2 * Copyright 2000 by Hans Reiser, licensing governed by reiserfs/README
4 * Trivial changes by Alan Cox to remove EHASHCOLLISION for compatibility
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.
14 #include <linux/config.h>
15 #include <linux/time.h>
16 #include <linux/bitops.h>
17 #include <linux/reiserfs_fs.h>
18 #include <linux/reiserfs_acl.h>
19 #include <linux/reiserfs_xattr.h>
20 #include <linux/smp_lock.h>
21 #include <linux/quotaops.h>
22 #include <linux/vserver/xid.h>
24 #define INC_DIR_INODE_NLINK(i) if (i->i_nlink != 1) { i->i_nlink++; if (i->i_nlink >= REISERFS_LINK_MAX) i->i_nlink=1; }
25 #define DEC_DIR_INODE_NLINK(i) if (i->i_nlink != 1) i->i_nlink--;
27 // directory item contains array of entry headers. This performs
28 // binary search through that array
29 static int bin_search_in_dir_item (struct reiserfs_dir_entry * de, loff_t off)
31 struct item_head * ih = de->de_ih;
32 struct reiserfs_de_head * deh = de->de_deh;
33 int rbound, lbound, j;
36 rbound = I_ENTRY_COUNT (ih) - 1;
38 for (j = (rbound + lbound) / 2; lbound <= rbound; j = (rbound + lbound) / 2) {
39 if (off < deh_offset (deh + j)) {
43 if (off > deh_offset (deh + j)) {
47 // this is not name found, but matched third key component
52 de->de_entry_num = lbound;
53 return NAME_NOT_FOUND;
57 // comment? maybe something like set de to point to what the path points to?
58 static inline void set_de_item_location (struct reiserfs_dir_entry * de, struct path * path)
60 de->de_bh = get_last_bh (path);
61 de->de_ih = get_ih (path);
62 de->de_deh = B_I_DEH (de->de_bh, de->de_ih);
63 de->de_item_num = PATH_LAST_POSITION (path);
67 // de_bh, de_ih, de_deh (points to first element of array), de_item_num is set
68 inline void set_de_name_and_namelen (struct reiserfs_dir_entry * de)
70 struct reiserfs_de_head * deh = de->de_deh + de->de_entry_num;
72 if (de->de_entry_num >= ih_entry_count (de->de_ih))
75 de->de_entrylen = entry_length (de->de_bh, de->de_ih, de->de_entry_num);
76 de->de_namelen = de->de_entrylen - (de_with_sd (deh) ? SD_SIZE : 0);
77 de->de_name = B_I_PITEM (de->de_bh, de->de_ih) + deh_location(deh);
78 if (de->de_name[de->de_namelen - 1] == 0)
79 de->de_namelen = strlen (de->de_name);
83 // what entry points to
84 static inline void set_de_object_key (struct reiserfs_dir_entry * de)
86 if (de->de_entry_num >= ih_entry_count (de->de_ih))
88 de->de_dir_id = deh_dir_id( &(de->de_deh[de->de_entry_num]));
89 de->de_objectid = deh_objectid( &(de->de_deh[de->de_entry_num]));
93 static inline void store_de_entry_key (struct reiserfs_dir_entry * de)
95 struct reiserfs_de_head * deh = de->de_deh + de->de_entry_num;
97 if (de->de_entry_num >= ih_entry_count (de->de_ih))
100 /* store key of the found entry */
101 de->de_entry_key.version = KEY_FORMAT_3_5;
102 de->de_entry_key.on_disk_key.k_dir_id = le32_to_cpu (de->de_ih->ih_key.k_dir_id);
103 de->de_entry_key.on_disk_key.k_objectid = le32_to_cpu (de->de_ih->ih_key.k_objectid);
104 set_cpu_key_k_offset (&(de->de_entry_key), deh_offset (deh));
105 set_cpu_key_k_type (&(de->de_entry_key), TYPE_DIRENTRY);
109 /* We assign a key to each directory item, and place multiple entries
110 in a single directory item. A directory item has a key equal to the
111 key of the first directory entry in it.
