup",
1046 "bad inode number: %lu", ino);
David Howells473043d2008-02-07 00:15:36 -08001047 return ERR_PTR(-EIO);
Vasily Averina6c15c22007-07-15 23:40:46 -07001048 }
David Howells473043d2008-02-07 00:15:36 -08001049 inode = ext3_iget(dir->i_sb, ino);
Bryan Donlande18f3b2009-04-02 16:57:15 -07001050 if (unlikely(IS_ERR(inode))) {
1051 if (PTR_ERR(inode) == -ESTALE) {
1052 ext3_error(dir->i_sb, __func__,
1053 "deleted inode referenced: %lu",
1054 ino);
1055 return ERR_PTR(-EIO);
1056 } else {
1057 return ERR_CAST(inode);
1058 }
1059 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 }
Pekka Enbergba7fe362006-01-14 13:21:08 -08001061 return d_splice_alias(inode, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062}
1063
1064
1065struct dentry *ext3_get_parent(struct dentry *child)
1066{
1067 unsigned long ino;
Al Viro734711a2008-08-24 07:26:48 -04001068 struct qstr dotdot = {.name = "..", .len = 2};
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 struct ext3_dir_entry_2 * de;
1070 struct buffer_head *bh;
1071
Al Viro734711a2008-08-24 07:26:48 -04001072 bh = ext3_find_entry(child->d_inode, &dotdot, &de);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 if (!bh)
1074 return ERR_PTR(-ENOENT);
1075 ino = le32_to_cpu(de->inode);
1076 brelse(bh);
Neil Brown2ccb48e2006-07-30 03:03:01 -07001077
1078 if (!ext3_valid_inum(child->d_inode->i_sb, ino)) {
1079 ext3_error(child->d_inode->i_sb, "ext3_get_parent",
1080 "bad inode number: %lu", ino);
David Howells473043d2008-02-07 00:15:36 -08001081 return ERR_PTR(-EIO);
Vasily Averina6c15c22007-07-15 23:40:46 -07001082 }
1083
Christoph Hellwig44003722008-08-11 15:49:04 +02001084 return d_obtain_alias(ext3_iget(child->d_inode->i_sb, ino));
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001085}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086
1087#define S_SHIFT 12
1088static unsigned char ext3_type_by_mode[S_IFMT >> S_SHIFT] = {
1089 [S_IFREG >> S_SHIFT] = EXT3_FT_REG_FILE,
1090 [S_IFDIR >> S_SHIFT] = EXT3_FT_DIR,
1091 [S_IFCHR >> S_SHIFT] = EXT3_FT_CHRDEV,
1092 [S_IFBLK >> S_SHIFT] = EXT3_FT_BLKDEV,
1093 [S_IFIFO >> S_SHIFT] = EXT3_FT_FIFO,
1094 [S_IFSOCK >> S_SHIFT] = EXT3_FT_SOCK,
1095 [S_IFLNK >> S_SHIFT] = EXT3_FT_SYMLINK,
1096};
1097
1098static inline void ext3_set_de_type(struct super_block *sb,
1099 struct ext3_dir_entry_2 *de,
1100 umode_t mode) {
1101 if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE))
1102 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1103}
1104
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001105/*
1106 * Move count entries from end of map between two memory locations.
1107 * Returns pointer to last entry moved.
1108 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109static struct ext3_dir_entry_2 *
1110dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1111{
1112 unsigned rec_len = 0;
1113
1114 while (count--) {
1115 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs);
1116 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1117 memcpy (to, de, rec_len);
1118 ((struct ext3_dir_entry_2 *) to)->rec_len =
Jan Kara7c06a8d2007-11-14 17:00:19 -08001119 ext3_rec_len_to_disk(rec_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 de->inode = 0;
1121 map++;
1122 to += rec_len;
1123 }
1124 return (struct ext3_dir_entry_2 *) (to - rec_len);
1125}
1126
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001127/*
1128 * Compact each dir entry in the range to the minimal rec_len.
1129 * Returns pointer to last entry in range.
1130 */
Wei Yongjun45f90212009-04-02 16:57:14 -07001131static struct ext3_dir_entry_2 *dx_pack_dirents(char *base, unsigned blocksize)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132{
Wei Yongjun45f90212009-04-02 16:57:14 -07001133 struct ext3_dir_entry_2 *next, *to, *prev;
1134 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *)base;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 unsigned rec_len = 0;
1136
1137 prev = to = de;
Wei Yongjun45f90212009-04-02 16:57:14 -07001138 while ((char *)de < base + blocksize) {
Jan Kara7c06a8d2007-11-14 17:00:19 -08001139 next = ext3_next_entry(de);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 if (de->inode && de->name_len) {
1141 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1142 if (de > to)
1143 memmove(to, de, rec_len);
Jan Kara7c06a8d2007-11-14 17:00:19 -08001144 to->rec_len = ext3_rec_len_to_disk(rec_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 prev = to;
1146 to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len);
1147 }
1148 de = next;
1149 }
1150 return prev;
1151}
1152
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001153/*
1154 * Split a full leaf block to make room for a new dir entry.
1155 * Allocate a new block, and move entries so that they are approx. equally full.
1156 * Returns pointer to de in block into which the new entry will be inserted.
1157 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1159 struct buffer_head **bh,struct dx_frame *frame,
1160 struct dx_hash_info *hinfo, int *error)
1161{
1162 unsigned blocksize = dir->i_sb->s_blocksize;
1163 unsigned count, continued;
1164 struct buffer_head *bh2;
1165 u32 newblock;
1166 u32 hash2;
1167 struct dx_map_entry *map;
1168 char *data1 = (*bh)->b_data, *data2;
Theodore Ts'o59e315b2008-12-06 16:58:39 -05001169 unsigned split, move, size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 struct ext3_dir_entry_2 *de = NULL, *de2;
Theodore Ts'o59e315b2008-12-06 16:58:39 -05001171 int err = 0, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001173 bh2 = ext3_append (handle, dir, &newblock, &err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 if (!(bh2)) {
1175 brelse(*bh);
1176 *bh = NULL;
1177 goto errout;
1178 }
1179
1180 BUFFER_TRACE(*bh, "get_write_access");
1181 err = ext3_journal_get_write_access(handle, *bh);
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001182 if (err)
1183 goto journal_error;
1184
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 BUFFER_TRACE(frame->bh, "get_write_access");
1186 err = ext3_journal_get_write_access(handle, frame->bh);
1187 if (err)
1188 goto journal_error;
1189
1190 data2 = bh2->b_data;
1191
1192 /* create map in the end of data2 block */
1193 map = (struct dx_map_entry *) (data2 + blocksize);
1194 count = dx_make_map ((struct ext3_dir_entry_2 *) data1,
1195 blocksize, hinfo, map);
1196 map -= count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 dx_sort_map (map, count);
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001198 /* Split the existing block in the middle, size-wise */
1199 size = 0;
1200 move = 0;
1201 for (i = count-1; i >= 0; i--) {
1202 /* is more than half of this entry in 2nd half of the block? */
1203 if (size + map[i].size/2 > blocksize/2)
1204 break;
1205 size += map[i].size;
1206 move++;
1207 }
1208 /* map index at which we will split */
1209 split = count - move;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 hash2 = map[split].hash;
1211 continued = hash2 == map[split - 1].hash;
1212 dxtrace(printk("Split block %i at %x, %i/%i\n",
1213 dx_get_block(frame->at), hash2, split, count-split));
1214
1215 /* Fancy dance to stay within two buffers */
1216 de2 = dx_move_dirents(data1, data2, map + split, count - split);
1217 de = dx_pack_dirents(data1,blocksize);
Jan Kara7c06a8d2007-11-14 17:00:19 -08001218 de->rec_len = ext3_rec_len_to_disk(data1 + blocksize - (char *) de);
1219 de2->rec_len = ext3_rec_len_to_disk(data2 + blocksize - (char *) de2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220 dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1));
1221 dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1));
1222
1223 /* Which block gets the new entry? */
1224 if (hinfo->hash >= hash2)
1225 {
1226 swap(*bh, bh2);
1227 de = de2;
1228 }
1229 dx_insert_block (frame, hash2 + continued, newblock);
1230 err = ext3_journal_dirty_metadata (handle, bh2);
1231 if (err)
1232 goto journal_error;
1233 err = ext3_journal_dirty_metadata (handle, frame->bh);
1234 if (err)
1235 goto journal_error;
1236 brelse (bh2);
1237 dxtrace(dx_show_index ("frame", frame->entries));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 return de;
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001239
1240journal_error:
1241 brelse(*bh);
1242 brelse(bh2);
1243 *bh = NULL;
1244 ext3_std_error(dir->i_sb, err);
1245errout:
1246 *error = err;
1247 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249
1250
1251/*
1252 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1253 * it points to a directory entry which is guaranteed to be large
1254 * enough for new directory entry. If de is NULL, then
1255 * add_dirent_to_buf will attempt search the directory block for
1256 * space. It will return -ENOSPC if no space is available, and -EIO
1257 * and -EEXIST if directory entry already exists.
