blob: 0b8cf80154f1a5c912a1b2d6cae18ccebea591c1 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/ext3/namei.c
3 *
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
8 *
9 * from
10 *
11 * linux/fs/minix/namei.c
12 *
13 * Copyright (C) 1991, 1992 Linus Torvalds
14 *
15 * Big-endian to little-endian byte-swapping/bitmaps by
16 * David S. Miller (davem@caip.rutgers.edu), 1995
17 * Directory entry file type support and forward compatibility hooks
Dave Kleikampe9ad5622006-09-27 01:49:35 -070018 * for B-tree directories by Theodore Ts'o (tytso@mit.edu), 1998
Linus Torvalds1da177e2005-04-16 15:20:36 -070019 * Hash Tree Directory indexing (c)
Dave Kleikampe9ad5622006-09-27 01:49:35 -070020 * Daniel Phillips, 2001
Linus Torvalds1da177e2005-04-16 15:20:36 -070021 * Hash Tree Directory indexing porting
Dave Kleikampe9ad5622006-09-27 01:49:35 -070022 * Christopher Li, 2002
Linus Torvalds1da177e2005-04-16 15:20:36 -070023 * Hash Tree Directory indexing cleanup
Dave Kleikampe9ad5622006-09-27 01:49:35 -070024 * Theodore Ts'o, 2002
Linus Torvalds1da177e2005-04-16 15:20:36 -070025 */
26
27#include <linux/fs.h>
28#include <linux/pagemap.h>
29#include <linux/jbd.h>
30#include <linux/time.h>
31#include <linux/ext3_fs.h>
32#include <linux/ext3_jbd.h>
33#include <linux/fcntl.h>
34#include <linux/stat.h>
35#include <linux/string.h>
36#include <linux/quotaops.h>
37#include <linux/buffer_head.h>
Jens Axboecaa38fb2006-07-23 01:41:26 +020038#include <linux/bio.h>
Ben Dooks381be252005-10-30 15:02:56 -080039
40#include "namei.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include "xattr.h"
42#include "acl.h"
43
44/*
45 * define how far ahead to read directories while searching them.
46 */
47#define NAMEI_RA_CHUNKS 2
48#define NAMEI_RA_BLOCKS 4
49#define NAMEI_RA_SIZE (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
50#define NAMEI_RA_INDEX(c,b) (((c) * NAMEI_RA_BLOCKS) + (b))
51
52static struct buffer_head *ext3_append(handle_t *handle,
53 struct inode *inode,
54 u32 *block, int *err)
55{
56 struct buffer_head *bh;
57
58 *block = inode->i_size >> inode->i_sb->s_blocksize_bits;
59
Akinobu Mita33575f82008-04-28 02:16:08 -070060 bh = ext3_bread(handle, inode, *block, 1, err);
61 if (bh) {
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 inode->i_size += inode->i_sb->s_blocksize;
63 EXT3_I(inode)->i_disksize = inode->i_size;
Akinobu Mita33575f82008-04-28 02:16:08 -070064 *err = ext3_journal_get_write_access(handle, bh);
65 if (*err) {
66 brelse(bh);
67 bh = NULL;
68 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 }
70 return bh;
71}
72
73#ifndef assert
74#define assert(test) J_ASSERT(test)
75#endif
76
77#ifndef swap
78#define swap(x, y) do { typeof(x) z = x; x = y; y = z; } while (0)
79#endif
80
81#ifdef DX_DEBUG
82#define dxtrace(command) command
83#else
Mingming Caoae6ddcc2006-09-27 01:49:27 -070084#define dxtrace(command)
Linus Torvalds1da177e2005-04-16 15:20:36 -070085#endif
86
87struct fake_dirent
88{
89 __le32 inode;
90 __le16 rec_len;
91 u8 name_len;
92 u8 file_type;
93};
94
95struct dx_countlimit
96{
97 __le16 limit;
98 __le16 count;
99};
100
101struct dx_entry
102{
103 __le32 hash;
104 __le32 block;
105};
106
107/*
108 * dx_root_info is laid out so that if it should somehow get overlaid by a
109 * dirent the two low bits of the hash version will be zero. Therefore, the
110 * hash version mod 4 should never be 0. Sincerely, the paranoia department.
111 */
112
113struct dx_root
114{
115 struct fake_dirent dot;
116 char dot_name[4];
117 struct fake_dirent dotdot;
118 char dotdot_name[4];
119 struct dx_root_info
120 {
121 __le32 reserved_zero;
122 u8 hash_version;
123 u8 info_length; /* 8 */
124 u8 indirect_levels;
125 u8 unused_flags;
126 }
127 info;
128 struct dx_entry entries[0];
129};
130
131struct dx_node
132{
133 struct fake_dirent fake;
134 struct dx_entry entries[0];
135};
136
137
138struct dx_frame
139{
140 struct buffer_head *bh;
141 struct dx_entry *entries;
142 struct dx_entry *at;
143};
144
145struct dx_map_entry
146{
147 u32 hash;
Eric Sandeenef2b02d2007-09-18 22:46:42 -0700148 u16 offs;
149 u16 size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150};
151
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152static inline unsigned dx_get_block (struct dx_entry *entry);
153static void dx_set_block (struct dx_entry *entry, unsigned value);
154static inline unsigned dx_get_hash (struct dx_entry *entry);
155static void dx_set_hash (struct dx_entry *entry, unsigned value);
156static unsigned dx_get_count (struct dx_entry *entries);
157static unsigned dx_get_limit (struct dx_entry *entries);
158static void dx_set_count (struct dx_entry *entries, unsigned value);
159static void dx_set_limit (struct dx_entry *entries, unsigned value);
160static unsigned dx_root_limit (struct inode *dir, unsigned infosize);
161static unsigned dx_node_limit (struct inode *dir);
162static struct dx_frame *dx_probe(struct dentry *dentry,
163 struct inode *dir,
164 struct dx_hash_info *hinfo,
165 struct dx_frame *frame,
166 int *err);
167static void dx_release (struct dx_frame *frames);
168static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
169 struct dx_hash_info *hinfo, struct dx_map_entry map[]);
170static void dx_sort_map(struct dx_map_entry *map, unsigned count);
171static struct ext3_dir_entry_2 *dx_move_dirents (char *from, char *to,
172 struct dx_map_entry *offsets, int count);
173static struct ext3_dir_entry_2* dx_pack_dirents (char *base, int size);
174static void dx_insert_block (struct dx_frame *frame, u32 hash, u32 block);
175static int ext3_htree_next_block(struct inode *dir, __u32 hash,
176 struct dx_frame *frame,
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700177 struct dx_frame *frames,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 __u32 *start_hash);
179static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
180 struct ext3_dir_entry_2 **res_dir, int *err);
181static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
182 struct inode *inode);
183
184/*
Jan Kara7c06a8d2007-11-14 17:00:19 -0800185 * p is at least 6 bytes before the end of page
186 */
187static inline struct ext3_dir_entry_2 *
188ext3_next_entry(struct ext3_dir_entry_2 *p)
189{
190 return (struct ext3_dir_entry_2 *)((char *)p +
191 ext3_rec_len_from_disk(p->rec_len));
192}
193
194/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 * Future: use high four bits of block for coalesce-on-delete flags
196 * Mask them off for now.
197 */
198
199static inline unsigned dx_get_block (struct dx_entry *entry)
200{
201 return le32_to_cpu(entry->block) & 0x00ffffff;
202}
203
204static inline void dx_set_block (struct dx_entry *entry, unsigned value)
205{
206 entry->block = cpu_to_le32(value);
207}
208
209static inline unsigned dx_get_hash (struct dx_entry *entry)
210{
211 return le32_to_cpu(entry->hash);
212}
213
214static inline void dx_set_hash (struct dx_entry *entry, unsigned value)
215{
216 entry->hash = cpu_to_le32(value);
217}
218
219static inline unsigned dx_get_count (struct dx_entry *entries)
220{
221 return le16_to_cpu(((struct dx_countlimit *) entries)->count);
222}
223
224static inline unsigned dx_get_limit (struct dx_entry *entries)
225{
226 return le16_to_cpu(((struct dx_countlimit *) entries)->limit);
227}
228
229static inline void dx_set_count (struct dx_entry *entries, unsigned value)
230{
231 ((struct dx_countlimit *) entries)->count = cpu_to_le16(value);
232}
233
234static inline void dx_set_limit (struct dx_entry *entries, unsigned value)
235{
236 ((struct dx_countlimit *) entries)->limit = cpu_to_le16(value);
237}
238
239static inline unsigned dx_root_limit (struct inode *dir, unsigned infosize)
240{
241 unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(1) -
242 EXT3_DIR_REC_LEN(2) - infosize;
243 return 0? 20: entry_space / sizeof(struct dx_entry);
244}
245
246static inline unsigned dx_node_limit (struct inode *dir)
247{
248 unsigned entry_space = dir->i_sb->s_blocksize - EXT3_DIR_REC_LEN(0);
249 return 0? 22: entry_space / sizeof(struct dx_entry);
250}
251
252/*
253 * Debug
254 */
255#ifdef DX_DEBUG
256static void dx_show_index (char * label, struct dx_entry *entries)
257{
258 int i, n = dx_get_count (entries);
259 printk("%s index ", label);
260 for (i = 0; i < n; i++)
261 {
262 printk("%x->%u ", i? dx_get_hash(entries + i): 0, dx_get_block(entries + i));
263 }
264 printk("\n");
265}
266
267struct stats
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700268{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 unsigned names;
270 unsigned space;
271 unsigned bcount;
272};
273
274static struct stats dx_show_leaf(struct dx_hash_info *hinfo, struct ext3_dir_entry_2 *de,
275 int size, int show_names)
276{
277 unsigned names = 0, space = 0;
278 char *base = (char *) de;
279 struct dx_hash_info h = *hinfo;
280
281 printk("names: ");
282 while ((char *) de < base + size)
283 {
284 if (de->inode)
285 {
286 if (show_names)
287 {
288 int len = de->name_len;
289 char *name = de->name;
290 while (len--) printk("%c", *name++);
291 ext3fs_dirhash(de->name, de->name_len, &h);
292 printk(":%x.%u ", h.hash,
293 ((char *) de - base));
294 }
295 space += EXT3_DIR_REC_LEN(de->name_len);
Dave Kleikampe9ad5622006-09-27 01:49:35 -0700296 names++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 }
Jan Kara7c06a8d2007-11-14 17:00:19 -0800298 de = ext3_next_entry(de);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 }
300 printk("(%i)\n", names);
301 return (struct stats) { names, space, 1 };
302}
303
304struct stats dx_show_entries(struct dx_hash_info *hinfo, struct inode *dir,
305 struct dx_entry *entries, int levels)
306{
307 unsigned blocksize = dir->i_sb->s_blocksize;
308 unsigned count = dx_get_count (entries), names = 0, space = 0, i;
309 unsigned bcount = 0;
310 struct buffer_head *bh;
311 int err;
312 printk("%i indexed blocks...\n", count);
313 for (i = 0; i < count; i++, entries++)
314 {
315 u32 block = dx_get_block(entries), hash = i? dx_get_hash(entries): 0;
316 u32 range = i < count - 1? (dx_get_hash(entries + 1) - hash): ~hash;
317 struct stats stats;
318 printk("%s%3u:%03u hash %8x/%8x ",levels?"":" ", i, block, hash, range);
319 if (!(bh = ext3_bread (NULL,dir, block, 0,&err))) continue;
320 stats = levels?
321 dx_show_entries(hinfo, dir, ((struct dx_node *) bh->b_data)->entries, levels - 1):
322 dx_show_leaf(hinfo, (struct ext3_dir_entry_2 *) bh->b_data, blocksize, 0);
323 names += stats.names;
324 space += stats.space;
325 bcount += stats.bcount;
326 brelse (bh);
327 }
328 if (bcount)
329 printk("%snames %u, fullness %u (%u%%)\n", levels?"":" ",
330 names, space/bcount,(space/bcount)*100/blocksize);
331 return (struct stats) { names, space, bcount};
332}
333#endif /* DX_DEBUG */
334
335/*
336 * Probe for a directory leaf block to search.
