namei.c revision 6d6b77f163c7eabedbba00ed2abb7d4a570bff76
1/*
2 * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README
3 *
4 * Trivial changes by Alan Cox to remove EHASHCOLLISION for compatibility
5 *
6 * Trivial Changes:
7 * Rights granted to Hans Reiser to redistribute under other terms providing
8 * he accepts all liability including but not limited to patent, fitness
9 * for purpose, and direct or indirect claims arising from failure to perform.
10 *
11 * NO WARRANTY
12 */
13
14#include <linux/time.h>
15#include <linux/bitops.h>
16#include <linux/slab.h>
17#include <linux/reiserfs_fs.h>
18#include <linux/reiserfs_acl.h>
19#include <linux/reiserfs_xattr.h>
20#include <linux/quotaops.h>
21
22#define INC_DIR_INODE_NLINK(i) if (i->i_nlink != 1) { inc_nlink(i); if (i->i_nlink >= REISERFS_LINK_MAX) i->i_nlink=1; }
23#define DEC_DIR_INODE_NLINK(i) if (i->i_nlink != 1) drop_nlink(i);
24
25// directory item contains array of entry headers. This performs
26// binary search through that array
27static int bin_search_in_dir_item(struct reiserfs_dir_entry *de, loff_t off)
28{
29	struct item_head *ih = de->de_ih;
30	struct reiserfs_de_head *deh = de->de_deh;
31	int rbound, lbound, j;
32
33	lbound = 0;
34	rbound = I_ENTRY_COUNT(ih) - 1;
35
36	for (j = (rbound + lbound) / 2; lbound <= rbound;
37	     j = (rbound + lbound) / 2) {
38		if (off < deh_offset(deh + j)) {
39			rbound = j - 1;
40			continue;
41		}
42		if (off > deh_offset(deh + j)) {
43			lbound = j + 1;
44			continue;
45		}
46		// this is not name found, but matched third key component
47		de->de_entry_num = j;
48		return NAME_FOUND;
49	}
50
51	de->de_entry_num = lbound;
52	return NAME_NOT_FOUND;
53}
54
55// comment?  maybe something like set de to point to what the path points to?
56static inline void set_de_item_location(struct reiserfs_dir_entry *de,
57					struct treepath *path)
58{
59	de->de_bh = get_last_bh(path);
60	de->de_ih = get_ih(path);
61	de->de_deh = B_I_DEH(de->de_bh, de->de_ih);
62	de->de_item_num = PATH_LAST_POSITION(path);
63}
64
65// de_bh, de_ih, de_deh (points to first element of array), de_item_num is set
66inline void set_de_name_and_namelen(struct reiserfs_dir_entry *de)
67{
68	struct reiserfs_de_head *deh = de->de_deh + de->de_entry_num;
69
70	BUG_ON(de->de_entry_num >= ih_entry_count(de->de_ih));
71
72	de->de_entrylen = entry_length(de->de_bh, de->de_ih, de->de_entry_num);
73	de->de_namelen = de->de_entrylen - (de_with_sd(deh) ? SD_SIZE : 0);
74	de->de_name = B_I_PITEM(de->de_bh, de->de_ih) + deh_location(deh);
75	if (de->de_name[de->de_namelen - 1] == 0)
76		de->de_namelen = strlen(de->de_name);
77}
78
79// what entry points to
80static inline void set_de_object_key(struct reiserfs_dir_entry *de)
81{
82	BUG_ON(de->de_entry_num >= ih_entry_count(de->de_ih));
83	de->de_dir_id = deh_dir_id(&(de->de_deh[de->de_entry_num]));
84	de->de_objectid = deh_objectid(&(de->de_deh[de->de_entry_num]));
85}
86
87static inline void store_de_entry_key(struct reiserfs_dir_entry *de)
88{
89	struct reiserfs_de_head *deh = de->de_deh + de->de_entry_num;
90
91	BUG_ON(de->de_entry_num >= ih_entry_count(de->de_ih));
92
93	/* store key of the found entry */
94	de->de_entry_key.version = KEY_FORMAT_3_5;
95	de->de_entry_key.on_disk_key.k_dir_id =
96	    le32_to_cpu(de->de_ih->ih_key.k_dir_id);
97	de->de_entry_key.on_disk_key.k_objectid =
98	    le32_to_cpu(de->de_ih->ih_key.k_objectid);
99	set_cpu_key_k_offset(&(de->de_entry_key), deh_offset(deh));
100	set_cpu_key_k_type(&(de->de_entry_key), TYPE_DIRENTRY);
101}
102
103/* We assign a key to each directory item, and place multiple entries
104in a single directory item.  A directory item has a key equal to the
105key of the first directory entry in it.
106
107This function first calls search_by_key, then, if item whose first
108entry matches is not found it looks for the entry inside directory
109item found by search_by_key. Fills the path to the entry, and to the
110entry position in the item
111
112*/
113
114/* The function is NOT SCHEDULE-SAFE! */
115int search_by_entry_key(struct super_block *sb, const struct cpu_key *key,
116			struct treepath *path, struct reiserfs_dir_entry *de)
117{
118	int retval;
119
120	retval = search_item(sb, key, path);
121	switch (retval) {
122	case ITEM_NOT_FOUND:
123		if (!PATH_LAST_POSITION(path)) {
124			reiserfs_error(sb, "vs-7000", "search_by_key "
125				       "returned item position == 0");
126			pathrelse(path);
127			return IO_ERROR;
128		}
129		PATH_LAST_POSITION(path)--;
130
131	case ITEM_FOUND:
132		break;
133
134	case IO_ERROR:
135		return retval;
136
137	default:
138		pathrelse(path);
139		reiserfs_error(sb, "vs-7002", "no path to here");
140		return IO_ERROR;
141	}
142
143	set_de_item_location(de, path);
144
145#ifdef CONFIG_REISERFS_CHECK
146	if (!is_direntry_le_ih(de->de_ih) ||
147	    COMP_SHORT_KEYS(&(de->de_ih->ih_key), key)) {
148		print_block(de->de_bh, 0, -1, -1);
149		reiserfs_panic(sb, "vs-7005", "found item %h is not directory "
150			       "item or does not belong to the same directory "
151			       "as key %K", de->de_ih, key);
152	}
153#endif				/* CONFIG_REISERFS_CHECK */
154
155	/* binary search in directory item by third componen t of the
156	   key. sets de->de_entry_num of de */
157	retval = bin_search_in_dir_item(de, cpu_key_k_offset(key));
158	path->pos_in_item = de->de_entry_num;
159	if (retval != NAME_NOT_FOUND) {
160		// ugly, but rename needs de_bh, de_deh, de_name, de_namelen, de_objectid set
161		set_de_name_and_namelen(de);
162		set_de_object_key(de);
163	}
164	return retval;
165}
166
167/* Keyed 32-bit hash function using TEA in a Davis-Meyer function */
168
169/* The third component is hashed, and you can choose from more than
170   one hash function.  Per directory hashes are not yet implemented
171   but are thought about. This function should be moved to hashes.c
172   Jedi, please do so.  -Hans */
173
174static __u32 get_third_component(struct super_block *s,
175				 const char *name, int len)
176{
177	__u32 res;
178
179	if (!len || (len == 1 && name[0] == '.'))
180		return DOT_OFFSET;
181	if (len == 2 && name[0] == '.' && name[1] == '.')
