blob: 94ee6f315dc17580c6de82f65d3f58735a29b870 [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,
54 u32 *block, int *err)
55{
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
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700147static inline unsigned dx_get_block (struct dx_entry *entry);
148static void dx_set_block (struct dx_entry *entry, unsigned value);
149static 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);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700169static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
Mingming Cao617ba132006-10-11 01:20:53 -0700170static int ext4_htree_next_block(struct inode *dir, __u32 hash,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700171 struct dx_frame *frame,
172 struct dx_frame *frames,
173 __u32 *start_hash);
Mingming Cao617ba132006-10-11 01:20:53 -0700174static struct buffer_head * ext4_dx_find_entry(struct dentry *dentry,
175 struct ext4_dir_entry_2 **res_dir, int *err);
176static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700177 struct inode *inode);
178
179/*
180 * Future: use high four bits of block for coalesce-on-delete flags
181 * Mask them off for now.
182 */
183
184static inline unsigned dx_get_block (struct dx_entry *entry)
185{
186 return le32_to_cpu(entry->block) & 0x00ffffff;
187}
188
189static inline void dx_set_block (struct dx_entry *entry, unsigned value)
190{
191 entry->block = cpu_to_le32(value);
192}
193
194static inline unsigned dx_get_hash (struct dx_entry *entry)
195{
196 return le32_to_cpu(entry->hash);
197}
198
199static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
200{
201 entry->hash = cpu_to_le32(value);
202}
203
204static inline unsigned dx_get_count (struct dx_entry *entries)
205{
206 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
207}
208
209static inline unsigned dx_get_limit (struct dx_entry *entries)
210{
211 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
212}
213
214static inline void dx_set_count (struct dx_entry *entries, unsigned value)
215{
216 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
217}
218
219static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
220{
221 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
222}
223
224static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
225{
Mingming Cao617ba132006-10-11 01:20:53 -0700226 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(1) -
227 EXT4_DIR_REC_LEN(2) - infosize;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700228 return 0? 20: entry_space / sizeof(struct dx_entry);
229}
230
231static inline unsigned dx_node_limit (struct inode *dir)
232{
Mingming Cao617ba132006-10-11 01:20:53 -0700233 unsigned entry_space = dir->i_sb->s_blocksize - EXT4_DIR_REC_LEN(0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700234 return 0? 22: entry_space / sizeof(struct dx_entry);
235}
236
237/*
238 * Debug
239 */
240#ifdef DX_DEBUG
241static void dx_show_index (char * label, struct dx_entry *entries)
242{
Andrew Morton63f57932006-10-11 01:21:24 -0700243 int i, n = dx_get_count (entries);
Dave Kleikamp8c55e202007-05-24 13:04:54 -0400244 printk("%s index ", label);
Andrew Morton63f57932006-10-11 01:21:24 -0700245 for (i = 0; i < n; i++) {
246 printk("%x->%u ", i? dx_get_hash(entries + i) :
247 0, dx_get_block(entries + i));
248 }
249 printk("\n");
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700250}
251
252struct stats
253{
254 unsigned names;
255 unsigned space;
256 unsigned bcount;
257};
258
Mingming Cao617ba132006-10-11 01:20:53 -0700259static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext4_dir_entry_2 *de,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700260 int size, int show_names)
261{
262 unsigned names = 0, space = 0;
263 char *base = (char *) de;
264 struct dx_hash_info h = *hinfo;
265
266 printk("names: ");
267 while ((char *) de < base + size)
268 {
269 if (de->inode)
270 {
271 if (show_names)
272 {
273 int len = de->name_len;
274 char *name = de->name;
275 while (len--) printk("%c", *name++);
Mingming Cao617ba132006-10-11 01:20:53 -0700276 ext4fs_dirhash(de->name, de->name_len, &h);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700277 printk(":%x.%u ", h.hash,
278 ((char *) de - base));
279 }
Mingming Cao617ba132006-10-11 01:20:53 -0700280 space += EXT4_DIR_REC_LEN(de->name_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700281 names++;
282 }
Mingming Cao617ba132006-10-11 01:20:53 -0700283 de = (struct ext4_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700284 }
285 printk("(%i)\n", names);
286 return (struct stats) { names, space, 1 };
287}
288
289struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
290 struct dx_entry *entries, int levels)
291{
292 unsigned blocksize = dir->i_sb->s_blocksize;
293 unsigned count = dx_get_count (entries), names = 0, space = 0, i;
294 unsigned bcount = 0;
295 struct buffer_head *bh;
296 int err;
297 printk("%i indexed blocks...\n", count);
298 for (i = 0; i < count; i++, entries++)
299 {
300 u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0;
301 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
302 struct stats stats;
303 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
Mingming Cao617ba132006-10-11 01:20:53 -0700304 if (!(bh = ext4_bread (NULL,dir, block, 0,&err))) continue;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700305 stats = levels?
306 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
Mingming Cao617ba132006-10-11 01:20:53 -0700307 dx_show_leaf(hinfo, (struct ext4_dir_entry_2 *) bh->b_data, blocksize, 0);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700308 names += stats.names;
309 space += stats.space;
310 bcount += stats.bcount;
311 brelse (bh);
312 }
313 if (bcount)
314 printk("%snames %u, fullness %u (%u%%)\n", levels?"":" ",
315 names, space/bcount,(space/bcount)*100/blocksize);
316 return (struct stats) { names, space, bcount};
317}
318#endif /* DX_DEBUG */
319
320/*
321 * Probe for a directory leaf block to search.
322 *
323 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
324 * error in the directory index, and the caller should fall back to
325 * searching the directory normally. The callers of dx_probe **MUST**
326 * check for this error code, and make sure it never gets reflected
327 * back to userspace.
328 */
329static struct dx_frame *
330dx_probe(struct dentry *dentry, struct inode *dir,
331 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
332{
333 unsigned count, indirect;
334 struct dx_entry *at, *entries, *p, *q, *m;
335 struct dx_root *root;
336 struct buffer_head *bh;
337 struct dx_frame *frame = frame_in;
338 u32 hash;
339
340 frame->bh = NULL;
341 if (dentry)
342 dir = dentry->d_parent->d_inode;
Mingming Cao617ba132006-10-11 01:20:53 -0700343 if (!(bh = ext4_bread (NULL,dir, 0, 0, err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700344 goto fail;
345 root = (struct dx_root *) bh->b_data;
346 if (root->info.hash_version != DX_HASH_TEA &&
347 root->info.hash_version != DX_HASH_HALF_MD4 &&
348 root->info.hash_version != DX_HASH_LEGACY) {
Mingming Cao617ba132006-10-11 01:20:53 -0700349 ext4_warning(dir->i_sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700350 "Unrecognised inode hash code %d",
351 root->info.hash_version);
352 brelse(bh);
353 *err = ERR_BAD_DX_DIR;
354 goto fail;
355 }
356 hinfo->hash_version = root->info.hash_version;
Mingming Cao617ba132006-10-11 01:20:53 -0700357 hinfo->seed = EXT4_SB(dir->i_sb)->s_hash_seed;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700358 if (dentry)
Mingming Cao617ba132006-10-11 01:20:53 -0700359 ext4fs_dirhash(dentry->d_name.name, dentry->d_name.len, hinfo);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700360 hash = hinfo->hash;
361
362 if (root->info.unused_flags & 1) {
Mingming Cao617ba132006-10-11 01:20:53 -0700363 ext4_warning(dir->i_sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700364 "Unimplemented inode hash flags: %#06x",
365 root->info.unused_flags);
366 brelse(bh);
367 *err = ERR_BAD_DX_DIR;
368 goto fail;
369 }
370
371 if ((indirect = root->info.indirect_levels) > 1) {
Mingming Cao617ba132006-10-11 01:20:53 -0700372 ext4_warning(dir->i_sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700373 "Unimplemented inode hash depth: %#06x",
374 root->info.indirect_levels);
375 brelse(bh);
376 *err = ERR_BAD_DX_DIR;
377 goto fail;
378 }
379
380 entries = (struct dx_entry *) (((char *)&root->info) +
381 root->info.info_length);
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700382
383 if (dx_get_limit(entries) != dx_root_limit(dir,
384 root->info.info_length)) {
385 ext4_warning(dir->i_sb, __FUNCTION__,
386 "dx entry: limit != root limit");
387 brelse(bh);
388 *err = ERR_BAD_DX_DIR;
389 goto fail;
390 }
391
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700392 dxtrace (printk("Look up %x", hash));
393 while (1)
394 {
395 count = dx_get_count(entries);
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700396 if (!count || count > dx_get_limit(entries)) {
397 ext4_warning(dir->i_sb, __FUNCTION__,
398 "dx entry: no count or count > limit");
399 brelse(bh);
400 *err = ERR_BAD_DX_DIR;
401 goto fail2;
402 }
403
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700404 p = entries + 1;
405 q = entries + count - 1;
406 while (p <= q)
407 {
408 m = p + (q - p)/2;
409 dxtrace(printk("."));
410 if (dx_get_hash(m) > hash)
411 q = m - 1;
412 else
413 p = m + 1;
414 }
415
416 if (0) // linear search cross check
417 {
418 unsigned n = count - 1;
419 at = entries;
420 while (n--)
421 {
422 dxtrace(printk(","));
423 if (dx_get_hash(++at) > hash)
424 {
425 at--;
426 break;
427 }
428 }
429 assert (at == p - 1);
430 }
431
432 at = p - 1;
433 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
434 frame->bh = bh;
435 frame->entries = entries;
436 frame->at = at;
437 if (!indirect--) return frame;
Mingming Cao617ba132006-10-11 01:20:53 -0700438 if (!(bh = ext4_bread (NULL,dir, dx_get_block(at), 0, err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700439 goto fail2;
440 at = entries = ((struct dx_node *) bh->b_data)->entries;
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700441 if (dx_get_limit(entries) != dx_node_limit (dir)) {
442 ext4_warning(dir->i_sb, __FUNCTION__,
443 "dx entry: limit != node limit");
444 brelse(bh);
445 *err = ERR_BAD_DX_DIR;
446 goto fail2;
447 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700448 frame++;
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700449 frame->bh = NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700450 }
451fail2:
452 while (frame >= frame_in) {
453 brelse(frame->bh);
454 frame--;
455 }
456fail:
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700457 if (*err == ERR_BAD_DX_DIR)
458 ext4_warning(dir->i_sb, __FUNCTION__,
459 "Corrupt dir inode %ld, running e2fsck is "
460 "recommended.", dir->i_ino);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700461 return NULL;
462}
463
464static void dx_release (struct dx_frame *frames)
465{
466 if (frames[0].bh == NULL)
467 return;
468
469 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
470 brelse(frames[1].bh);
471 brelse(frames[0].bh);
472}
473
474/*
475 * This function increments the frame pointer to search the next leaf
476 * block, and reads in the necessary intervening nodes if the search
477 * should be necessary. Whether or not the search is necessary is
478 * controlled by the hash parameter. If the hash value is even, then
479 * the search is only continued if the next block starts with that
480 * hash value. This is used if we are searching for a specific file.
481 *
482 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
483 *
484 * This function returns 1 if the caller should continue to search,
485 * or 0 if it should not. If there is an error reading one of the
486 * index blocks, it will a negative error code.
487 *
488 * If start_hash is non-null, it will be filled in with the starting
489 * hash of the next page.
490 */
Mingming Cao617ba132006-10-11 01:20:53 -0700491static int ext4_htree_next_block(struct inode *dir, __u32 hash,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700492 struct dx_frame *frame,
493 struct dx_frame *frames,
494 __u32 *start_hash)
495{
496 struct dx_frame *p;
497 struct buffer_head *bh;
498 int err, num_frames = 0;
499 __u32 bhash;
500
501 p = frame;
502 /*
503 * Find the next leaf page by incrementing the frame pointer.
504 * If we run out of entries in the interior node, loop around and
505 * increment pointer in the parent node. When we break out of
506 * this loop, num_frames indicates the number of interior
507 * nodes need to be read.