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001258 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 * NOTE! bh is NOT released in the case where ENOSPC is returned. In
1260 * all other cases bh is released.
1261 */
1262static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1263 struct inode *inode, struct ext3_dir_entry_2 *de,
1264 struct buffer_head * bh)
1265{
1266 struct inode *dir = dentry->d_parent->d_inode;
1267 const char *name = dentry->d_name.name;
1268 int namelen = dentry->d_name.len;
1269 unsigned long offset = 0;
1270 unsigned short reclen;
1271 int nlen, rlen, err;
1272 char *top;
1273
1274 reclen = EXT3_DIR_REC_LEN(namelen);
1275 if (!de) {
1276 de = (struct ext3_dir_entry_2 *)bh->b_data;
1277 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1278 while ((char *) de <= top) {
1279 if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
1280 bh, offset)) {
1281 brelse (bh);
1282 return -EIO;
1283 }
1284 if (ext3_match (namelen, name, de)) {
1285 brelse (bh);
1286 return -EEXIST;
1287 }
1288 nlen = EXT3_DIR_REC_LEN(de->name_len);
Jan Kara7c06a8d2007-11-14 17:00:19 -08001289 rlen = ext3_rec_len_from_disk(de->rec_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 if ((de->inode? rlen - nlen: rlen) >= reclen)
1291 break;
1292 de = (struct ext3_dir_entry_2 *)((char *)de + rlen);
1293 offset += rlen;
1294 }
1295 if ((char *) de > top)
1296 return -ENOSPC;
1297 }
1298 BUFFER_TRACE(bh, "get_write_access");
1299 err = ext3_journal_get_write_access(handle, bh);
1300 if (err) {
1301 ext3_std_error(dir->i_sb, err);
1302 brelse(bh);
1303 return err;
1304 }
1305
1306 /* By now the buffer is marked for journaling */
1307 nlen = EXT3_DIR_REC_LEN(de->name_len);
Jan Kara7c06a8d2007-11-14 17:00:19 -08001308 rlen = ext3_rec_len_from_disk(de->rec_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 if (de->inode) {
1310 struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen);
Jan Kara7c06a8d2007-11-14 17:00:19 -08001311 de1->rec_len = ext3_rec_len_to_disk(rlen - nlen);
1312 de->rec_len = ext3_rec_len_to_disk(nlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 de = de1;
1314 }
1315 de->file_type = EXT3_FT_UNKNOWN;
1316 if (inode) {
1317 de->inode = cpu_to_le32(inode->i_ino);
1318 ext3_set_de_type(dir->i_sb, de, inode->i_mode);
1319 } else
1320 de->inode = 0;
1321 de->name_len = namelen;
1322 memcpy (de->name, name, namelen);
1323 /*
1324 * XXX shouldn't update any times until successful
1325 * completion of syscall, but too many callers depend
1326 * on this.
1327 *
1328 * XXX similarly, too many callers depend on
1329 * ext3_new_inode() setting the times, but error
1330 * recovery deletes the inode, so the worst that can
1331 * happen is that the times are slightly out of date
1332 * and/or different from the directory change time.
1333 */
1334 dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
1335 ext3_update_dx_flag(dir);
1336 dir->i_version++;
1337 ext3_mark_inode_dirty(handle, dir);
1338 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1339 err = ext3_journal_dirty_metadata(handle, bh);
1340 if (err)
1341 ext3_std_error(dir->i_sb, err);
1342 brelse(bh);
1343 return 0;
1344}
1345
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346/*
1347 * This converts a one block unindexed directory to a 3 block indexed
1348 * directory, and adds the dentry to the indexed directory.