182		return DOT_DOT_OFFSET;
183
184	res = REISERFS_SB(s)->s_hash_function(name, len);
185
186	// take bits from 7-th to 30-th including both bounds
187	res = GET_HASH_VALUE(res);
188	if (res == 0)
189		// needed to have no names before "." and ".." those have hash
190		// value == 0 and generation conters 1 and 2 accordingly
191		res = 128;
192	return res + MAX_GENERATION_NUMBER;
193}
194
195static int reiserfs_match(struct reiserfs_dir_entry *de,
196			  const char *name, int namelen)
197{
198	int retval = NAME_NOT_FOUND;
199
200	if ((namelen == de->de_namelen) &&
201	    !memcmp(de->de_name, name, de->de_namelen))
202		retval =
203		    (de_visible(de->de_deh + de->de_entry_num) ? NAME_FOUND :
204		     NAME_FOUND_INVISIBLE);
205
206	return retval;
207}
208
209/* de's de_bh, de_ih, de_deh, de_item_num, de_entry_num are set already */
210
211				/* used when hash collisions exist */
212
213static int linear_search_in_dir_item(struct cpu_key *key,
214				     struct reiserfs_dir_entry *de,
215				     const char *name, int namelen)
216{
217	struct reiserfs_de_head *deh = de->de_deh;
218	int retval;
219	int i;
220
221	i = de->de_entry_num;
222
223	if (i == I_ENTRY_COUNT(de->de_ih) ||
224	    GET_HASH_VALUE(deh_offset(deh + i)) !=
225	    GET_HASH_VALUE(cpu_key_k_offset(key))) {
226		i--;
227	}
228
229	RFALSE(de->de_deh != B_I_DEH(de->de_bh, de->de_ih),
230	       "vs-7010: array of entry headers not found");
231
232	deh += i;
233
234	for (; i >= 0; i--, deh--) {
235		if (GET_HASH_VALUE(deh_offset(deh)) !=
236		    GET_HASH_VALUE(cpu_key_k_offset(key))) {
237			// hash value does not match, no need to check whole name
238			return NAME_NOT_FOUND;
239		}
240
241		/* mark, that this generation number is used */
242		if (de->de_gen_number_bit_string)
243			set_bit(GET_GENERATION_NUMBER(deh_offset(deh)),
244				de->de_gen_number_bit_string);
245
246		// calculate pointer to name and namelen
247		de->de_entry_num = i;
248		set_de_name_and_namelen(de);
249
250		if ((retval =
251		     reiserfs_match(de, name, namelen)) != NAME_NOT_FOUND) {
252			// de's de_name, de_namelen, de_recordlen are set. Fill the rest:
253
254			// key of pointed object
255			set_de_object_key(de);
256
257			store_de_entry_key(de);
258
259			// retval can be NAME_FOUND or NAME_FOUND_INVISIBLE
260			return retval;
261		}
262	}
263
264	if (GET_GENERATION_NUMBER(le_ih_k_offset(de->de_ih)) == 0)
265		/* we have reached left most entry in the node. In common we
266		   have to go to the left neighbor, but if generation counter
267		   is 0 already, we know for sure, that there is no name with
268		   the same hash value */
269		// FIXME: this work correctly only because hash value can not
270		// be 0. Btw, in case of Yura's hash it is probably possible,
271		// so, this is a bug
272		return NAME_NOT_FOUND;
273
274	RFALSE(de->de_item_num,
275	       "vs-7015: two diritems of the same directory in one node?");
276
277	return GOTO_PREVIOUS_ITEM;
278}
279
280// may return NAME_FOUND, NAME_FOUND_INVISIBLE, NAME_NOT_FOUND
281// FIXME: should add something like IOERROR
282static int reiserfs_find_entry(struct inode *dir, const char *name, int namelen,
283			       struct treepath *path_to_entry,
284			       struct reiserfs_dir_entry *de)
285{
286	struct cpu_key key_to_search;
287	int retval;
288
289	if (namelen > REISERFS_MAX_NAME(dir->i_sb->s_blocksize))
290		return NAME_NOT_FOUND;
291
292	/* we will search for this key in the tree */
293	make_cpu_key(&key_to_search, dir,
294		     get_third_component(dir->i_sb, name, namelen),
295		     TYPE_DIRENTRY, 3);
296
297	while (1) {
298		retval =
299		    search_by_entry_key(dir->i_sb, &key_to_search,
300					path_to_entry, de);
301		if (retval == IO_ERROR) {
302			reiserfs_error(dir->i_sb, "zam-7001", "io error");
303			return IO_ERROR;
304		}
305
306		/* compare names for all entries having given hash value */
307		retval =
308		    linear_search_in_dir_item(&key_to_search, de, name,
309					      namelen);
310		if (retval != GOTO_PREVIOUS_ITEM) {
311			/* there is no need to scan directory anymore. Given entry found or does not exist */
312			path_to_entry->pos_in_item = de->de_entry_num;
313			return retval;
314		}
315
316		/* there is left neighboring item of this directory and given entry can be there */
317		set_cpu_key_k_offset(&key_to_search,
318				     le_ih_k_offset(de->de_ih) - 1);
319		pathrelse(path_to_entry);
320
321	}			/* while (1) */
322}
323
324static struct dentry *reiserfs_lookup(struct inode *dir, struct dentry *dentry,
325				      struct nameidata *nd)
326{
327	int retval;
328	int lock_depth;
329	struct inode *inode = NULL;
330	struct reiserfs_dir_entry de;
331	INITIALIZE_PATH(path_to_entry);
332
333	if (REISERFS_MAX_NAME(dir->i_sb->s_blocksize) < dentry->d_name.len)
334		return ERR_PTR(-ENAMETOOLONG);
335
336	/*
337	 * Might be called with or without the write lock, must be careful
338	 * to not recursively hold it in case we want to release the lock
339	 * before rescheduling.
340	 */
341	lock_depth = reiserfs_write_lock_once(dir->i_sb);
342
343	de.de_gen_number_bit_string = NULL;
344	retval =
345	    reiserfs_find_entry(dir, dentry->d_name.name, dentry->d_name.len,
346				&path_to_entry, &de);
347	pathrelse(&path_to_entry);
348	if (retval == NAME_FOUND) {
349		inode = reiserfs_iget(dir->i_sb,
350				      (struct cpu_key *)&(de.de_dir_id));
351		if (!inode || IS_ERR(inode)) {
352			reiserfs_write_unlock_once(dir->i_sb, lock_depth);
353			return ERR_PTR(-EACCES);
354		}
355
356		/* Propagate the private flag so we know we're
357		 * in the priv tree */
358		if (IS_PRIVATE(dir))
359			inode->i_flags |= S_PRIVATE;
360	}
361	reiserfs_write_unlock_once(dir->i_sb, lock_depth);
362	if (retval == IO_ERROR) {
363		return ERR_PTR(-EIO);
364	}
365
366	return d_splice_alias(inode, dentry);
367}
368
369/*
370** looks up the dentry of the parent directory for child.
371** taken from ext2_get_parent
372*/
373struct dentry *reiserfs_get_parent(struct dentry *child)
374{
375	int retval;
376	struct inode *inode = NULL;
377	struct reiserfs_dir_entry de;
378	INITIALIZE_PATH(path_to_entry);
379	struct inode *dir = child->d_inode;
380
381	if (dir->i_nlink == 0) {
382		return ERR_PTR(-ENOENT);
383	}
384	de.de_gen_number_bit_string = NULL;
385
386	reiserfs_write_lock(dir->i_sb);
387	retval = reiserfs_find_entry(dir, "..", 2, &path_to_entry, &de);
388	pathrelse(&path_to_entry);
389	if (retval != NAME_FOUND) {
390		reiserfs_write_unlock(dir->i_sb);
391		return ERR_PTR(-ENOENT);
392	}
393	inode = reiserfs_iget(dir->i_sb, (struct cpu_key *)&(de.de_dir_id));
394	reiserfs_write_unlock(dir->i_sb);
395
396	return d_obtain_alias(inode);
397}
398
399/* add entry to the directory (entry can be hidden).