508 */
509 while (1) {
510 if (++(p->at) < p->entries + dx_get_count(p->entries))
511 break;
512 if (p == frames)
513 return 0;
514 num_frames++;
515 p--;
516 }
517
518 /*
519 * If the hash is 1, then continue only if the next page has a
520 * continuation hash of any value. This is used for readdir
521 * handling. Otherwise, check to see if the hash matches the
522 * desired contiuation hash. If it doesn't, return since
523 * there's no point to read in the successive index pages.
524 */
525 bhash = dx_get_hash(p->at);
526 if (start_hash)
527 *start_hash = bhash;
528 if ((hash & 1) == 0) {
529 if ((bhash & ~1) != hash)
530 return 0;
531 }
532 /*
533 * If the hash is HASH_NB_ALWAYS, we always go to the next
534 * block so no check is necessary
535 */
536 while (num_frames--) {
Mingming Cao617ba132006-10-11 01:20:53 -0700537 if (!(bh = ext4_bread(NULL, dir, dx_get_block(p->at),
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700538 0, &err)))
539 return err; /* Failure */
540 p++;
541 brelse (p->bh);
542 p->bh = bh;
543 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
544 }
545 return 1;
546}
547
548
549/*
550 * p is at least 6 bytes before the end of page
551 */
Mingming Cao617ba132006-10-11 01:20:53 -0700552static inline struct ext4_dir_entry_2 *ext4_next_entry(struct ext4_dir_entry_2 *p)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700553{
Mingming Cao617ba132006-10-11 01:20:53 -0700554 return (struct ext4_dir_entry_2 *)((char*)p + le16_to_cpu(p->rec_len));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700555}
556
557/*
558 * This function fills a red-black tree with information from a
559 * directory block. It returns the number directory entries loaded
560 * into the tree. If there is an error it is returned in err.
561 */
562static int htree_dirblock_to_tree(struct file *dir_file,
563 struct inode *dir, int block,
564 struct dx_hash_info *hinfo,
565 __u32 start_hash, __u32 start_minor_hash)
566{
567 struct buffer_head *bh;
Mingming Cao617ba132006-10-11 01:20:53 -0700568 struct ext4_dir_entry_2 *de, *top;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700569 int err, count = 0;
570
571 dxtrace(printk("In htree dirblock_to_tree: block %d\n", block));
Mingming Cao617ba132006-10-11 01:20:53 -0700572 if (!(bh = ext4_bread (NULL, dir, block, 0, &err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700573 return err;
574
Mingming Cao617ba132006-10-11 01:20:53 -0700575 de = (struct ext4_dir_entry_2 *) bh->b_data;
576 top = (struct ext4_dir_entry_2 *) ((char *) de +
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700577 dir->i_sb->s_blocksize -
Mingming Cao617ba132006-10-11 01:20:53 -0700578 EXT4_DIR_REC_LEN(0));
579 for (; de < top; de = ext4_next_entry(de)) {
Eric Sandeene6c40212006-12-06 20:36:28 -0800580 if (!ext4_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
581 (block<<EXT4_BLOCK_SIZE_BITS(dir->i_sb))
582 +((char *)de - bh->b_data))) {
583 /* On error, skip the f_pos to the next block. */
584 dir_file->f_pos = (dir_file->f_pos |
585 (dir->i_sb->s_blocksize - 1)) + 1;
586 brelse (bh);
587 return count;
588 }
Mingming Cao617ba132006-10-11 01:20:53 -0700589 ext4fs_dirhash(de->name, de->name_len, hinfo);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700590 if ((hinfo->hash < start_hash) ||
591 ((hinfo->hash == start_hash) &&
592 (hinfo->minor_hash < start_minor_hash)))
593 continue;
594 if (de->inode == 0)
595 continue;
Mingming Cao617ba132006-10-11 01:20:53 -0700596 if ((err = ext4_htree_store_dirent(dir_file,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700597 hinfo->hash, hinfo->minor_hash, de)) != 0) {
598 brelse(bh);
599 return err;
600 }
601 count++;
602 }
603 brelse(bh);
604 return count;
605}
606
607
608/*
609 * This function fills a red-black tree with information from a
610 * directory. We start scanning the directory in hash order, starting
611 * at start_hash and start_minor_hash.
612 *
613 * This function returns the number of entries inserted into the tree,
614 * or a negative error code.
615 */
Mingming Cao617ba132006-10-11 01:20:53 -0700616int ext4_htree_fill_tree(struct file *dir_file, __u32 start_hash,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700617 __u32 start_minor_hash, __u32 *next_hash)
618{
619 struct dx_hash_info hinfo;
Mingming Cao617ba132006-10-11 01:20:53 -0700620 struct ext4_dir_entry_2 *de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700621 struct dx_frame frames[2], *frame;
622 struct inode *dir;
623 int block, err;
624 int count = 0;
625 int ret;
626 __u32 hashval;
627
628 dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
629 start_minor_hash));
Josef "Jeff" Sipek9d549892006-12-08 02:36:38 -0800630 dir = dir_file->f_path.dentry->d_inode;
Mingming Cao617ba132006-10-11 01:20:53 -0700631 if (!(EXT4_I(dir)->i_flags & EXT4_INDEX_FL)) {
632 hinfo.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
633 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700634 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
635 start_hash, start_minor_hash);
636 *next_hash = ~0;
637 return count;
638 }
639 hinfo.hash = start_hash;
640 hinfo.minor_hash = 0;
Josef "Jeff" Sipek9d549892006-12-08 02:36:38 -0800641 frame = dx_probe(NULL, dir_file->f_path.dentry->d_inode, &hinfo, frames, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700642 if (!frame)
643 return err;
644
645 /* Add '.' and '..' from the htree header */
646 if (!start_hash && !start_minor_hash) {
Mingming Cao617ba132006-10-11 01:20:53 -0700647 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
648 if ((err = ext4_htree_store_dirent(dir_file, 0, 0, de)) != 0)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700649 goto errout;
650 count++;
651 }
652 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
Mingming Cao617ba132006-10-11 01:20:53 -0700653 de = (struct ext4_dir_entry_2 *) frames[0].bh->b_data;
654 de = ext4_next_entry(de);
655 if ((err = ext4_htree_store_dirent(dir_file, 2, 0, de)) != 0)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700656 goto errout;
657 count++;
658 }
659
660 while (1) {
661 block = dx_get_block(frame->at);
662 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
663 start_hash, start_minor_hash);
664 if (ret < 0) {
665 err = ret;
666 goto errout;
667 }
668 count += ret;
669 hashval = ~0;
Mingming Cao617ba132006-10-11 01:20:53 -0700670 ret = ext4_htree_next_block(dir, HASH_NB_ALWAYS,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700671 frame, frames, &hashval);
672 *next_hash = hashval;
673 if (ret < 0) {
674 err = ret;
675 goto errout;
676 }
677 /*
678 * Stop if: (a) there are no more entries, or
679 * (b) we have inserted at least one entry and the
680 * next hash value is not a continuation
681 */
682 if ((ret == 0) ||
683 (count && ((hashval & 1) == 0)))
684 break;
685 }
686 dx_release(frames);
687 dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
688 count, *next_hash));
689 return count;
690errout:
691 dx_release(frames);
692 return (err);
693}
694
695
696/*
697 * Directory block splitting, compacting
698 */
699
Eric Sandeenef2b02d2007-09-18 22:46:42 -0700700/*
701 * Create map of hash values, offsets, and sizes, stored at end of block.
702 * Returns number of entries mapped.
703 */
Mingming Cao617ba132006-10-11 01:20:53 -0700704static int dx_make_map (struct ext4_dir_entry_2 *de, int size,
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700705 struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
706{
707 int count = 0;
708 char *base = (char *) de;
709 struct dx_hash_info h = *hinfo;
710
711 while ((char *) de < base + size)
712 {
713 if (de->name_len && de->inode) {
Mingming Cao617ba132006-10-11 01:20:53 -0700714 ext4fs_dirhash(de->name, de->name_len, &h);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700715 map_tail--;
716 map_tail->hash = h.hash;
Eric Sandeenef2b02d2007-09-18 22:46:42 -0700717 map_tail->offs = (u16) ((char *) de - base);
718 map_tail->size = le16_to_cpu(de->rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700719 count++;
720 cond_resched();
721 }
722 /* XXX: do we need to check rec_len == 0 case? -Chris */
Mingming Cao617ba132006-10-11 01:20:53 -0700723 de = (struct ext4_dir_entry_2 *) ((char *) de + le16_to_cpu(de->rec_len));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700724 }
725 return count;
726}
727
Eric Sandeenef2b02d2007-09-18 22:46:42 -0700728/* Sort map by hash value */
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700729static void dx_sort_map (struct dx_map_entry *map, unsigned count)
730{
Andrew Morton63f57932006-10-11 01:21:24 -0700731 struct dx_map_entry *p, *q, *top = map + count - 1;
732 int more;
733 /* Combsort until bubble sort doesn't suck */
734 while (count > 2) {
735 count = count*10/13;
736 if (count - 9 < 2) /* 9, 10 -> 11 */
737 count = 11;
738 for (p = top, q = p - count; q >= map; p--, q--)
739 if (p->hash < q->hash)
740 swap(*p, *q);
741 }
742 /* Garden variety bubble sort */
743 do {
744 more = 0;
745 q = top;
746 while (q-- > map) {
747 if (q[1].hash >= q[0].hash)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700748 continue;
Andrew Morton63f57932006-10-11 01:21:24 -0700749 swap(*(q+1), *q);
750 more = 1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700751 }
752 } while(more);
753}
754
755static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
756{
757 struct dx_entry *entries = frame->entries;
758 struct dx_entry *old = frame->at, *new = old + 1;
759 int count = dx_get_count(entries);
760
761 assert(count < dx_get_limit(entries));
762 assert(old < entries + count);
763 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
764 dx_set_hash(new, hash);
765 dx_set_block(new, block);
766 dx_set_count(entries, count + 1);
767}
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700768
Mingming Cao617ba132006-10-11 01:20:53 -0700769static void ext4_update_dx_flag(struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700770{
Mingming Cao617ba132006-10-11 01:20:53 -0700771 if (!EXT4_HAS_COMPAT_FEATURE(inode->i_sb,
772 EXT4_FEATURE_COMPAT_DIR_INDEX))
773 EXT4_I(inode)->i_flags &= ~EXT4_INDEX_FL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700774}
775
776/*
Mingming Cao617ba132006-10-11 01:20:53 -0700777 * NOTE! unlike strncmp, ext4_match returns 1 for success, 0 for failure.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700778 *
Mingming Cao617ba132006-10-11 01:20:53 -0700779 * `len <= EXT4_NAME_LEN' is guaranteed by caller.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700780 * `de != NULL' is guaranteed by caller.
781 */
Mingming Cao617ba132006-10-11 01:20:53 -0700782static inline int ext4_match (int len, const char * const name,
783 struct ext4_dir_entry_2 * de)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700784{
785 if (len != de->name_len)
786 return 0;
787 if (!de->inode)
788 return 0;
789 return !memcmp(name, de->name, len);
790}
791
792/*
793 * Returns 0 if not found, -1 on failure, and 1 on success
794 */
795static inline int search_dirblock(struct buffer_head * bh,
796 struct inode *dir,
797 struct dentry *dentry,
798 unsigned long offset,
Mingming Cao617ba132006-10-11 01:20:53 -0700799 struct ext4_dir_entry_2 ** res_dir)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700800{
Mingming Cao617ba132006-10-11 01:20:53 -0700801 struct ext4_dir_entry_2 * de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700802 char * dlimit;
803 int de_len;
804 const char *name = dentry->d_name.name;
805 int namelen = dentry->d_name.len;
806
Mingming Cao617ba132006-10-11 01:20:53 -0700807 de = (struct ext4_dir_entry_2 *) bh->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700808 dlimit = bh->b_data + dir->i_sb->s_blocksize;
809 while ((char *) de < dlimit) {
810 /* this code is executed quadratically often */
811 /* do minimal checking `by hand' */
812
813 if ((char *) de + namelen <= dlimit &&
Mingming Cao617ba132006-10-11 01:20:53 -0700814 ext4_match (namelen, name, de)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700815 /* found a match - just to be sure, do a full check */
Mingming Cao617ba132006-10-11 01:20:53 -0700816 if (!ext4_check_dir_entry("ext4_find_entry",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700817 dir, de, bh, offset))
818 return -1;
819 *res_dir = de;
820 return 1;
821 }
822 /* prevent looping on a bad block */
823 de_len = le16_to_cpu(de->rec_len);
824 if (de_len <= 0)
825 return -1;
826 offset += de_len;
Mingming Cao617ba132006-10-11 01:20:53 -0700827 de = (struct ext4_dir_entry_2 *) ((char *) de + de_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700828 }
829 return 0;
830}
831
832
833/*
Mingming Cao617ba132006-10-11 01:20:53 -0700834 * ext4_find_entry()
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700835 *
836 * finds an entry in the specified directory with the wanted name. It
837 * returns the cache buffer in which the entry was found, and the entry
838 * itself (as a parameter - res_dir). It does NOT read the inode of the
839 * entry - you'll have to do that yourself if you want to.