1349 */
1350static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1351 struct inode *inode, struct buffer_head *bh)
1352{
1353 struct inode *dir = dentry->d_parent->d_inode;
1354 const char *name = dentry->d_name.name;
1355 int namelen = dentry->d_name.len;
1356 struct buffer_head *bh2;
1357 struct dx_root *root;
1358 struct dx_frame frames[2], *frame;
1359 struct dx_entry *entries;
1360 struct ext3_dir_entry_2 *de, *de2;
1361 char *data1, *top;
1362 unsigned len;
1363 int retval;
1364 unsigned blocksize;
1365 struct dx_hash_info hinfo;
1366 u32 block;
1367 struct fake_dirent *fde;
1368
1369 blocksize = dir->i_sb->s_blocksize;
Theodore Ts'oa21102b2009-01-16 11:13:47 -05001370 dxtrace(printk(KERN_DEBUG "Creating index: inode %lu\n", dir->i_ino));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 retval = ext3_journal_get_write_access(handle, bh);
1372 if (retval) {
1373 ext3_std_error(dir->i_sb, retval);
1374 brelse(bh);
1375 return retval;
1376 }
1377 root = (struct dx_root *) bh->b_data;
1378
Theodore Ts'oa21102b2009-01-16 11:13:47 -05001379 /* The 0th block becomes the root, move the dirents out */
1380 fde = &root->dotdot;
1381 de = (struct ext3_dir_entry_2 *)((char *)fde +
1382 ext3_rec_len_from_disk(fde->rec_len));
1383 if ((char *) de >= (((char *) root) + blocksize)) {
1384 ext3_error(dir->i_sb, __func__,
1385 "invalid rec_len for '..' in inode %lu",
1386 dir->i_ino);
1387 brelse(bh);
1388 return -EIO;
1389 }
1390 len = ((char *) root) + blocksize - (char *) de;
1391
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 bh2 = ext3_append (handle, dir, &block, &retval);
1393 if (!(bh2)) {
1394 brelse(bh);
1395 return retval;
1396 }
1397 EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1398 data1 = bh2->b_data;
1399
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 memcpy (data1, de, len);
1401 de = (struct ext3_dir_entry_2 *) data1;
1402 top = data1 + len;
Jan Kara7c06a8d2007-11-14 17:00:19 -08001403 while ((char *)(de2 = ext3_next_entry(de)) < top)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 de = de2;
Jan Kara7c06a8d2007-11-14 17:00:19 -08001405 de->rec_len = ext3_rec_len_to_disk(data1 + blocksize - (char *) de);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 /* Initialize the root; the dot dirents already exist */
1407 de = (struct ext3_dir_entry_2 *) (&root->dotdot);
Jan Kara7c06a8d2007-11-14 17:00:19 -08001408 de->rec_len = ext3_rec_len_to_disk(blocksize - EXT3_DIR_REC_LEN(2));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 memset (&root->info, 0, sizeof(root->info));
1410 root->info.info_length = sizeof(root->info);
1411 root->info.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
1412 entries = root->entries;
1413 dx_set_block (entries, 1);
1414 dx_set_count (entries, 1);
1415 dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1416
1417 /* Initialize as for dx_probe */
1418 hinfo.hash_version = root->info.hash_version;
Theodore Ts'o5e1f8c92008-10-28 13:21:55 -04001419 if (hinfo.hash_version <= DX_HASH_TEA)
1420 hinfo.hash_version += EXT3_SB(dir->i_sb)->s_hash_unsigned;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
1422 ext3fs_dirhash(name, namelen, &hinfo);
1423 frame = frames;
1424 frame->entries = entries;
1425 frame->at = entries;
1426 frame->bh = bh;
1427 bh = bh2;
1428 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1429 dx_release (frames);
1430 if (!(de))
1431 return retval;
1432
1433 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1434}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435
1436/*
1437 * ext3_add_entry()
1438 *
1439 * adds a file entry to the specified directory, using the same
1440 * semantics as ext3_find_entry(). It returns NULL if it failed.
1441 *
1442 * NOTE!! The inode part of 'de' is left at 0 - which means you
1443 * may not sleep between calling this and putting something into
1444 * the entry, as someone else might have used it while you slept.
1445 */
1446static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
1447 struct inode *inode)
1448{
1449 struct inode *dir = dentry->d_parent->d_inode;
1450 unsigned long offset;
1451 struct buffer_head * bh;
1452 struct ext3_dir_entry_2 *de;
1453 struct super_block * sb;
1454 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 int dx_fallback=0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 unsigned blocksize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 u32 block, blocks;
1458
1459 sb = dir->i_sb;
1460 blocksize = sb->s_blocksize;
1461 if (!dentry->d_name.len)
1462 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 if (is_dx(dir)) {
1464 retval = ext3_dx_add_entry(handle, dentry, inode);
1465 if (!retval || (retval != ERR_BAD_DX_DIR))
1466 return retval;
1467 EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL;
1468 dx_fallback++;
1469 ext3_mark_inode_dirty(handle, dir);
1470 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 blocks = dir->i_size >> sb->s_blocksize_bits;
1472 for (block = 0, offset = 0; block < blocks; block++) {
1473 bh = ext3_bread(handle, dir, block, 0, &retval);
1474 if(!bh)
1475 return retval;
1476 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1477 if (retval != -ENOSPC)
1478 return retval;
1479
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 if (blocks == 1 && !dx_fallback &&
1481 EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1482 return make_indexed_dir(handle, dentry, inode, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 brelse(bh);
1484 }
1485 bh = ext3_append(handle, dir, &block, &retval);
1486 if (!bh)
1487 return retval;
1488 de = (struct ext3_dir_entry_2 *) bh->b_data;
1489 de->inode = 0;
Jan Kara7c06a8d2007-11-14 17:00:19 -08001490 de->rec_len = ext3_rec_len_to_disk(blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1492}
1493
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494/*
1495 * Returns 0 for success, or a negative error value
1496 */
1497static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
1498 struct inode *inode)
1499{
1500 struct dx_frame frames[2], *frame;
1501 struct dx_entry *entries, *at;
1502 struct dx_hash_info hinfo;
1503 struct buffer_head * bh;
1504 struct inode *dir = dentry->d_parent->d_inode;
1505 struct super_block * sb = dir->i_sb;
1506 struct ext3_dir_entry_2 *de;
1507 int err;
1508
Al Viro734711a2008-08-24 07:26:48 -04001509 frame = dx_probe(&dentry->d_name, dir, &hinfo, frames, &err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 if (!frame)
1511 return err;
1512 entries = frame->entries;
1513 at = frame->at;
1514
1515 if (!(bh = ext3_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1516 goto cleanup;
1517
1518 BUFFER_TRACE(bh, "get_write_access");
1519 err = ext3_journal_get_write_access(handle, bh);
1520 if (err)
1521 goto journal_error;
1522
1523 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1524 if (err != -ENOSPC) {
1525 bh = NULL;
1526 goto cleanup;
1527 }
1528
1529 /* Block full, should compress but for now just split */
1530 dxtrace(printk("using %u of %u node entries\n",
1531 dx_get_count(entries), dx_get_limit(entries)));
1532 /* Need to split index? */
1533 if (dx_get_count(entries) == dx_get_limit(entries)) {
1534 u32 newblock;
1535 unsigned icount = dx_get_count(entries);
1536 int levels = frame - frames;
1537 struct dx_entry *entries2;
1538 struct dx_node *node2;
1539 struct buffer_head *bh2;
1540
1541 if (levels && (dx_get_count(frames->entries) ==
1542 dx_get_limit(frames->entries))) {
Harvey Harrisone05b6b52008-04-28 02:16:15 -07001543 ext3_warning(sb, __func__,