400
401insert definition of when hidden directories are used here -Hans
402
403 Does not mark dir   inode dirty, do it after successesfull call to it */
404
405static int reiserfs_add_entry(struct reiserfs_transaction_handle *th,
406			      struct inode *dir, const char *name, int namelen,
407			      struct inode *inode, int visible)
408{
409	struct cpu_key entry_key;
410	struct reiserfs_de_head *deh;
411	INITIALIZE_PATH(path);
412	struct reiserfs_dir_entry de;
413	DECLARE_BITMAP(bit_string, MAX_GENERATION_NUMBER + 1);
414	int gen_number;
415	char small_buf[32 + DEH_SIZE];	/* 48 bytes now and we avoid kmalloc
416					   if we create file with short name */
417	char *buffer;
418	int buflen, paste_size;
419	int retval;
420
421	BUG_ON(!th->t_trans_id);
422
423	/* cannot allow items to be added into a busy deleted directory */
424	if (!namelen)
425		return -EINVAL;
426
427	if (namelen > REISERFS_MAX_NAME(dir->i_sb->s_blocksize))
428		return -ENAMETOOLONG;
429
430	/* each entry has unique key. compose it */
431	make_cpu_key(&entry_key, dir,
432		     get_third_component(dir->i_sb, name, namelen),
433		     TYPE_DIRENTRY, 3);
434
435	/* get memory for composing the entry */
436	buflen = DEH_SIZE + ROUND_UP(namelen);
437	if (buflen > sizeof(small_buf)) {
438		buffer = kmalloc(buflen, GFP_NOFS);
439		if (!buffer)
440			return -ENOMEM;
441	} else
442		buffer = small_buf;
443
444	paste_size =
445	    (get_inode_sd_version(dir) ==
446	     STAT_DATA_V1) ? (DEH_SIZE + namelen) : buflen;
447
448	/* fill buffer : directory entry head, name[, dir objectid | , stat data | ,stat data, dir objectid ] */
449	deh = (struct reiserfs_de_head *)buffer;
450	deh->deh_location = 0;	/* JDM Endian safe if 0 */
451	put_deh_offset(deh, cpu_key_k_offset(&entry_key));
452	deh->deh_state = 0;	/* JDM Endian safe if 0 */
453	/* put key (ino analog) to de */
454	deh->deh_dir_id = INODE_PKEY(inode)->k_dir_id;	/* safe: k_dir_id is le */
455	deh->deh_objectid = INODE_PKEY(inode)->k_objectid;	/* safe: k_objectid is le */
456
457	/* copy name */
458	memcpy((char *)(deh + 1), name, namelen);
459	/* padd by 0s to the 4 byte boundary */
460	padd_item((char *)(deh + 1), ROUND_UP(namelen), namelen);
461
462	/* entry is ready to be pasted into tree, set 'visibility' and 'stat data in entry' attributes */
463	mark_de_without_sd(deh);
464	visible ? mark_de_visible(deh) : mark_de_hidden(deh);
465
466	/* find the proper place for the new entry */
467	memset(bit_string, 0, sizeof(bit_string));
468	de.de_gen_number_bit_string = bit_string;
469	retval = reiserfs_find_entry(dir, name, namelen, &path, &de);
470	if (retval != NAME_NOT_FOUND) {
471		if (buffer != small_buf)
472			kfree(buffer);
473		pathrelse(&path);
474
475		if (retval == IO_ERROR) {
476			return -EIO;
477		}
478
479		if (retval != NAME_FOUND) {
480			reiserfs_error(dir->i_sb, "zam-7002",
481				       "reiserfs_find_entry() returned "
482				       "unexpected value (%d)", retval);
483		}
484
485		return -EEXIST;
486	}
487
488	gen_number =
489	    find_first_zero_bit(bit_string,
490				MAX_GENERATION_NUMBER + 1);
491	if (gen_number > MAX_GENERATION_NUMBER) {
492		/* there is no free generation number */
493		reiserfs_warning(dir->i_sb, "reiserfs-7010",
494				 "Congratulations! we have got hash function "
495				 "screwed up");
496		if (buffer != small_buf)
497			kfree(buffer);
498		pathrelse(&path);
499		return -EBUSY;
500	}
501	/* adjust offset of directory enrty */
502	put_deh_offset(deh, SET_GENERATION_NUMBER(deh_offset(deh), gen_number));
503	set_cpu_key_k_offset(&entry_key, deh_offset(deh));
504
505	/* update max-hash-collisions counter in reiserfs_sb_info */
506	PROC_INFO_MAX(th->t_super, max_hash_collisions, gen_number);
507
508	if (gen_number != 0) {	/* we need to re-search for the insertion point */
509		if (search_by_entry_key(dir->i_sb, &entry_key, &path, &de) !=
510		    NAME_NOT_FOUND) {
511			reiserfs_warning(dir->i_sb, "vs-7032",
512					 "entry with this key (%K) already "
513					 "exists", &entry_key);
514
515			if (buffer != small_buf)
516				kfree(buffer);
517			pathrelse(&path);
518			return -EBUSY;
519		}
520	}
521
522	/* perform the insertion of the entry that we have prepared */
523	retval =
524	    reiserfs_paste_into_item(th, &path, &entry_key, dir, buffer,
525				     paste_size);
526	if (buffer != small_buf)
527		kfree(buffer);
528	if (retval) {
529		reiserfs_check_path(&path);
530		return retval;
531	}
532
533	dir->i_size += paste_size;
534	dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
535	if (!S_ISDIR(inode->i_mode) && visible)
536		// reiserfs_mkdir or reiserfs_rename will do that by itself
537		reiserfs_update_sd(th, dir);
538
539	reiserfs_check_path(&path);
540	return 0;
541}
542
543/* quota utility function, call if you've had to abort after calling
544** new_inode_init, and have not called reiserfs_new_inode yet.
545** This should only be called on inodes that do not have stat data
546** inserted into the tree yet.
547*/
548static int drop_new_inode(struct inode *inode)
549{
550	dquot_drop(inode);
551	make_bad_inode(inode);
552	inode->i_flags |= S_NOQUOTA;
553	iput(inode);
554	return 0;
555}
556
557/* utility function that does setup for reiserfs_new_inode.
558** dquot_initialize needs lots of credits so it's better to have it
559** outside of a transaction, so we had to pull some bits of
560** reiserfs_new_inode out into this func.