840 *
841 * The returned buffer_head has ->b_count elevated. The caller is expected
842 * to brelse() it when appropriate.
843 */
Mingming Cao617ba132006-10-11 01:20:53 -0700844static struct buffer_head * ext4_find_entry (struct dentry *dentry,
845 struct ext4_dir_entry_2 ** res_dir)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700846{
847 struct super_block * sb;
848 struct buffer_head * bh_use[NAMEI_RA_SIZE];
849 struct buffer_head * bh, *ret = NULL;
850 unsigned long start, block, b;
851 int ra_max = 0; /* Number of bh's in the readahead
852 buffer, bh_use[] */
853 int ra_ptr = 0; /* Current index into readahead
854 buffer */
855 int num = 0;
856 int nblocks, i, err;
857 struct inode *dir = dentry->d_parent->d_inode;
858 int namelen;
859 const u8 *name;
860 unsigned blocksize;
861
862 *res_dir = NULL;
863 sb = dir->i_sb;
864 blocksize = sb->s_blocksize;
865 namelen = dentry->d_name.len;
866 name = dentry->d_name.name;
Mingming Cao617ba132006-10-11 01:20:53 -0700867 if (namelen > EXT4_NAME_LEN)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700868 return NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700869 if (is_dx(dir)) {
Mingming Cao617ba132006-10-11 01:20:53 -0700870 bh = ext4_dx_find_entry(dentry, res_dir, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700871 /*
872 * On success, or if the error was file not found,
873 * return. Otherwise, fall back to doing a search the
874 * old fashioned way.
875 */
876 if (bh || (err != ERR_BAD_DX_DIR))
877 return bh;
Mingming Cao617ba132006-10-11 01:20:53 -0700878 dxtrace(printk("ext4_find_entry: dx failed, falling back\n"));
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700879 }
Mingming Cao617ba132006-10-11 01:20:53 -0700880 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
881 start = EXT4_I(dir)->i_dir_start_lookup;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700882 if (start >= nblocks)
883 start = 0;
884 block = start;
885restart:
886 do {
887 /*
888 * We deal with the read-ahead logic here.
889 */
890 if (ra_ptr >= ra_max) {
891 /* Refill the readahead buffer */
892 ra_ptr = 0;
893 b = block;
894 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
895 /*
896 * Terminate if we reach the end of the
897 * directory and must wrap, or if our
898 * search has finished at this block.
899 */
900 if (b >= nblocks || (num && block == start)) {
901 bh_use[ra_max] = NULL;
902 break;
903 }
904 num++;
Mingming Cao617ba132006-10-11 01:20:53 -0700905 bh = ext4_getblk(NULL, dir, b++, 0, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700906 bh_use[ra_max] = bh;
907 if (bh)
908 ll_rw_block(READ_META, 1, &bh);
909 }
910 }
911 if ((bh = bh_use[ra_ptr++]) == NULL)
912 goto next;
913 wait_on_buffer(bh);
914 if (!buffer_uptodate(bh)) {
915 /* read error, skip block & hope for the best */
Mingming Cao617ba132006-10-11 01:20:53 -0700916 ext4_error(sb, __FUNCTION__, "reading directory #%lu "
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700917 "offset %lu", dir->i_ino, block);
918 brelse(bh);
919 goto next;
920 }
921 i = search_dirblock(bh, dir, dentry,
Mingming Cao617ba132006-10-11 01:20:53 -0700922 block << EXT4_BLOCK_SIZE_BITS(sb), res_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700923 if (i == 1) {
Mingming Cao617ba132006-10-11 01:20:53 -0700924 EXT4_I(dir)->i_dir_start_lookup = block;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700925 ret = bh;
926 goto cleanup_and_exit;
927 } else {
928 brelse(bh);
929 if (i < 0)
930 goto cleanup_and_exit;
931 }
932 next:
933 if (++block >= nblocks)
934 block = 0;
935 } while (block != start);
936
937 /*
938 * If the directory has grown while we were searching, then
939 * search the last part of the directory before giving up.
940 */
941 block = nblocks;
Mingming Cao617ba132006-10-11 01:20:53 -0700942 nblocks = dir->i_size >> EXT4_BLOCK_SIZE_BITS(sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700943 if (block < nblocks) {
944 start = 0;
945 goto restart;
946 }
947
948cleanup_and_exit:
949 /* Clean up the read-ahead blocks */
950 for (; ra_ptr < ra_max; ra_ptr++)
951 brelse (bh_use[ra_ptr]);
952 return ret;
953}
954
Mingming Cao617ba132006-10-11 01:20:53 -0700955static struct buffer_head * ext4_dx_find_entry(struct dentry *dentry,
956 struct ext4_dir_entry_2 **res_dir, int *err)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700957{
958 struct super_block * sb;
959 struct dx_hash_info hinfo;
960 u32 hash;
961 struct dx_frame frames[2], *frame;
Mingming Cao617ba132006-10-11 01:20:53 -0700962 struct ext4_dir_entry_2 *de, *top;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700963 struct buffer_head *bh;
964 unsigned long block;
965 int retval;
966 int namelen = dentry->d_name.len;
967 const u8 *name = dentry->d_name.name;
968 struct inode *dir = dentry->d_parent->d_inode;
969
970 sb = dir->i_sb;
971 /* NFS may look up ".." - look at dx_root directory block */
972 if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
973 if (!(frame = dx_probe(dentry, NULL, &hinfo, frames, err)))
974 return NULL;
975 } else {
976 frame = frames;
977 frame->bh = NULL; /* for dx_release() */
978 frame->at = (struct dx_entry *)frames; /* hack for zero entry*/
979 dx_set_block(frame->at, 0); /* dx_root block is 0 */
980 }
981 hash = hinfo.hash;
982 do {
983 block = dx_get_block(frame->at);
Mingming Cao617ba132006-10-11 01:20:53 -0700984 if (!(bh = ext4_bread (NULL,dir, block, 0, err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700985 goto errout;
Mingming Cao617ba132006-10-11 01:20:53 -0700986 de = (struct ext4_dir_entry_2 *) bh->b_data;
987 top = (struct ext4_dir_entry_2 *) ((char *) de + sb->s_blocksize -
988 EXT4_DIR_REC_LEN(0));
989 for (; de < top; de = ext4_next_entry(de))
990 if (ext4_match (namelen, name, de)) {
991 if (!ext4_check_dir_entry("ext4_find_entry",
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700992 dir, de, bh,
Mingming Cao617ba132006-10-11 01:20:53 -0700993 (block<<EXT4_BLOCK_SIZE_BITS(sb))
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700994 +((char *)de - bh->b_data))) {
995 brelse (bh);
Dmitriy Monakhovfedee542007-05-08 00:25:34 -0700996 *err = ERR_BAD_DX_DIR;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700997 goto errout;
998 }
999 *res_dir = de;
1000 dx_release (frames);
1001 return bh;
1002 }
1003 brelse (bh);
1004 /* Check to see if we should continue to search */
Mingming Cao617ba132006-10-11 01:20:53 -07001005 retval = ext4_htree_next_block(dir, hash, frame,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001006 frames, NULL);
1007 if (retval < 0) {
Mingming Cao617ba132006-10-11 01:20:53 -07001008 ext4_warning(sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001009 "error reading index page in directory #%lu",
1010 dir->i_ino);
1011 *err = retval;
1012 goto errout;
1013 }
1014 } while (retval == 1);
1015
1016 *err = -ENOENT;
1017errout:
1018 dxtrace(printk("%s not found\n", name));
1019 dx_release (frames);
1020 return NULL;
1021}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001022
Mingming Cao617ba132006-10-11 01:20:53 -07001023static struct dentry *ext4_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001024{
1025 struct inode * inode;
Mingming Cao617ba132006-10-11 01:20:53 -07001026 struct ext4_dir_entry_2 * de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001027 struct buffer_head * bh;
1028
Mingming Cao617ba132006-10-11 01:20:53 -07001029 if (dentry->d_name.len > EXT4_NAME_LEN)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001030 return ERR_PTR(-ENAMETOOLONG);
1031
Mingming Cao617ba132006-10-11 01:20:53 -07001032 bh = ext4_find_entry(dentry, &de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001033 inode = NULL;
1034 if (bh) {
1035 unsigned long ino = le32_to_cpu(de->inode);
1036 brelse (bh);
Mingming Cao617ba132006-10-11 01:20:53 -07001037 if (!ext4_valid_inum(dir->i_sb, ino)) {
1038 ext4_error(dir->i_sb, "ext4_lookup",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001039 "bad inode number: %lu", ino);
1040 inode = NULL;
1041 } else
1042 inode = iget(dir->i_sb, ino);
1043
1044 if (!inode)
1045 return ERR_PTR(-EACCES);
Vasily Averina6c15c22007-07-15 23:40:46 -07001046
1047 if (is_bad_inode(inode)) {
1048 iput(inode);
1049 return ERR_PTR(-ENOENT);
1050 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001051 }
1052 return d_splice_alias(inode, dentry);
1053}
1054
1055
Mingming Cao617ba132006-10-11 01:20:53 -07001056struct dentry *ext4_get_parent(struct dentry *child)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001057{
1058 unsigned long ino;
1059 struct dentry *parent;
1060 struct inode *inode;
1061 struct dentry dotdot;
Mingming Cao617ba132006-10-11 01:20:53 -07001062 struct ext4_dir_entry_2 * de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001063 struct buffer_head *bh;
1064
1065 dotdot.d_name.name = "..";
1066 dotdot.d_name.len = 2;
1067 dotdot.d_parent = child; /* confusing, isn't it! */
1068
Mingming Cao617ba132006-10-11 01:20:53 -07001069 bh = ext4_find_entry(&dotdot, &de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001070 inode = NULL;
1071 if (!bh)
1072 return ERR_PTR(-ENOENT);
1073 ino = le32_to_cpu(de->inode);
1074 brelse(bh);
1075
Mingming Cao617ba132006-10-11 01:20:53 -07001076 if (!ext4_valid_inum(child->d_inode->i_sb, ino)) {
1077 ext4_error(child->d_inode->i_sb, "ext4_get_parent",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001078 "bad inode number: %lu", ino);
1079 inode = NULL;
1080 } else
1081 inode = iget(child->d_inode->i_sb, ino);
1082
1083 if (!inode)
1084 return ERR_PTR(-EACCES);
1085
Vasily Averina6c15c22007-07-15 23:40:46 -07001086 if (is_bad_inode(inode)) {
1087 iput(inode);
1088 return ERR_PTR(-ENOENT);
1089 }
1090
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001091 parent = d_alloc_anon(inode);
1092 if (!parent) {
1093 iput(inode);
1094 parent = ERR_PTR(-ENOMEM);
1095 }
1096 return parent;
1097}
1098
1099#define S_SHIFT 12
Mingming Cao617ba132006-10-11 01:20:53 -07001100static unsigned char ext4_type_by_mode[S_IFMT >> S_SHIFT] = {
1101 [S_IFREG >> S_SHIFT] = EXT4_FT_REG_FILE,
1102 [S_IFDIR >> S_SHIFT] = EXT4_FT_DIR,
1103 [S_IFCHR >> S_SHIFT] = EXT4_FT_CHRDEV,
1104 [S_IFBLK >> S_SHIFT] = EXT4_FT_BLKDEV,
1105 [S_IFIFO >> S_SHIFT] = EXT4_FT_FIFO,
1106 [S_IFSOCK >> S_SHIFT] = EXT4_FT_SOCK,
1107 [S_IFLNK >> S_SHIFT] = EXT4_FT_SYMLINK,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001108};
1109
Mingming Cao617ba132006-10-11 01:20:53 -07001110static inline void ext4_set_de_type(struct super_block *sb,
1111 struct ext4_dir_entry_2 *de,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001112 umode_t mode) {
Mingming Cao617ba132006-10-11 01:20:53 -07001113 if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FILETYPE))
1114 de->file_type = ext4_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001115}
1116
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001117/*
1118 * Move count entries from end of map between two memory locations.