113 This function first calls search_by_key, then, if item whose first
114 entry matches is not found it looks for the entry inside directory
115 item found by search_by_key. Fills the path to the entry, and to the
116 entry position in the item
120 /* The function is NOT SCHEDULE-SAFE! */
121 int search_by_entry_key (struct super_block * sb, const struct cpu_key * key,
122 struct path * path, struct reiserfs_dir_entry * de)
126 retval = search_item (sb, key, path);
129 if (!PATH_LAST_POSITION (path)) {
130 reiserfs_warning (sb, "vs-7000: search_by_entry_key: search_by_key returned item position == 0");
134 PATH_LAST_POSITION (path) --;
144 reiserfs_warning (sb, "vs-7002: search_by_entry_key: no path to here");
148 set_de_item_location (de, path);
150 #ifdef CONFIG_REISERFS_CHECK
151 if (!is_direntry_le_ih (de->de_ih) ||
152 COMP_SHORT_KEYS (&(de->de_ih->ih_key), key)) {
153 print_block (de->de_bh, 0, -1, -1);
154 reiserfs_panic (sb, "vs-7005: search_by_entry_key: found item %h is not directory item or "
155 "does not belong to the same directory as key %K", de->de_ih, key);
157 #endif /* CONFIG_REISERFS_CHECK */
159 /* binary search in directory item by third componen t of the
160 key. sets de->de_entry_num of de */
161 retval = bin_search_in_dir_item (de, cpu_key_k_offset (key));
162 path->pos_in_item = de->de_entry_num;
163 if (retval != NAME_NOT_FOUND) {
164 // ugly, but rename needs de_bh, de_deh, de_name, de_namelen, de_objectid set
165 set_de_name_and_namelen (de);
166 set_de_object_key (de);
173 /* Keyed 32-bit hash function using TEA in a Davis-Meyer function */
175 /* The third component is hashed, and you can choose from more than
176 one hash function. Per directory hashes are not yet implemented
177 but are thought about. This function should be moved to hashes.c
178 Jedi, please do so. -Hans */
180 static __u32 get_third_component (struct super_block * s,
181 const char * name, int len)
185 if (!len || (len == 1 && name[0] == '.'))
187 if (len == 2 && name[0] == '.' && name[1] == '.')
188 return DOT_DOT_OFFSET;
190 res = REISERFS_SB(s)->s_hash_function (name, len);
192 // take bits from 7-th to 30-th including both bounds
193 res = GET_HASH_VALUE(res);
195 // needed to have no names before "." and ".." those have hash
196 // value == 0 and generation conters 1 and 2 accordingly
198 return res + MAX_GENERATION_NUMBER;
202 static int reiserfs_match (struct reiserfs_dir_entry * de,
203 const char * name, int namelen)
205 int retval = NAME_NOT_FOUND;
207 if ((namelen == de->de_namelen) &&
208 !memcmp(de->de_name, name, de->de_namelen))
209 retval = (de_visible (de->de_deh + de->de_entry_num) ? NAME_FOUND : NAME_FOUND_INVISIBLE);
215 /* de's de_bh, de_ih, de_deh, de_item_num, de_entry_num are set already */
217 /* used when hash collisions exist */
220 static int linear_search_in_dir_item (struct cpu_key * key, struct reiserfs_dir_entry * de,
221 const char * name, int namelen)
223 struct reiserfs_de_head * deh = de->de_deh;
227 i = de->de_entry_num;
229 if (i == I_ENTRY_COUNT (de->de_ih) ||
230 GET_HASH_VALUE (deh_offset (deh + i)) != GET_HASH_VALUE (cpu_key_k_offset (key))) {
234 RFALSE( de->de_deh != B_I_DEH (de->de_bh, de->de_ih),
235 "vs-7010: array of entry headers not found");
239 for (; i >= 0; i --, deh --) {
240 if (GET_HASH_VALUE (deh_offset (deh)) !=
241 GET_HASH_VALUE (cpu_key_k_offset (key))) {
242 // hash value does not match, no need to check whole name
243 return NAME_NOT_FOUND;
246 /* mark, that this generation number is used */
247 if (de->de_gen_number_bit_string)
248 set_bit (GET_GENERATION_NUMBER (deh_offset (deh)), (unsigned long *)de->de_gen_number_bit_string);