561*/
562static int new_inode_init(struct inode *inode, struct inode *dir, int mode)
563{
564	/* Make inode invalid - just in case we are going to drop it before
565	 * the initialization happens */
566	INODE_PKEY(inode)->k_objectid = 0;
567	/* the quota init calls have to know who to charge the quota to, so
568	 ** we have to set uid and gid here
569	 */
570	inode_init_owner(inode, dir, mode);
571	dquot_initialize(inode);
572	return 0;
573}
574
575static int reiserfs_create(struct inode *dir, struct dentry *dentry, int mode,
576			   struct nameidata *nd)
577{
578	int retval;
579	struct inode *inode;
580	/* We need blocks for transaction + (user+group)*(quotas for new inode + update of quota for directory owner) */
581	int jbegin_count =
582	    JOURNAL_PER_BALANCE_CNT * 2 +
583	    2 * (REISERFS_QUOTA_INIT_BLOCKS(dir->i_sb) +
584		 REISERFS_QUOTA_TRANS_BLOCKS(dir->i_sb));
585	struct reiserfs_transaction_handle th;
586	struct reiserfs_security_handle security;
587
588	dquot_initialize(dir);
589
590	if (!(inode = new_inode(dir->i_sb))) {
591		return -ENOMEM;
592	}
593	new_inode_init(inode, dir, mode);
594
595	jbegin_count += reiserfs_cache_default_acl(dir);
596	retval = reiserfs_security_init(dir, inode, &dentry->d_name, &security);
597	if (retval < 0) {
598		drop_new_inode(inode);
599		return retval;
600	}
601	jbegin_count += retval;
602	reiserfs_write_lock(dir->i_sb);
603
604	retval = journal_begin(&th, dir->i_sb, jbegin_count);
605	if (retval) {
606		drop_new_inode(inode);
607		goto out_failed;
608	}
609
610	retval =
611	    reiserfs_new_inode(&th, dir, mode, NULL, 0 /*i_size */ , dentry,
612			       inode, &security);
613	if (retval)
614		goto out_failed;
615
616	inode->i_op = &reiserfs_file_inode_operations;
617	inode->i_fop = &reiserfs_file_operations;
618	inode->i_mapping->a_ops = &reiserfs_address_space_operations;
619
620	retval =
621	    reiserfs_add_entry(&th, dir, dentry->d_name.name,
622			       dentry->d_name.len, inode, 1 /*visible */ );
623	if (retval) {
624		int err;
625		drop_nlink(inode);
626		reiserfs_update_sd(&th, inode);
627		err = journal_end(&th, dir->i_sb, jbegin_count);
628		if (err)
629			retval = err;
630		unlock_new_inode(inode);
631		iput(inode);
632		goto out_failed;
633	}
634	reiserfs_update_inode_transaction(inode);
635	reiserfs_update_inode_transaction(dir);
636
637	d_instantiate(dentry, inode);
638	unlock_new_inode(inode);
639	retval = journal_end(&th, dir->i_sb, jbegin_count);
640
641      out_failed:
642	reiserfs_write_unlock(dir->i_sb);
643	return retval;
644}
645
646static int reiserfs_mknod(struct inode *dir, struct dentry *dentry, int mode,
647			  dev_t rdev)
648{
649	int retval;
650	struct inode *inode;
651	struct reiserfs_transaction_handle th;
652	struct reiserfs_security_handle security;
653	/* We need blocks for transaction + (user+group)*(quotas for new inode + update of quota for directory owner) */
654	int jbegin_count =
655	    JOURNAL_PER_BALANCE_CNT * 3 +
656	    2 * (REISERFS_QUOTA_INIT_BLOCKS(dir->i_sb) +
657		 REISERFS_QUOTA_TRANS_BLOCKS(dir->i_sb));
658
659	if (!new_valid_dev(rdev))
660		return -EINVAL;
661
662	dquot_initialize(dir);
663
664	if (!(inode = new_inode(dir->i_sb))) {
665		return -ENOMEM;
666	}
667	new_inode_init(inode, dir, mode);
668
669	jbegin_count += reiserfs_cache_default_acl(dir);
670	retval = reiserfs_security_init(dir, inode, &dentry->d_name, &security);
671	if (retval < 0) {
672		drop_new_inode(inode);
673		return retval;
674	}
675	jbegin_count += retval;
676	reiserfs_write_lock(dir->i_sb);
677
678	retval = journal_begin(&th, dir->i_sb, jbegin_count);
679	if (retval) {
680		drop_new_inode(inode);
681		goto out_failed;
682	}
683
684	retval =
685	    reiserfs_new_inode(&th, dir, mode, NULL, 0 /*i_size */ , dentry,
686			       inode, &security);
687	if (retval) {
688		goto out_failed;
689	}
690
691	inode->i_op = &reiserfs_special_inode_operations;
692	init_special_inode(inode, inode->i_mode, rdev);
693
694	//FIXME: needed for block and char devices only
695	reiserfs_update_sd(&th, inode);
696
697	reiserfs_update_inode_transaction(inode);
698	reiserfs_update_inode_transaction(dir);
699
700	retval =
701	    reiserfs_add_entry(&th, dir, dentry->d_name.name,
702			       dentry->d_name.len, inode, 1 /*visible */ );
703	if (retval) {
704		int err;
705		drop_nlink(inode);
706		reiserfs_update_sd(&th, inode);
707		err = journal_end(&th, dir->i_sb, jbegin_count);
708		if (err)
709			retval = err;
710		unlock_new_inode(inode);
711		iput(inode);
712		goto out_failed;
713	}
714
715	d_instantiate(dentry, inode);
716	unlock_new_inode(inode);
717	retval = journal_end(&th, dir->i_sb, jbegin_count);
718
719      out_failed:
720	reiserfs_write_unlock(dir->i_sb);
721	return retval;
722}
723
724static int reiserfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
725{
726	int retval;
727	struct inode *inode;
728	struct reiserfs_transaction_handle th;
729	struct reiserfs_security_handle security;
730	int lock_depth;
731	/* We need blocks for transaction + (user+group)*(quotas for new inode + update of quota for directory owner) */
732	int jbegin_count =
733	    JOURNAL_PER_BALANCE_CNT * 3 +
734	    2 * (REISERFS_QUOTA_INIT_BLOCKS(dir->i_sb) +
735		 REISERFS_QUOTA_TRANS_BLOCKS(dir->i_sb));
736
737	dquot_initialize(dir);
738
739#ifdef DISPLACE_NEW_PACKING_LOCALITIES
740	/* set flag that new packing locality created and new blocks for the content     * of that directory are not displaced yet */
741	REISERFS_I(dir)->new_packing_locality = 1;
742#endif
743	mode = S_IFDIR | mode;
744	if (!(inode = new_inode(dir->i_sb))) {
745		return -ENOMEM;
746	}
747	new_inode_init(inode, dir, mode);
748
749	jbegin_count += reiserfs_cache_default_acl(dir);
750	retval = reiserfs_security_init(dir, inode, &dentry->d_name, &security);
751	if (retval < 0) {
752		drop_new_inode(inode);
753		return retval;
754	}
755	jbegin_count += retval;
756	lock_depth = reiserfs_write_lock_once(dir->i_sb);
757
758	retval = journal_begin(&th, dir->i_sb, jbegin_count);
759	if (retval) {
760		drop_new_inode(inode);
761		goto out_failed;
762	}
763
764	/* inc the link count now, so another writer doesn't overflow it while
765	 ** we sleep later on.
766	 */
767	INC_DIR_INODE_NLINK(dir)
768
769	    retval = reiserfs_new_inode(&th, dir, mode, NULL /*symlink */ ,
770					old_format_only(dir->i_sb) ?
771					EMPTY_DIR_SIZE_V1 : EMPTY_DIR_SIZE,
772					dentry, inode, &security);
773	if (retval) {
774		DEC_DIR_INODE_NLINK(dir)
775		goto out_failed;
776	}
777
778	reiserfs_update_inode_transaction(inode);
779	reiserfs_update_inode_transaction(dir);
780
781	inode->i_op = &reiserfs_dir_inode_operations;
782	inode->i_fop = &reiserfs_dir_operations;
783
784	// note, _this_ add_entry will not update dir's stat data
785	retval =
786	    reiserfs_add_entry(&th, dir, dentry->d_name.name,
787			       dentry->d_name.len, inode, 1 /*visible */ );
788	if (retval) {
789		int err;
790		clear_nlink(inode);
791		DEC_DIR_INODE_NLINK(dir);
792		reiserfs_update_sd(&th, inode);
793		err = journal_end(&th, dir->i_sb, jbegin_count);
794		if (err)
795			retval = err;
796		unlock_new_inode(inode);
797		iput(inode);
798		goto out_failed;
799	}
800	// the above add_entry did not update dir's stat data
801	reiserfs_update_sd(&th, dir);
802
803	d_instantiate(dentry, inode);
804	unlock_new_inode(inode);
805	retval = journal_end(&th, dir->i_sb, jbegin_count);
806out_failed:
807	reiserfs_write_unlock_once(dir->i_sb, lock_depth);
808	return retval;
809}
810
811static inline int reiserfs_empty_dir(struct inode *inode)
812{
813	/* we can cheat because an old format dir cannot have
814	 ** EMPTY_DIR_SIZE, and a new format dir cannot have
815	 ** EMPTY_DIR_SIZE_V1.  So, if the inode is either size,
816	 ** regardless of disk format version, the directory is empty.