337 *
338 * dx_probe can return ERR_BAD_DX_DIR, which means there was a format
339 * error in the directory index, and the caller should fall back to
340 * searching the directory normally. The callers of dx_probe **MUST**
341 * check for this error code, and make sure it never gets reflected
342 * back to userspace.
343 */
344static struct dx_frame *
345dx_probe(struct dentry *dentry, struct inode *dir,
346 struct dx_hash_info *hinfo, struct dx_frame *frame_in, int *err)
347{
348 unsigned count, indirect;
349 struct dx_entry *at, *entries, *p, *q, *m;
350 struct dx_root *root;
351 struct buffer_head *bh;
352 struct dx_frame *frame = frame_in;
353 u32 hash;
354
355 frame->bh = NULL;
356 if (dentry)
357 dir = dentry->d_parent->d_inode;
358 if (!(bh = ext3_bread (NULL,dir, 0, 0, err)))
359 goto fail;
360 root = (struct dx_root *) bh->b_data;
361 if (root->info.hash_version != DX_HASH_TEA &&
362 root->info.hash_version != DX_HASH_HALF_MD4 &&
363 root->info.hash_version != DX_HASH_LEGACY) {
Harvey Harrisone05b6b52008-04-28 02:16:15 -0700364 ext3_warning(dir->i_sb, __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 "Unrecognised inode hash code %d",
366 root->info.hash_version);
367 brelse(bh);
368 *err = ERR_BAD_DX_DIR;
369 goto fail;
370 }
371 hinfo->hash_version = root->info.hash_version;
372 hinfo->seed = EXT3_SB(dir->i_sb)->s_hash_seed;
373 if (dentry)
374 ext3fs_dirhash(dentry->d_name.name, dentry->d_name.len, hinfo);
375 hash = hinfo->hash;
376
377 if (root->info.unused_flags & 1) {
Harvey Harrisone05b6b52008-04-28 02:16:15 -0700378 ext3_warning(dir->i_sb, __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 "Unimplemented inode hash flags: %#06x",
380 root->info.unused_flags);
381 brelse(bh);
382 *err = ERR_BAD_DX_DIR;
383 goto fail;
384 }
385
386 if ((indirect = root->info.indirect_levels) > 1) {
Harvey Harrisone05b6b52008-04-28 02:16:15 -0700387 ext3_warning(dir->i_sb, __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 "Unimplemented inode hash depth: %#06x",
389 root->info.indirect_levels);
390 brelse(bh);
391 *err = ERR_BAD_DX_DIR;
392 goto fail;
393 }
394
395 entries = (struct dx_entry *) (((char *)&root->info) +
396 root->info.info_length);
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700397
398 if (dx_get_limit(entries) != dx_root_limit(dir,
399 root->info.info_length)) {
Harvey Harrisone05b6b52008-04-28 02:16:15 -0700400 ext3_warning(dir->i_sb, __func__,
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700401 "dx entry: limit != root limit");
402 brelse(bh);
403 *err = ERR_BAD_DX_DIR;
404 goto fail;
405 }
406
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 dxtrace (printk("Look up %x", hash));
408 while (1)
409 {
410 count = dx_get_count(entries);
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700411 if (!count || count > dx_get_limit(entries)) {
Harvey Harrisone05b6b52008-04-28 02:16:15 -0700412 ext3_warning(dir->i_sb, __func__,
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700413 "dx entry: no count or count > limit");
414 brelse(bh);
415 *err = ERR_BAD_DX_DIR;
416 goto fail2;
417 }
418
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 p = entries + 1;
420 q = entries + count - 1;
421 while (p <= q)
422 {
423 m = p + (q - p)/2;
424 dxtrace(printk("."));
425 if (dx_get_hash(m) > hash)
426 q = m - 1;
427 else
428 p = m + 1;
429 }
430
431 if (0) // linear search cross check
432 {
433 unsigned n = count - 1;
434 at = entries;
435 while (n--)
436 {
437 dxtrace(printk(","));
438 if (dx_get_hash(++at) > hash)
439 {
440 at--;
441 break;
442 }
443 }
444 assert (at == p - 1);
445 }
446
447 at = p - 1;
448 dxtrace(printk(" %x->%u\n", at == entries? 0: dx_get_hash(at), dx_get_block(at)));
449 frame->bh = bh;
450 frame->entries = entries;
451 frame->at = at;
452 if (!indirect--) return frame;
453 if (!(bh = ext3_bread (NULL,dir, dx_get_block(at), 0, err)))
454 goto fail2;
455 at = entries = ((struct dx_node *) bh->b_data)->entries;
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700456 if (dx_get_limit(entries) != dx_node_limit (dir)) {
Harvey Harrisone05b6b52008-04-28 02:16:15 -0700457 ext3_warning(dir->i_sb, __func__,
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700458 "dx entry: limit != node limit");
459 brelse(bh);
460 *err = ERR_BAD_DX_DIR;
461 goto fail2;
462 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 frame++;
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700464 frame->bh = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 }
466fail2:
467 while (frame >= frame_in) {
468 brelse(frame->bh);
469 frame--;
470 }
471fail:
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700472 if (*err == ERR_BAD_DX_DIR)
Harvey Harrisone05b6b52008-04-28 02:16:15 -0700473 ext3_warning(dir->i_sb, __func__,
Eric Sandeen3d82aba2007-09-18 22:46:38 -0700474 "Corrupt dir inode %ld, running e2fsck is "
475 "recommended.", dir->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 return NULL;
477}
478
479static void dx_release (struct dx_frame *frames)
480{
481 if (frames[0].bh == NULL)
482 return;
483
484 if (((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels)
485 brelse(frames[1].bh);
486 brelse(frames[0].bh);
487}
488
489/*
490 * This function increments the frame pointer to search the next leaf
491 * block, and reads in the necessary intervening nodes if the search
492 * should be necessary. Whether or not the search is necessary is
493 * controlled by the hash parameter. If the hash value is even, then
494 * the search is only continued if the next block starts with that
495 * hash value. This is used if we are searching for a specific file.
496 *
497 * If the hash value is HASH_NB_ALWAYS, then always go to the next block.
498 *
499 * This function returns 1 if the caller should continue to search,
500 * or 0 if it should not. If there is an error reading one of the
501 * index blocks, it will a negative error code.
502 *
503 * If start_hash is non-null, it will be filled in with the starting
504 * hash of the next page.
505 */
506static int ext3_htree_next_block(struct inode *dir, __u32 hash,
507 struct dx_frame *frame,
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700508 struct dx_frame *frames,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 __u32 *start_hash)
510{
511 struct dx_frame *p;
512 struct buffer_head *bh;
513 int err, num_frames = 0;
514 __u32 bhash;
515
516 p = frame;
517 /*
518 * Find the next leaf page by incrementing the frame pointer.
519 * If we run out of entries in the interior node, loop around and
520 * increment pointer in the parent node. When we break out of
521 * this loop, num_frames indicates the number of interior
522 * nodes need to be read.
523 */
524 while (1) {
525 if (++(p->at) < p->entries + dx_get_count(p->entries))
526 break;
527 if (p == frames)
528 return 0;
529 num_frames++;
530 p--;
531 }
532
533 /*
534 * If the hash is 1, then continue only if the next page has a
535 * continuation hash of any value. This is used for readdir
536 * handling. Otherwise, check to see if the hash matches the
537 * desired contiuation hash. If it doesn't, return since
538 * there's no point to read in the successive index pages.
539 */
540 bhash = dx_get_hash(p->at);
541 if (start_hash)
542 *start_hash = bhash;
543 if ((hash & 1) == 0) {
544 if ((bhash & ~1) != hash)
545 return 0;
546 }
547 /*
548 * If the hash is HASH_NB_ALWAYS, we always go to the next
549 * block so no check is necessary
550 */
551 while (num_frames--) {
552 if (!(bh = ext3_bread(NULL, dir, dx_get_block(p->at),
553 0, &err)))
554 return err; /* Failure */
555 p++;
556 brelse (p->bh);
557 p->bh = bh;
558 p->at = p->entries = ((struct dx_node *) bh->b_data)->entries;
559 }
560 return 1;
561}
562
563
564/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 * This function fills a red-black tree with information from a
566 * directory block. It returns the number directory entries loaded
567 * into the tree. If there is an error it is returned in err.
568 */
569static int htree_dirblock_to_tree(struct file *dir_file,
570 struct inode *dir, int block,
571 struct dx_hash_info *hinfo,
572 __u32 start_hash, __u32 start_minor_hash)
573{
574 struct buffer_head *bh;
575 struct ext3_dir_entry_2 *de, *top;
576 int err, count = 0;
577
578 dxtrace(printk("In htree dirblock_to_tree: block %d\n", block));
579 if (!(bh = ext3_bread (NULL, dir, block, 0, &err)))
580 return err;
581
582 de = (struct ext3_dir_entry_2 *) bh->b_data;
583 top = (struct ext3_dir_entry_2 *) ((char *) de +
584 dir->i_sb->s_blocksize -
585 EXT3_DIR_REC_LEN(0));
586 for (; de < top; de = ext3_next_entry(de)) {
Eric Sandeen40b85132006-12-06 20:36:26 -0800587 if (!ext3_check_dir_entry("htree_dirblock_to_tree", dir, de, bh,
588 (block<<EXT3_BLOCK_SIZE_BITS(dir->i_sb))
589 +((char *)de - bh->b_data))) {
590 /* On error, skip the f_pos to the next block. */
591 dir_file->f_pos = (dir_file->f_pos |
592 (dir->i_sb->s_blocksize - 1)) + 1;
593 brelse (bh);
594 return count;
595 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 ext3fs_dirhash(de->name, de->name_len, hinfo);
597 if ((hinfo->hash < start_hash) ||
598 ((hinfo->hash == start_hash) &&
599 (hinfo->minor_hash < start_minor_hash)))
600 continue;
601 if (de->inode == 0)
602 continue;
603 if ((err = ext3_htree_store_dirent(dir_file,
604 hinfo->hash, hinfo->minor_hash, de)) != 0) {
605 brelse(bh);
606 return err;
607 }
608 count++;
609 }
610 brelse(bh);
611 return count;
612}
613
614
615/*
616 * This function fills a red-black tree with information from a
617 * directory. We start scanning the directory in hash order, starting
618 * at start_hash and start_minor_hash.
619 *
620 * This function returns the number of entries inserted into the tree,
621 * or a negative error code.
622 */
623int ext3_htree_fill_tree(struct file *dir_file, __u32 start_hash,
624 __u32 start_minor_hash, __u32 *next_hash)
625{
626 struct dx_hash_info hinfo;
627 struct ext3_dir_entry_2 *de;
628 struct dx_frame frames[2], *frame;
629 struct inode *dir;
630 int block, err;
631 int count = 0;
632 int ret;
633 __u32 hashval;
634
635 dxtrace(printk("In htree_fill_tree, start hash: %x:%x\n", start_hash,
636 start_minor_hash));
Josef "Jeff" Sipekfe21a692006-12-08 02:36:38 -0800637 dir = dir_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 if (!(EXT3_I(dir)->i_flags & EXT3_INDEX_FL)) {
639 hinfo.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
640 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
641 count = htree_dirblock_to_tree(dir_file, dir, 0, &hinfo,
642 start_hash, start_minor_hash);
643 *next_hash = ~0;
644 return count;
645 }
646 hinfo.hash = start_hash;
647 hinfo.minor_hash = 0;
Josef "Jeff" Sipekfe21a692006-12-08 02:36:38 -0800648 frame = dx_probe(NULL, dir_file->f_path.dentry->d_inode, &hinfo, frames, &err);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 if (!frame)
650 return err;
651
652 /* Add '.' and '..' from the htree header */
653 if (!start_hash && !start_minor_hash) {
654 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
655 if ((err = ext3_htree_store_dirent(dir_file, 0, 0, de)) != 0)
656 goto errout;
657 count++;
658 }
659 if (start_hash < 2 || (start_hash ==2 && start_minor_hash==0)) {
660 de = (struct ext3_dir_entry_2 *) frames[0].bh->b_data;
661 de = ext3_next_entry(de);
662 if ((err = ext3_htree_store_dirent(dir_file, 2, 0, de)) != 0)
663 goto errout;
664 count++;
665 }
666
667 while (1) {
668 block = dx_get_block(frame->at);
669 ret = htree_dirblock_to_tree(dir_file, dir, block, &hinfo,
670 start_hash, start_minor_hash);
671 if (ret < 0) {
672 err = ret;
673 goto errout;
674 }
675 count += ret;
676 hashval = ~0;
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700677 ret = ext3_htree_next_block(dir, HASH_NB_ALWAYS,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 frame, frames, &hashval);
679 *next_hash = hashval;
680 if (ret < 0) {
681 err = ret;
682 goto errout;
683 }
684 /*
685 * Stop if: (a) there are no more entries, or
686 * (b) we have inserted at least one entry and the
687 * next hash value is not a continuation
688 */
689 if ((ret == 0) ||
690 (count && ((hashval & 1) == 0)))
691 break;
692 }
693 dx_release(frames);
Mingming Caoae6ddcc2006-09-27 01:49:27 -0700694 dxtrace(printk("Fill tree: returned %d entries, next hash: %x\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 count, *next_hash));
696 return count;
697errout:
698 dx_release(frames);
699 return (err);
700}
701
702
703/*
704 * Directory block splitting, compacting
705 */
706
Eric Sandeenef2b02d2007-09-18 22:46:42 -0700707/*
708 * Create map of hash values, offsets, and sizes, stored at end of block.