250 // calculate pointer to name and namelen
251 de->de_entry_num = i;
252 set_de_name_and_namelen (de);
254 if ((retval = reiserfs_match (de, name, namelen)) != NAME_NOT_FOUND) {
255 // de's de_name, de_namelen, de_recordlen are set. Fill the rest:
257 // key of pointed object
258 set_de_object_key (de);
260 store_de_entry_key (de);
262 // retval can be NAME_FOUND or NAME_FOUND_INVISIBLE
267 if (GET_GENERATION_NUMBER (le_ih_k_offset (de->de_ih)) == 0)
268 /* we have reached left most entry in the node. In common we
269 have to go to the left neighbor, but if generation counter
270 is 0 already, we know for sure, that there is no name with
271 the same hash value */
272 // FIXME: this work correctly only because hash value can not
273 // be 0. Btw, in case of Yura's hash it is probably possible,
275 return NAME_NOT_FOUND;
277 RFALSE( de->de_item_num,
278 "vs-7015: two diritems of the same directory in one node?");
280 return GOTO_PREVIOUS_ITEM;
284 // may return NAME_FOUND, NAME_FOUND_INVISIBLE, NAME_NOT_FOUND
285 // FIXME: should add something like IOERROR
286 static int reiserfs_find_entry (struct inode * dir, const char * name, int namelen,
287 struct path * path_to_entry, struct reiserfs_dir_entry * de)
289 struct cpu_key key_to_search;
293 if (namelen > REISERFS_MAX_NAME (dir->i_sb->s_blocksize))
294 return NAME_NOT_FOUND;
296 /* we will search for this key in the tree */
297 make_cpu_key (&key_to_search, dir,
298 get_third_component (dir->i_sb, name, namelen), TYPE_DIRENTRY, 3);
301 retval = search_by_entry_key (dir->i_sb, &key_to_search, path_to_entry, de);
302 if (retval == IO_ERROR) {
303 reiserfs_warning (dir->i_sb, "zam-7001: io error in %s",
308 /* compare names for all entries having given hash value */
309 retval = linear_search_in_dir_item (&key_to_search, de, name, 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;
316 /* there is left neighboring item of this directory and given entry can be there */
317 set_cpu_key_k_offset (&key_to_search, le_ih_k_offset (de->de_ih) - 1);
318 pathrelse (path_to_entry);
324 static struct dentry * reiserfs_lookup (struct inode * dir, struct dentry * dentry, struct nameidata *nd)
327 struct inode * inode = NULL;
328 struct reiserfs_dir_entry de;
329 INITIALIZE_PATH (path_to_entry);
331 if (REISERFS_MAX_NAME (dir->i_sb->s_blocksize) < dentry->d_name.len)
332 return ERR_PTR(-ENAMETOOLONG);
334 reiserfs_write_lock(dir->i_sb);
335 de.de_gen_number_bit_string = NULL;
336 retval = reiserfs_find_entry (dir, dentry->d_name.name, dentry->d_name.len, &path_to_entry, &de);
337 pathrelse (&path_to_entry);
338 if (retval == NAME_FOUND) {
339 /* Hide the .reiserfs_priv directory */
340 if (reiserfs_xattrs (dir->i_sb) &&
341 !old_format_only(dir->i_sb) &&
342 REISERFS_SB(dir->i_sb)->priv_root &&
343 REISERFS_SB(dir->i_sb)->priv_root->d_inode &&
344 de.de_objectid == le32_to_cpu (INODE_PKEY(REISERFS_SB(dir->i_sb)->priv_root->d_inode)->k_objectid)) {
345 reiserfs_write_unlock (dir->i_sb);
346 return ERR_PTR (-EACCES);
349 inode = reiserfs_iget (dir->i_sb, (struct cpu_key *)&(de.de_dir_id));
350 if (!inode || IS_ERR(inode)) {
351 reiserfs_write_unlock(dir->i_sb);
352 return ERR_PTR(-EACCES);
354 vx_propagate_xid(nd, inode);
356 /* Propogate the priv_object flag so we know we're in the priv tree */
357 if (is_reiserfs_priv_object (dir))
358 REISERFS_I(inode)->i_flags |= i_priv_object;
360 reiserfs_write_unlock(dir->i_sb);
361 if ( retval == IO_ERROR ) {
362 return ERR_PTR(-EIO);
366 return d_splice_alias(inode, dentry);
368 d_add(dentry, inode);
374 ** looks up the dentry of the parent directory for child.