817	 */
818	if (inode->i_size != EMPTY_DIR_SIZE &&
819	    inode->i_size != EMPTY_DIR_SIZE_V1) {
820		return 0;
821	}
822	return 1;
823}
824
825static int reiserfs_rmdir(struct inode *dir, struct dentry *dentry)
826{
827	int retval, err;
828	struct inode *inode;
829	struct reiserfs_transaction_handle th;
830	int jbegin_count;
831	INITIALIZE_PATH(path);
832	struct reiserfs_dir_entry de;
833
834	/* we will be doing 2 balancings and update 2 stat data, we change quotas
835	 * of the owner of the directory and of the owner of the parent directory.
836	 * The quota structure is possibly deleted only on last iput => outside
837	 * of this transaction */
838	jbegin_count =
839	    JOURNAL_PER_BALANCE_CNT * 2 + 2 +
840	    4 * REISERFS_QUOTA_TRANS_BLOCKS(dir->i_sb);
841
842	dquot_initialize(dir);
843
844	reiserfs_write_lock(dir->i_sb);
845	retval = journal_begin(&th, dir->i_sb, jbegin_count);
846	if (retval)
847		goto out_rmdir;
848
849	de.de_gen_number_bit_string = NULL;
850	if ((retval =
851	     reiserfs_find_entry(dir, dentry->d_name.name, dentry->d_name.len,
852				 &path, &de)) == NAME_NOT_FOUND) {
853		retval = -ENOENT;
854		goto end_rmdir;
855	} else if (retval == IO_ERROR) {
856		retval = -EIO;
857		goto end_rmdir;
858	}
859
860	inode = dentry->d_inode;
861
862	reiserfs_update_inode_transaction(inode);
863	reiserfs_update_inode_transaction(dir);
864
865	if (de.de_objectid != inode->i_ino) {
866		// FIXME: compare key of an object and a key found in the
867		// entry
868		retval = -EIO;
869		goto end_rmdir;
870	}
871	if (!reiserfs_empty_dir(inode)) {
872		retval = -ENOTEMPTY;
873		goto end_rmdir;
874	}
875
876	/* cut entry from dir directory */
877	retval = reiserfs_cut_from_item(&th, &path, &(de.de_entry_key), dir, NULL,	/* page */
878					0 /*new file size - not used here */ );
879	if (retval < 0)
880		goto end_rmdir;
881
882	if (inode->i_nlink != 2 && inode->i_nlink != 1)
883		reiserfs_error(inode->i_sb, "reiserfs-7040",
884			       "empty directory has nlink != 2 (%d)",
885			       inode->i_nlink);
886
887	clear_nlink(inode);
888	inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
889	reiserfs_update_sd(&th, inode);
890
891	DEC_DIR_INODE_NLINK(dir)
892	    dir->i_size -= (DEH_SIZE + de.de_entrylen);
893	reiserfs_update_sd(&th, dir);
894
895	/* prevent empty directory from getting lost */
896	add_save_link(&th, inode, 0 /* not truncate */ );
897
898	retval = journal_end(&th, dir->i_sb, jbegin_count);
899	reiserfs_check_path(&path);
900      out_rmdir:
901	reiserfs_write_unlock(dir->i_sb);
902	return retval;
903
904      end_rmdir:
905	/* we must release path, because we did not call
906	   reiserfs_cut_from_item, or reiserfs_cut_from_item does not
907	   release path if operation was not complete */
908	pathrelse(&path);
909	err = journal_end(&th, dir->i_sb, jbegin_count);
910	reiserfs_write_unlock(dir->i_sb);
911	return err ? err : retval;
912}
913
914static int reiserfs_unlink(struct inode *dir, struct dentry *dentry)
915{
916	int retval, err;
917	struct inode *inode;
918	struct reiserfs_dir_entry de;
919	INITIALIZE_PATH(path);
920	struct reiserfs_transaction_handle th;
921	int jbegin_count;
922	unsigned long savelink;
923	int depth;
924
925	dquot_initialize(dir);
926
927	inode = dentry->d_inode;
928
929	/* in this transaction we can be doing at max two balancings and update
930	 * two stat datas, we change quotas of the owner of the directory and of
931	 * the owner of the parent directory. The quota structure is possibly
932	 * deleted only on iput => outside of this transaction */
933	jbegin_count =
934	    JOURNAL_PER_BALANCE_CNT * 2 + 2 +
935	    4 * REISERFS_QUOTA_TRANS_BLOCKS(dir->i_sb);
936
937	depth = reiserfs_write_lock_once(dir->i_sb);
938	retval = journal_begin(&th, dir->i_sb, jbegin_count);
939	if (retval)
940		goto out_unlink;
941
942	de.de_gen_number_bit_string = NULL;
943	if ((retval =
944	     reiserfs_find_entry(dir, dentry->d_name.name, dentry->d_name.len,
945				 &path, &de)) == NAME_NOT_FOUND) {
946		retval = -ENOENT;
947		goto end_unlink;
948	} else if (retval == IO_ERROR) {
949		retval = -EIO;
950		goto end_unlink;
951	}
952
953	reiserfs_update_inode_transaction(inode);
954	reiserfs_update_inode_transaction(dir);
955
956	if (de.de_objectid != inode->i_ino) {
957		// FIXME: compare key of an object and a key found in the
958		// entry
959		retval = -EIO;
960		goto end_unlink;
961	}
962
963	if (!inode->i_nlink) {
964		reiserfs_warning(inode->i_sb, "reiserfs-7042",
965				 "deleting nonexistent file (%lu), %d",
966				 inode->i_ino, inode->i_nlink);
967		inode->i_nlink = 1;
968	}
969
970	drop_nlink(inode);
971
972	/*
973	 * we schedule before doing the add_save_link call, save the link
974	 * count so we don't race
975	 */
976	savelink = inode->i_nlink;
977
978	retval =
979	    reiserfs_cut_from_item(&th, &path, &(de.de_entry_key), dir, NULL,
980				   0);
981	if (retval < 0) {
982		inc_nlink(inode);
983		goto end_unlink;
984	}
985	inode->i_ctime = CURRENT_TIME_SEC;
986	reiserfs_update_sd(&th, inode);
987
988	dir->i_size -= (de.de_entrylen + DEH_SIZE);
989	dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
990	reiserfs_update_sd(&th, dir);
991
992	if (!savelink)
993		/* prevent file from getting lost */
994		add_save_link(&th, inode, 0 /* not truncate */ );
995
996	retval = journal_end(&th, dir->i_sb, jbegin_count);
997	reiserfs_check_path(&path);
998	reiserfs_write_unlock_once(dir->i_sb, depth);
999	return retval;
1000
1001      end_unlink:
1002	pathrelse(&path);
1003	err = journal_end(&th, dir->i_sb, jbegin_count);
1004	reiserfs_check_path(&path);
1005	if (err)
1006		retval = err;
1007      out_unlink:
1008	reiserfs_write_unlock_once(dir->i_sb, depth);
1009	return retval;
1010}
1011
1012static int reiserfs_symlink(struct inode *parent_dir,
1013			    struct dentry *dentry, const char *symname)
1014{
1015	int retval;
1016	struct inode *inode;
1017	char *name;
1018	int item_len;
1019	struct reiserfs_transaction_handle th;
1020	struct reiserfs_security_handle security;
1021	int mode = S_IFLNK | S_IRWXUGO;
1022	/* We need blocks for transaction + (user+group)*(quotas for new inode + update of quota for directory owner) */
1023	int jbegin_count =
1024	    JOURNAL_PER_BALANCE_CNT * 3 +