Glauber de Oliveira Costa9f406682006-01-08 01:03:22 -08001544 "Directory index full!");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545 err = -ENOSPC;
1546 goto cleanup;
1547 }
1548 bh2 = ext3_append (handle, dir, &newblock, &err);
1549 if (!(bh2))
1550 goto cleanup;
1551 node2 = (struct dx_node *)(bh2->b_data);
1552 entries2 = node2->entries;
Jan Kara7c06a8d2007-11-14 17:00:19 -08001553 node2->fake.rec_len = ext3_rec_len_to_disk(sb->s_blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 node2->fake.inode = 0;
1555 BUFFER_TRACE(frame->bh, "get_write_access");
1556 err = ext3_journal_get_write_access(handle, frame->bh);
1557 if (err)
1558 goto journal_error;
1559 if (levels) {
1560 unsigned icount1 = icount/2, icount2 = icount - icount1;
1561 unsigned hash2 = dx_get_hash(entries + icount1);
1562 dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1563
1564 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1565 err = ext3_journal_get_write_access(handle,
1566 frames[0].bh);
1567 if (err)
1568 goto journal_error;
1569
1570 memcpy ((char *) entries2, (char *) (entries + icount1),
1571 icount2 * sizeof(struct dx_entry));
1572 dx_set_count (entries, icount1);
1573 dx_set_count (entries2, icount2);
1574 dx_set_limit (entries2, dx_node_limit(dir));
1575
1576 /* Which index block gets the new entry? */
1577 if (at - entries >= icount1) {
1578 frame->at = at = at - entries - icount1 + entries2;
1579 frame->entries = entries = entries2;
1580 swap(frame->bh, bh2);
1581 }
1582 dx_insert_block (frames + 0, hash2, newblock);
1583 dxtrace(dx_show_index ("node", frames[1].entries));
1584 dxtrace(dx_show_index ("node",
1585 ((struct dx_node *) bh2->b_data)->entries));
1586 err = ext3_journal_dirty_metadata(handle, bh2);
1587 if (err)
1588 goto journal_error;
1589 brelse (bh2);
1590 } else {
1591 dxtrace(printk("Creating second level index...\n"));
1592 memcpy((char *) entries2, (char *) entries,
1593 icount * sizeof(struct dx_entry));
1594 dx_set_limit(entries2, dx_node_limit(dir));
1595
1596 /* Set up root */
1597 dx_set_count(entries, 1);
1598 dx_set_block(entries + 0, newblock);
1599 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1600
1601 /* Add new access path frame */
1602 frame = frames + 1;
1603 frame->at = at = at - entries + entries2;
1604 frame->entries = entries = entries2;
1605 frame->bh = bh2;
1606 err = ext3_journal_get_write_access(handle,
1607 frame->bh);
1608 if (err)
1609 goto journal_error;
1610 }
1611 ext3_journal_dirty_metadata(handle, frames[0].bh);
1612 }
1613 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1614 if (!de)
1615 goto cleanup;
1616 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1617 bh = NULL;
1618 goto cleanup;
1619
1620journal_error:
1621 ext3_std_error(dir->i_sb, err);
1622cleanup:
1623 if (bh)
1624 brelse(bh);
1625 dx_release(frames);
1626 return err;
1627}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628
1629/*
1630 * ext3_delete_entry deletes a directory entry by merging it with the
1631 * previous entry
1632 */
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001633static int ext3_delete_entry (handle_t *handle,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 struct inode * dir,
1635 struct ext3_dir_entry_2 * de_del,
1636 struct buffer_head * bh)
1637{
1638 struct ext3_dir_entry_2 * de, * pde;
1639 int i;
1640
1641 i = 0;
1642 pde = NULL;
1643 de = (struct ext3_dir_entry_2 *) bh->b_data;
1644 while (i < bh->b_size) {
1645 if (!ext3_check_dir_entry("ext3_delete_entry", dir, de, bh, i))
1646 return -EIO;
1647 if (de == de_del) {
1648 BUFFER_TRACE(bh, "get_write_access");
1649 ext3_journal_get_write_access(handle, bh);
1650 if (pde)
Jan Kara7c06a8d2007-11-14 17:00:19 -08001651 pde->rec_len = ext3_rec_len_to_disk(
1652 ext3_rec_len_from_disk(pde->rec_len) +
1653 ext3_rec_len_from_disk(de->rec_len));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 else
1655 de->inode = 0;
1656 dir->i_version++;
1657 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1658 ext3_journal_dirty_metadata(handle, bh);
1659 return 0;
1660 }
Jan Kara7c06a8d2007-11-14 17:00:19 -08001661 i += ext3_rec_len_from_disk(de->rec_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 pde = de;
Jan Kara7c06a8d2007-11-14 17:00:19 -08001663 de = ext3_next_entry(de);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 }
1665 return -ENOENT;
1666}
1667
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668static int ext3_add_nondir(handle_t *handle,
1669 struct dentry *dentry, struct inode *inode)
1670{
1671 int err = ext3_add_entry(handle, dentry, inode);
1672 if (!err) {
1673 ext3_mark_inode_dirty(handle, inode);
1674 d_instantiate(dentry, inode);
Al Viroc38012d2008-12-30 02:02:50 -05001675 unlock_new_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 return 0;
1677 }
Eric Sandeen731b9a52007-02-10 01:46:16 -08001678 drop_nlink(inode);
Al Viroc38012d2008-12-30 02:02:50 -05001679 unlock_new_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 iput(inode);
1681 return err;
1682}
1683
1684/*
1685 * By the time this is called, we already have created
1686 * the directory cache entry for the new file, but it
1687 * is so far negative - it has no inode.
1688 *
1689 * If the create succeeds, we fill in the inode information
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001690 * with d_instantiate().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691 */
1692static int ext3_create (struct inode * dir, struct dentry * dentry, int mode,
1693 struct nameidata *nd)
1694{
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001695 handle_t *handle;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 struct inode * inode;
1697 int err, retries = 0;
1698
1699retry:
Jan Kara1f545872005-06-23 22:01:04 -07001700 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
Dmitry Monakhovc4590012009-12-09 03:05:30 +03001702 EXT3_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 if (IS_ERR(handle))
1704 return PTR_ERR(handle);
1705
1706 if (IS_DIRSYNC(dir))
1707 handle->h_sync = 1;
1708
1709 inode = ext3_new_inode (handle, dir, mode);
1710 err = PTR_ERR(inode);
1711 if (!IS_ERR(inode)) {
1712 inode->i_op = &ext3_file_inode_operations;
1713 inode->i_fop = &ext3_file_operations;
1714 ext3_set_aops(inode);
1715 err = ext3_add_nondir(handle, dentry, inode);
1716 }
1717 ext3_journal_stop(handle);
1718 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1719 goto retry;
1720 return err;
1721}
1722
1723static int ext3_mknod (struct inode * dir, struct dentry *dentry,
1724 int mode, dev_t rdev)
1725{
1726 handle_t *handle;
1727 struct inode *inode;
1728 int err, retries = 0;
1729
1730 if (!new_valid_dev(rdev))
1731 return -EINVAL;
1732
1733retry:
Jan Kara1f545872005-06-23 22:01:04 -07001734 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
Dave Kleikampe9ad5622006-09-27 01:49:35 -07001735 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
Dmitry Monakhovc4590012009-12-09 03:05:30 +03001736 EXT3_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 if (IS_ERR(handle))
1738 return PTR_ERR(handle);
1739
1740 if (IS_DIRSYNC(dir))
1741 handle->h_sync = 1;
1742
1743 inode = ext3_new_inode (handle, dir, mode);
1744 err = PTR_ERR(inode);
1745 if (!IS_ERR(inode)) {
1746 init_special_inode(inode, inode->i_mode, rdev);
1747#ifdef CONFIG_EXT3_FS_XATTR
1748 inode->i_op = &ext3_special_inode_operations;
1749#endif
1750 err = ext3_add_nondir(handle, dentry, inode);
1751 }
1752 ext3_journal_stop(handle);
1753 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1754 goto retry;