375 ** taken from ext2_get_parent
377 struct dentry *reiserfs_get_parent(struct dentry *child)
380 struct inode * inode = NULL;
381 struct reiserfs_dir_entry de;
382 INITIALIZE_PATH (path_to_entry);
383 struct dentry *parent;
384 struct inode *dir = child->d_inode ;
387 if (dir->i_nlink == 0) {
388 return ERR_PTR(-ENOENT);
390 de.de_gen_number_bit_string = NULL;
392 reiserfs_write_lock(dir->i_sb);
393 retval = reiserfs_find_entry (dir, "..", 2, &path_to_entry, &de);
394 pathrelse (&path_to_entry);
395 if (retval != NAME_FOUND) {
396 reiserfs_write_unlock(dir->i_sb);
397 return ERR_PTR(-ENOENT);
399 inode = reiserfs_iget (dir->i_sb, (struct cpu_key *)&(de.de_dir_id));
400 reiserfs_write_unlock(dir->i_sb);
402 if (!inode || IS_ERR(inode)) {
403 return ERR_PTR(-EACCES);
405 parent = d_alloc_anon(inode);
408 parent = ERR_PTR(-ENOMEM);
414 /* add entry to the directory (entry can be hidden).
416 insert definition of when hidden directories are used here -Hans
418 Does not mark dir inode dirty, do it after successesfull call to it */
420 static int reiserfs_add_entry (struct reiserfs_transaction_handle *th, struct inode * dir,
421 const char * name, int namelen, struct inode * inode,
424 struct cpu_key entry_key;
425 struct reiserfs_de_head * deh;
426 INITIALIZE_PATH (path);
427 struct reiserfs_dir_entry de;
428 int bit_string [MAX_GENERATION_NUMBER / (sizeof(int) * 8) + 1];
430 char small_buf[32+DEH_SIZE] ; /* 48 bytes now and we avoid kmalloc
431 if we create file with short name */
433 int buflen, paste_size;
436 BUG_ON (!th->t_trans_id);
438 /* cannot allow items to be added into a busy deleted directory */
442 if (namelen > REISERFS_MAX_NAME (dir->i_sb->s_blocksize))
443 return -ENAMETOOLONG;
445 /* each entry has unique key. compose it */
446 make_cpu_key (&entry_key, dir,
447 get_third_component (dir->i_sb, name, namelen), TYPE_DIRENTRY, 3);
449 /* get memory for composing the entry */
450 buflen = DEH_SIZE + ROUND_UP (namelen);
451 if (buflen > sizeof (small_buf)) {
452 buffer = reiserfs_kmalloc (buflen, GFP_NOFS, dir->i_sb);
458 paste_size = (get_inode_sd_version (dir) == STAT_DATA_V1) ? (DEH_SIZE + namelen) : buflen;
460 /* fill buffer : directory entry head, name[, dir objectid | , stat data | ,stat data, dir objectid ] */
461 deh = (struct reiserfs_de_head *)buffer;
462 deh->deh_location = 0; /* JDM Endian safe if 0 */
463 put_deh_offset( deh, cpu_key_k_offset( &entry_key ) );
464 deh->deh_state = 0; /* JDM Endian safe if 0 */
465 /* put key (ino analog) to de */
466 deh->deh_dir_id = INODE_PKEY (inode)->k_dir_id; /* safe: k_dir_id is le */
467 deh->deh_objectid = INODE_PKEY (inode)->k_objectid; /* safe: k_objectid is le */
470 memcpy ((char *)(deh + 1), name, namelen);
471 /* padd by 0s to the 4 byte boundary */
472 padd_item ((char *)(deh + 1), ROUND_UP (namelen), namelen);