1119 * Returns pointer to last entry moved.
1120 */
Mingming Cao617ba132006-10-11 01:20:53 -07001121static struct ext4_dir_entry_2 *
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001122dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1123{
1124 unsigned rec_len = 0;
1125
1126 while (count--) {
Mingming Cao617ba132006-10-11 01:20:53 -07001127 struct ext4_dir_entry_2 *de = (struct ext4_dir_entry_2 *) (from + map->offs);
1128 rec_len = EXT4_DIR_REC_LEN(de->name_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001129 memcpy (to, de, rec_len);
Mingming Cao617ba132006-10-11 01:20:53 -07001130 ((struct ext4_dir_entry_2 *) to)->rec_len =
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001131 cpu_to_le16(rec_len);
1132 de->inode = 0;
1133 map++;
1134 to += rec_len;
1135 }
Mingming Cao617ba132006-10-11 01:20:53 -07001136 return (struct ext4_dir_entry_2 *) (to - rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001137}
1138
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001139/*
1140 * Compact each dir entry in the range to the minimal rec_len.
1141 * Returns pointer to last entry in range.
1142 */
Mingming Cao617ba132006-10-11 01:20:53 -07001143static struct ext4_dir_entry_2* dx_pack_dirents(char *base, int size)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001144{
Mingming Cao617ba132006-10-11 01:20:53 -07001145 struct ext4_dir_entry_2 *next, *to, *prev, *de = (struct ext4_dir_entry_2 *) base;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001146 unsigned rec_len = 0;
1147
1148 prev = to = de;
1149 while ((char*)de < base + size) {
Mingming Cao617ba132006-10-11 01:20:53 -07001150 next = (struct ext4_dir_entry_2 *) ((char *) de +
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001151 le16_to_cpu(de->rec_len));
1152 if (de->inode && de->name_len) {
Mingming Cao617ba132006-10-11 01:20:53 -07001153 rec_len = EXT4_DIR_REC_LEN(de->name_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001154 if (de > to)
1155 memmove(to, de, rec_len);
1156 to->rec_len = cpu_to_le16(rec_len);
1157 prev = to;
Mingming Cao617ba132006-10-11 01:20:53 -07001158 to = (struct ext4_dir_entry_2 *) (((char *) to) + rec_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001159 }
1160 de = next;
1161 }
1162 return prev;
1163}
1164
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001165/*
1166 * Split a full leaf block to make room for a new dir entry.
1167 * Allocate a new block, and move entries so that they are approx. equally full.
1168 * Returns pointer to de in block into which the new entry will be inserted.
1169 */
Mingming Cao617ba132006-10-11 01:20:53 -07001170static struct ext4_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001171 struct buffer_head **bh,struct dx_frame *frame,
1172 struct dx_hash_info *hinfo, int *error)
1173{
1174 unsigned blocksize = dir->i_sb->s_blocksize;
1175 unsigned count, continued;
1176 struct buffer_head *bh2;
1177 u32 newblock;
1178 u32 hash2;
1179 struct dx_map_entry *map;
1180 char *data1 = (*bh)->b_data, *data2;
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001181 unsigned split, move, size, i;
Mingming Cao617ba132006-10-11 01:20:53 -07001182 struct ext4_dir_entry_2 *de = NULL, *de2;
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001183 int err = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001184
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001185 bh2 = ext4_append (handle, dir, &newblock, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001186 if (!(bh2)) {
1187 brelse(*bh);
1188 *bh = NULL;
1189 goto errout;
1190 }
1191
1192 BUFFER_TRACE(*bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001193 err = ext4_journal_get_write_access(handle, *bh);
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001194 if (err)
1195 goto journal_error;
1196
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001197 BUFFER_TRACE(frame->bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001198 err = ext4_journal_get_write_access(handle, frame->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001199 if (err)
1200 goto journal_error;
1201
1202 data2 = bh2->b_data;
1203
1204 /* create map in the end of data2 block */
1205 map = (struct dx_map_entry *) (data2 + blocksize);
Mingming Cao617ba132006-10-11 01:20:53 -07001206 count = dx_make_map ((struct ext4_dir_entry_2 *) data1,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001207 blocksize, hinfo, map);
1208 map -= count;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001209 dx_sort_map (map, count);
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001210 /* Split the existing block in the middle, size-wise */
1211 size = 0;
1212 move = 0;
1213 for (i = count-1; i >= 0; i--) {
1214 /* is more than half of this entry in 2nd half of the block? */
1215 if (size + map[i].size/2 > blocksize/2)
1216 break;
1217 size += map[i].size;
1218 move++;
1219 }
1220 /* map index at which we will split */
1221 split = count - move;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001222 hash2 = map[split].hash;
1223 continued = hash2 == map[split - 1].hash;
1224 dxtrace(printk("Split block %i at %x, %i/%i\n",
1225 dx_get_block(frame->at), hash2, split, count-split));
1226
1227 /* Fancy dance to stay within two buffers */
1228 de2 = dx_move_dirents(data1, data2, map + split, count - split);
1229 de = dx_pack_dirents(data1,blocksize);
1230 de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1231 de2->rec_len = cpu_to_le16(data2 + blocksize - (char *) de2);
Mingming Cao617ba132006-10-11 01:20:53 -07001232 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data1, blocksize, 1));
1233 dxtrace(dx_show_leaf (hinfo, (struct ext4_dir_entry_2 *) data2, blocksize, 1));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001234
1235 /* Which block gets the new entry? */
1236 if (hinfo->hash >= hash2)
1237 {
1238 swap(*bh, bh2);
1239 de = de2;
1240 }
1241 dx_insert_block (frame, hash2 + continued, newblock);
Mingming Cao617ba132006-10-11 01:20:53 -07001242 err = ext4_journal_dirty_metadata (handle, bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001243 if (err)
1244 goto journal_error;
Mingming Cao617ba132006-10-11 01:20:53 -07001245 err = ext4_journal_dirty_metadata (handle, frame->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001246 if (err)
1247 goto journal_error;
1248 brelse (bh2);
1249 dxtrace(dx_show_index ("frame", frame->entries));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001250 return de;
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001251
1252journal_error:
1253 brelse(*bh);
1254 brelse(bh2);
1255 *bh = NULL;
1256 ext4_std_error(dir->i_sb, err);
1257errout:
1258 *error = err;
1259 return NULL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001260}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001261
1262/*
1263 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1264 * it points to a directory entry which is guaranteed to be large
1265 * enough for new directory entry. If de is NULL, then
1266 * add_dirent_to_buf will attempt search the directory block for
1267 * space. It will return -ENOSPC if no space is available, and -EIO
1268 * and -EEXIST if directory entry already exists.
1269 *
1270 * NOTE! bh is NOT released in the case where ENOSPC is returned. In
1271 * all other cases bh is released.
1272 */
1273static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
Mingming Cao617ba132006-10-11 01:20:53 -07001274 struct inode *inode, struct ext4_dir_entry_2 *de,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001275 struct buffer_head * bh)
1276{
1277 struct inode *dir = dentry->d_parent->d_inode;
1278 const char *name = dentry->d_name.name;
1279 int namelen = dentry->d_name.len;
1280 unsigned long offset = 0;
1281 unsigned short reclen;
1282 int nlen, rlen, err;
1283 char *top;
1284
Mingming Cao617ba132006-10-11 01:20:53 -07001285 reclen = EXT4_DIR_REC_LEN(namelen);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001286 if (!de) {
Mingming Cao617ba132006-10-11 01:20:53 -07001287 de = (struct ext4_dir_entry_2 *)bh->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001288 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1289 while ((char *) de <= top) {
Mingming Cao617ba132006-10-11 01:20:53 -07001290 if (!ext4_check_dir_entry("ext4_add_entry", dir, de,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001291 bh, offset)) {
1292 brelse (bh);
1293 return -EIO;
1294 }
Mingming Cao617ba132006-10-11 01:20:53 -07001295 if (ext4_match (namelen, name, de)) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001296 brelse (bh);
1297 return -EEXIST;
1298 }
Mingming Cao617ba132006-10-11 01:20:53 -07001299 nlen = EXT4_DIR_REC_LEN(de->name_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001300 rlen = le16_to_cpu(de->rec_len);
1301 if ((de->inode? rlen - nlen: rlen) >= reclen)
1302 break;
Mingming Cao617ba132006-10-11 01:20:53 -07001303 de = (struct ext4_dir_entry_2 *)((char *)de + rlen);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001304 offset += rlen;
1305 }
1306 if ((char *) de > top)
1307 return -ENOSPC;
1308 }
1309 BUFFER_TRACE(bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001310 err = ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001311 if (err) {
Mingming Cao617ba132006-10-11 01:20:53 -07001312 ext4_std_error(dir->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001313 brelse(bh);
1314 return err;
1315 }
1316
1317 /* By now the buffer is marked for journaling */
Mingming Cao617ba132006-10-11 01:20:53 -07001318 nlen = EXT4_DIR_REC_LEN(de->name_len);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001319 rlen = le16_to_cpu(de->rec_len);
1320 if (de->inode) {
Mingming Cao617ba132006-10-11 01:20:53 -07001321 struct ext4_dir_entry_2 *de1 = (struct ext4_dir_entry_2 *)((char *)de + nlen);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001322 de1->rec_len = cpu_to_le16(rlen - nlen);
1323 de->rec_len = cpu_to_le16(nlen);
1324 de = de1;
1325 }
Mingming Cao617ba132006-10-11 01:20:53 -07001326 de->file_type = EXT4_FT_UNKNOWN;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001327 if (inode) {
1328 de->inode = cpu_to_le32(inode->i_ino);
Mingming Cao617ba132006-10-11 01:20:53 -07001329 ext4_set_de_type(dir->i_sb, de, inode->i_mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001330 } else
1331 de->inode = 0;
1332 de->name_len = namelen;
1333 memcpy (de->name, name, namelen);
1334 /*
1335 * XXX shouldn't update any times until successful
1336 * completion of syscall, but too many callers depend
1337 * on this.
1338 *
1339 * XXX similarly, too many callers depend on
Mingming Cao617ba132006-10-11 01:20:53 -07001340 * ext4_new_inode() setting the times, but error
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001341 * recovery deletes the inode, so the worst that can
1342 * happen is that the times are slightly out of date
1343 * and/or different from the directory change time.
1344 */
Kalpak Shahef7f3832007-07-18 09:15:20 -04001345 dir->i_mtime = dir->i_ctime = ext4_current_time(dir);
Mingming Cao617ba132006-10-11 01:20:53 -07001346 ext4_update_dx_flag(dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001347 dir->i_version++;
Mingming Cao617ba132006-10-11 01:20:53 -07001348 ext4_mark_inode_dirty(handle, dir);
1349 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1350 err = ext4_journal_dirty_metadata(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001351 if (err)
Mingming Cao617ba132006-10-11 01:20:53 -07001352 ext4_std_error(dir->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001353 brelse(bh);
1354 return 0;
1355}
1356
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001357/*
1358 * This converts a one block unindexed directory to a 3 block indexed
1359 * directory, and adds the dentry to the indexed directory.