1025	    2 * (REISERFS_QUOTA_INIT_BLOCKS(parent_dir->i_sb) +
1026		 REISERFS_QUOTA_TRANS_BLOCKS(parent_dir->i_sb));
1027
1028	dquot_initialize(parent_dir);
1029
1030	if (!(inode = new_inode(parent_dir->i_sb))) {
1031		return -ENOMEM;
1032	}
1033	new_inode_init(inode, parent_dir, mode);
1034
1035	retval = reiserfs_security_init(parent_dir, inode, &dentry->d_name,
1036					&security);
1037	if (retval < 0) {
1038		drop_new_inode(inode);
1039		return retval;
1040	}
1041	jbegin_count += retval;
1042
1043	reiserfs_write_lock(parent_dir->i_sb);
1044	item_len = ROUND_UP(strlen(symname));
1045	if (item_len > MAX_DIRECT_ITEM_LEN(parent_dir->i_sb->s_blocksize)) {
1046		retval = -ENAMETOOLONG;
1047		drop_new_inode(inode);
1048		goto out_failed;
1049	}
1050
1051	name = kmalloc(item_len, GFP_NOFS);
1052	if (!name) {
1053		drop_new_inode(inode);
1054		retval = -ENOMEM;
1055		goto out_failed;
1056	}
1057	memcpy(name, symname, strlen(symname));
1058	padd_item(name, item_len, strlen(symname));
1059
1060	retval = journal_begin(&th, parent_dir->i_sb, jbegin_count);
1061	if (retval) {
1062		drop_new_inode(inode);
1063		kfree(name);
1064		goto out_failed;
1065	}
1066
1067	retval =
1068	    reiserfs_new_inode(&th, parent_dir, mode, name, strlen(symname),
1069			       dentry, inode, &security);
1070	kfree(name);
1071	if (retval) {		/* reiserfs_new_inode iputs for us */
1072		goto out_failed;
1073	}
1074
1075	reiserfs_update_inode_transaction(inode);
1076	reiserfs_update_inode_transaction(parent_dir);
1077
1078	inode->i_op = &reiserfs_symlink_inode_operations;
1079	inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1080
1081	// must be sure this inode is written with this transaction
1082	//
1083	//reiserfs_update_sd (&th, inode, READ_BLOCKS);
1084
1085	retval = reiserfs_add_entry(&th, parent_dir, dentry->d_name.name,
1086				    dentry->d_name.len, inode, 1 /*visible */ );
1087	if (retval) {
1088		int err;
1089		drop_nlink(inode);
1090		reiserfs_update_sd(&th, inode);
1091		err = journal_end(&th, parent_dir->i_sb, jbegin_count);
1092		if (err)
1093			retval = err;
1094		unlock_new_inode(inode);
1095		iput(inode);
1096		goto out_failed;
1097	}
1098
1099	d_instantiate(dentry, inode);
1100	unlock_new_inode(inode);
1101	retval = journal_end(&th, parent_dir->i_sb, jbegin_count);
1102      out_failed:
1103	reiserfs_write_unlock(parent_dir->i_sb);
1104	return retval;
1105}
1106
1107static int reiserfs_link(struct dentry *old_dentry, struct inode *dir,
1108			 struct dentry *dentry)
1109{
1110	int retval;
1111	struct inode *inode = old_dentry->d_inode;
1112	struct reiserfs_transaction_handle th;
1113	/* We need blocks for transaction + update of quotas for the owners of the directory */
1114	int jbegin_count =
1115	    JOURNAL_PER_BALANCE_CNT * 3 +
1116	    2 * REISERFS_QUOTA_TRANS_BLOCKS(dir->i_sb);
1117
1118	dquot_initialize(dir);
1119
1120	reiserfs_write_lock(dir->i_sb);
1121	if (inode->i_nlink >= REISERFS_LINK_MAX) {
1122		//FIXME: sd_nlink is 32 bit for new files
1123		reiserfs_write_unlock(dir->i_sb);
1124		return -EMLINK;
1125	}
1126
1127	/* inc before scheduling so reiserfs_unlink knows we are here */
1128	inc_nlink(inode);
1129
1130	retval = journal_begin(&th, dir->i_sb, jbegin_count);
1131	if (retval) {
1132		drop_nlink(inode);
1133		reiserfs_write_unlock(dir->i_sb);
1134		return retval;
1135	}
1136
1137	/* create new entry */
1138	retval =
1139	    reiserfs_add_entry(&th, dir, dentry->d_name.name,
1140			       dentry->d_name.len, inode, 1 /*visible */ );
1141
1142	reiserfs_update_inode_transaction(inode);
1143	reiserfs_update_inode_transaction(dir);
1144
1145	if (retval) {
1146		int err;
1147		drop_nlink(inode);
1148		err = journal_end(&th, dir->i_sb, jbegin_count);
1149		reiserfs_write_unlock(dir->i_sb);
1150		return err ? err : retval;
1151	}
1152
1153	inode->i_ctime = CURRENT_TIME_SEC;
1154	reiserfs_update_sd(&th, inode);
1155
1156	ihold(inode);
1157	d_instantiate(dentry, inode);
1158	retval = journal_end(&th, dir->i_sb, jbegin_count);
1159	reiserfs_write_unlock(dir->i_sb);
1160	return retval;
1161}
1162
1163/* de contains information pointing to an entry which */
1164static int de_still_valid(const char *name, int len,
1165			  struct reiserfs_dir_entry *de)
1166{
1167	struct reiserfs_dir_entry tmp = *de;
1168
1169	// recalculate pointer to name and name length
1170	set_de_name_and_namelen(&tmp);
1171	// FIXME: could check more
1172	if (tmp.de_namelen != len || memcmp(name, de->de_name, len))
1173		return 0;
1174	return 1;
1175}
1176
1177static int entry_points_to_object(const char *name, int len,
1178				  struct reiserfs_dir_entry *de,
1179				  struct inode *inode)
1180{
1181	if (!de_still_valid(name, len, de))
1182		return 0;
1183
1184	if (inode) {
1185		if (!de_visible(de->de_deh + de->de_entry_num))
1186			reiserfs_panic(inode->i_sb, "vs-7042",
1187				       "entry must be visible");
1188		return (de->de_objectid == inode->i_ino) ? 1 : 0;
1189	}
1190
1191	/* this must be added hidden entry */
1192	if (de_visible(de->de_deh + de->de_entry_num))
1193		reiserfs_panic(NULL, "vs-7043", "entry must be visible");
1194
1195	return 1;
1196}
1197
1198/* sets key of objectid the entry has to point to */
1199static void set_ino_in_dir_entry(struct reiserfs_dir_entry *de,
1200				 struct reiserfs_key *key)
1201{
1202	/* JDM These operations are endian safe - both are le */
1203	de->de_deh[de->de_entry_num].deh_dir_id = key->k_dir_id;
1204	de->de_deh[de->de_entry_num].deh_objectid = key->k_objectid;
1205}
1206
1207/*
1208 * process, that is going to call fix_nodes/do_balance must hold only
1209 * one path. If it holds 2 or more, it can get into endless waiting in
1210 * get_empty_nodes or its clones
1211 */
1212static int reiserfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1213			   struct inode *new_dir, struct dentry *new_dentry)
1214{
1215	int retval;
1216	INITIALIZE_PATH(old_entry_path);
1217	INITIALIZE_PATH(new_entry_path);
1218	INITIALIZE_PATH(dot_dot_entry_path);
1219	struct item_head new_entry_ih, old_entry_ih, dot_dot_ih;
1220	struct reiserfs_dir_entry old_de, new_de, dot_dot_de;
1221	struct inode *old_inode, *new_dentry_inode;
1222	struct reiserfs_transaction_handle th;