709 * Returns number of entries mapped.
710 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711static int dx_make_map (struct ext3_dir_entry_2 *de, int size,
712 struct dx_hash_info *hinfo, struct dx_map_entry *map_tail)
713{
714 int count = 0;
715 char *base = (char *) de;
716 struct dx_hash_info h = *hinfo;
717
718 while ((char *) de < base + size)
719 {
720 if (de->name_len && de->inode) {
721 ext3fs_dirhash(de->name, de->name_len, &h);
722 map_tail--;
723 map_tail->hash = h.hash;
Eric Sandeenef2b02d2007-09-18 22:46:42 -0700724 map_tail->offs = (u16) ((char *) de - base);
725 map_tail->size = le16_to_cpu(de->rec_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 count++;
727 cond_resched();
728 }
729 /* XXX: do we need to check rec_len == 0 case? -Chris */
Jan Kara7c06a8d2007-11-14 17:00:19 -0800730 de = ext3_next_entry(de);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 }
732 return count;
733}
734
Eric Sandeenef2b02d2007-09-18 22:46:42 -0700735/* Sort map by hash value */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736static void dx_sort_map (struct dx_map_entry *map, unsigned count)
737{
738 struct dx_map_entry *p, *q, *top = map + count - 1;
739 int more;
740 /* Combsort until bubble sort doesn't suck */
741 while (count > 2)
742 {
743 count = count*10/13;
744 if (count - 9 < 2) /* 9, 10 -> 11 */
745 count = 11;
746 for (p = top, q = p - count; q >= map; p--, q--)
747 if (p->hash < q->hash)
748 swap(*p, *q);
749 }
750 /* Garden variety bubble sort */
751 do {
752 more = 0;
753 q = top;
754 while (q-- > map)
755 {
756 if (q[1].hash >= q[0].hash)
757 continue;
758 swap(*(q+1), *q);
759 more = 1;
760 }
761 } while(more);
762}
763
764static void dx_insert_block(struct dx_frame *frame, u32 hash, u32 block)
765{
766 struct dx_entry *entries = frame->entries;
767 struct dx_entry *old = frame->at, *new = old + 1;
768 int count = dx_get_count(entries);
769
770 assert(count < dx_get_limit(entries));
771 assert(old < entries + count);
772 memmove(new + 1, new, (char *)(entries + count) - (char *)(new));
773 dx_set_hash(new, hash);
774 dx_set_block(new, block);
775 dx_set_count(entries, count + 1);
776}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777
778static void ext3_update_dx_flag(struct inode *inode)
779{
780 if (!EXT3_HAS_COMPAT_FEATURE(inode->i_sb,
781 EXT3_FEATURE_COMPAT_DIR_INDEX))
782 EXT3_I(inode)->i_flags &= ~EXT3_INDEX_FL;
783}
784
785/*
786 * NOTE! unlike strncmp, ext3_match returns 1 for success, 0 for failure.
787 *
788 * `len <= EXT3_NAME_LEN' is guaranteed by caller.
789 * `de != NULL' is guaranteed by caller.
790 */
791static inline int ext3_match (int len, const char * const name,
792 struct ext3_dir_entry_2 * de)
793{
794 if (len != de->name_len)
795 return 0;
796 if (!de->inode)
797 return 0;
798 return !memcmp(name, de->name, len);
799}
800
801/*
802 * Returns 0 if not found, -1 on failure, and 1 on success
803 */
804static inline int search_dirblock(struct buffer_head * bh,
805 struct inode *dir,
806 struct dentry *dentry,
807 unsigned long offset,
808 struct ext3_dir_entry_2 ** res_dir)
809{
810 struct ext3_dir_entry_2 * de;
811 char * dlimit;
812 int de_len;
813 const char *name = dentry->d_name.name;
814 int namelen = dentry->d_name.len;
815
816 de = (struct ext3_dir_entry_2 *) bh->b_data;
817 dlimit = bh->b_data + dir->i_sb->s_blocksize;
818 while ((char *) de < dlimit) {
819 /* this code is executed quadratically often */
820 /* do minimal checking `by hand' */
821
822 if ((char *) de + namelen <= dlimit &&
823 ext3_match (namelen, name, de)) {
824 /* found a match - just to be sure, do a full check */
825 if (!ext3_check_dir_entry("ext3_find_entry",
826 dir, de, bh, offset))
827 return -1;
828 *res_dir = de;
829 return 1;
830 }
831 /* prevent looping on a bad block */
Jan Kara7c06a8d2007-11-14 17:00:19 -0800832 de_len = ext3_rec_len_from_disk(de->rec_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 if (de_len <= 0)
834 return -1;
835 offset += de_len;
836 de = (struct ext3_dir_entry_2 *) ((char *) de + de_len);
837 }
838 return 0;
839}
840
841
842/*
843 * ext3_find_entry()
844 *
845 * finds an entry in the specified directory with the wanted name. It
846 * returns the cache buffer in which the entry was found, and the entry
847 * itself (as a parameter - res_dir). It does NOT read the inode of the
848 * entry - you'll have to do that yourself if you want to.
849 *
850 * The returned buffer_head has ->b_count elevated. The caller is expected
851 * to brelse() it when appropriate.
852 */
853static struct buffer_head * ext3_find_entry (struct dentry *dentry,
854 struct ext3_dir_entry_2 ** res_dir)
855{
856 struct super_block * sb;
857 struct buffer_head * bh_use[NAMEI_RA_SIZE];
858 struct buffer_head * bh, *ret = NULL;
859 unsigned long start, block, b;
860 int ra_max = 0; /* Number of bh's in the readahead
861 buffer, bh_use[] */
862 int ra_ptr = 0; /* Current index into readahead
863 buffer */
864 int num = 0;
865 int nblocks, i, err;
866 struct inode *dir = dentry->d_parent->d_inode;
867 int namelen;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868
869 *res_dir = NULL;
870 sb = dir->i_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 namelen = dentry->d_name.len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 if (namelen > EXT3_NAME_LEN)
873 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 if (is_dx(dir)) {
875 bh = ext3_dx_find_entry(dentry, res_dir, &err);
876 /*
877 * On success, or if the error was file not found,
878 * return. Otherwise, fall back to doing a search the
879 * old fashioned way.
880 */
881 if (bh || (err != ERR_BAD_DX_DIR))
882 return bh;
883 dxtrace(printk("ext3_find_entry: dx failed, falling back\n"));
884 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
886 start = EXT3_I(dir)->i_dir_start_lookup;
887 if (start >= nblocks)
888 start = 0;
889 block = start;
890restart:
891 do {
892 /*
893 * We deal with the read-ahead logic here.
894 */
895 if (ra_ptr >= ra_max) {
896 /* Refill the readahead buffer */
897 ra_ptr = 0;
898 b = block;
899 for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
900 /*
901 * Terminate if we reach the end of the
902 * directory and must wrap, or if our
903 * search has finished at this block.
904 */
905 if (b >= nblocks || (num && block == start)) {
906 bh_use[ra_max] = NULL;
907 break;
908 }
909 num++;
910 bh = ext3_getblk(NULL, dir, b++, 0, &err);
911 bh_use[ra_max] = bh;
912 if (bh)
Jens Axboecaa38fb2006-07-23 01:41:26 +0200913 ll_rw_block(READ_META, 1, &bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 }
915 }
916 if ((bh = bh_use[ra_ptr++]) == NULL)
917 goto next;
918 wait_on_buffer(bh);
919 if (!buffer_uptodate(bh)) {
920 /* read error, skip block & hope for the best */
Harvey Harrisone05b6b52008-04-28 02:16:15 -0700921 ext3_error(sb, __func__, "reading directory #%lu "
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 "offset %lu", dir->i_ino, block);
923 brelse(bh);
924 goto next;
925 }
926 i = search_dirblock(bh, dir, dentry,
927 block << EXT3_BLOCK_SIZE_BITS(sb), res_dir);
928 if (i == 1) {
929 EXT3_I(dir)->i_dir_start_lookup = block;
930 ret = bh;
931 goto cleanup_and_exit;
932 } else {
933 brelse(bh);
934 if (i < 0)
935 goto cleanup_and_exit;
936 }
937 next:
938 if (++block >= nblocks)
939 block = 0;
940 } while (block != start);
941
942 /*
943 * If the directory has grown while we were searching, then
944 * search the last part of the directory before giving up.