1360 */
1361static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1362 struct inode *inode, struct buffer_head *bh)
1363{
1364 struct inode *dir = dentry->d_parent->d_inode;
1365 const char *name = dentry->d_name.name;
1366 int namelen = dentry->d_name.len;
1367 struct buffer_head *bh2;
1368 struct dx_root *root;
1369 struct dx_frame frames[2], *frame;
1370 struct dx_entry *entries;
Mingming Cao617ba132006-10-11 01:20:53 -07001371 struct ext4_dir_entry_2 *de, *de2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001372 char *data1, *top;
1373 unsigned len;
1374 int retval;
1375 unsigned blocksize;
1376 struct dx_hash_info hinfo;
1377 u32 block;
1378 struct fake_dirent *fde;
1379
1380 blocksize = dir->i_sb->s_blocksize;
1381 dxtrace(printk("Creating index\n"));
Mingming Cao617ba132006-10-11 01:20:53 -07001382 retval = ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001383 if (retval) {
Mingming Cao617ba132006-10-11 01:20:53 -07001384 ext4_std_error(dir->i_sb, retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001385 brelse(bh);
1386 return retval;
1387 }
1388 root = (struct dx_root *) bh->b_data;
1389
Mingming Cao617ba132006-10-11 01:20:53 -07001390 bh2 = ext4_append (handle, dir, &block, &retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001391 if (!(bh2)) {
1392 brelse(bh);
1393 return retval;
1394 }
Mingming Cao617ba132006-10-11 01:20:53 -07001395 EXT4_I(dir)->i_flags |= EXT4_INDEX_FL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001396 data1 = bh2->b_data;
1397
1398 /* The 0th block becomes the root, move the dirents out */
1399 fde = &root->dotdot;
Mingming Cao617ba132006-10-11 01:20:53 -07001400 de = (struct ext4_dir_entry_2 *)((char *)fde + le16_to_cpu(fde->rec_len));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001401 len = ((char *) root) + blocksize - (char *) de;
1402 memcpy (data1, de, len);
Mingming Cao617ba132006-10-11 01:20:53 -07001403 de = (struct ext4_dir_entry_2 *) data1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001404 top = data1 + len;
1405 while ((char *)(de2=(void*)de+le16_to_cpu(de->rec_len)) < top)
1406 de = de2;
1407 de->rec_len = cpu_to_le16(data1 + blocksize - (char *) de);
1408 /* Initialize the root; the dot dirents already exist */
Mingming Cao617ba132006-10-11 01:20:53 -07001409 de = (struct ext4_dir_entry_2 *) (&root->dotdot);
1410 de->rec_len = cpu_to_le16(blocksize - EXT4_DIR_REC_LEN(2));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001411 memset (&root->info, 0, sizeof(root->info));
1412 root->info.info_length = sizeof(root->info);
Mingming Cao617ba132006-10-11 01:20:53 -07001413 root->info.hash_version = EXT4_SB(dir->i_sb)->s_def_hash_version;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001414 entries = root->entries;
1415 dx_set_block (entries, 1);
1416 dx_set_count (entries, 1);
1417 dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1418
1419 /* Initialize as for dx_probe */
1420 hinfo.hash_version = root->info.hash_version;
Mingming Cao617ba132006-10-11 01:20:53 -07001421 hinfo.seed = EXT4_SB(dir->i_sb)->s_hash_seed;
1422 ext4fs_dirhash(name, namelen, &hinfo);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001423 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}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001435
1436/*
Mingming Cao617ba132006-10-11 01:20:53 -07001437 * ext4_add_entry()
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001438 *
1439 * adds a file entry to the specified directory, using the same
Mingming Cao617ba132006-10-11 01:20:53 -07001440 * semantics as ext4_find_entry(). It returns NULL if it failed.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001441 *
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 */
Mingming Cao617ba132006-10-11 01:20:53 -07001446static int ext4_add_entry (handle_t *handle, struct dentry *dentry,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001447 struct inode *inode)
1448{
1449 struct inode *dir = dentry->d_parent->d_inode;
1450 unsigned long offset;
1451 struct buffer_head * bh;
Mingming Cao617ba132006-10-11 01:20:53 -07001452 struct ext4_dir_entry_2 *de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001453 struct super_block * sb;
1454 int retval;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001455 int dx_fallback=0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001456 unsigned blocksize;
1457 u32 block, blocks;
1458
1459 sb = dir->i_sb;
1460 blocksize = sb->s_blocksize;
1461 if (!dentry->d_name.len)
1462 return -EINVAL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001463 if (is_dx(dir)) {
Mingming Cao617ba132006-10-11 01:20:53 -07001464 retval = ext4_dx_add_entry(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001465 if (!retval || (retval != ERR_BAD_DX_DIR))
1466 return retval;
Mingming Cao617ba132006-10-11 01:20:53 -07001467 EXT4_I(dir)->i_flags &= ~EXT4_INDEX_FL;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001468 dx_fallback++;
Mingming Cao617ba132006-10-11 01:20:53 -07001469 ext4_mark_inode_dirty(handle, dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001470 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001471 blocks = dir->i_size >> sb->s_blocksize_bits;
1472 for (block = 0, offset = 0; block < blocks; block++) {
Mingming Cao617ba132006-10-11 01:20:53 -07001473 bh = ext4_bread(handle, dir, block, 0, &retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001474 if(!bh)
1475 return retval;
1476 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1477 if (retval != -ENOSPC)
1478 return retval;
1479
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001480 if (blocks == 1 && !dx_fallback &&
Mingming Cao617ba132006-10-11 01:20:53 -07001481 EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001482 return make_indexed_dir(handle, dentry, inode, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001483 brelse(bh);
1484 }
Mingming Cao617ba132006-10-11 01:20:53 -07001485 bh = ext4_append(handle, dir, &block, &retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001486 if (!bh)
1487 return retval;
Mingming Cao617ba132006-10-11 01:20:53 -07001488 de = (struct ext4_dir_entry_2 *) bh->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001489 de->inode = 0;
1490 de->rec_len = cpu_to_le16(blocksize);
1491 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1492}
1493
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001494/*
1495 * Returns 0 for success, or a negative error value
1496 */
Mingming Cao617ba132006-10-11 01:20:53 -07001497static int ext4_dx_add_entry(handle_t *handle, struct dentry *dentry,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001498 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;
Mingming Cao617ba132006-10-11 01:20:53 -07001506 struct ext4_dir_entry_2 *de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001507 int err;
1508
1509 frame = dx_probe(dentry, NULL, &hinfo, frames, &err);
1510 if (!frame)
1511 return err;
1512 entries = frame->entries;
1513 at = frame->at;
1514
Mingming Cao617ba132006-10-11 01:20:53 -07001515 if (!(bh = ext4_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001516 goto cleanup;
1517
1518 BUFFER_TRACE(bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001519 err = ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001520 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))) {
Mingming Cao617ba132006-10-11 01:20:53 -07001543 ext4_warning(sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001544 "Directory index full!");
1545 err = -ENOSPC;
1546 goto cleanup;
1547 }
Mingming Cao617ba132006-10-11 01:20:53 -07001548 bh2 = ext4_append (handle, dir, &newblock, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001549 if (!(bh2))
1550 goto cleanup;
1551 node2 = (struct dx_node *)(bh2->b_data);
1552 entries2 = node2->entries;
1553 node2->fake.rec_len = cpu_to_le16(sb->s_blocksize);
1554 node2->fake.inode = 0;
1555 BUFFER_TRACE(frame->bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001556 err = ext4_journal_get_write_access(handle, frame->bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001557 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 */
Mingming Cao617ba132006-10-11 01:20:53 -07001565 err = ext4_journal_get_write_access(handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001566 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));
Mingming Cao617ba132006-10-11 01:20:53 -07001586 err = ext4_journal_dirty_metadata(handle, bh2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001587 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;
Mingming Cao617ba132006-10-11 01:20:53 -07001606 err = ext4_journal_get_write_access(handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001607 frame->bh);
1608 if (err)
1609 goto journal_error;
1610 }
Mingming Cao617ba132006-10-11 01:20:53 -07001611 ext4_journal_dirty_metadata(handle, frames[0].bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001612 }
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:
Mingming Cao617ba132006-10-11 01:20:53 -07001621 ext4_std_error(dir->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001622cleanup:
1623 if (bh)
1624 brelse(bh);
1625 dx_release(frames);
1626 return err;
1627}
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001628
1629/*
Mingming Cao617ba132006-10-11 01:20:53 -07001630 * ext4_delete_entry deletes a directory entry by merging it with the
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001631 * previous entry
1632 */
Mingming Cao617ba132006-10-11 01:20:53 -07001633static int ext4_delete_entry (handle_t *handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001634 struct inode * dir,
Mingming Cao617ba132006-10-11 01:20:53 -07001635 struct ext4_dir_entry_2 * de_del,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001636 struct buffer_head * bh)
1637{
Mingming Cao617ba132006-10-11 01:20:53 -07001638 struct ext4_dir_entry_2 * de, * pde;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001639 int i;
1640
1641 i = 0;
1642 pde = NULL;
Mingming Cao617ba132006-10-11 01:20:53 -07001643 de = (struct ext4_dir_entry_2 *) bh->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001644 while (i < bh->b_size) {
Mingming Cao617ba132006-10-11 01:20:53 -07001645 if (!ext4_check_dir_entry("ext4_delete_entry", dir, de, bh, i))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001646 return -EIO;
1647 if (de == de_del) {
1648 BUFFER_TRACE(bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001649 ext4_journal_get_write_access(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001650 if (pde)
1651 pde->rec_len =
1652 cpu_to_le16(le16_to_cpu(pde->rec_len) +
1653 le16_to_cpu(de->rec_len));
1654 else
1655 de->inode = 0;
1656 dir->i_version++;
Mingming Cao617ba132006-10-11 01:20:53 -07001657 BUFFER_TRACE(bh, "call ext4_journal_dirty_metadata");
1658 ext4_journal_dirty_metadata(handle, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001659 return 0;
1660 }
1661 i += le16_to_cpu(de->rec_len);
1662 pde = de;
Mingming Cao617ba132006-10-11 01:20:53 -07001663 de = (struct ext4_dir_entry_2 *)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001664 ((char *) de + le16_to_cpu(de->rec_len));
1665 }
1666 return -ENOENT;
1667}
1668
Andreas Dilgerf8628a12007-07-18 08:38:01 -04001669/*
1670 * DIR_NLINK feature is set if 1) nlinks > EXT4_LINK_MAX or 2) nlinks == 2,
1671 * since this indicates that nlinks count was previously 1.
1672 */
1673static void ext4_inc_count(handle_t *handle, struct inode *inode)
1674{
1675 inc_nlink(inode);
1676 if (is_dx(inode) && inode->i_nlink > 1) {
1677 /* limit is 16-bit i_links_count */
1678 if (inode->i_nlink >= EXT4_LINK_MAX || inode->i_nlink == 2) {
1679 inode->i_nlink = 1;
1680 EXT4_SET_RO_COMPAT_FEATURE(inode->i_sb,
1681 EXT4_FEATURE_RO_COMPAT_DIR_NLINK);
1682 }
1683 }
1684}
1685
1686/*
1687 * If a directory had nlink == 1, then we should let it be 1. This indicates
1688 * directory has >EXT4_LINK_MAX subdirs.
1689 */
1690static void ext4_dec_count(handle_t *handle, struct inode *inode)
1691{
1692 drop_nlink(inode);
1693 if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
1694 inc_nlink(inode);
1695}
1696
1697
Mingming Cao617ba132006-10-11 01:20:53 -07001698static int ext4_add_nondir(handle_t *handle,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001699 struct dentry *dentry, struct inode *inode)
1700{
Mingming Cao617ba132006-10-11 01:20:53 -07001701 int err = ext4_add_entry(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001702 if (!err) {
Mingming Cao617ba132006-10-11 01:20:53 -07001703 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001704 d_instantiate(dentry, inode);
1705 return 0;
1706 }
Eric Sandeen731b9a52007-02-10 01:46:16 -08001707 drop_nlink(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001708 iput(inode);
1709 return err;
1710}
1711
1712/*
1713 * By the time this is called, we already have created
1714 * the directory cache entry for the new file, but it
1715 * is so far negative - it has no inode.