474 /* entry is ready to be pasted into tree, set 'visibility' and 'stat data in entry' attributes */
475 mark_de_without_sd (deh);
476 visible ? mark_de_visible (deh) : mark_de_hidden (deh);
478 /* find the proper place for the new entry */
479 memset (bit_string, 0, sizeof (bit_string));
480 de.de_gen_number_bit_string = (char *)bit_string;
481 retval = reiserfs_find_entry (dir, name, namelen, &path, &de);
482 if( retval != NAME_NOT_FOUND ) {
483 if (buffer != small_buf)
484 reiserfs_kfree (buffer, buflen, dir->i_sb);
487 if ( retval == IO_ERROR ) {
491 if (retval != NAME_FOUND) {
492 reiserfs_warning (dir->i_sb, "zam-7002:%s: \"reiserfs_find_entry\" "
493 "has returned unexpected value (%d)",
494 __FUNCTION__, retval);
500 gen_number = find_first_zero_bit ((unsigned long *)bit_string, MAX_GENERATION_NUMBER + 1);
501 if (gen_number > MAX_GENERATION_NUMBER) {
502 /* there is no free generation number */
503 reiserfs_warning (dir->i_sb, "reiserfs_add_entry: Congratulations! we have got hash function screwed up");
504 if (buffer != small_buf)
505 reiserfs_kfree (buffer, buflen, dir->i_sb);
509 /* adjust offset of directory enrty */
510 put_deh_offset(deh, SET_GENERATION_NUMBER(deh_offset(deh), gen_number));
511 set_cpu_key_k_offset (&entry_key, deh_offset(deh));
513 /* update max-hash-collisions counter in reiserfs_sb_info */
514 PROC_INFO_MAX( th -> t_super, max_hash_collisions, gen_number );
516 if (gen_number != 0) { /* we need to re-search for the insertion point */
517 if (search_by_entry_key (dir->i_sb, &entry_key, &path, &de) != NAME_NOT_FOUND) {
518 reiserfs_warning (dir->i_sb, "vs-7032: reiserfs_add_entry: "
519 "entry with this key (%K) already exists",
522 if (buffer != small_buf)
523 reiserfs_kfree (buffer, buflen, dir->i_sb);
529 /* perform the insertion of the entry that we have prepared */
530 retval = reiserfs_paste_into_item (th, &path, &entry_key, dir, buffer, paste_size);
531 if (buffer != small_buf)
532 reiserfs_kfree (buffer, buflen, dir->i_sb);
534 reiserfs_check_path(&path) ;
538 dir->i_size += paste_size;
539 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
540 if (!S_ISDIR (inode->i_mode) && visible)
541 // reiserfs_mkdir or reiserfs_rename will do that by itself
542 reiserfs_update_sd (th, dir);
544 reiserfs_check_path(&path) ;
548 /* quota utility function, call if you've had to abort after calling
549 ** new_inode_init, and have not called reiserfs_new_inode yet.
550 ** This should only be called on inodes that do not hav stat data
551 ** inserted into the tree yet.
553 static int drop_new_inode(struct inode *inode) {
555 make_bad_inode(inode) ;
556 inode->i_flags |= S_NOQUOTA;
561 /* utility function that does setup for reiserfs_new_inode.
562 ** DQUOT_ALLOC_INODE cannot be called inside a transaction, so we had
563 ** to pull some bits of reiserfs_new_inode out into this func.