945 */
946 block = nblocks;
947 nblocks = dir->i_size >> EXT3_BLOCK_SIZE_BITS(sb);
948 if (block < nblocks) {
949 start = 0;
950 goto restart;
951 }
952
953cleanup_and_exit:
954 /* Clean up the read-ahead blocks */
955 for (; ra_ptr < ra_max; ra_ptr++)
956 brelse (bh_use[ra_ptr]);
957 return ret;
958}
959
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960static struct buffer_head * ext3_dx_find_entry(struct dentry *dentry,
961 struct ext3_dir_entry_2 **res_dir, int *err)
962{
963 struct super_block * sb;
964 struct dx_hash_info hinfo;
965 u32 hash;
966 struct dx_frame frames[2], *frame;
967 struct ext3_dir_entry_2 *de, *top;
968 struct buffer_head *bh;
969 unsigned long block;
970 int retval;
971 int namelen = dentry->d_name.len;
972 const u8 *name = dentry->d_name.name;
973 struct inode *dir = dentry->d_parent->d_inode;
974
975 sb = dir->i_sb;
Andreas Dilgeracfa1823d2005-06-23 00:09:45 -0700976 /* NFS may look up ".." - look at dx_root directory block */
977 if (namelen > 2 || name[0] != '.'||(name[1] != '.' && name[1] != '\0')){
978 if (!(frame = dx_probe(dentry, NULL, &hinfo, frames, err)))
979 return NULL;
980 } else {
981 frame = frames;
982 frame->bh = NULL; /* for dx_release() */
983 frame->at = (struct dx_entry *)frames; /* hack for zero entry*/
984 dx_set_block(frame->at, 0); /* dx_root block is 0 */
985 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 hash = hinfo.hash;
987 do {
988 block = dx_get_block(frame->at);
989 if (!(bh = ext3_bread (NULL,dir, block, 0, err)))
990 goto errout;
991 de = (struct ext3_dir_entry_2 *) bh->b_data;
992 top = (struct ext3_dir_entry_2 *) ((char *) de + sb->s_blocksize -
993 EXT3_DIR_REC_LEN(0));
994 for (; de < top; de = ext3_next_entry(de))
995 if (ext3_match (namelen, name, de)) {
996 if (!ext3_check_dir_entry("ext3_find_entry",
997 dir, de, bh,
998 (block<<EXT3_BLOCK_SIZE_BITS(sb))
999 +((char *)de - bh->b_data))) {
1000 brelse (bh);
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001001 *err = ERR_BAD_DX_DIR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 goto errout;
1003 }
1004 *res_dir = de;
1005 dx_release (frames);
1006 return bh;
1007 }
1008 brelse (bh);
1009 /* Check to see if we should continue to search */
1010 retval = ext3_htree_next_block(dir, hash, frame,
1011 frames, NULL);
1012 if (retval < 0) {
Harvey Harrisone05b6b52008-04-28 02:16:15 -07001013 ext3_warning(sb, __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 "error reading index page in directory #%lu",
1015 dir->i_ino);
1016 *err = retval;
1017 goto errout;
1018 }
1019 } while (retval == 1);
1020
1021 *err = -ENOENT;
1022errout:
1023 dxtrace(printk("%s not found\n", name));
1024 dx_release (frames);
1025 return NULL;
1026}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027
1028static struct dentry *ext3_lookup(struct inode * dir, struct dentry *dentry, struct nameidata *nd)
1029{
1030 struct inode * inode;
1031 struct ext3_dir_entry_2 * de;
1032 struct buffer_head * bh;
1033
1034 if (dentry->d_name.len > EXT3_NAME_LEN)
1035 return ERR_PTR(-ENAMETOOLONG);
1036
1037 bh = ext3_find_entry(dentry, &de);
1038 inode = NULL;
1039 if (bh) {
1040 unsigned long ino = le32_to_cpu(de->inode);
1041 brelse (bh);
Neil Brown2ccb48e2006-07-30 03:03:01 -07001042 if (!ext3_valid_inum(dir->i_sb, ino)) {
1043 ext3_error(dir->i_sb, "ext3_lookup",
1044 "bad inode number: %lu", ino);
David Howells473043d2008-02-07 00:15:36 -08001045 return ERR_PTR(-EIO);
Vasily Averina6c15c22007-07-15 23:40:46 -07001046 }
David Howells473043d2008-02-07 00:15:36 -08001047 inode = ext3_iget(dir->i_sb, ino);
1048 if (IS_ERR(inode))
1049 return ERR_CAST(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 }
Pekka Enbergba7fe362006-01-14 13:21:08 -08001051 return d_splice_alias(inode, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052}
1053
1054
1055struct dentry *ext3_get_parent(struct dentry *child)
1056{
1057 unsigned long ino;
1058 struct dentry *parent;
1059 struct inode *inode;
1060 struct dentry dotdot;
1061 struct ext3_dir_entry_2 * de;
1062 struct buffer_head *bh;
1063
1064 dotdot.d_name.name = "..";
1065 dotdot.d_name.len = 2;
1066 dotdot.d_parent = child; /* confusing, isn't it! */
1067
1068 bh = ext3_find_entry(&dotdot, &de);
1069 inode = NULL;
1070 if (!bh)
1071 return ERR_PTR(-ENOENT);
1072 ino = le32_to_cpu(de->inode);
1073 brelse(bh);
Neil Brown2ccb48e2006-07-30 03:03:01 -07001074
1075 if (!ext3_valid_inum(child->d_inode->i_sb, ino)) {
1076 ext3_error(child->d_inode->i_sb, "ext3_get_parent",
1077 "bad inode number: %lu", ino);
David Howells473043d2008-02-07 00:15:36 -08001078 return ERR_PTR(-EIO);
Vasily Averina6c15c22007-07-15 23:40:46 -07001079 }
1080
David Howells473043d2008-02-07 00:15:36 -08001081 inode = ext3_iget(child->d_inode->i_sb, ino);
1082 if (IS_ERR(inode))
1083 return ERR_CAST(inode);
1084
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 parent = d_alloc_anon(inode);
1086 if (!parent) {
1087 iput(inode);
1088 parent = ERR_PTR(-ENOMEM);
1089 }
1090 return parent;
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001091}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092
1093#define S_SHIFT 12
1094static unsigned char ext3_type_by_mode[S_IFMT >> S_SHIFT] = {
1095 [S_IFREG >> S_SHIFT] = EXT3_FT_REG_FILE,
1096 [S_IFDIR >> S_SHIFT] = EXT3_FT_DIR,
1097 [S_IFCHR >> S_SHIFT] = EXT3_FT_CHRDEV,
1098 [S_IFBLK >> S_SHIFT] = EXT3_FT_BLKDEV,
1099 [S_IFIFO >> S_SHIFT] = EXT3_FT_FIFO,
1100 [S_IFSOCK >> S_SHIFT] = EXT3_FT_SOCK,
1101 [S_IFLNK >> S_SHIFT] = EXT3_FT_SYMLINK,
1102};
1103
1104static inline void ext3_set_de_type(struct super_block *sb,
1105 struct ext3_dir_entry_2 *de,
1106 umode_t mode) {
1107 if (EXT3_HAS_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_FILETYPE))
1108 de->file_type = ext3_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
1109}
1110
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001111/*
1112 * Move count entries from end of map between two memory locations.
1113 * Returns pointer to last entry moved.
1114 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115static struct ext3_dir_entry_2 *
1116dx_move_dirents(char *from, char *to, struct dx_map_entry *map, int count)
1117{
1118 unsigned rec_len = 0;
1119
1120 while (count--) {
1121 struct ext3_dir_entry_2 *de = (struct ext3_dir_entry_2 *) (from + map->offs);
1122 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1123 memcpy (to, de, rec_len);
1124 ((struct ext3_dir_entry_2 *) to)->rec_len =
Jan Kara7c06a8d2007-11-14 17:00:19 -08001125 ext3_rec_len_to_disk(rec_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 de->inode = 0;
1127 map++;
1128 to += rec_len;
1129 }
1130 return (struct ext3_dir_entry_2 *) (to - rec_len);
1131}
1132
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001133/*
1134 * Compact each dir entry in the range to the minimal rec_len.
1135 * Returns pointer to last entry in range.
1136 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137static struct ext3_dir_entry_2* dx_pack_dirents(char *base, int size)
1138{
1139 struct ext3_dir_entry_2 *next, *to, *prev, *de = (struct ext3_dir_entry_2 *) base;
1140 unsigned rec_len = 0;
1141
1142 prev = to = de;
1143 while ((char*)de < base + size) {
Jan Kara7c06a8d2007-11-14 17:00:19 -08001144 next = ext3_next_entry(de);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 if (de->inode && de->name_len) {
1146 rec_len = EXT3_DIR_REC_LEN(de->name_len);
1147 if (de > to)
1148 memmove(to, de, rec_len);
Jan Kara7c06a8d2007-11-14 17:00:19 -08001149 to->rec_len = ext3_rec_len_to_disk(rec_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 prev = to;
1151 to = (struct ext3_dir_entry_2 *) (((char *) to) + rec_len);
1152 }
1153 de = next;
1154 }
1155 return prev;
1156}
1157
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001158/*
1159 * Split a full leaf block to make room for a new dir entry.
1160 * Allocate a new block, and move entries so that they are approx. equally full.
1161 * Returns pointer to de in block into which the new entry will be inserted.
1162 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163static struct ext3_dir_entry_2 *do_split(handle_t *handle, struct inode *dir,
1164 struct buffer_head **bh,struct dx_frame *frame,
1165 struct dx_hash_info *hinfo, int *error)
1166{
1167 unsigned blocksize = dir->i_sb->s_blocksize;
1168 unsigned count, continued;
1169 struct buffer_head *bh2;
1170 u32 newblock;
1171 u32 hash2;
1172 struct dx_map_entry *map;
1173 char *data1 = (*bh)->b_data, *data2;
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001174 unsigned split, move, size, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 struct ext3_dir_entry_2 *de = NULL, *de2;
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001176 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001178 bh2 = ext3_append (handle, dir, &newblock, &err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 if (!(bh2)) {
1180 brelse(*bh);
1181 *bh = NULL;
1182 goto errout;
1183 }
1184
1185 BUFFER_TRACE(*bh, "get_write_access");
1186 err = ext3_journal_get_write_access(handle, *bh);
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001187 if (err)
1188 goto journal_error;
1189
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 BUFFER_TRACE(frame->bh, "get_write_access");
1191 err = ext3_journal_get_write_access(handle, frame->bh);
1192 if (err)
1193 goto journal_error;
1194
1195 data2 = bh2->b_data;
1196
1197 /* create map in the end of data2 block */
1198 map = (struct dx_map_entry *) (data2 + blocksize);
1199 count = dx_make_map ((struct ext3_dir_entry_2 *) data1,
1200 blocksize, hinfo, map);
1201 map -= count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 dx_sort_map (map, count);
Eric Sandeenef2b02d2007-09-18 22:46:42 -07001203 /* Split the existing block in the middle, size-wise */
1204 size = 0;
1205 move = 0;
1206 for (i = count-1; i >= 0; i--) {
1207 /* is more than half of this entry in 2nd half of the block? */
1208 if (size + map[i].size/2 > blocksize/2)
1209 break;
1210 size += map[i].size;
1211 move++;
1212 }
1213 /* map index at which we will split */
1214 split = count - move;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 hash2 = map[split].hash;
1216 continued = hash2 == map[split - 1].hash;
1217 dxtrace(printk("Split block %i at %x, %i/%i\n",
1218 dx_get_block(frame->at), hash2, split, count-split));
1219
1220 /* Fancy dance to stay within two buffers */
1221 de2 = dx_move_dirents(data1, data2, map + split, count - split);
1222 de = dx_pack_dirents(data1,blocksize);
Jan Kara7c06a8d2007-11-14 17:00:19 -08001223 de->rec_len = ext3_rec_len_to_disk(data1 + blocksize - (char *) de);
1224 de2->rec_len = ext3_rec_len_to_disk(data2 + blocksize - (char *) de2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data1, blocksize, 1));
1226 dxtrace(dx_show_leaf (hinfo, (struct ext3_dir_entry_2 *) data2, blocksize, 1));
1227
1228 /* Which block gets the new entry? */
1229 if (hinfo->hash >= hash2)
1230 {
1231 swap(*bh, bh2);
1232 de = de2;
1233 }
1234 dx_insert_block (frame, hash2 + continued, newblock);
1235 err = ext3_journal_dirty_metadata (handle, bh2);
1236 if (err)
1237 goto journal_error;
1238 err = ext3_journal_dirty_metadata (handle, frame->bh);
1239 if (err)
1240 goto journal_error;
1241 brelse (bh2);
1242 dxtrace(dx_show_index ("frame", frame->entries));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 return de;
Dmitriy Monakhovfedee542007-05-08 00:25:34 -07001244
1245journal_error:
1246 brelse(*bh);
1247 brelse(bh2);
1248 *bh = NULL;
1249 ext3_std_error(dir->i_sb, err);
1250errout:
1251 *error = err;
1252 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254
1255
1256/*
1257 * Add a new entry into a directory (leaf) block. If de is non-NULL,
1258 * it points to a directory entry which is guaranteed to be large
1259 * enough for new directory entry. If de is NULL, then
1260 * add_dirent_to_buf will attempt search the directory block for
1261 * space. It will return -ENOSPC if no space is available, and -EIO
1262 * and -EEXIST if directory entry already exists.
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001263 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 * NOTE! bh is NOT released in the case where ENOSPC is returned. In
1265 * all other cases bh is released.