1716 *
1717 * If the create succeeds, we fill in the inode information
1718 * with d_instantiate().
1719 */
Mingming Cao617ba132006-10-11 01:20:53 -07001720static int ext4_create (struct inode * dir, struct dentry * dentry, int mode,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001721 struct nameidata *nd)
1722{
1723 handle_t *handle;
1724 struct inode * inode;
1725 int err, retries = 0;
1726
1727retry:
Mingming Cao617ba132006-10-11 01:20:53 -07001728 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1729 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1730 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001731 if (IS_ERR(handle))
1732 return PTR_ERR(handle);
1733
1734 if (IS_DIRSYNC(dir))
1735 handle->h_sync = 1;
1736
Mingming Cao617ba132006-10-11 01:20:53 -07001737 inode = ext4_new_inode (handle, dir, mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001738 err = PTR_ERR(inode);
1739 if (!IS_ERR(inode)) {
Mingming Cao617ba132006-10-11 01:20:53 -07001740 inode->i_op = &ext4_file_inode_operations;
1741 inode->i_fop = &ext4_file_operations;
1742 ext4_set_aops(inode);
1743 err = ext4_add_nondir(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001744 }
Mingming Cao617ba132006-10-11 01:20:53 -07001745 ext4_journal_stop(handle);
1746 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001747 goto retry;
1748 return err;
1749}
1750
Mingming Cao617ba132006-10-11 01:20:53 -07001751static int ext4_mknod (struct inode * dir, struct dentry *dentry,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001752 int mode, dev_t rdev)
1753{
1754 handle_t *handle;
1755 struct inode *inode;
1756 int err, retries = 0;
1757
1758 if (!new_valid_dev(rdev))
1759 return -EINVAL;
1760
1761retry:
Mingming Cao617ba132006-10-11 01:20:53 -07001762 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1763 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1764 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001765 if (IS_ERR(handle))
1766 return PTR_ERR(handle);
1767
1768 if (IS_DIRSYNC(dir))
1769 handle->h_sync = 1;
1770
Mingming Cao617ba132006-10-11 01:20:53 -07001771 inode = ext4_new_inode (handle, dir, mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001772 err = PTR_ERR(inode);
1773 if (!IS_ERR(inode)) {
1774 init_special_inode(inode, inode->i_mode, rdev);
Mingming Cao617ba132006-10-11 01:20:53 -07001775#ifdef CONFIG_EXT4DEV_FS_XATTR
1776 inode->i_op = &ext4_special_inode_operations;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001777#endif
Mingming Cao617ba132006-10-11 01:20:53 -07001778 err = ext4_add_nondir(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001779 }
Mingming Cao617ba132006-10-11 01:20:53 -07001780 ext4_journal_stop(handle);
1781 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001782 goto retry;
1783 return err;
1784}
1785
Mingming Cao617ba132006-10-11 01:20:53 -07001786static int ext4_mkdir(struct inode * dir, struct dentry * dentry, int mode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001787{
1788 handle_t *handle;
1789 struct inode * inode;
1790 struct buffer_head * dir_block;
Mingming Cao617ba132006-10-11 01:20:53 -07001791 struct ext4_dir_entry_2 * de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001792 int err, retries = 0;
1793
Andreas Dilgerf8628a12007-07-18 08:38:01 -04001794 if (EXT4_DIR_LINK_MAX(dir))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001795 return -EMLINK;
1796
1797retry:
Mingming Cao617ba132006-10-11 01:20:53 -07001798 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
1799 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 3 +
1800 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001801 if (IS_ERR(handle))
1802 return PTR_ERR(handle);
1803
1804 if (IS_DIRSYNC(dir))
1805 handle->h_sync = 1;
1806
Mingming Cao617ba132006-10-11 01:20:53 -07001807 inode = ext4_new_inode (handle, dir, S_IFDIR | mode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001808 err = PTR_ERR(inode);
1809 if (IS_ERR(inode))
1810 goto out_stop;
1811
Mingming Cao617ba132006-10-11 01:20:53 -07001812 inode->i_op = &ext4_dir_inode_operations;
1813 inode->i_fop = &ext4_dir_operations;
1814 inode->i_size = EXT4_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1815 dir_block = ext4_bread (handle, inode, 0, 1, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001816 if (!dir_block) {
Andreas Dilgerf8628a12007-07-18 08:38:01 -04001817 ext4_dec_count(handle, inode); /* is this nlink == 0? */
Mingming Cao617ba132006-10-11 01:20:53 -07001818 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001819 iput (inode);
1820 goto out_stop;
1821 }
1822 BUFFER_TRACE(dir_block, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07001823 ext4_journal_get_write_access(handle, dir_block);
1824 de = (struct ext4_dir_entry_2 *) dir_block->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001825 de->inode = cpu_to_le32(inode->i_ino);
1826 de->name_len = 1;
Mingming Cao617ba132006-10-11 01:20:53 -07001827 de->rec_len = cpu_to_le16(EXT4_DIR_REC_LEN(de->name_len));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001828 strcpy (de->name, ".");
Mingming Cao617ba132006-10-11 01:20:53 -07001829 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
1830 de = (struct ext4_dir_entry_2 *)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001831 ((char *) de + le16_to_cpu(de->rec_len));
1832 de->inode = cpu_to_le32(dir->i_ino);
Mingming Cao617ba132006-10-11 01:20:53 -07001833 de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize-EXT4_DIR_REC_LEN(1));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001834 de->name_len = 2;
1835 strcpy (de->name, "..");
Mingming Cao617ba132006-10-11 01:20:53 -07001836 ext4_set_de_type(dir->i_sb, de, S_IFDIR);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001837 inode->i_nlink = 2;
Mingming Cao617ba132006-10-11 01:20:53 -07001838 BUFFER_TRACE(dir_block, "call ext4_journal_dirty_metadata");
1839 ext4_journal_dirty_metadata(handle, dir_block);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001840 brelse (dir_block);
Mingming Cao617ba132006-10-11 01:20:53 -07001841 ext4_mark_inode_dirty(handle, inode);
1842 err = ext4_add_entry (handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001843 if (err) {
1844 inode->i_nlink = 0;
Mingming Cao617ba132006-10-11 01:20:53 -07001845 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001846 iput (inode);
1847 goto out_stop;
1848 }
Andreas Dilgerf8628a12007-07-18 08:38:01 -04001849 ext4_inc_count(handle, dir);
Mingming Cao617ba132006-10-11 01:20:53 -07001850 ext4_update_dx_flag(dir);
1851 ext4_mark_inode_dirty(handle, dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001852 d_instantiate(dentry, inode);
1853out_stop:
Mingming Cao617ba132006-10-11 01:20:53 -07001854 ext4_journal_stop(handle);
1855 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001856 goto retry;
1857 return err;
1858}
1859
1860/*
1861 * routine to check that the specified directory is empty (for rmdir)
1862 */
1863static int empty_dir (struct inode * inode)
1864{
1865 unsigned long offset;
1866 struct buffer_head * bh;
Mingming Cao617ba132006-10-11 01:20:53 -07001867 struct ext4_dir_entry_2 * de, * de1;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001868 struct super_block * sb;
1869 int err = 0;
1870
1871 sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -07001872 if (inode->i_size < EXT4_DIR_REC_LEN(1) + EXT4_DIR_REC_LEN(2) ||
1873 !(bh = ext4_bread (NULL, inode, 0, 0, &err))) {
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001874 if (err)
Mingming Cao617ba132006-10-11 01:20:53 -07001875 ext4_error(inode->i_sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001876 "error %d reading directory #%lu offset 0",
1877 err, inode->i_ino);
1878 else
Mingming Cao617ba132006-10-11 01:20:53 -07001879 ext4_warning(inode->i_sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001880 "bad directory (dir #%lu) - no data block",
1881 inode->i_ino);
1882 return 1;
1883 }
Mingming Cao617ba132006-10-11 01:20:53 -07001884 de = (struct ext4_dir_entry_2 *) bh->b_data;
1885 de1 = (struct ext4_dir_entry_2 *)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001886 ((char *) de + le16_to_cpu(de->rec_len));
1887 if (le32_to_cpu(de->inode) != inode->i_ino ||
1888 !le32_to_cpu(de1->inode) ||
1889 strcmp (".", de->name) ||
1890 strcmp ("..", de1->name)) {
Mingming Cao617ba132006-10-11 01:20:53 -07001891 ext4_warning (inode->i_sb, "empty_dir",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001892 "bad directory (dir #%lu) - no `.' or `..'",
1893 inode->i_ino);
1894 brelse (bh);
1895 return 1;
1896 }
1897 offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
Mingming Cao617ba132006-10-11 01:20:53 -07001898 de = (struct ext4_dir_entry_2 *)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001899 ((char *) de1 + le16_to_cpu(de1->rec_len));
1900 while (offset < inode->i_size ) {
1901 if (!bh ||
1902 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1903 err = 0;
1904 brelse (bh);
Mingming Cao617ba132006-10-11 01:20:53 -07001905 bh = ext4_bread (NULL, inode,
1906 offset >> EXT4_BLOCK_SIZE_BITS(sb), 0, &err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001907 if (!bh) {
1908 if (err)
Mingming Cao617ba132006-10-11 01:20:53 -07001909 ext4_error(sb, __FUNCTION__,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001910 "error %d reading directory"
1911 " #%lu offset %lu",
1912 err, inode->i_ino, offset);
1913 offset += sb->s_blocksize;
1914 continue;
1915 }
Mingming Cao617ba132006-10-11 01:20:53 -07001916 de = (struct ext4_dir_entry_2 *) bh->b_data;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001917 }
Mingming Cao617ba132006-10-11 01:20:53 -07001918 if (!ext4_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1919 de = (struct ext4_dir_entry_2 *)(bh->b_data +
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001920 sb->s_blocksize);
1921 offset = (offset | (sb->s_blocksize - 1)) + 1;
1922 continue;
1923 }
1924 if (le32_to_cpu(de->inode)) {
1925 brelse (bh);
1926 return 0;
1927 }
1928 offset += le16_to_cpu(de->rec_len);
Mingming Cao617ba132006-10-11 01:20:53 -07001929 de = (struct ext4_dir_entry_2 *)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001930 ((char *) de + le16_to_cpu(de->rec_len));
1931 }
1932 brelse (bh);
1933 return 1;
1934}
1935
Mingming Cao617ba132006-10-11 01:20:53 -07001936/* ext4_orphan_add() links an unlinked or truncated inode into a list of
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001937 * such inodes, starting at the superblock, in case we crash before the
1938 * file is closed/deleted, or in case the inode truncate spans multiple
1939 * transactions and the last transaction is not recovered after a crash.
1940 *
1941 * At filesystem recovery time, we walk this list deleting unlinked
Mingming Cao617ba132006-10-11 01:20:53 -07001942 * inodes and truncating linked inodes in ext4_orphan_cleanup().
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001943 */
Mingming Cao617ba132006-10-11 01:20:53 -07001944int ext4_orphan_add(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001945{
1946 struct super_block *sb = inode->i_sb;
Mingming Cao617ba132006-10-11 01:20:53 -07001947 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001948 int err = 0, rc;
1949
1950 lock_super(sb);
Mingming Cao617ba132006-10-11 01:20:53 -07001951 if (!list_empty(&EXT4_I(inode)->i_orphan))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001952 goto out_unlock;
1953
1954 /* Orphan handling is only valid for files with data blocks
1955 * being truncated, or files being unlinked. */
1956
1957 /* @@@ FIXME: Observation from aviro:
Mingming Cao617ba132006-10-11 01:20:53 -07001958 * I think I can trigger J_ASSERT in ext4_orphan_add(). We block
1959 * here (on lock_super()), so race with ext4_link() which might bump
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001960 * ->i_nlink. For, say it, character device. Not a regular file,
1961 * not a directory, not a symlink and ->i_nlink > 0.