1223	int jbegin_count;
1224	umode_t old_inode_mode;
1225	unsigned long savelink = 1;
1226	struct timespec ctime;
1227
1228	/* three balancings: (1) old name removal, (2) new name insertion
1229	   and (3) maybe "save" link insertion
1230	   stat data updates: (1) old directory,
1231	   (2) new directory and (3) maybe old object stat data (when it is
1232	   directory) and (4) maybe stat data of object to which new entry
1233	   pointed initially and (5) maybe block containing ".." of
1234	   renamed directory
1235	   quota updates: two parent directories */
1236	jbegin_count =
1237	    JOURNAL_PER_BALANCE_CNT * 3 + 5 +
1238	    4 * REISERFS_QUOTA_TRANS_BLOCKS(old_dir->i_sb);
1239
1240	dquot_initialize(old_dir);
1241	dquot_initialize(new_dir);
1242
1243	old_inode = old_dentry->d_inode;
1244	new_dentry_inode = new_dentry->d_inode;
1245
1246	// make sure, that oldname still exists and points to an object we
1247	// are going to rename
1248	old_de.de_gen_number_bit_string = NULL;
1249	reiserfs_write_lock(old_dir->i_sb);
1250	retval =
1251	    reiserfs_find_entry(old_dir, old_dentry->d_name.name,
1252				old_dentry->d_name.len, &old_entry_path,
1253				&old_de);
1254	pathrelse(&old_entry_path);
1255	if (retval == IO_ERROR) {
1256		reiserfs_write_unlock(old_dir->i_sb);
1257		return -EIO;
1258	}
1259
1260	if (retval != NAME_FOUND || old_de.de_objectid != old_inode->i_ino) {
1261		reiserfs_write_unlock(old_dir->i_sb);
1262		return -ENOENT;
1263	}
1264
1265	old_inode_mode = old_inode->i_mode;
1266	if (S_ISDIR(old_inode_mode)) {
1267		// make sure, that directory being renamed has correct ".."
1268		// and that its new parent directory has not too many links
1269		// already
1270
1271		if (new_dentry_inode) {
1272			if (!reiserfs_empty_dir(new_dentry_inode)) {
1273				reiserfs_write_unlock(old_dir->i_sb);
1274				return -ENOTEMPTY;
1275			}
1276		}
1277
1278		/* directory is renamed, its parent directory will be changed,
1279		 ** so find ".." entry
1280		 */
1281		dot_dot_de.de_gen_number_bit_string = NULL;
1282		retval =
1283		    reiserfs_find_entry(old_inode, "..", 2, &dot_dot_entry_path,
1284					&dot_dot_de);
1285		pathrelse(&dot_dot_entry_path);
1286		if (retval != NAME_FOUND) {
1287			reiserfs_write_unlock(old_dir->i_sb);
1288			return -EIO;
1289		}
1290
1291		/* inode number of .. must equal old_dir->i_ino */
1292		if (dot_dot_de.de_objectid != old_dir->i_ino) {
1293			reiserfs_write_unlock(old_dir->i_sb);
1294			return -EIO;
1295		}
1296	}
1297
1298	retval = journal_begin(&th, old_dir->i_sb, jbegin_count);
1299	if (retval) {
1300		reiserfs_write_unlock(old_dir->i_sb);
1301		return retval;
1302	}
1303
1304	/* add new entry (or find the existing one) */
1305	retval =
1306	    reiserfs_add_entry(&th, new_dir, new_dentry->d_name.name,
1307			       new_dentry->d_name.len, old_inode, 0);
1308	if (retval == -EEXIST) {
1309		if (!new_dentry_inode) {
1310			reiserfs_panic(old_dir->i_sb, "vs-7050",
1311				       "new entry is found, new inode == 0");
1312		}
1313	} else if (retval) {
1314		int err = journal_end(&th, old_dir->i_sb, jbegin_count);
1315		reiserfs_write_unlock(old_dir->i_sb);
1316		return err ? err : retval;
1317	}
1318
1319	reiserfs_update_inode_transaction(old_dir);
1320	reiserfs_update_inode_transaction(new_dir);
1321
1322	/* this makes it so an fsync on an open fd for the old name will
1323	 ** commit the rename operation
1324	 */
1325	reiserfs_update_inode_transaction(old_inode);
1326
1327	if (new_dentry_inode)
1328		reiserfs_update_inode_transaction(new_dentry_inode);
1329
1330	while (1) {
1331		// look for old name using corresponding entry key (found by reiserfs_find_entry)
1332		if ((retval =
1333		     search_by_entry_key(new_dir->i_sb, &old_de.de_entry_key,
1334					 &old_entry_path,
1335					 &old_de)) != NAME_FOUND) {
1336			pathrelse(&old_entry_path);
1337			journal_end(&th, old_dir->i_sb, jbegin_count);
1338			reiserfs_write_unlock(old_dir->i_sb);
1339			return -EIO;
1340		}
1341
1342		copy_item_head(&old_entry_ih, get_ih(&old_entry_path));
1343
1344		reiserfs_prepare_for_journal(old_inode->i_sb, old_de.de_bh, 1);
1345
1346		// look for new name by reiserfs_find_entry
1347		new_de.de_gen_number_bit_string = NULL;
1348		retval =
1349		    reiserfs_find_entry(new_dir, new_dentry->d_name.name,
1350					new_dentry->d_name.len, &new_entry_path,
1351					&new_de);
1352		// reiserfs_add_entry should not return IO_ERROR, because it is called with essentially same parameters from
1353		// reiserfs_add_entry above, and we'll catch any i/o errors before we get here.
1354		if (retval != NAME_FOUND_INVISIBLE && retval != NAME_FOUND) {
1355			pathrelse(&new_entry_path);
1356			pathrelse(&old_entry_path);
1357			journal_end(&th, old_dir->i_sb, jbegin_count);
1358			reiserfs_write_unlock(old_dir->i_sb);
1359			return -EIO;
1360		}
1361
1362		copy_item_head(&new_entry_ih, get_ih(&new_entry_path));
1363
1364		reiserfs_prepare_for_journal(old_inode->i_sb, new_de.de_bh, 1);
1365
1366		if (S_ISDIR(old_inode->i_mode)) {
1367			if ((retval =
1368			     search_by_entry_key(new_dir->i_sb,
1369						 &dot_dot_de.de_entry_key,
1370						 &dot_dot_entry_path,
1371						 &dot_dot_de)) != NAME_FOUND) {
1372				pathrelse(&dot_dot_entry_path);
1373				pathrelse(&new_entry_path);
1374				pathrelse(&old_entry_path);
1375				journal_end(&th, old_dir->i_sb, jbegin_count);
1376				reiserfs_write_unlock(old_dir->i_sb);
1377				return -EIO;
1378			}
1379			copy_item_head(&dot_dot_ih,
1380				       get_ih(&dot_dot_entry_path));
1381			// node containing ".." gets into transaction
1382			reiserfs_prepare_for_journal(old_inode->i_sb,
1383						     dot_dot_de.de_bh, 1);
1384		}
1385		/* we should check seals here, not do
1386		   this stuff, yes? Then, having
1387		   gathered everything into RAM we
1388		   should lock the buffers, yes?  -Hans */
1389		/* probably.  our rename needs to hold more
1390		 ** than one path at once.  The seals would
1391		 ** have to be written to deal with multi-path
1392		 ** issues -chris
1393		 */
1394		/* sanity checking before doing the rename - avoid races many
1395		 ** of the above checks could have scheduled.  We have to be
1396		 ** sure our items haven't been shifted by another process.