1266 */
1267static int add_dirent_to_buf(handle_t *handle, struct dentry *dentry,
1268 struct inode *inode, struct ext3_dir_entry_2 *de,
1269 struct buffer_head * bh)
1270{
1271 struct inode *dir = dentry->d_parent->d_inode;
1272 const char *name = dentry->d_name.name;
1273 int namelen = dentry->d_name.len;
1274 unsigned long offset = 0;
1275 unsigned short reclen;
1276 int nlen, rlen, err;
1277 char *top;
1278
1279 reclen = EXT3_DIR_REC_LEN(namelen);
1280 if (!de) {
1281 de = (struct ext3_dir_entry_2 *)bh->b_data;
1282 top = bh->b_data + dir->i_sb->s_blocksize - reclen;
1283 while ((char *) de <= top) {
1284 if (!ext3_check_dir_entry("ext3_add_entry", dir, de,
1285 bh, offset)) {
1286 brelse (bh);
1287 return -EIO;
1288 }
1289 if (ext3_match (namelen, name, de)) {
1290 brelse (bh);
1291 return -EEXIST;
1292 }
1293 nlen = EXT3_DIR_REC_LEN(de->name_len);
Jan Kara7c06a8d2007-11-14 17:00:19 -08001294 rlen = ext3_rec_len_from_disk(de->rec_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 if ((de->inode? rlen - nlen: rlen) >= reclen)
1296 break;
1297 de = (struct ext3_dir_entry_2 *)((char *)de + rlen);
1298 offset += rlen;
1299 }
1300 if ((char *) de > top)
1301 return -ENOSPC;
1302 }
1303 BUFFER_TRACE(bh, "get_write_access");
1304 err = ext3_journal_get_write_access(handle, bh);
1305 if (err) {
1306 ext3_std_error(dir->i_sb, err);
1307 brelse(bh);
1308 return err;
1309 }
1310
1311 /* By now the buffer is marked for journaling */
1312 nlen = EXT3_DIR_REC_LEN(de->name_len);
Jan Kara7c06a8d2007-11-14 17:00:19 -08001313 rlen = ext3_rec_len_from_disk(de->rec_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 if (de->inode) {
1315 struct ext3_dir_entry_2 *de1 = (struct ext3_dir_entry_2 *)((char *)de + nlen);
Jan Kara7c06a8d2007-11-14 17:00:19 -08001316 de1->rec_len = ext3_rec_len_to_disk(rlen - nlen);
1317 de->rec_len = ext3_rec_len_to_disk(nlen);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 de = de1;
1319 }
1320 de->file_type = EXT3_FT_UNKNOWN;
1321 if (inode) {
1322 de->inode = cpu_to_le32(inode->i_ino);
1323 ext3_set_de_type(dir->i_sb, de, inode->i_mode);
1324 } else
1325 de->inode = 0;
1326 de->name_len = namelen;
1327 memcpy (de->name, name, namelen);
1328 /*
1329 * XXX shouldn't update any times until successful
1330 * completion of syscall, but too many callers depend
1331 * on this.
1332 *
1333 * XXX similarly, too many callers depend on
1334 * ext3_new_inode() setting the times, but error
1335 * recovery deletes the inode, so the worst that can
1336 * happen is that the times are slightly out of date
1337 * and/or different from the directory change time.
1338 */
1339 dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
1340 ext3_update_dx_flag(dir);
1341 dir->i_version++;
1342 ext3_mark_inode_dirty(handle, dir);
1343 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1344 err = ext3_journal_dirty_metadata(handle, bh);
1345 if (err)
1346 ext3_std_error(dir->i_sb, err);
1347 brelse(bh);
1348 return 0;
1349}
1350
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351/*
1352 * This converts a one block unindexed directory to a 3 block indexed
1353 * directory, and adds the dentry to the indexed directory.
1354 */
1355static int make_indexed_dir(handle_t *handle, struct dentry *dentry,
1356 struct inode *inode, struct buffer_head *bh)
1357{
1358 struct inode *dir = dentry->d_parent->d_inode;
1359 const char *name = dentry->d_name.name;
1360 int namelen = dentry->d_name.len;
1361 struct buffer_head *bh2;
1362 struct dx_root *root;
1363 struct dx_frame frames[2], *frame;
1364 struct dx_entry *entries;
1365 struct ext3_dir_entry_2 *de, *de2;
1366 char *data1, *top;
1367 unsigned len;
1368 int retval;
1369 unsigned blocksize;
1370 struct dx_hash_info hinfo;
1371 u32 block;
1372 struct fake_dirent *fde;
1373
1374 blocksize = dir->i_sb->s_blocksize;
1375 dxtrace(printk("Creating index\n"));
1376 retval = ext3_journal_get_write_access(handle, bh);
1377 if (retval) {
1378 ext3_std_error(dir->i_sb, retval);
1379 brelse(bh);
1380 return retval;
1381 }
1382 root = (struct dx_root *) bh->b_data;
1383
1384 bh2 = ext3_append (handle, dir, &block, &retval);
1385 if (!(bh2)) {
1386 brelse(bh);
1387 return retval;
1388 }
1389 EXT3_I(dir)->i_flags |= EXT3_INDEX_FL;
1390 data1 = bh2->b_data;
1391
1392 /* The 0th block becomes the root, move the dirents out */
1393 fde = &root->dotdot;
Jan Kara7c06a8d2007-11-14 17:00:19 -08001394 de = (struct ext3_dir_entry_2 *)((char *)fde +
1395 ext3_rec_len_from_disk(fde->rec_len));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 len = ((char *) root) + blocksize - (char *) de;
1397 memcpy (data1, de, len);
1398 de = (struct ext3_dir_entry_2 *) data1;
1399 top = data1 + len;
Jan Kara7c06a8d2007-11-14 17:00:19 -08001400 while ((char *)(de2 = ext3_next_entry(de)) < top)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 de = de2;
Jan Kara7c06a8d2007-11-14 17:00:19 -08001402 de->rec_len = ext3_rec_len_to_disk(data1 + blocksize - (char *) de);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 /* Initialize the root; the dot dirents already exist */
1404 de = (struct ext3_dir_entry_2 *) (&root->dotdot);
Jan Kara7c06a8d2007-11-14 17:00:19 -08001405 de->rec_len = ext3_rec_len_to_disk(blocksize - EXT3_DIR_REC_LEN(2));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 memset (&root->info, 0, sizeof(root->info));
1407 root->info.info_length = sizeof(root->info);
1408 root->info.hash_version = EXT3_SB(dir->i_sb)->s_def_hash_version;
1409 entries = root->entries;
1410 dx_set_block (entries, 1);
1411 dx_set_count (entries, 1);
1412 dx_set_limit (entries, dx_root_limit(dir, sizeof(root->info)));
1413
1414 /* Initialize as for dx_probe */
1415 hinfo.hash_version = root->info.hash_version;
1416 hinfo.seed = EXT3_SB(dir->i_sb)->s_hash_seed;
1417 ext3fs_dirhash(name, namelen, &hinfo);
1418 frame = frames;
1419 frame->entries = entries;
1420 frame->at = entries;
1421 frame->bh = bh;
1422 bh = bh2;
1423 de = do_split(handle,dir, &bh, frame, &hinfo, &retval);
1424 dx_release (frames);
1425 if (!(de))
1426 return retval;
1427
1428 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1429}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430
1431/*
1432 * ext3_add_entry()
1433 *
1434 * adds a file entry to the specified directory, using the same
1435 * semantics as ext3_find_entry(). It returns NULL if it failed.
1436 *
1437 * NOTE!! The inode part of 'de' is left at 0 - which means you
1438 * may not sleep between calling this and putting something into
1439 * the entry, as someone else might have used it while you slept.
1440 */
1441static int ext3_add_entry (handle_t *handle, struct dentry *dentry,
1442 struct inode *inode)
1443{
1444 struct inode *dir = dentry->d_parent->d_inode;
1445 unsigned long offset;
1446 struct buffer_head * bh;
1447 struct ext3_dir_entry_2 *de;
1448 struct super_block * sb;
1449 int retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 int dx_fallback=0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 unsigned blocksize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 u32 block, blocks;
1453
1454 sb = dir->i_sb;
1455 blocksize = sb->s_blocksize;
1456 if (!dentry->d_name.len)
1457 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 if (is_dx(dir)) {
1459 retval = ext3_dx_add_entry(handle, dentry, inode);
1460 if (!retval || (retval != ERR_BAD_DX_DIR))
1461 return retval;
1462 EXT3_I(dir)->i_flags &= ~EXT3_INDEX_FL;
1463 dx_fallback++;
1464 ext3_mark_inode_dirty(handle, dir);
1465 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 blocks = dir->i_size >> sb->s_blocksize_bits;
1467 for (block = 0, offset = 0; block < blocks; block++) {
1468 bh = ext3_bread(handle, dir, block, 0, &retval);
1469 if(!bh)
1470 return retval;
1471 retval = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1472 if (retval != -ENOSPC)
1473 return retval;
1474
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 if (blocks == 1 && !dx_fallback &&
1476 EXT3_HAS_COMPAT_FEATURE(sb, EXT3_FEATURE_COMPAT_DIR_INDEX))
1477 return make_indexed_dir(handle, dentry, inode, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478 brelse(bh);
1479 }
1480 bh = ext3_append(handle, dir, &block, &retval);
1481 if (!bh)
1482 return retval;
1483 de = (struct ext3_dir_entry_2 *) bh->b_data;
1484 de->inode = 0;
Jan Kara7c06a8d2007-11-14 17:00:19 -08001485 de->rec_len = ext3_rec_len_to_disk(blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 return add_dirent_to_buf(handle, dentry, inode, de, bh);
1487}
1488
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489/*
1490 * Returns 0 for success, or a negative error value
1491 */
1492static int ext3_dx_add_entry(handle_t *handle, struct dentry *dentry,
1493 struct inode *inode)
1494{
1495 struct dx_frame frames[2], *frame;
1496 struct dx_entry *entries, *at;
1497 struct dx_hash_info hinfo;
1498 struct buffer_head * bh;
1499 struct inode *dir = dentry->d_parent->d_inode;
1500 struct super_block * sb = dir->i_sb;
1501 struct ext3_dir_entry_2 *de;
1502 int err;
1503
1504 frame = dx_probe(dentry, NULL, &hinfo, frames, &err);
1505 if (!frame)
1506 return err;
1507 entries = frame->entries;
1508 at = frame->at;
1509
1510 if (!(bh = ext3_bread(handle,dir, dx_get_block(frame->at), 0, &err)))
1511 goto cleanup;
1512
1513 BUFFER_TRACE(bh, "get_write_access");
1514 err = ext3_journal_get_write_access(handle, bh);
1515 if (err)
1516 goto journal_error;
1517
1518 err = add_dirent_to_buf(handle, dentry, inode, NULL, bh);
1519 if (err != -ENOSPC) {
1520 bh = NULL;
1521 goto cleanup;
1522 }
1523
1524 /* Block full, should compress but for now just split */
1525 dxtrace(printk("using %u of %u node entries\n",
1526 dx_get_count(entries), dx_get_limit(entries)));
1527 /* Need to split index? */
1528 if (dx_get_count(entries) == dx_get_limit(entries)) {
1529 u32 newblock;
1530 unsigned icount = dx_get_count(entries);
1531 int levels = frame - frames;
1532 struct dx_entry *entries2;
1533 struct dx_node *node2;
1534 struct buffer_head *bh2;
1535
1536 if (levels && (dx_get_count(frames->entries) ==
1537 dx_get_limit(frames->entries))) {
Harvey Harrisone05b6b52008-04-28 02:16:15 -07001538 ext3_warning(sb, __func__,
Glauber de Oliveira Costa9f406682006-01-08 01:03:22 -08001539 "Directory index full!");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 err = -ENOSPC;
1541 goto cleanup;
1542 }
1543 bh2 = ext3_append (handle, dir, &newblock, &err);
1544 if (!(bh2))
1545 goto cleanup;
1546 node2 = (struct dx_node *)(bh2->b_data);
1547 entries2 = node2->entries;
Jan Kara7c06a8d2007-11-14 17:00:19 -08001548 node2->fake.rec_len = ext3_rec_len_to_disk(sb->s_blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 node2->fake.inode = 0;
1550 BUFFER_TRACE(frame->bh, "get_write_access");
1551 err = ext3_journal_get_write_access(handle, frame->bh);
1552 if (err)
1553 goto journal_error;
1554 if (levels) {
1555 unsigned icount1 = icount/2, icount2 = icount - icount1;
1556 unsigned hash2 = dx_get_hash(entries + icount1);
1557 dxtrace(printk("Split index %i/%i\n", icount1, icount2));
1558
1559 BUFFER_TRACE(frame->bh, "get_write_access"); /* index root */
1560 err = ext3_journal_get_write_access(handle,
1561 frames[0].bh);
1562 if (err)
1563 goto journal_error;
1564
1565 memcpy ((char *) entries2, (char *) (entries + icount1),
1566 icount2 * sizeof(struct dx_entry));
1567 dx_set_count (entries, icount1);
1568 dx_set_count (entries2, icount2);
1569 dx_set_limit (entries2, dx_node_limit(dir));
1570
1571 /* Which index block gets the new entry? */
1572 if (at - entries >= icount1) {
1573 frame->at = at = at - entries - icount1 + entries2;
1574 frame->entries = entries = entries2;
1575 swap(frame->bh, bh2);
1576 }
1577 dx_insert_block (frames + 0, hash2, newblock);
1578 dxtrace(dx_show_index ("node", frames[1].entries));
1579 dxtrace(dx_show_index ("node",