1962 */
1963 J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1964 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1965
Mingming Cao617ba132006-10-11 01:20:53 -07001966 BUFFER_TRACE(EXT4_SB(sb)->s_sbh, "get_write_access");
1967 err = ext4_journal_get_write_access(handle, EXT4_SB(sb)->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001968 if (err)
1969 goto out_unlock;
1970
Mingming Cao617ba132006-10-11 01:20:53 -07001971 err = ext4_reserve_inode_write(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001972 if (err)
1973 goto out_unlock;
1974
1975 /* Insert this inode at the head of the on-disk orphan list... */
Mingming Cao617ba132006-10-11 01:20:53 -07001976 NEXT_ORPHAN(inode) = le32_to_cpu(EXT4_SB(sb)->s_es->s_last_orphan);
1977 EXT4_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1978 err = ext4_journal_dirty_metadata(handle, EXT4_SB(sb)->s_sbh);
1979 rc = ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001980 if (!err)
1981 err = rc;
1982
1983 /* Only add to the head of the in-memory list if all the
1984 * previous operations succeeded. If the orphan_add is going to
1985 * fail (possibly taking the journal offline), we can't risk
1986 * leaving the inode on the orphan list: stray orphan-list
1987 * entries can cause panics at unmount time.
1988 *
1989 * This is safe: on error we're going to ignore the orphan list
1990 * anyway on the next recovery. */
1991 if (!err)
Mingming Cao617ba132006-10-11 01:20:53 -07001992 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001993
1994 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1995 jbd_debug(4, "orphan inode %lu will point to %d\n",
1996 inode->i_ino, NEXT_ORPHAN(inode));
1997out_unlock:
1998 unlock_super(sb);
Mingming Cao617ba132006-10-11 01:20:53 -07001999 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002000 return err;
2001}
2002
2003/*
Mingming Cao617ba132006-10-11 01:20:53 -07002004 * ext4_orphan_del() removes an unlinked or truncated inode from the list
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002005 * of such inodes stored on disk, because it is finally being cleaned up.
2006 */
Mingming Cao617ba132006-10-11 01:20:53 -07002007int ext4_orphan_del(handle_t *handle, struct inode *inode)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002008{
2009 struct list_head *prev;
Mingming Cao617ba132006-10-11 01:20:53 -07002010 struct ext4_inode_info *ei = EXT4_I(inode);
2011 struct ext4_sb_info *sbi;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002012 unsigned long ino_next;
Mingming Cao617ba132006-10-11 01:20:53 -07002013 struct ext4_iloc iloc;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002014 int err = 0;
2015
2016 lock_super(inode->i_sb);
2017 if (list_empty(&ei->i_orphan)) {
2018 unlock_super(inode->i_sb);
2019 return 0;
2020 }
2021
2022 ino_next = NEXT_ORPHAN(inode);
2023 prev = ei->i_orphan.prev;
Mingming Cao617ba132006-10-11 01:20:53 -07002024 sbi = EXT4_SB(inode->i_sb);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002025
2026 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
2027
2028 list_del_init(&ei->i_orphan);
2029
2030 /* If we're on an error path, we may not have a valid
2031 * transaction handle with which to update the orphan list on
2032 * disk, but we still need to remove the inode from the linked
2033 * list in memory. */
2034 if (!handle)
2035 goto out;
2036
Mingming Cao617ba132006-10-11 01:20:53 -07002037 err = ext4_reserve_inode_write(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002038 if (err)
2039 goto out_err;
2040
2041 if (prev == &sbi->s_orphan) {
2042 jbd_debug(4, "superblock will point to %lu\n", ino_next);
2043 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07002044 err = ext4_journal_get_write_access(handle, sbi->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002045 if (err)
2046 goto out_brelse;
2047 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
Mingming Cao617ba132006-10-11 01:20:53 -07002048 err = ext4_journal_dirty_metadata(handle, sbi->s_sbh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002049 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07002050 struct ext4_iloc iloc2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002051 struct inode *i_prev =
Mingming Cao617ba132006-10-11 01:20:53 -07002052 &list_entry(prev, struct ext4_inode_info, i_orphan)->vfs_inode;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002053
2054 jbd_debug(4, "orphan inode %lu will point to %lu\n",
2055 i_prev->i_ino, ino_next);
Mingming Cao617ba132006-10-11 01:20:53 -07002056 err = ext4_reserve_inode_write(handle, i_prev, &iloc2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002057 if (err)
2058 goto out_brelse;
2059 NEXT_ORPHAN(i_prev) = ino_next;
Mingming Cao617ba132006-10-11 01:20:53 -07002060 err = ext4_mark_iloc_dirty(handle, i_prev, &iloc2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002061 }
2062 if (err)
2063 goto out_brelse;
2064 NEXT_ORPHAN(inode) = 0;
Mingming Cao617ba132006-10-11 01:20:53 -07002065 err = ext4_mark_iloc_dirty(handle, inode, &iloc);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002066
2067out_err:
Mingming Cao617ba132006-10-11 01:20:53 -07002068 ext4_std_error(inode->i_sb, err);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002069out:
2070 unlock_super(inode->i_sb);
2071 return err;
2072
2073out_brelse:
2074 brelse(iloc.bh);
2075 goto out_err;
2076}
2077
Mingming Cao617ba132006-10-11 01:20:53 -07002078static int ext4_rmdir (struct inode * dir, struct dentry *dentry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002079{
2080 int retval;
2081 struct inode * inode;
2082 struct buffer_head * bh;
Mingming Cao617ba132006-10-11 01:20:53 -07002083 struct ext4_dir_entry_2 * de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002084 handle_t *handle;
2085
2086 /* Initialize quotas before so that eventual writes go in
2087 * separate transaction */
2088 DQUOT_INIT(dentry->d_inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002089 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002090 if (IS_ERR(handle))
2091 return PTR_ERR(handle);
2092
2093 retval = -ENOENT;
Mingming Cao617ba132006-10-11 01:20:53 -07002094 bh = ext4_find_entry (dentry, &de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002095 if (!bh)
2096 goto end_rmdir;
2097
2098 if (IS_DIRSYNC(dir))
2099 handle->h_sync = 1;
2100
2101 inode = dentry->d_inode;
2102
2103 retval = -EIO;
2104 if (le32_to_cpu(de->inode) != inode->i_ino)
2105 goto end_rmdir;
2106
2107 retval = -ENOTEMPTY;
2108 if (!empty_dir (inode))
2109 goto end_rmdir;
2110
Mingming Cao617ba132006-10-11 01:20:53 -07002111 retval = ext4_delete_entry(handle, dir, de, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002112 if (retval)
2113 goto end_rmdir;
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002114 if (!EXT4_DIR_LINK_EMPTY(inode))
Mingming Cao617ba132006-10-11 01:20:53 -07002115 ext4_warning (inode->i_sb, "ext4_rmdir",
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002116 "empty directory has too many links (%d)",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002117 inode->i_nlink);
2118 inode->i_version++;
2119 clear_nlink(inode);
2120 /* There's no need to set i_disksize: the fact that i_nlink is
2121 * zero will ensure that the right thing happens during any
2122 * recovery. */
2123 inode->i_size = 0;
Mingming Cao617ba132006-10-11 01:20:53 -07002124 ext4_orphan_add(handle, inode);
Kalpak Shahef7f3832007-07-18 09:15:20 -04002125 inode->i_ctime = dir->i_ctime = dir->i_mtime = ext4_current_time(inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002126 ext4_mark_inode_dirty(handle, inode);
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002127 ext4_dec_count(handle, dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002128 ext4_update_dx_flag(dir);
2129 ext4_mark_inode_dirty(handle, dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002130
2131end_rmdir:
Mingming Cao617ba132006-10-11 01:20:53 -07002132 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002133 brelse (bh);
2134 return retval;
2135}
2136
Mingming Cao617ba132006-10-11 01:20:53 -07002137static int ext4_unlink(struct inode * dir, struct dentry *dentry)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002138{
2139 int retval;
2140 struct inode * inode;
2141 struct buffer_head * bh;
Mingming Cao617ba132006-10-11 01:20:53 -07002142 struct ext4_dir_entry_2 * de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002143 handle_t *handle;
2144
2145 /* Initialize quotas before so that eventual writes go
2146 * in separate transaction */
2147 DQUOT_INIT(dentry->d_inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002148 handle = ext4_journal_start(dir, EXT4_DELETE_TRANS_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002149 if (IS_ERR(handle))
2150 return PTR_ERR(handle);
2151
2152 if (IS_DIRSYNC(dir))
2153 handle->h_sync = 1;
2154
2155 retval = -ENOENT;
Mingming Cao617ba132006-10-11 01:20:53 -07002156 bh = ext4_find_entry (dentry, &de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002157 if (!bh)
2158 goto end_unlink;
2159
2160 inode = dentry->d_inode;
2161
2162 retval = -EIO;
2163 if (le32_to_cpu(de->inode) != inode->i_ino)
2164 goto end_unlink;
2165
2166 if (!inode->i_nlink) {
Mingming Cao617ba132006-10-11 01:20:53 -07002167 ext4_warning (inode->i_sb, "ext4_unlink",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002168 "Deleting nonexistent file (%lu), %d",
2169 inode->i_ino, inode->i_nlink);
2170 inode->i_nlink = 1;
2171 }
Mingming Cao617ba132006-10-11 01:20:53 -07002172 retval = ext4_delete_entry(handle, dir, de, bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002173 if (retval)
2174 goto end_unlink;
Kalpak Shahef7f3832007-07-18 09:15:20 -04002175 dir->i_ctime = dir->i_mtime = ext4_current_time(dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002176 ext4_update_dx_flag(dir);
2177 ext4_mark_inode_dirty(handle, dir);
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002178 ext4_dec_count(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002179 if (!inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07002180 ext4_orphan_add(handle, inode);
Kalpak Shahef7f3832007-07-18 09:15:20 -04002181 inode->i_ctime = ext4_current_time(inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002182 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002183 retval = 0;
2184
2185end_unlink:
Mingming Cao617ba132006-10-11 01:20:53 -07002186 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002187 brelse (bh);
2188 return retval;
2189}
2190
Mingming Cao617ba132006-10-11 01:20:53 -07002191static int ext4_symlink (struct inode * dir,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002192 struct dentry *dentry, const char * symname)
2193{
2194 handle_t *handle;
2195 struct inode * inode;
2196 int l, err, retries = 0;
2197
2198 l = strlen(symname)+1;
2199 if (l > dir->i_sb->s_blocksize)
2200 return -ENAMETOOLONG;
2201
2202retry:
Mingming Cao617ba132006-10-11 01:20:53 -07002203 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2204 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 5 +
2205 2*EXT4_QUOTA_INIT_BLOCKS(dir->i_sb));
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002206 if (IS_ERR(handle))
2207 return PTR_ERR(handle);
2208
2209 if (IS_DIRSYNC(dir))
2210 handle->h_sync = 1;
2211
Mingming Cao617ba132006-10-11 01:20:53 -07002212 inode = ext4_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002213 err = PTR_ERR(inode);
2214 if (IS_ERR(inode))
2215 goto out_stop;
2216
Mingming Cao617ba132006-10-11 01:20:53 -07002217 if (l > sizeof (EXT4_I(inode)->i_data)) {
2218 inode->i_op = &ext4_symlink_inode_operations;
2219 ext4_set_aops(inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002220 /*
Mingming Cao617ba132006-10-11 01:20:53 -07002221 * page_symlink() calls into ext4_prepare/commit_write.
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002222 * We have a transaction open. All is sweetness. It also sets
2223 * i_size in generic_commit_write().