1755 return err;
1756}
1757
1758static int ext3_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1759{
1760 handle_t *handle;
1761 struct inode * inode;
1762 struct buffer_head * dir_block;
1763 struct ext3_dir_entry_2 * de;
1764 int err, retries = 0;
1765
1766 if (dir->i_nlink >= EXT3_LINK_MAX)
1767 return -EMLINK;
1768
1769retry:
Jan Kara1f545872005-06-23 22:01:04 -07001770 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
Dmitry Monakhovc4590012009-12-09 03:05:30 +03001772 EXT3_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773 if (IS_ERR(handle))
1774 return PTR_ERR(handle);
1775
1776 if (IS_DIRSYNC(dir))
1777 handle->h_sync = 1;
1778
1779 inode = ext3_new_inode (handle, dir, S_IFDIR | mode);
1780 err = PTR_ERR(inode);
1781 if (IS_ERR(inode))
1782 goto out_stop;
1783
1784 inode->i_op = &ext3_dir_inode_operations;
1785 inode->i_fop = &ext3_dir_operations;
1786 inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1787 dir_block = ext3_bread (handle, inode, 0, 1, &err);
1788 if (!dir_block) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001789 drop_nlink(inode); /* is this nlink == 0? */
Al Viroc38012d2008-12-30 02:02:50 -05001790 unlock_new_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 ext3_mark_inode_dirty(handle, inode);
1792 iput (inode);
1793 goto out_stop;
1794 }
1795 BUFFER_TRACE(dir_block, "get_write_access");
1796 ext3_journal_get_write_access(handle, dir_block);
1797 de = (struct ext3_dir_entry_2 *) dir_block->b_data;
1798 de->inode = cpu_to_le32(inode->i_ino);
1799 de->name_len = 1;
Jan Kara7c06a8d2007-11-14 17:00:19 -08001800 de->rec_len = ext3_rec_len_to_disk(EXT3_DIR_REC_LEN(de->name_len));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801 strcpy (de->name, ".");
1802 ext3_set_de_type(dir->i_sb, de, S_IFDIR);
Jan Kara7c06a8d2007-11-14 17:00:19 -08001803 de = ext3_next_entry(de);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804 de->inode = cpu_to_le32(dir->i_ino);
Jan Kara7c06a8d2007-11-14 17:00:19 -08001805 de->rec_len = ext3_rec_len_to_disk(inode->i_sb->s_blocksize -
1806 EXT3_DIR_REC_LEN(1));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807 de->name_len = 2;
1808 strcpy (de->name, "..");
1809 ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1810 inode->i_nlink = 2;
1811 BUFFER_TRACE(dir_block, "call ext3_journal_dirty_metadata");
1812 ext3_journal_dirty_metadata(handle, dir_block);
1813 brelse (dir_block);
1814 ext3_mark_inode_dirty(handle, inode);
1815 err = ext3_add_entry (handle, dentry, inode);
1816 if (err) {
1817 inode->i_nlink = 0;
Al Viroc38012d2008-12-30 02:02:50 -05001818 unlock_new_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 ext3_mark_inode_dirty(handle, inode);
1820 iput (inode);
1821 goto out_stop;
1822 }
Dave Hansend8c76e62006-09-30 23:29:04 -07001823 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824 ext3_update_dx_flag(dir);
1825 ext3_mark_inode_dirty(handle, dir);
1826 d_instantiate(dentry, inode);
Al Viroc38012d2008-12-30 02:02:50 -05001827 unlock_new_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828out_stop:
1829 ext3_journal_stop(handle);
1830 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1831 goto retry;
1832 return err;
1833}
1834
1835/*
1836 * routine to check that the specified directory is empty (for rmdir)
1837 */
1838static int empty_dir (struct inode * inode)
1839{
1840 unsigned long offset;
1841 struct buffer_head * bh;
1842 struct ext3_dir_entry_2 * de, * de1;
1843 struct super_block * sb;
1844 int err = 0;
1845
1846 sb = inode->i_sb;
1847 if (inode->i_size < EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1848 !(bh = ext3_bread (NULL, inode, 0, 0, &err))) {
1849 if (err)
Harvey Harrisone05b6b52008-04-28 02:16:15 -07001850 ext3_error(inode->i_sb, __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851 "error %d reading directory #%lu offset 0",
1852 err, inode->i_ino);
1853 else
Harvey Harrisone05b6b52008-04-28 02:16:15 -07001854 ext3_warning(inode->i_sb, __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 "bad directory (dir #%lu) - no data block",
1856 inode->i_ino);
1857 return 1;
1858 }
1859 de = (struct ext3_dir_entry_2 *) bh->b_data;
Jan Kara7c06a8d2007-11-14 17:00:19 -08001860 de1 = ext3_next_entry(de);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 if (le32_to_cpu(de->inode) != inode->i_ino ||
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001862 !le32_to_cpu(de1->inode) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863 strcmp (".", de->name) ||
1864 strcmp ("..", de1->name)) {
Dave Kleikampe9ad5622006-09-27 01:49:35 -07001865 ext3_warning (inode->i_sb, "empty_dir",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866 "bad directory (dir #%lu) - no `.' or `..'",
1867 inode->i_ino);
1868 brelse (bh);
1869 return 1;
1870 }
Jan Kara7c06a8d2007-11-14 17:00:19 -08001871 offset = ext3_rec_len_from_disk(de->rec_len) +
1872 ext3_rec_len_from_disk(de1->rec_len);
1873 de = ext3_next_entry(de1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874 while (offset < inode->i_size ) {
1875 if (!bh ||
1876 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1877 err = 0;
1878 brelse (bh);
1879 bh = ext3_bread (NULL, inode,
1880 offset >> EXT3_BLOCK_SIZE_BITS(sb), 0, &err);
1881 if (!bh) {
1882 if (err)
Harvey Harrisone05b6b52008-04-28 02:16:15 -07001883 ext3_error(sb, __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884 "error %d reading directory"
1885 " #%lu offset %lu",
1886 err, inode->i_ino, offset);
1887 offset += sb->s_blocksize;
1888 continue;
1889 }
1890 de = (struct ext3_dir_entry_2 *) bh->b_data;
1891 }
1892 if (!ext3_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1893 de = (struct ext3_dir_entry_2 *)(bh->b_data +
1894 sb->s_blocksize);
1895 offset = (offset | (sb->s_blocksize - 1)) + 1;
1896 continue;
1897 }
1898 if (le32_to_cpu(de->inode)) {
1899 brelse (bh);
1900 return 0;
1901 }
Jan Kara7c06a8d2007-11-14 17:00:19 -08001902 offset += ext3_rec_len_from_disk(de->rec_len);
1903 de = ext3_next_entry(de);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904 }
1905 brelse (bh);
1906 return 1;
1907}
1908
1909/* ext3_orphan_add() links an unlinked or truncated inode into a list of
1910 * such inodes, starting at the superblock, in case we crash before the
1911 * file is closed/deleted, or in case the inode truncate spans multiple
1912 * transactions and the last transaction is not recovered after a crash.
1913 *
1914 * At filesystem recovery time, we walk this list deleting unlinked
1915 * inodes and truncating linked inodes in ext3_orphan_cleanup().
1916 */
1917int ext3_orphan_add(handle_t *handle, struct inode *inode)
1918{
1919 struct super_block *sb = inode->i_sb;
1920 struct ext3_iloc iloc;
1921 int err = 0, rc;
1922
Eric Sandeenb8a052d2009-12-14 13:00:30 -06001923 mutex_lock(&EXT3_SB(sb)->s_orphan_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924 if (!list_empty(&EXT3_I(inode)->i_orphan))
1925 goto out_unlock;
1926
1927 /* Orphan handling is only valid for files with data blocks
1928 * being truncated, or files being unlinked. */
1929
1930 /* @@@ FIXME: Observation from aviro:
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001931 * I think I can trigger J_ASSERT in ext3_orphan_add(). We block
Eric Sandeenb8a052d2009-12-14 13:00:30 -06001932 * here (on s_orphan_lock), so race with ext3_link() which might bump
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933 * ->i_nlink. For, say it, character device. Not a regular file,
1934 * not a directory, not a symlink and ->i_nlink > 0.