1580 ((struct dx_node *) bh2->b_data)->entries));
1581 err = ext3_journal_dirty_metadata(handle, bh2);
1582 if (err)
1583 goto journal_error;
1584 brelse (bh2);
1585 } else {
1586 dxtrace(printk("Creating second level index...\n"));
1587 memcpy((char *) entries2, (char *) entries,
1588 icount * sizeof(struct dx_entry));
1589 dx_set_limit(entries2, dx_node_limit(dir));
1590
1591 /* Set up root */
1592 dx_set_count(entries, 1);
1593 dx_set_block(entries + 0, newblock);
1594 ((struct dx_root *) frames[0].bh->b_data)->info.indirect_levels = 1;
1595
1596 /* Add new access path frame */
1597 frame = frames + 1;
1598 frame->at = at = at - entries + entries2;
1599 frame->entries = entries = entries2;
1600 frame->bh = bh2;
1601 err = ext3_journal_get_write_access(handle,
1602 frame->bh);
1603 if (err)
1604 goto journal_error;
1605 }
1606 ext3_journal_dirty_metadata(handle, frames[0].bh);
1607 }
1608 de = do_split(handle, dir, &bh, frame, &hinfo, &err);
1609 if (!de)
1610 goto cleanup;
1611 err = add_dirent_to_buf(handle, dentry, inode, de, bh);
1612 bh = NULL;
1613 goto cleanup;
1614
1615journal_error:
1616 ext3_std_error(dir->i_sb, err);
1617cleanup:
1618 if (bh)
1619 brelse(bh);
1620 dx_release(frames);
1621 return err;
1622}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623
1624/*
1625 * ext3_delete_entry deletes a directory entry by merging it with the
1626 * previous entry
1627 */
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001628static int ext3_delete_entry (handle_t *handle,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 struct inode * dir,
1630 struct ext3_dir_entry_2 * de_del,
1631 struct buffer_head * bh)
1632{
1633 struct ext3_dir_entry_2 * de, * pde;
1634 int i;
1635
1636 i = 0;
1637 pde = NULL;
1638 de = (struct ext3_dir_entry_2 *) bh->b_data;
1639 while (i < bh->b_size) {
1640 if (!ext3_check_dir_entry("ext3_delete_entry", dir, de, bh, i))
1641 return -EIO;
1642 if (de == de_del) {
1643 BUFFER_TRACE(bh, "get_write_access");
1644 ext3_journal_get_write_access(handle, bh);
1645 if (pde)
Jan Kara7c06a8d2007-11-14 17:00:19 -08001646 pde->rec_len = ext3_rec_len_to_disk(
1647 ext3_rec_len_from_disk(pde->rec_len) +
1648 ext3_rec_len_from_disk(de->rec_len));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 else
1650 de->inode = 0;
1651 dir->i_version++;
1652 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1653 ext3_journal_dirty_metadata(handle, bh);
1654 return 0;
1655 }
Jan Kara7c06a8d2007-11-14 17:00:19 -08001656 i += ext3_rec_len_from_disk(de->rec_len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 pde = de;
Jan Kara7c06a8d2007-11-14 17:00:19 -08001658 de = ext3_next_entry(de);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659 }
1660 return -ENOENT;
1661}
1662
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663static int ext3_add_nondir(handle_t *handle,
1664 struct dentry *dentry, struct inode *inode)
1665{
1666 int err = ext3_add_entry(handle, dentry, inode);
1667 if (!err) {
1668 ext3_mark_inode_dirty(handle, inode);
1669 d_instantiate(dentry, inode);
1670 return 0;
1671 }
Eric Sandeen731b9a52007-02-10 01:46:16 -08001672 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 iput(inode);
1674 return err;
1675}
1676
1677/*
1678 * By the time this is called, we already have created
1679 * the directory cache entry for the new file, but it
1680 * is so far negative - it has no inode.
1681 *
1682 * If the create succeeds, we fill in the inode information
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001683 * with d_instantiate().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 */
1685static int ext3_create (struct inode * dir, struct dentry * dentry, int mode,
1686 struct nameidata *nd)
1687{
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001688 handle_t *handle;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 struct inode * inode;
1690 int err, retries = 0;
1691
1692retry:
Jan Kara1f545872005-06-23 22:01:04 -07001693 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
Jan Kara1f545872005-06-23 22:01:04 -07001695 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 if (IS_ERR(handle))
1697 return PTR_ERR(handle);
1698
1699 if (IS_DIRSYNC(dir))
1700 handle->h_sync = 1;
1701
1702 inode = ext3_new_inode (handle, dir, mode);
1703 err = PTR_ERR(inode);
1704 if (!IS_ERR(inode)) {
1705 inode->i_op = &ext3_file_inode_operations;
1706 inode->i_fop = &ext3_file_operations;
1707 ext3_set_aops(inode);
1708 err = ext3_add_nondir(handle, dentry, inode);
1709 }
1710 ext3_journal_stop(handle);
1711 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1712 goto retry;
1713 return err;
1714}
1715
1716static int ext3_mknod (struct inode * dir, struct dentry *dentry,
1717 int mode, dev_t rdev)
1718{
1719 handle_t *handle;
1720 struct inode *inode;
1721 int err, retries = 0;
1722
1723 if (!new_valid_dev(rdev))
1724 return -EINVAL;
1725
1726retry:
Jan Kara1f545872005-06-23 22:01:04 -07001727 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
Dave Kleikampe9ad5622006-09-27 01:49:35 -07001728 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
Jan Kara1f545872005-06-23 22:01:04 -07001729 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 if (IS_ERR(handle))
1731 return PTR_ERR(handle);
1732
1733 if (IS_DIRSYNC(dir))
1734 handle->h_sync = 1;
1735
1736 inode = ext3_new_inode (handle, dir, mode);
1737 err = PTR_ERR(inode);
1738 if (!IS_ERR(inode)) {
1739 init_special_inode(inode, inode->i_mode, rdev);
1740#ifdef CONFIG_EXT3_FS_XATTR
1741 inode->i_op = &ext3_special_inode_operations;
1742#endif
1743 err = ext3_add_nondir(handle, dentry, inode);
1744 }
1745 ext3_journal_stop(handle);
1746 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1747 goto retry;
1748 return err;
1749}
1750
1751static int ext3_mkdir(struct inode * dir, struct dentry * dentry, int mode)
1752{
1753 handle_t *handle;
1754 struct inode * inode;
1755 struct buffer_head * dir_block;
1756 struct ext3_dir_entry_2 * de;
1757 int err, retries = 0;
1758
1759 if (dir->i_nlink >= EXT3_LINK_MAX)
1760 return -EMLINK;
1761
1762retry:
Jan Kara1f545872005-06-23 22:01:04 -07001763 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 3 +
Jan Kara1f545872005-06-23 22:01:04 -07001765 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 if (IS_ERR(handle))
1767 return PTR_ERR(handle);
1768
1769 if (IS_DIRSYNC(dir))
1770 handle->h_sync = 1;
1771
1772 inode = ext3_new_inode (handle, dir, S_IFDIR | mode);
1773 err = PTR_ERR(inode);
1774 if (IS_ERR(inode))
1775 goto out_stop;
1776
1777 inode->i_op = &ext3_dir_inode_operations;
1778 inode->i_fop = &ext3_dir_operations;
1779 inode->i_size = EXT3_I(inode)->i_disksize = inode->i_sb->s_blocksize;
1780 dir_block = ext3_bread (handle, inode, 0, 1, &err);
1781 if (!dir_block) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001782 drop_nlink(inode); /* is this nlink == 0? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 ext3_mark_inode_dirty(handle, inode);
1784 iput (inode);
1785 goto out_stop;
1786 }
1787 BUFFER_TRACE(dir_block, "get_write_access");
1788 ext3_journal_get_write_access(handle, dir_block);
1789 de = (struct ext3_dir_entry_2 *) dir_block->b_data;
1790 de->inode = cpu_to_le32(inode->i_ino);
1791 de->name_len = 1;
Jan Kara7c06a8d2007-11-14 17:00:19 -08001792 de->rec_len = ext3_rec_len_to_disk(EXT3_DIR_REC_LEN(de->name_len));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 strcpy (de->name, ".");
1794 ext3_set_de_type(dir->i_sb, de, S_IFDIR);
Jan Kara7c06a8d2007-11-14 17:00:19 -08001795 de = ext3_next_entry(de);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796 de->inode = cpu_to_le32(dir->i_ino);
Jan Kara7c06a8d2007-11-14 17:00:19 -08001797 de->rec_len = ext3_rec_len_to_disk(inode->i_sb->s_blocksize -
1798 EXT3_DIR_REC_LEN(1));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 de->name_len = 2;
1800 strcpy (de->name, "..");
1801 ext3_set_de_type(dir->i_sb, de, S_IFDIR);
1802 inode->i_nlink = 2;
1803 BUFFER_TRACE(dir_block, "call ext3_journal_dirty_metadata");
1804 ext3_journal_dirty_metadata(handle, dir_block);
1805 brelse (dir_block);
1806 ext3_mark_inode_dirty(handle, inode);
1807 err = ext3_add_entry (handle, dentry, inode);
1808 if (err) {
1809 inode->i_nlink = 0;
1810 ext3_mark_inode_dirty(handle, inode);
1811 iput (inode);
1812 goto out_stop;
1813 }
Dave Hansend8c76e62006-09-30 23:29:04 -07001814 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 ext3_update_dx_flag(dir);
1816 ext3_mark_inode_dirty(handle, dir);
1817 d_instantiate(dentry, inode);
1818out_stop:
1819 ext3_journal_stop(handle);
1820 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
1821 goto retry;
1822 return err;
1823}
1824
1825/*
1826 * routine to check that the specified directory is empty (for rmdir)
1827 */
1828static int empty_dir (struct inode * inode)
1829{
1830 unsigned long offset;
1831 struct buffer_head * bh;
1832 struct ext3_dir_entry_2 * de, * de1;
1833 struct super_block * sb;
1834 int err = 0;
1835
1836 sb = inode->i_sb;
1837 if (inode->i_size < EXT3_DIR_REC_LEN(1) + EXT3_DIR_REC_LEN(2) ||
1838 !(bh = ext3_bread (NULL, inode, 0, 0, &err))) {
1839 if (err)
Harvey Harrisone05b6b52008-04-28 02:16:15 -07001840 ext3_error(inode->i_sb, __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841 "error %d reading directory #%lu offset 0",
1842 err, inode->i_ino);
1843 else
Harvey Harrisone05b6b52008-04-28 02:16:15 -07001844 ext3_warning(inode->i_sb, __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 "bad directory (dir #%lu) - no data block",
1846 inode->i_ino);
1847 return 1;
1848 }
1849 de = (struct ext3_dir_entry_2 *) bh->b_data;
Jan Kara7c06a8d2007-11-14 17:00:19 -08001850 de1 = ext3_next_entry(de);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851 if (le32_to_cpu(de->inode) != inode->i_ino ||
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001852 !le32_to_cpu(de1->inode) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853 strcmp (".", de->name) ||
1854 strcmp ("..", de1->name)) {
Dave Kleikampe9ad5622006-09-27 01:49:35 -07001855 ext3_warning (inode->i_sb, "empty_dir",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856 "bad directory (dir #%lu) - no `.' or `..'",
1857 inode->i_ino);
1858 brelse (bh);
1859 return 1;
1860 }
Jan Kara7c06a8d2007-11-14 17:00:19 -08001861 offset = ext3_rec_len_from_disk(de->rec_len) +
1862 ext3_rec_len_from_disk(de1->rec_len);
1863 de = ext3_next_entry(de1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864 while (offset < inode->i_size ) {
1865 if (!bh ||
1866 (void *) de >= (void *) (bh->b_data+sb->s_blocksize)) {
1867 err = 0;
1868 brelse (bh);
1869 bh = ext3_bread (NULL, inode,
1870 offset >> EXT3_BLOCK_SIZE_BITS(sb), 0, &err);
1871 if (!bh) {
1872 if (err)
Harvey Harrisone05b6b52008-04-28 02:16:15 -07001873 ext3_error(sb, __func__,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874 "error %d reading directory"
1875 " #%lu offset %lu",
1876 err, inode->i_ino, offset);
1877 offset += sb->s_blocksize;
1878 continue;
1879 }
1880 de = (struct ext3_dir_entry_2 *) bh->b_data;
1881 }
1882 if (!ext3_check_dir_entry("empty_dir", inode, de, bh, offset)) {
1883 de = (struct ext3_dir_entry_2 *)(bh->b_data +
1884 sb->s_blocksize);
1885 offset = (offset | (sb->s_blocksize - 1)) + 1;
1886 continue;
1887 }
1888 if (le32_to_cpu(de->inode)) {
1889 brelse (bh);
1890 return 0;
1891 }
Jan Kara7c06a8d2007-11-14 17:00:19 -08001892 offset += ext3_rec_len_from_disk(de->rec_len);
1893 de = ext3_next_entry(de);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 }
1895 brelse (bh);
1896 return 1;
1897}
1898
1899/* ext3_orphan_add() links an unlinked or truncated inode into a list of
1900 * such inodes, starting at the superblock, in case we crash before the
1901 * file is closed/deleted, or in case the inode truncate spans multiple
1902 * transactions and the last transaction is not recovered after a crash.