564 ** Yes, the actual quota calls are missing, they are part of the quota
567 static int new_inode_init(struct inode *inode, struct inode *dir, int mode) {
569 /* the quota init calls have to know who to charge the quota to, so
570 ** we have to set uid and gid here
572 inode->i_uid = current->fsuid;
573 inode->i_mode = mode;
575 if (dir->i_mode & S_ISGID) {
576 inode->i_gid = dir->i_gid;
578 inode->i_mode |= S_ISGID;
580 inode->i_gid = current->fsgid;
583 if (DQUOT_ALLOC_INODE(inode)) {
584 drop_new_inode(inode);
590 static int reiserfs_create (struct inode * dir, struct dentry *dentry, int mode,
591 struct nameidata *nd)
594 struct inode * inode;
595 int jbegin_count = JOURNAL_PER_BALANCE_CNT * 2 ;
596 struct reiserfs_transaction_handle th ;
599 if (!(inode = new_inode(dir->i_sb))) {
602 retval = new_inode_init(inode, dir, mode);
606 locked = reiserfs_cache_default_acl (dir);
608 reiserfs_write_lock(dir->i_sb);
611 reiserfs_write_lock_xattrs (dir->i_sb);
613 retval = journal_begin(&th, dir->i_sb, jbegin_count);
615 drop_new_inode (inode);
619 retval = reiserfs_new_inode (&th, dir, mode, 0, 0/*i_size*/, dentry, inode);
624 reiserfs_write_unlock_xattrs (dir->i_sb);
628 inode->i_op = &reiserfs_file_inode_operations;
629 inode->i_fop = &reiserfs_file_operations;
630 inode->i_mapping->a_ops = &reiserfs_address_space_operations ;
632 retval = reiserfs_add_entry (&th, dir, dentry->d_name.name, dentry->d_name.len,
633 inode, 1/*visible*/);
637 reiserfs_update_sd (&th, inode);
638 err = journal_end(&th, dir->i_sb, jbegin_count) ;
644 reiserfs_update_inode_transaction(inode) ;
645 reiserfs_update_inode_transaction(dir) ;
647 d_instantiate(dentry, inode);
648 retval = journal_end(&th, dir->i_sb, jbegin_count) ;
652 reiserfs_write_unlock_xattrs (dir->i_sb);
653 reiserfs_write_unlock(dir->i_sb);
658 static int reiserfs_mknod (struct inode * dir, struct dentry *dentry, int mode, dev_t rdev)
661 struct inode * inode;
662 struct reiserfs_transaction_handle th ;
663 int jbegin_count = JOURNAL_PER_BALANCE_CNT * 3;
666 if (!new_valid_dev(rdev))
669 if (!(inode = new_inode(dir->i_sb))) {
672 retval = new_inode_init(inode, dir, mode);
676 locked = reiserfs_cache_default_acl (dir);
678 reiserfs_write_lock(dir->i_sb);
681 reiserfs_write_lock_xattrs (dir->i_sb);
683 retval = journal_begin(&th, dir->i_sb, jbegin_count) ;
685 drop_new_inode (inode);
689 retval = reiserfs_new_inode (&th, dir, mode, NULL, 0/*i_size*/, dentry, inode);
695 reiserfs_write_unlock_xattrs (dir->i_sb);
700 inode->i_op = &reiserfs_special_inode_operations;
701 init_special_inode(inode, inode->i_mode, rdev) ;
703 //FIXME: needed for block and char devices only
704 reiserfs_update_sd (&th, inode);
706 reiserfs_update_inode_transaction(inode) ;
707 reiserfs_update_inode_transaction(dir) ;
709 retval = reiserfs_add_entry (&th, dir, dentry->d_name.name, dentry->d_name.len,
710 inode, 1/*visible*/);
714 reiserfs_update_sd (&th, inode);
715 err = journal_end(&th, dir->i_sb, jbegin_count) ;
722 d_instantiate(dentry, inode);
723 retval = journal_end(&th, dir->i_sb, jbegin_count) ;
727 reiserfs_write_unlock_xattrs (dir->i_sb);
728 reiserfs_write_unlock(dir->i_sb);
733 static int reiserfs_mkdir (struct inode * dir, struct dentry *dentry, int mode)
736 struct inode * inode;
737 struct reiserfs_transaction_handle th ;
738 int jbegin_count = JOURNAL_PER_BALANCE_CNT * 3;
741 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
742 /* set flag that new packing locality created and new blocks for the content * of that directory are not displaced yet */
743 REISERFS_I(dir)->new_packing_locality = 1;
745 mode = S_IFDIR | mode;
746 if (!(inode = new_inode(dir->i_sb))) {
749 retval = new_inode_init(inode, dir, mode);
753 locked = reiserfs_cache_default_acl (dir);
755 reiserfs_write_lock(dir->i_sb);
757 reiserfs_write_lock_xattrs (dir->i_sb);
759 retval = journal_begin(&th, dir->i_sb, jbegin_count) ;
761 drop_new_inode (inode);
766 /* inc the link count now, so another writer doesn't overflow it while
767 ** we sleep later on.