1397		 */
1398		if (item_moved(&new_entry_ih, &new_entry_path) ||
1399		    !entry_points_to_object(new_dentry->d_name.name,
1400					    new_dentry->d_name.len,
1401					    &new_de, new_dentry_inode) ||
1402		    item_moved(&old_entry_ih, &old_entry_path) ||
1403		    !entry_points_to_object(old_dentry->d_name.name,
1404					    old_dentry->d_name.len,
1405					    &old_de, old_inode)) {
1406			reiserfs_restore_prepared_buffer(old_inode->i_sb,
1407							 new_de.de_bh);
1408			reiserfs_restore_prepared_buffer(old_inode->i_sb,
1409							 old_de.de_bh);
1410			if (S_ISDIR(old_inode_mode))
1411				reiserfs_restore_prepared_buffer(old_inode->
1412								 i_sb,
1413								 dot_dot_de.
1414								 de_bh);
1415			continue;
1416		}
1417		if (S_ISDIR(old_inode_mode)) {
1418			if (item_moved(&dot_dot_ih, &dot_dot_entry_path) ||
1419			    !entry_points_to_object("..", 2, &dot_dot_de,
1420						    old_dir)) {
1421				reiserfs_restore_prepared_buffer(old_inode->
1422								 i_sb,
1423								 old_de.de_bh);
1424				reiserfs_restore_prepared_buffer(old_inode->
1425								 i_sb,
1426								 new_de.de_bh);
1427				reiserfs_restore_prepared_buffer(old_inode->
1428								 i_sb,
1429								 dot_dot_de.
1430								 de_bh);
1431				continue;
1432			}
1433		}
1434
1435		RFALSE(S_ISDIR(old_inode_mode) &&
1436		       !buffer_journal_prepared(dot_dot_de.de_bh), "");
1437
1438		break;
1439	}
1440
1441	/* ok, all the changes can be done in one fell swoop when we
1442	   have claimed all the buffers needed. */
1443
1444	mark_de_visible(new_de.de_deh + new_de.de_entry_num);
1445	set_ino_in_dir_entry(&new_de, INODE_PKEY(old_inode));
1446	journal_mark_dirty(&th, old_dir->i_sb, new_de.de_bh);
1447
1448	mark_de_hidden(old_de.de_deh + old_de.de_entry_num);
1449	journal_mark_dirty(&th, old_dir->i_sb, old_de.de_bh);
1450	ctime = CURRENT_TIME_SEC;
1451	old_dir->i_ctime = old_dir->i_mtime = ctime;
1452	new_dir->i_ctime = new_dir->i_mtime = ctime;
1453	/* thanks to Alex Adriaanse <alex_a@caltech.edu> for patch which adds ctime update of
1454	   renamed object */
1455	old_inode->i_ctime = ctime;
1456
1457	if (new_dentry_inode) {
1458		// adjust link number of the victim
1459		if (S_ISDIR(new_dentry_inode->i_mode)) {
1460			clear_nlink(new_dentry_inode);
1461		} else {
1462			drop_nlink(new_dentry_inode);
1463		}
1464		new_dentry_inode->i_ctime = ctime;
1465		savelink = new_dentry_inode->i_nlink;
1466	}
1467
1468	if (S_ISDIR(old_inode_mode)) {
1469		/* adjust ".." of renamed directory */
1470		set_ino_in_dir_entry(&dot_dot_de, INODE_PKEY(new_dir));
1471		journal_mark_dirty(&th, new_dir->i_sb, dot_dot_de.de_bh);
1472
1473		if (!new_dentry_inode)
1474			/* there (in new_dir) was no directory, so it got new link
1475			   (".."  of renamed directory) */
1476			INC_DIR_INODE_NLINK(new_dir);
1477
1478		/* old directory lost one link - ".. " of renamed directory */
1479		DEC_DIR_INODE_NLINK(old_dir);
1480	}
1481	// looks like in 2.3.99pre3 brelse is atomic. so we can use pathrelse
1482	pathrelse(&new_entry_path);
1483	pathrelse(&dot_dot_entry_path);
1484
1485	// FIXME: this reiserfs_cut_from_item's return value may screw up
1486	// anybody, but it will panic if will not be able to find the
1487	// entry. This needs one more clean up
1488	if (reiserfs_cut_from_item
1489	    (&th, &old_entry_path, &(old_de.de_entry_key), old_dir, NULL,
1490	     0) < 0)
1491		reiserfs_error(old_dir->i_sb, "vs-7060",
1492			       "couldn't not cut old name. Fsck later?");
1493
1494	old_dir->i_size -= DEH_SIZE + old_de.de_entrylen;
1495
1496	reiserfs_update_sd(&th, old_dir);
1497	reiserfs_update_sd(&th, new_dir);
1498	reiserfs_update_sd(&th, old_inode);
1499
1500	if (new_dentry_inode) {
1501		if (savelink == 0)
1502			add_save_link(&th, new_dentry_inode,
1503				      0 /* not truncate */ );
1504		reiserfs_update_sd(&th, new_dentry_inode);
1505	}
1506
1507	retval = journal_end(&th, old_dir->i_sb, jbegin_count);
1508	reiserfs_write_unlock(old_dir->i_sb);
1509	return retval;
1510}
1511
1512/*
1513 * directories can handle most operations...
1514 */
1515const struct inode_operations reiserfs_dir_inode_operations = {
1516	//&reiserfs_dir_operations,   /* default_file_ops */
1517	.create = reiserfs_create,
1518	.lookup = reiserfs_lookup,
1519	.link = reiserfs_link,
1520	.unlink = reiserfs_unlink,
1521	.symlink = reiserfs_symlink,
1522	.mkdir = reiserfs_mkdir,
1523	.rmdir = reiserfs_rmdir,
1524	.mknod = reiserfs_mknod,
1525	.rename = reiserfs_rename,
1526	.setattr = reiserfs_setattr,
1527	.setxattr = reiserfs_setxattr,
1528	.getxattr = reiserfs_getxattr,
1529	.listxattr = reiserfs_listxattr,
1530	.removexattr = reiserfs_removexattr,
1531	.permission = reiserfs_permission,
1532	.get_acl = reiserfs_get_acl,
1533};
1534
1535/*
1536 * symlink operations.. same as page_symlink_inode_operations, with xattr
1537 * stuff added
1538 */
1539const struct inode_operations reiserfs_symlink_inode_operations = {
1540	.readlink = generic_readlink,
1541	.follow_link = page_follow_link_light,
1542	.put_link = page_put_link,
1543	.setattr = reiserfs_setattr,
1544	.setxattr = reiserfs_setxattr,
1545	.getxattr = reiserfs_getxattr,
1546	.listxattr = reiserfs_listxattr,
1547	.removexattr = reiserfs_removexattr,
1548	.permission = reiserfs_permission,
1549	.get_acl = reiserfs_get_acl,
1550
1551};
1552
1553/*
1554 * special file operations.. just xattr/acl stuff
1555 */
1556const struct inode_operations reiserfs_special_inode_operations = {
1557	.setattr = reiserfs_setattr,
1558	.setxattr = reiserfs_setxattr,
1559	.getxattr = reiserfs_getxattr,
1560	.listxattr = reiserfs_listxattr,
1561	.removexattr = reiserfs_removexattr,
1562	.permission = reiserfs_permission,
1563	.get_acl = reiserfs_get_acl,
1564};
1565