1903 *
1904 * At filesystem recovery time, we walk this list deleting unlinked
1905 * inodes and truncating linked inodes in ext3_orphan_cleanup().
1906 */
1907int ext3_orphan_add(handle_t *handle, struct inode *inode)
1908{
1909 struct super_block *sb = inode->i_sb;
1910 struct ext3_iloc iloc;
1911 int err = 0, rc;
1912
1913 lock_super(sb);
1914 if (!list_empty(&EXT3_I(inode)->i_orphan))
1915 goto out_unlock;
1916
1917 /* Orphan handling is only valid for files with data blocks
1918 * being truncated, or files being unlinked. */
1919
1920 /* @@@ FIXME: Observation from aviro:
Mingming Caoae6ddcc2006-09-27 01:49:27 -07001921 * I think I can trigger J_ASSERT in ext3_orphan_add(). We block
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 * here (on lock_super()), so race with ext3_link() which might bump
1923 * ->i_nlink. For, say it, character device. Not a regular file,
1924 * not a directory, not a symlink and ->i_nlink > 0.
1925 */
1926 J_ASSERT ((S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1927 S_ISLNK(inode->i_mode)) || inode->i_nlink == 0);
1928
1929 BUFFER_TRACE(EXT3_SB(sb)->s_sbh, "get_write_access");
1930 err = ext3_journal_get_write_access(handle, EXT3_SB(sb)->s_sbh);
1931 if (err)
1932 goto out_unlock;
1933
1934 err = ext3_reserve_inode_write(handle, inode, &iloc);
1935 if (err)
1936 goto out_unlock;
1937
1938 /* Insert this inode at the head of the on-disk orphan list... */
1939 NEXT_ORPHAN(inode) = le32_to_cpu(EXT3_SB(sb)->s_es->s_last_orphan);
1940 EXT3_SB(sb)->s_es->s_last_orphan = cpu_to_le32(inode->i_ino);
1941 err = ext3_journal_dirty_metadata(handle, EXT3_SB(sb)->s_sbh);
1942 rc = ext3_mark_iloc_dirty(handle, inode, &iloc);
1943 if (!err)
1944 err = rc;
1945
1946 /* Only add to the head of the in-memory list if all the
1947 * previous operations succeeded. If the orphan_add is going to
1948 * fail (possibly taking the journal offline), we can't risk
1949 * leaving the inode on the orphan list: stray orphan-list
1950 * entries can cause panics at unmount time.
1951 *
1952 * This is safe: on error we're going to ignore the orphan list
1953 * anyway on the next recovery. */
1954 if (!err)
1955 list_add(&EXT3_I(inode)->i_orphan, &EXT3_SB(sb)->s_orphan);
1956
Eric Sandeeneee194e2006-09-27 01:49:30 -07001957 jbd_debug(4, "superblock will point to %lu\n", inode->i_ino);
1958 jbd_debug(4, "orphan inode %lu will point to %d\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959 inode->i_ino, NEXT_ORPHAN(inode));
1960out_unlock:
1961 unlock_super(sb);
1962 ext3_std_error(inode->i_sb, err);
1963 return err;
1964}
1965
1966/*
1967 * ext3_orphan_del() removes an unlinked or truncated inode from the list
1968 * of such inodes stored on disk, because it is finally being cleaned up.
1969 */
1970int ext3_orphan_del(handle_t *handle, struct inode *inode)
1971{
1972 struct list_head *prev;
1973 struct ext3_inode_info *ei = EXT3_I(inode);
1974 struct ext3_sb_info *sbi;
1975 unsigned long ino_next;
1976 struct ext3_iloc iloc;
1977 int err = 0;
1978
1979 lock_super(inode->i_sb);
1980 if (list_empty(&ei->i_orphan)) {
1981 unlock_super(inode->i_sb);
1982 return 0;
1983 }
1984
1985 ino_next = NEXT_ORPHAN(inode);
1986 prev = ei->i_orphan.prev;
1987 sbi = EXT3_SB(inode->i_sb);
1988
1989 jbd_debug(4, "remove inode %lu from orphan list\n", inode->i_ino);
1990
1991 list_del_init(&ei->i_orphan);
1992
1993 /* If we're on an error path, we may not have a valid
1994 * transaction handle with which to update the orphan list on
1995 * disk, but we still need to remove the inode from the linked
1996 * list in memory. */
1997 if (!handle)
1998 goto out;
1999
2000 err = ext3_reserve_inode_write(handle, inode, &iloc);
2001 if (err)
2002 goto out_err;
2003
2004 if (prev == &sbi->s_orphan) {
2005 jbd_debug(4, "superblock will point to %lu\n", ino_next);
2006 BUFFER_TRACE(sbi->s_sbh, "get_write_access");
2007 err = ext3_journal_get_write_access(handle, sbi->s_sbh);
2008 if (err)
2009 goto out_brelse;
2010 sbi->s_es->s_last_orphan = cpu_to_le32(ino_next);
2011 err = ext3_journal_dirty_metadata(handle, sbi->s_sbh);
2012 } else {
2013 struct ext3_iloc iloc2;
2014 struct inode *i_prev =
2015 &list_entry(prev, struct ext3_inode_info, i_orphan)->vfs_inode;
2016
2017 jbd_debug(4, "orphan inode %lu will point to %lu\n",
2018 i_prev->i_ino, ino_next);
2019 err = ext3_reserve_inode_write(handle, i_prev, &iloc2);
2020 if (err)
2021 goto out_brelse;
2022 NEXT_ORPHAN(i_prev) = ino_next;
2023 err = ext3_mark_iloc_dirty(handle, i_prev, &iloc2);
2024 }
2025 if (err)
2026 goto out_brelse;
2027 NEXT_ORPHAN(inode) = 0;
2028 err = ext3_mark_iloc_dirty(handle, inode, &iloc);
2029
2030out_err:
2031 ext3_std_error(inode->i_sb, err);
2032out:
2033 unlock_super(inode->i_sb);
2034 return err;
2035
2036out_brelse:
2037 brelse(iloc.bh);
2038 goto out_err;
2039}
2040
2041static int ext3_rmdir (struct inode * dir, struct dentry *dentry)
2042{
2043 int retval;
2044 struct inode * inode;
2045 struct buffer_head * bh;
2046 struct ext3_dir_entry_2 * de;
2047 handle_t *handle;
2048
2049 /* Initialize quotas before so that eventual writes go in
2050 * separate transaction */
2051 DQUOT_INIT(dentry->d_inode);
Jan Kara1f545872005-06-23 22:01:04 -07002052 handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 if (IS_ERR(handle))
2054 return PTR_ERR(handle);
2055
2056 retval = -ENOENT;
2057 bh = ext3_find_entry (dentry, &de);
2058 if (!bh)
2059 goto end_rmdir;
2060
2061 if (IS_DIRSYNC(dir))
2062 handle->h_sync = 1;
2063
2064 inode = dentry->d_inode;
2065
2066 retval = -EIO;
2067 if (le32_to_cpu(de->inode) != inode->i_ino)
2068 goto end_rmdir;
2069
2070 retval = -ENOTEMPTY;
2071 if (!empty_dir (inode))
2072 goto end_rmdir;
2073
2074 retval = ext3_delete_entry(handle, dir, de, bh);
2075 if (retval)
2076 goto end_rmdir;
2077 if (inode->i_nlink != 2)
2078 ext3_warning (inode->i_sb, "ext3_rmdir",
2079 "empty directory has nlink!=2 (%d)",
2080 inode->i_nlink);
2081 inode->i_version++;
Dave Hansence71ec32006-09-30 23:29:06 -07002082 clear_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083 /* There's no need to set i_disksize: the fact that i_nlink is
2084 * zero will ensure that the right thing happens during any
2085 * recovery. */
2086 inode->i_size = 0;
2087 ext3_orphan_add(handle, inode);
2088 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2089 ext3_mark_inode_dirty(handle, inode);
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002090 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091 ext3_update_dx_flag(dir);
2092 ext3_mark_inode_dirty(handle, dir);
2093
2094end_rmdir:
2095 ext3_journal_stop(handle);
2096 brelse (bh);
2097 return retval;
2098}
2099
2100static int ext3_unlink(struct inode * dir, struct dentry *dentry)
2101{
2102 int retval;
2103 struct inode * inode;
2104 struct buffer_head * bh;
2105 struct ext3_dir_entry_2 * de;
2106 handle_t *handle;
2107
2108 /* Initialize quotas before so that eventual writes go
2109 * in separate transaction */
2110 DQUOT_INIT(dentry->d_inode);
Jan Kara1f545872005-06-23 22:01:04 -07002111 handle = ext3_journal_start(dir, EXT3_DELETE_TRANS_BLOCKS(dir->i_sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112 if (IS_ERR(handle))
2113 return PTR_ERR(handle);
2114
2115 if (IS_DIRSYNC(dir))
2116 handle->h_sync = 1;
2117
2118 retval = -ENOENT;
2119 bh = ext3_find_entry (dentry, &de);
2120 if (!bh)
2121 goto end_unlink;
2122
2123 inode = dentry->d_inode;
2124
2125 retval = -EIO;
2126 if (le32_to_cpu(de->inode) != inode->i_ino)
2127 goto end_unlink;
2128
2129 if (!inode->i_nlink) {
2130 ext3_warning (inode->i_sb, "ext3_unlink",
2131 "Deleting nonexistent file (%lu), %d",
2132 inode->i_ino, inode->i_nlink);
2133 inode->i_nlink = 1;
2134 }
2135 retval = ext3_delete_entry(handle, dir, de, bh);
2136 if (retval)
2137 goto end_unlink;
2138 dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
2139 ext3_update_dx_flag(dir);
2140 ext3_mark_inode_dirty(handle, dir);
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002141 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142 if (!inode->i_nlink)
2143 ext3_orphan_add(handle, inode);
2144 inode->i_ctime = dir->i_ctime;
2145 ext3_mark_inode_dirty(handle, inode);
2146 retval = 0;
2147
2148end_unlink:
2149 ext3_journal_stop(handle);
2150 brelse (bh);
2151 return retval;
2152}
2153
2154static int ext3_symlink (struct inode * dir,
2155 struct dentry *dentry, const char * symname)
2156{
2157 handle_t *handle;
2158 struct inode * inode;
2159 int l, err, retries = 0;
2160
2161 l = strlen(symname)+1;
2162 if (l > dir->i_sb->s_blocksize)
2163 return -ENAMETOOLONG;
2164
2165retry:
Jan Kara1f545872005-06-23 22:01:04 -07002166 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
Dave Kleikampe9ad5622006-09-27 01:49:35 -07002167 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 5 +
Jan Kara1f545872005-06-23 22:01:04 -07002168 2*EXT3_QUOTA_INIT_BLOCKS(dir->i_sb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169 if (IS_ERR(handle))
2170 return PTR_ERR(handle);
2171
2172 if (IS_DIRSYNC(dir))
2173 handle->h_sync = 1;
2174
2175 inode = ext3_new_inode (handle, dir, S_IFLNK|S_IRWXUGO);
2176 err = PTR_ERR(inode);
2177 if (IS_ERR(inode))
2178 goto out_stop;
2179
2180 if (l > sizeof (EXT3_I(inode)->i_data)) {
2181 inode->i_op = &ext3_symlink_inode_operations;
2182 ext3_set_aops(inode);
2183 /*
2184 * page_symlink() calls into ext3_prepare/commit_write.