2224 */
2225 err = __page_symlink(inode, symname, l,
2226 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
2227 if (err) {
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002228 ext4_dec_count(handle, inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002229 ext4_mark_inode_dirty(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002230 iput (inode);
2231 goto out_stop;
2232 }
2233 } else {
Mingming Cao617ba132006-10-11 01:20:53 -07002234 inode->i_op = &ext4_fast_symlink_inode_operations;
2235 memcpy((char*)&EXT4_I(inode)->i_data,symname,l);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002236 inode->i_size = l-1;
2237 }
Mingming Cao617ba132006-10-11 01:20:53 -07002238 EXT4_I(inode)->i_disksize = inode->i_size;
2239 err = ext4_add_nondir(handle, dentry, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002240out_stop:
Mingming Cao617ba132006-10-11 01:20:53 -07002241 ext4_journal_stop(handle);
2242 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002243 goto retry;
2244 return err;
2245}
2246
Mingming Cao617ba132006-10-11 01:20:53 -07002247static int ext4_link (struct dentry * old_dentry,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002248 struct inode * dir, struct dentry *dentry)
2249{
2250 handle_t *handle;
2251 struct inode *inode = old_dentry->d_inode;
2252 int err, retries = 0;
2253
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002254 if (EXT4_DIR_LINK_MAX(inode))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002255 return -EMLINK;
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002256
Eric Sandeen2988a772007-02-10 01:46:16 -08002257 /*
2258 * Return -ENOENT if we've raced with unlink and i_nlink is 0. Doing
2259 * otherwise has the potential to corrupt the orphan inode list.
2260 */
2261 if (inode->i_nlink == 0)
2262 return -ENOENT;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002263
2264retry:
Mingming Cao617ba132006-10-11 01:20:53 -07002265 handle = ext4_journal_start(dir, EXT4_DATA_TRANS_BLOCKS(dir->i_sb) +
2266 EXT4_INDEX_EXTRA_TRANS_BLOCKS);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002267 if (IS_ERR(handle))
2268 return PTR_ERR(handle);
2269
2270 if (IS_DIRSYNC(dir))
2271 handle->h_sync = 1;
2272
Kalpak Shahef7f3832007-07-18 09:15:20 -04002273 inode->i_ctime = ext4_current_time(inode);
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002274 ext4_inc_count(handle, inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002275 atomic_inc(&inode->i_count);
2276
Mingming Cao617ba132006-10-11 01:20:53 -07002277 err = ext4_add_nondir(handle, dentry, inode);
2278 ext4_journal_stop(handle);
2279 if (err == -ENOSPC && ext4_should_retry_alloc(dir->i_sb, &retries))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002280 goto retry;
2281 return err;
2282}
2283
2284#define PARENT_INO(buffer) \
Mingming Cao617ba132006-10-11 01:20:53 -07002285 ((struct ext4_dir_entry_2 *) ((char *) buffer + \
2286 le16_to_cpu(((struct ext4_dir_entry_2 *) buffer)->rec_len)))->inode
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002287
2288/*
2289 * Anybody can rename anything with this: the permission checks are left to the
2290 * higher-level routines.
2291 */
Mingming Cao617ba132006-10-11 01:20:53 -07002292static int ext4_rename (struct inode * old_dir, struct dentry *old_dentry,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002293 struct inode * new_dir,struct dentry *new_dentry)
2294{
2295 handle_t *handle;
2296 struct inode * old_inode, * new_inode;
2297 struct buffer_head * old_bh, * new_bh, * dir_bh;
Mingming Cao617ba132006-10-11 01:20:53 -07002298 struct ext4_dir_entry_2 * old_de, * new_de;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002299 int retval;
2300
2301 old_bh = new_bh = dir_bh = NULL;
2302
2303 /* Initialize quotas before so that eventual writes go
2304 * in separate transaction */
2305 if (new_dentry->d_inode)
2306 DQUOT_INIT(new_dentry->d_inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002307 handle = ext4_journal_start(old_dir, 2 *
2308 EXT4_DATA_TRANS_BLOCKS(old_dir->i_sb) +
2309 EXT4_INDEX_EXTRA_TRANS_BLOCKS + 2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002310 if (IS_ERR(handle))
2311 return PTR_ERR(handle);
2312
2313 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2314 handle->h_sync = 1;
2315
Mingming Cao617ba132006-10-11 01:20:53 -07002316 old_bh = ext4_find_entry (old_dentry, &old_de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002317 /*
2318 * Check for inode number is _not_ due to possible IO errors.
2319 * We might rmdir the source, keep it as pwd of some process
2320 * and merrily kill the link to whatever was created under the
2321 * same name. Goodbye sticky bit ;-<
2322 */
2323 old_inode = old_dentry->d_inode;
2324 retval = -ENOENT;
2325 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2326 goto end_rename;
2327
2328 new_inode = new_dentry->d_inode;
Mingming Cao617ba132006-10-11 01:20:53 -07002329 new_bh = ext4_find_entry (new_dentry, &new_de);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002330 if (new_bh) {
2331 if (!new_inode) {
2332 brelse (new_bh);
2333 new_bh = NULL;
2334 }
2335 }
2336 if (S_ISDIR(old_inode->i_mode)) {
2337 if (new_inode) {
2338 retval = -ENOTEMPTY;
2339 if (!empty_dir (new_inode))
2340 goto end_rename;
2341 }
2342 retval = -EIO;
Mingming Cao617ba132006-10-11 01:20:53 -07002343 dir_bh = ext4_bread (handle, old_inode, 0, 0, &retval);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002344 if (!dir_bh)
2345 goto end_rename;
2346 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2347 goto end_rename;
2348 retval = -EMLINK;
2349 if (!new_inode && new_dir!=old_dir &&
Mingming Cao617ba132006-10-11 01:20:53 -07002350 new_dir->i_nlink >= EXT4_LINK_MAX)
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002351 goto end_rename;
2352 }
2353 if (!new_bh) {
Mingming Cao617ba132006-10-11 01:20:53 -07002354 retval = ext4_add_entry (handle, new_dentry, old_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002355 if (retval)
2356 goto end_rename;
2357 } else {
2358 BUFFER_TRACE(new_bh, "get write access");
Mingming Cao617ba132006-10-11 01:20:53 -07002359 ext4_journal_get_write_access(handle, new_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002360 new_de->inode = cpu_to_le32(old_inode->i_ino);
Mingming Cao617ba132006-10-11 01:20:53 -07002361 if (EXT4_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2362 EXT4_FEATURE_INCOMPAT_FILETYPE))
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002363 new_de->file_type = old_de->file_type;
2364 new_dir->i_version++;
Mingming Cao617ba132006-10-11 01:20:53 -07002365 BUFFER_TRACE(new_bh, "call ext4_journal_dirty_metadata");
2366 ext4_journal_dirty_metadata(handle, new_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002367 brelse(new_bh);
2368 new_bh = NULL;
2369 }
2370
2371 /*
2372 * Like most other Unix systems, set the ctime for inodes on a
2373 * rename.
2374 */
Kalpak Shahef7f3832007-07-18 09:15:20 -04002375 old_inode->i_ctime = ext4_current_time(old_inode);
Mingming Cao617ba132006-10-11 01:20:53 -07002376 ext4_mark_inode_dirty(handle, old_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002377
2378 /*
2379 * ok, that's it
2380 */
2381 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2382 old_de->name_len != old_dentry->d_name.len ||
2383 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
Mingming Cao617ba132006-10-11 01:20:53 -07002384 (retval = ext4_delete_entry(handle, old_dir,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002385 old_de, old_bh)) == -ENOENT) {
2386 /* old_de could have moved from under us during htree split, so
2387 * make sure that we are deleting the right entry. We might
2388 * also be pointing to a stale entry in the unused part of
2389 * old_bh so just checking inum and the name isn't enough. */
2390 struct buffer_head *old_bh2;
Mingming Cao617ba132006-10-11 01:20:53 -07002391 struct ext4_dir_entry_2 *old_de2;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002392
Mingming Cao617ba132006-10-11 01:20:53 -07002393 old_bh2 = ext4_find_entry(old_dentry, &old_de2);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002394 if (old_bh2) {
Mingming Cao617ba132006-10-11 01:20:53 -07002395 retval = ext4_delete_entry(handle, old_dir,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002396 old_de2, old_bh2);
2397 brelse(old_bh2);
2398 }
2399 }
2400 if (retval) {
Mingming Cao617ba132006-10-11 01:20:53 -07002401 ext4_warning(old_dir->i_sb, "ext4_rename",
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002402 "Deleting old file (%lu), %d, error=%d",
2403 old_dir->i_ino, old_dir->i_nlink, retval);
2404 }
2405
2406 if (new_inode) {
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002407 ext4_dec_count(handle, new_inode);
Kalpak Shahef7f3832007-07-18 09:15:20 -04002408 new_inode->i_ctime = ext4_current_time(new_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002409 }
Kalpak Shahef7f3832007-07-18 09:15:20 -04002410 old_dir->i_ctime = old_dir->i_mtime = ext4_current_time(old_dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002411 ext4_update_dx_flag(old_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002412 if (dir_bh) {
2413 BUFFER_TRACE(dir_bh, "get_write_access");
Mingming Cao617ba132006-10-11 01:20:53 -07002414 ext4_journal_get_write_access(handle, dir_bh);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002415 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
Mingming Cao617ba132006-10-11 01:20:53 -07002416 BUFFER_TRACE(dir_bh, "call ext4_journal_dirty_metadata");
2417 ext4_journal_dirty_metadata(handle, dir_bh);
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002418 ext4_dec_count(handle, old_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002419 if (new_inode) {
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002420 /* checked empty_dir above, can't have another parent,
2421 * ext3_dec_count() won't work for many-linked dirs */
2422 new_inode->i_nlink = 0;
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002423 } else {
Andreas Dilgerf8628a12007-07-18 08:38:01 -04002424 ext4_inc_count(handle, new_dir);
Mingming Cao617ba132006-10-11 01:20:53 -07002425 ext4_update_dx_flag(new_dir);
2426 ext4_mark_inode_dirty(handle, new_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002427 }
2428 }
Mingming Cao617ba132006-10-11 01:20:53 -07002429 ext4_mark_inode_dirty(handle, old_dir);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002430 if (new_inode) {
Mingming Cao617ba132006-10-11 01:20:53 -07002431 ext4_mark_inode_dirty(handle, new_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002432 if (!new_inode->i_nlink)
Mingming Cao617ba132006-10-11 01:20:53 -07002433 ext4_orphan_add(handle, new_inode);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002434 }
2435 retval = 0;
2436
2437end_rename:
2438 brelse (dir_bh);
2439 brelse (old_bh);
2440 brelse (new_bh);
Mingming Cao617ba132006-10-11 01:20:53 -07002441 ext4_journal_stop(handle);
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002442 return retval;
2443}
2444
2445/*
2446 * directories can handle most operations...
2447 */
Arjan van de Ven754661f2007-02-12 00:55:38 -08002448const struct inode_operations ext4_dir_inode_operations = {
Mingming Cao617ba132006-10-11 01:20:53 -07002449 .create = ext4_create,
2450 .lookup = ext4_lookup,
2451 .link = ext4_link,
2452 .unlink = ext4_unlink,
2453 .symlink = ext4_symlink,
2454 .mkdir = ext4_mkdir,
2455 .rmdir = ext4_rmdir,
2456 .mknod = ext4_mknod,
2457 .rename = ext4_rename,
2458 .setattr = ext4_setattr,
2459#ifdef CONFIG_EXT4DEV_FS_XATTR
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002460 .setxattr = generic_setxattr,
2461 .getxattr = generic_getxattr,
Mingming Cao617ba132006-10-11 01:20:53 -07002462 .listxattr = ext4_listxattr,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002463 .removexattr = generic_removexattr,
2464#endif
Mingming Cao617ba132006-10-11 01:20:53 -07002465 .permission = ext4_permission,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002466};
2467
Arjan van de Ven754661f2007-02-12 00:55:38 -08002468const struct inode_operations ext4_special_inode_operations = {
Mingming Cao617ba132006-10-11 01:20:53 -07002469 .setattr = ext4_setattr,
2470#ifdef CONFIG_EXT4DEV_FS_XATTR
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002471 .setxattr = generic_setxattr,
2472 .getxattr = generic_getxattr,
Mingming Cao617ba132006-10-11 01:20:53 -07002473 .listxattr = ext4_listxattr,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002474 .removexattr = generic_removexattr,
2475#endif
Mingming Cao617ba132006-10-11 01:20:53 -07002476 .permission = ext4_permission,
Dave Kleikampac27a0e2006-10-11 01:20:50 -07002477};