Eric Sandeenb8a052d2009-12-14 13:00:30 -06001935 *
1936 * tytso, 4/25/2009: I'm not sure how that could happen;
1937 * shouldn't the fs core protect us from these sort of
1938 * unlink()/link() races?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 */
1940 J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1941 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1942
1943 BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "get_write_access");
1944 err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
1945 if (err)
1946 goto out_unlock;
1947
1948 err = ext3_reserve_inode_write(handle, inode, &iloc);
1949 if (err)
1950 goto out_unlock;
1951
1952 /* Insert this inode at the head of the on-disk orphan list... */
1953 NEXT_ORPHAN(inode) = le32_to_cpu(EXT3_SB(sb)->s_es->s_last_orphan);
1954 EXT3_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1955 err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
1956 rc = ext3_mark_iloc_dirty(handle, inode, &iloc);
1957 if (!err)
1958 err = rc;
1959
1960 /* Only add to the head of the in-memory list if all the
1961 * previous operations succeeded. If the orphan_add is going to
1962 * fail (possibly taking the journal offline), we can't risk
1963 * leaving the inode on the orphan list: stray orphan-list
1964 * entries can cause panics at unmount time.
1965 *
1966 * This is safe: on error we're going to ignore the orphan list
1967 * anyway on the next recovery. */
1968 if (!err)
1969 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1970
Eric Sandeeneee194e2006-09-27 01:49:30 -07001971 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1972 jbd_debug(4, "orphan inode %lu will point to %d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973 inode->i_ino, NEXT_ORPHAN(inode));
1974out_unlock:
Eric Sandeenb8a052d2009-12-14 13:00:30 -06001975 mutex_unlock(&EXT3_SB(sb)->s_orphan_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 ext3_std_error(inode->i_sb, err);
1977 return err;
1978}
1979
1980/*
1981 * ext3_orphan_del() removes an unlinked or truncated inode from the list
1982 * of such inodes stored on disk, because it is finally being cleaned up.
1983 */
1984int ext3_orphan_del(handle_t *handle, struct inode *inode)
1985{
1986 struct list_head *prev;
1987 struct ext3_inode_info *ei = EXT3_I(inode);
1988 struct ext3_sb_info *sbi;
1989 unsigned long ino_next;
1990 struct ext3_iloc iloc;
1991 int err = 0;
1992
Eric Sandeenb8a052d2009-12-14 13:00:30 -06001993 mutex_lock(&EXT3_SB(inode->i_sb)->s_orphan_lock);
1994 if (list_empty(&ei->i_orphan))
1995 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996
1997 ino_next = NEXT_ORPHAN(inode);
1998 prev = ei->i_orphan.prev;
1999 sbi = EXT3_SB(inode->i_sb);
2000
2001 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2002
2003 list_del_init(&ei->i_orphan);
2004
2005 /* If we're on an error path, we may not have a valid
2006 * transaction handle with which to update the orphan list on
2007 * disk, but we still need to remove the inode from the linked
2008 * list in memory. */
2009 if (!handle)
2010 goto out;
2011
2012 err = ext3_reserve_inode_write(handle, inode, &iloc);
2013 if (err)
2014 goto out_err;
2015
2016 if (prev == &sbi->s_orphan) {
2017 jbd_debug(4, "superblock will point to %lu\n", ino_next);
2018 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2019 err = ext3_journal_get_write_access(handle, sbi->s_sbh);
2020 if (err)
2021 goto out_brelse;
2022 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2023 err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
2024 } else {
2025 struct ext3_iloc iloc2;
2026 struct inode *i_prev =
2027 &list_entry(prev, struct ext3_inode_info, i_orphan)->vfs_inode;
2028
2029 jbd_debug(4, "orphan inode %lu will point to %lu\n",
2030 i_prev->i_ino, ino_next);
2031 err = ext3_reserve_inode_write(handle, i_prev, &iloc2);
2032 if (err)
2033 goto out_brelse;
2034 NEXT_ORPHAN(i_prev) = ino_next;
2035 err = ext3_mark_iloc_dirty(handle, i_prev, &iloc2);
2036 }
2037 if (err)
2038 goto out_brelse;
2039 NEXT_ORPHAN(inode) = 0;
2040 err = ext3_mark_iloc_dirty(handle, inode, &iloc);
2041
2042out_err:
2043 ext3_std_error(inode->i_sb, err);
2044out:
Eric Sandeenb8a052d2009-12-14 13:00:30 -06002045 mutex_unlock(&EXT3_SB(inode->i_sb)->s_orphan_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 return err;
2047
2048out_brelse:
2049 brelse(iloc.bh);
2050 goto out_err;
2051}
2052
2053static int ext3_rmdir (struct inode * dir, struct dentry *dentry)
2054{
2055 int retval;
2056 struct inode * inode;
2057 struct buffer_head * bh;
2058 struct ext3_dir_entry_2 * de;
2059 handle_t *handle;
2060
2061 /* Initialize quotas before so that eventual writes go in
2062 * separate transaction */
Jan Kara81a05222009-01-26 16:58:01 +01002063 vfs_dq_init(dentry->d_inode);
Jan Kara1f545872005-06-23 22:01:04 -07002064 handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065 if (IS_ERR(handle))
2066 return PTR_ERR(handle);
2067
2068 retval = -ENOENT;
Al Viro734711a2008-08-24 07:26:48 -04002069 bh = ext3_find_entry(dir, &dentry->d_name, &de);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070 if (!bh)
2071 goto end_rmdir;
2072
2073 if (IS_DIRSYNC(dir))
2074 handle->h_sync = 1;
2075
2076 inode = dentry->d_inode;
2077
2078 retval = -EIO;
2079 if (le32_to_cpu(de->inode) != inode->i_ino)
2080 goto end_rmdir;
2081
2082 retval = -ENOTEMPTY;
2083 if (!empty_dir (inode))
2084 goto end_rmdir;
2085
2086 retval = ext3_delete_entry(handle, dir, de, bh);
2087 if (retval)
2088 goto end_rmdir;
2089 if (inode->i_nlink != 2)
2090 ext3_warning (inode->i_sb, "ext3_rmdir",
2091 "empty directory has nlink!=2 (%d)",
2092 inode->i_nlink);
2093 inode->i_version++;
Dave Hansence71ec32006-09-30 23:29:06 -07002094 clear_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 /* There's no need to set i_disksize: the fact that i_nlink is
2096 * zero will ensure that the right thing happens during any
2097 * recovery. */
2098 inode->i_size = 0;
2099 ext3_orphan_add(handle, inode);
2100 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2101 ext3_mark_inode_dirty(handle, inode);
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002102 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103 ext3_update_dx_flag(dir);
2104 ext3_mark_inode_dirty(handle, dir);
2105
2106end_rmdir:
2107 ext3_journal_stop(handle);
2108 brelse (bh);
2109 return retval;
2110}
2111
2112static int ext3_unlink(struct inode * dir, struct dentry *dentry)
2113{
2114 int retval;
2115 struct inode * inode;
2116 struct buffer_head * bh;
2117 struct ext3_dir_entry_2 * de;
2118 handle_t *handle;
2119