769 INC_DIR_INODE_NLINK(dir)
771 retval = reiserfs_new_inode (&th, dir, mode, NULL/*symlink*/,
772 old_format_only (dir->i_sb) ?
773 EMPTY_DIR_SIZE_V1 : EMPTY_DIR_SIZE,
781 reiserfs_write_unlock_xattrs (dir->i_sb);
785 reiserfs_update_inode_transaction(inode) ;
786 reiserfs_update_inode_transaction(dir) ;
788 inode->i_op = &reiserfs_dir_inode_operations;
789 inode->i_fop = &reiserfs_dir_operations;
791 // note, _this_ add_entry will not update dir's stat data
792 retval = reiserfs_add_entry (&th, dir, dentry->d_name.name, dentry->d_name.len,
793 inode, 1/*visible*/);
797 DEC_DIR_INODE_NLINK(dir);
798 reiserfs_update_sd (&th, inode);
799 err = journal_end(&th, dir->i_sb, jbegin_count) ;
806 // the above add_entry did not update dir's stat data
807 reiserfs_update_sd (&th, dir);
809 d_instantiate(dentry, inode);
810 retval = journal_end(&th, dir->i_sb, jbegin_count) ;
813 reiserfs_write_unlock_xattrs (dir->i_sb);
814 reiserfs_write_unlock(dir->i_sb);
818 static inline int reiserfs_empty_dir(struct inode *inode) {
819 /* we can cheat because an old format dir cannot have
820 ** EMPTY_DIR_SIZE, and a new format dir cannot have
821 ** EMPTY_DIR_SIZE_V1. So, if the inode is either size,
822 ** regardless of disk format version, the directory is empty.
824 if (inode->i_size != EMPTY_DIR_SIZE &&
825 inode->i_size != EMPTY_DIR_SIZE_V1) {
831 static int reiserfs_rmdir (struct inode * dir, struct dentry *dentry)
834 struct inode * inode;
835 struct reiserfs_transaction_handle th ;
837 INITIALIZE_PATH (path);
838 struct reiserfs_dir_entry de;
841 /* we will be doing 2 balancings and update 2 stat data */
842 jbegin_count = JOURNAL_PER_BALANCE_CNT * 2 + 2;
844 reiserfs_write_lock(dir->i_sb);
845 retval = journal_begin(&th, dir->i_sb, jbegin_count) ;
849 de.de_gen_number_bit_string = NULL;
850 if ( (retval = reiserfs_find_entry (dir, dentry->d_name.name, dentry->d_name.len, &path, &de)) == NAME_NOT_FOUND) {
853 } else if ( retval == IO_ERROR) {
858 inode = dentry->d_inode;
860 reiserfs_update_inode_transaction(inode) ;
861 reiserfs_update_inode_transaction(dir) ;
863 if (de.de_objectid != inode->i_ino) {
864 // FIXME: compare key of an object and a key found in the
869 if (!reiserfs_empty_dir(inode)) {
874 /* cut entry from dir directory */
875 retval = reiserfs_cut_from_item (&th, &path, &(de.de_entry_key), dir,
877 0/*new file size - not used here*/);
881 if ( inode->i_nlink != 2 && inode->i_nlink != 1 )
882 reiserfs_warning (inode->i_sb, "%s: empty directory has nlink "
883 "!= 2 (%d)", __FUNCTION__, inode->i_nlink);
886 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
887 reiserfs_update_sd (&th, inode);
889 DEC_DIR_INODE_NLINK(dir)
890 dir->i_size -= (DEH_SIZE + de.de_entrylen);
891 reiserfs_update_sd (&th, dir);
893 /* prevent empty directory from getting lost */
894 add_save_link (&th, inode, 0/* not truncate */);
896 retval = journal_end(&th, dir->i_sb, jbegin_count) ;
897 reiserfs_check_path(&path) ;
899 reiserfs_write_unlock(dir->i_sb);
903 /* we must release path, because we did not call