2185 * We have a transaction open. All is sweetness. It also sets
2186 * i_size in generic_commit_write().
2187 */
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002188 err = __page_symlink(inode, symname, l,
2189 mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190 if (err) {
Eric Sandeen731b9a52007-02-10 01:46:16 -08002191 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 ext3_mark_inode_dirty(handle, inode);
2193 iput (inode);
2194 goto out_stop;
2195 }
2196 } else {
2197 inode->i_op = &ext3_fast_symlink_inode_operations;
2198 memcpy((char*)&EXT3_I(inode)->i_data,symname,l);
2199 inode->i_size = l-1;
2200 }
2201 EXT3_I(inode)->i_disksize = inode->i_size;
2202 err = ext3_add_nondir(handle, dentry, inode);
2203out_stop:
2204 ext3_journal_stop(handle);
2205 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2206 goto retry;
2207 return err;
2208}
2209
2210static int ext3_link (struct dentry * old_dentry,
2211 struct inode * dir, struct dentry *dentry)
2212{
2213 handle_t *handle;
2214 struct inode *inode = old_dentry->d_inode;
2215 int err, retries = 0;
2216
2217 if (inode->i_nlink >= EXT3_LINK_MAX)
2218 return -EMLINK;
Eric Sandeen2988a772007-02-10 01:46:16 -08002219 /*
2220 * Return -ENOENT if we've raced with unlink and i_nlink is 0. Doing
2221 * otherwise has the potential to corrupt the orphan inode list.
2222 */
2223 if (inode->i_nlink == 0)
2224 return -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225
2226retry:
Jan Kara1f545872005-06-23 22:01:04 -07002227 handle = ext3_journal_start(dir, EXT3_DATA_TRANS_BLOCKS(dir->i_sb) +
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228 EXT3_INDEX_EXTRA_TRANS_BLOCKS);
2229 if (IS_ERR(handle))
2230 return PTR_ERR(handle);
2231
2232 if (IS_DIRSYNC(dir))
2233 handle->h_sync = 1;
2234
2235 inode->i_ctime = CURRENT_TIME_SEC;
Eric Sandeen731b9a52007-02-10 01:46:16 -08002236 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237 atomic_inc(&inode->i_count);
2238
2239 err = ext3_add_nondir(handle, dentry, inode);
2240 ext3_journal_stop(handle);
2241 if (err == -ENOSPC && ext3_should_retry_alloc(dir->i_sb, &retries))
2242 goto retry;
2243 return err;
2244}
2245
2246#define PARENT_INO(buffer) \
Jan Kara7c06a8d2007-11-14 17:00:19 -08002247 (ext3_next_entry((struct ext3_dir_entry_2 *)(buffer))->inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248
2249/*
2250 * Anybody can rename anything with this: the permission checks are left to the
2251 * higher-level routines.
2252 */
2253static int ext3_rename (struct inode * old_dir, struct dentry *old_dentry,
2254 struct inode * new_dir,struct dentry *new_dentry)
2255{
2256 handle_t *handle;
2257 struct inode * old_inode, * new_inode;
2258 struct buffer_head * old_bh, * new_bh, * dir_bh;
2259 struct ext3_dir_entry_2 * old_de, * new_de;
2260 int retval;
2261
2262 old_bh = new_bh = dir_bh = NULL;
2263
2264 /* Initialize quotas before so that eventual writes go
2265 * in separate transaction */
2266 if (new_dentry->d_inode)
2267 DQUOT_INIT(new_dentry->d_inode);
Jan Kara1f545872005-06-23 22:01:04 -07002268 handle = ext3_journal_start(old_dir, 2 *
2269 EXT3_DATA_TRANS_BLOCKS(old_dir->i_sb) +
Dave Kleikampe9ad5622006-09-27 01:49:35 -07002270 EXT3_INDEX_EXTRA_TRANS_BLOCKS + 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271 if (IS_ERR(handle))
2272 return PTR_ERR(handle);
2273
2274 if (IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir))
2275 handle->h_sync = 1;
2276
2277 old_bh = ext3_find_entry (old_dentry, &old_de);
2278 /*
2279 * Check for inode number is _not_ due to possible IO errors.
2280 * We might rmdir the source, keep it as pwd of some process
2281 * and merrily kill the link to whatever was created under the
2282 * same name. Goodbye sticky bit ;-<
2283 */
2284 old_inode = old_dentry->d_inode;
2285 retval = -ENOENT;
2286 if (!old_bh || le32_to_cpu(old_de->inode) != old_inode->i_ino)
2287 goto end_rename;
2288
2289 new_inode = new_dentry->d_inode;
2290 new_bh = ext3_find_entry (new_dentry, &new_de);
2291 if (new_bh) {
2292 if (!new_inode) {
2293 brelse (new_bh);
2294 new_bh = NULL;
2295 }
2296 }
2297 if (S_ISDIR(old_inode->i_mode)) {
2298 if (new_inode) {
2299 retval = -ENOTEMPTY;
2300 if (!empty_dir (new_inode))
2301 goto end_rename;
2302 }
2303 retval = -EIO;
2304 dir_bh = ext3_bread (handle, old_inode, 0, 0, &retval);
2305 if (!dir_bh)
2306 goto end_rename;
2307 if (le32_to_cpu(PARENT_INO(dir_bh->b_data)) != old_dir->i_ino)
2308 goto end_rename;
2309 retval = -EMLINK;
2310 if (!new_inode && new_dir!=old_dir &&
2311 new_dir->i_nlink >= EXT3_LINK_MAX)
2312 goto end_rename;
2313 }
2314 if (!new_bh) {
2315 retval = ext3_add_entry (handle, new_dentry, old_inode);
2316 if (retval)
2317 goto end_rename;
2318 } else {
2319 BUFFER_TRACE(new_bh, "get write access");
2320 ext3_journal_get_write_access(handle, new_bh);
2321 new_de->inode = cpu_to_le32(old_inode->i_ino);
2322 if (EXT3_HAS_INCOMPAT_FEATURE(new_dir->i_sb,
2323 EXT3_FEATURE_INCOMPAT_FILETYPE))
2324 new_de->file_type = old_de->file_type;
2325 new_dir->i_version++;
Jan Kara0b230762008-04-28 02:16:12 -07002326 new_dir->i_ctime = new_dir->i_mtime = CURRENT_TIME_SEC;
2327 ext3_mark_inode_dirty(handle, new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328 BUFFER_TRACE(new_bh, "call ext3_journal_dirty_metadata");
2329 ext3_journal_dirty_metadata(handle, new_bh);
2330 brelse(new_bh);
2331 new_bh = NULL;
2332 }
2333
2334 /*
2335 * Like most other Unix systems, set the ctime for inodes on a
2336 * rename.
2337 */
2338 old_inode->i_ctime = CURRENT_TIME_SEC;
2339 ext3_mark_inode_dirty(handle, old_inode);
2340
2341 /*
2342 * ok, that's it
2343 */
2344 if (le32_to_cpu(old_de->inode) != old_inode->i_ino ||
2345 old_de->name_len != old_dentry->d_name.len ||
2346 strncmp(old_de->name, old_dentry->d_name.name, old_de->name_len) ||
2347 (retval = ext3_delete_entry(handle, old_dir,
2348 old_de, old_bh)) == -ENOENT) {
2349 /* old_de could have moved from under us during htree split, so
2350 * make sure that we are deleting the right entry. We might
2351 * also be pointing to a stale entry in the unused part of
2352 * old_bh so just checking inum and the name isn't enough. */
2353 struct buffer_head *old_bh2;
2354 struct ext3_dir_entry_2 *old_de2;
2355
2356 old_bh2 = ext3_find_entry(old_dentry, &old_de2);
2357 if (old_bh2) {
2358 retval = ext3_delete_entry(handle, old_dir,
2359 old_de2, old_bh2);
2360 brelse(old_bh2);
2361 }
2362 }
2363 if (retval) {
2364 ext3_warning(old_dir->i_sb, "ext3_rename",
2365 "Deleting old file (%lu), %d, error=%d",
2366 old_dir->i_ino, old_dir->i_nlink, retval);
2367 }
2368
2369 if (new_inode) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002370 drop_nlink(new_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371 new_inode->i_ctime = CURRENT_TIME_SEC;
2372 }
2373 old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME_SEC;
2374 ext3_update_dx_flag(old_dir);
2375 if (dir_bh) {
2376 BUFFER_TRACE(dir_bh, "get_write_access");
2377 ext3_journal_get_write_access(handle, dir_bh);
2378 PARENT_INO(dir_bh->b_data) = cpu_to_le32(new_dir->i_ino);
2379 BUFFER_TRACE(dir_bh, "call ext3_journal_dirty_metadata");
2380 ext3_journal_dirty_metadata(handle, dir_bh);
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002381 drop_nlink(old_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 if (new_inode) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002383 drop_nlink(new_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384 } else {
Dave Hansend8c76e62006-09-30 23:29:04 -07002385 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386 ext3_update_dx_flag(new_dir);
2387 ext3_mark_inode_dirty(handle, new_dir);
2388 }
2389 }
2390 ext3_mark_inode_dirty(handle, old_dir);
2391 if (new_inode) {
2392 ext3_mark_inode_dirty(handle, new_inode);
2393 if (!new_inode->i_nlink)
2394 ext3_orphan_add(handle, new_inode);
2395 }
2396 retval = 0;
2397
2398end_rename:
2399 brelse (dir_bh);
2400 brelse (old_bh);
2401 brelse (new_bh);
2402 ext3_journal_stop(handle);
2403 return retval;
2404}
2405
2406/*
2407 * directories can handle most operations...
2408 */
Arjan van de Ven754661f2007-02-12 00:55:38 -08002409const struct inode_operations ext3_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410 .create = ext3_create,
2411 .lookup = ext3_lookup,
2412 .link = ext3_link,
2413 .unlink = ext3_unlink,
2414 .symlink = ext3_symlink,
2415 .mkdir = ext3_mkdir,
2416 .rmdir = ext3_rmdir,
2417 .mknod = ext3_mknod,
2418 .rename = ext3_rename,
2419 .setattr = ext3_setattr,
2420#ifdef CONFIG_EXT3_FS_XATTR
2421 .setxattr = generic_setxattr,
2422 .getxattr = generic_getxattr,
2423 .listxattr = ext3_listxattr,
2424 .removexattr = generic_removexattr,
2425#endif
2426 .permission = ext3_permission,
2427};
2428
Arjan van de Ven754661f2007-02-12 00:55:38 -08002429const struct inode_operations ext3_special_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430 .setattr = ext3_setattr,
2431#ifdef CONFIG_EXT3_FS_XATTR
2432 .setxattr = generic_setxattr,
2433 .getxattr = generic_getxattr,
2434 .listxattr = ext3_listxattr,
2435 .removexattr = generic_removexattr,
2436#endif
2437 .permission = ext3_permission,
Mingming Caoae6ddcc2006-09-27 01:49:27 -07002438};