2120 /* Initialize quotas before so that eventual writes go
2121 * in separate transaction */
Jan Kara81a05222009-01-26 16:58:01 +01002122 vfs_dq_init(dentry->d_inode);
Jan Kara1f545872005-06-23 22:01:04 -07002123 handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 if (IS_ERR(handle))
2125 return PTR_ERR(handle);
2126
2127 if (IS_DIRSYNC(dir))
2128 handle->h_sync = 1;
2129
2130 retval = -ENOENT;
Al Viro734711a2008-08-24 07:26:48 -04002131 bh = ext3_find_entry(dir, &dentry->d_name, &de);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132 if (!bh)
2133 goto end_unlink;
2134
2135 inode = dentry->d_inode;
2136
2137 retval = -EIO;
2138 if (le32_to_cpu(de->inode) != inode->i_ino)
2139 goto end_unlink;
2140
2141 if (!inode->i_nlink) {
2142 ext3_warning (inode->i_sb, "ext3_unlink",
2143 "Deleting nonexistent file (%lu), %d",
2144 inode->i_ino, inode->i_nlink);
2145 inode->i_nlink = 1;
2146 }
2147 retval = ext3_delete_entry(handle, dir, de, bh);
2148 if (retval)
2149 goto end_unlink;
2150 dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2151 ext3_update_dx_flag(dir);
2152 ext3_mark_inode_dirty(handle, dir);
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002153 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154 if (!inode->i_nlink)
2155 ext3_orphan_add(handle, inode);
2156 inode->i_ctime = dir->i_ctime;
2157 ext3_mark_inode_dirty(handle, inode);
2158 retval = 0;
2159
2160end_unlink:
2161 ext3_journal_stop(handle);
2162 brelse (bh);
2163 return retval;
2164}
2165
2166static int ext3_symlink (struct inode * dir,
2167 struct dentry *dentry, const char * symname)
2168{
2169 handle_t *handle;
2170 struct inode * inode;
2171 int l, err, retries = 0;
2172
2173 l = strlen(symname)+1;
2174 if (l > dir->i_sb->s_blocksize)
2175 return -ENAMETOOLONG;
2176
2177retry:
Jan Kara1f545872005-06-23 22:01:04 -07002178 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
Dave Kleikampe9ad5622006-09-27 01:49:35 -07002179 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 5 +
Dmitry Monakhovc4590012009-12-09 03:05:30 +03002180 EXT3_MAXQUOTAS_INIT_BLOCKS(dir->i_sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181 if (IS_ERR(handle))
2182 return PTR_ERR(handle);
2183
2184 if (IS_DIRSYNC(dir))
2185 handle->h_sync = 1;
2186
2187 inode = ext3_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
2188 err = PTR_ERR(inode);
2189 if (IS_ERR(inode))
2190 goto out_stop;
2191
2192 if (l > sizeof (EXT3_I(inode)->i_data)) {
2193 inode->i_op = &ext3_symlink_inode_operations;
2194 ext3_set_aops(inode);
2195 /*
2196 * page_symlink() calls into ext3_prepare/commit_write.
2197 * We have a transaction open. All is sweetness. It also sets
2198 * i_size in generic_commit_write().
2199 */
Nick Piggin54566b22009-01-04 12:00:53 -08002200 err = __page_symlink(inode, symname, l, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201 if (err) {
Eric Sandeen731b9a52007-02-10 01:46:16 -08002202 drop_nlink(inode);
Al Viroc38012d2008-12-30 02:02:50 -05002203 unlock_new_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 ext3_mark_inode_dirty(handle, inode);
2205 iput (inode);
2206 goto out_stop;
2207 }
2208 } else {
2209 inode->i_op = &ext3_fast_symlink_inode_operations;
2210 memcpy((char*)&EXT3_I(inode)->i_data,symname,l);
2211 inode->i_size = l-1;
2212 }
2213 EXT3_I(inode)->i_disksize = inode->i_size;
2214 err = ext3_add_nondir(handle, dentry, inode);
2215out_stop:
2216 ext3_journal_stop(handle);
2217 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2218 goto retry;
2219 return err;
2220}
2221
2222static int ext3_link (struct dentry * old_dentry,
2223 struct inode * dir, struct dentry *dentry)
2224{
2225 handle_t *handle;
2226 struct inode *inode = old_dentry->d_inode;
2227 int err, retries = 0;
2228
2229 if (inode->i_nlink >= EXT3_LINK_MAX)
2230 return -EMLINK;
Eric Sandeen2988a772007-02-10 01:46:16 -08002231 /*
2232 * Return -ENOENT if we've raced with unlink and i_nlink is 0. Doing
2233 * otherwise has the potential to corrupt the orphan inode list.
2234 */
2235 if (inode->i_nlink == 0)
2236 return -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237
2238retry:
Jan Kara1f545872005-06-23 22:01:04 -07002239 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 EXT3_INDEX_EXTRA_TRANS_BLOCKS);
2241 if (IS_ERR(handle))
2242 return PTR_ERR(handle);
2243
2244 if (IS_DIRSYNC(dir))
2245 handle->h_sync = 1;
2246
2247 inode->i_ctime = CURRENT_TIME_SEC;
Eric Sandeen731b9a52007-02-10 01:46:16 -08002248 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249 atomic_inc(&inode->i_count);
2250
Al Viroc38012d2008-12-30 02:02:50 -05002251 err = ext3_add_entry(handle, dentry, inode);
2252 if (!err) {
2253 ext3_mark_inode_dirty(handle, inode);
2254 d_instantiate(dentry, inode);
2255 } else {
2256 drop_nlink(inode);
2257 iput(inode);
2258 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259 ext3_journal_stop(handle);
2260 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2261 goto retry;
2262 return err;
2263}
2264
2265#define PARENT_INO(buffer) \
Jan Kara7c06a8d2007-11-14 17:00:19 -08002266 (ext3_next_entry((struct ext3_dir_entry_2 *)(buffer))->inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267
2268/*
2269 * Anybody can rename anything with this: the permission checks are left to the
2270 * higher-level routines.
2271 */
2272static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry,
2273 struct inode * new_dir,struct dentry *new_dentry)
2274{
2275 handle_t *handle;
2276 struct inode * old_inode, * new_inode;
2277 struct buffer_head * old_bh, * new_bh, * dir_bh;
2278 struct ext3_dir_entry_2 * old_de, * new_de;
Theodore Ts'oe7c8f502009-04-03 01:34:49 -04002279 int retval, flush_file = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280
2281 old_bh = new_bh = dir_bh = NULL;
2282
2283 /* Initialize quotas before so that eventual writes go
2284 * in separate transaction */
2285 if (new_dentry->d_inode)
Jan Kara81a05222009-01-26 16:58:01 +01002286 vfs_dq_init(new_dentry->d_inode);
Jan Kara1f545872005-06-23 22:01:04 -07002287 handle = ext3_journal_start(old_dir, 2 *
2288 EXT3_DATA_TRANS_BLOCKS(old_dir->i_sb) +
Dave Kleikampe9ad5622006-09-27 01:49:35 -07002289 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290 if (IS_ERR(handle))
2291 return PTR_ERR(handle);
2292
2293 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2294 handle->h_sync = 1;
2295
Al Viro734711a2008-08-24 07:26:48 -04002296 old_bh = ext3_find_entry(old_dir, &old_dentry->d_name, &old_de);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297 /*
2298 * Check for inode number is _not_ due to possible IO errors.
2299 * We might rmdir the source, keep it as pwd of some process
2300 * and merrily kill the link to whatever was created under the
2301 * same name. Goodbye sticky bit ;-<
2302 */
2303 old_inode = old_dentry->d_inode;
2304 retval = -ENOENT;
2305 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2306 goto end_rename;
2307
2308 new_inode = new_dentry->d_inode;
Al Viro734711a2008-08-24 07:26:48 -04002309 new_bh = ext3_find_entry(new_dir, &new_dentry->d_name, &new_de);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310 if (new_bh) {
2311 if (!new_inode) {
2312 brelse (new_bh);
2313 new_bh = NULL;
2314 }
2315 }
2316 if (S_ISDIR(old_inode->i_mode)) {
2317 if (new_inode) {
2318 retval = -ENOTEMPTY;
2319 if (!empty_dir (new_inode))
2320 goto end_rename;
2321 }
2322 retval = -EIO;
2323 dir_bh = ext3_bread (handle, old_inode, 0, 0, &retval);
2324 if (!dir_bh)
2325 goto end_rename;
2326 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2327 goto end_rename;
2328 retval = -EMLINK;
2329 if (!new_inode && new_dir!=old_dir &&
2330 new_dir->i_nlink >= EXT3_LINK_MAX)
2331 goto end_rename;
2332 }
2333 if (!new_bh) {
2334 retval = ext3_add_entry (handle, new_dentry, old_inode);
2335 if (retval)
2336 goto end_rename;
2337 } else {
2338 BUFFER_TRACE(new_bh, "get write access");
2339 ext3_journal_get_write_access(handle, new_bh);
2340 new_de->inode = cpu_to_le32(old_inode->i_ino);
2341 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2342 EXT3_FEATURE_INCOMPAT_FILETYPE))
2343 new_de->file_type = old_de->file_type;
2344 new_dir->i_version++;
Jan Kara0b230762008-04-28 02:16:12 -07002345 new_dir->i_ctime = new_dir->i_mtime = CURRENT_TIME_SEC;
2346 ext3_mark_inode_dirty(handle, new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347 BUFFER_TRACE(new_bh, "call ext3_journal_dirty_metadata");
2348 ext3_journal_dirty_metadata(handle, new_bh);
2349 brelse(new_bh);
2350 new_bh = NULL;
2351 }
2352
2353 /*
2354 * Like most other Unix systems, set the ctime for inodes on a
2355 * rename.
2356 */
2357 old_inode->i_ctime = CURRENT_TIME_SEC;
2358 ext3_mark_inode_dirty(handle, old_inode);
2359
2360 /*
2361 * ok, that's it
2362 */
2363 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2364 old_de->name_len != old_dentry->d_name.len ||
2365 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2366 (retval = ext3_delete_entry(handle, old_dir,
2367 old_de, old_bh)) == -ENOENT) {
2368 /* old_de could have moved from under us during htree split, so
2369 * make sure that we are deleting the right entry. We might
2370 * also be pointing to a stale entry in the unused part of
2371 * old_bh so just checking inum and the name isn't enough. */
2372 struct buffer_head *old_bh2;
2373 struct ext3_dir_entry_2 *old_de2;
2374
Al Viro734711a2008-08-24 07:26:48 -04002375 old_bh2 = ext3_find_entry(old_dir, &old_dentry->d_name,
2376 &old_de2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377 if (old_bh2) {
2378 retval = ext3_delete_entry(handle, old_dir,
2379 old_de2, old_bh2);
2380 brelse(old_bh2);
2381 }
2382 }
2383 if (retval) {
2384 ext3_warning(old_dir->i_sb, "ext3_rename",
2385 "Deleting old file (%lu), %d, error=%d",
2386 old_dir->i_ino, old_dir->i_nlink, retval);
2387 }
2388
2389 if (new_inode) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002390 drop_nlink(new_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391 new_inode->i_ctime = CURRENT_TIME_SEC;
2392 }
2393 old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC;
2394 ext3_update_dx_flag(old_dir);
2395 if (dir_bh) {
2396 BUFFER_TRACE(dir_bh, "get_write_access");
2397 ext3_journal_get_write_access(handle, dir_bh);
2398 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2399 BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata");
2400 ext3_journal_dirty_metadata(handle, dir_bh);
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002401 drop_nlink(old_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402 if (new_inode) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002403 drop_nlink(new_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404 } else {
Dave Hansend8c76e62006-09-30 23:29:04 -07002405 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406 ext3_update_dx_flag(new_dir);
2407 ext3_mark_inode_dirty(handle, new_dir);
2408 }
2409 }
2410 ext3_mark_inode_dirty(handle, old_dir);
2411 if (new_inode) {
2412 ext3_mark_inode_dirty(handle, new_inode);
2413 if (!new_inode->i_nlink)
2414 ext3_orphan_add(handle, new_inode);
Theodore Ts'oe7c8f502009-04-03 01:34:49 -04002415 if (ext3_should_writeback_data(new_inode))
2416 flush_file = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417 }
2418 retval = 0;
2419
2420end_rename:
2421 brelse (dir_bh);
2422 brelse (old_bh);
2423 brelse (new_bh);
2424 ext3_journal_stop(handle);
Theodore Ts'oe7c8f502009-04-03 01:34:49 -04002425 if (retval == 0 && flush_file)
2426 filemap_flush(old_inode->i_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427 return retval;
2428}
2429
2430/*
2431 * directories can handle most operations...
2432 */
Arjan van de Ven754661f2007-02-12 00:55:38 -08002433const struct inode_operations ext3_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002434 .create = ext3_create,
2435 .lookup = ext3_lookup,
2436 .link = ext3_link,
2437 .unlink = ext3_unlink,
2438 .symlink = ext3_symlink,
2439 .mkdir = ext3_mkdir,
2440 .rmdir = ext3_rmdir,
2441 .mknod = ext3_mknod,
2442 .rename = ext3_rename,
2443 .setattr = ext3_setattr,
2444#ifdef CONFIG_EXT3_FS_XATTR
2445 .setxattr = generic_setxattr,
2446 .getxattr = generic_getxattr,
2447 .listxattr = ext3_listxattr,
2448 .removexattr = generic_removexattr,
2449#endif
Linus Torvalds1d5ccd12009-08-28 12:12:24 -07002450 .check_acl = ext3_check_acl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451};
2452
Arjan van de Ven754661f2007-02-12 00:55:38 -08002453const struct inode_operations ext3_special_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454 .setattr = ext3_setattr,
2455#ifdef CONFIG_EXT3_FS_XATTR
2456 .setxattr = generic_setxattr,
2457 .getxattr = generic_getxattr,
2458 .listxattr = ext3_listxattr,
2459 .removexattr = generic_removexattr,
2460#endif
Linus Torvalds1d5ccd12009-08-28 12:12:24 -07002461 .check_acl = ext3_check_acl,
Mingming Caoae6ddcc2006-09-27 01:49:27 -07002462};