904 reiserfs_cut_from_item, or reiserfs_cut_from_item does not
905 release path if operation was not complete */
907 err = journal_end(&th, dir->i_sb, jbegin_count) ;
908 reiserfs_write_unlock(dir->i_sb);
909 return err ? err : retval;
912 static int reiserfs_unlink (struct inode * dir, struct dentry *dentry)
915 struct inode * inode;
916 struct reiserfs_dir_entry de;
917 INITIALIZE_PATH (path);
918 struct reiserfs_transaction_handle th ;
920 unsigned long savelink;
922 inode = dentry->d_inode;
924 /* in this transaction we can be doing at max two balancings and update
926 jbegin_count = JOURNAL_PER_BALANCE_CNT * 2 + 2;
928 reiserfs_write_lock(dir->i_sb);
929 retval = journal_begin(&th, dir->i_sb, jbegin_count) ;
933 de.de_gen_number_bit_string = NULL;
934 if ( (retval = reiserfs_find_entry (dir, dentry->d_name.name, dentry->d_name.len, &path, &de)) == NAME_NOT_FOUND) {
937 } else if (retval == IO_ERROR) {
942 reiserfs_update_inode_transaction(inode) ;
943 reiserfs_update_inode_transaction(dir) ;
945 if (de.de_objectid != inode->i_ino) {
946 // FIXME: compare key of an object and a key found in the
952 if (!inode->i_nlink) {
953 reiserfs_warning (inode->i_sb, "%s: deleting nonexistent file "
954 "(%s:%lu), %d", __FUNCTION__,
955 reiserfs_bdevname (inode->i_sb), inode->i_ino,
963 * we schedule before doing the add_save_link call, save the link
964 * count so we don't race
966 savelink = inode->i_nlink;
969 retval = reiserfs_cut_from_item (&th, &path, &(de.de_entry_key), dir, NULL, 0);
974 inode->i_ctime = CURRENT_TIME;
975 reiserfs_update_sd (&th, inode);
977 dir->i_size -= (de.de_entrylen + DEH_SIZE);
978 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
979 reiserfs_update_sd (&th, dir);
982 /* prevent file from getting lost */
983 add_save_link (&th, inode, 0/* not truncate */);
985 retval = journal_end(&th, dir->i_sb, jbegin_count) ;
986 reiserfs_check_path(&path) ;
987 reiserfs_write_unlock(dir->i_sb);
992 err = journal_end(&th, dir->i_sb, jbegin_count) ;
993 reiserfs_check_path(&path) ;
997 reiserfs_write_unlock(dir->i_sb);
1001 static int reiserfs_symlink (struct inode * parent_dir,
1002 struct dentry * dentry, const char * symname)
1005 struct inode * inode;
1008 struct reiserfs_transaction_handle th ;
1009 int mode = S_IFLNK | S_IRWXUGO;
1010 int jbegin_count = JOURNAL_PER_BALANCE_CNT * 3;
1012 if (!(inode = new_inode(parent_dir->i_sb))) {
1015 retval = new_inode_init(inode, parent_dir, mode);
1020 reiserfs_write_lock(parent_dir->i_sb);
1021 item_len = ROUND_UP (strlen (symname));
1022 if (item_len > MAX_DIRECT_ITEM_LEN (parent_dir->i_sb->s_blocksize)) {
1023 retval = -ENAMETOOLONG;
1024 drop_new_inode(inode);
1028 name = reiserfs_kmalloc (item_len, GFP_NOFS, parent_dir->i_sb);
1030 drop_new_inode(inode);
1034 memcpy (name, symname, strlen (symname));
1035 padd_item (name, item_len, strlen (symname));
1037 /* We would inherit the default ACL here, but symlinks don't get ACLs */
1039 retval = journal_begin(&th, parent_dir->i_sb, jbegin_count) ;
1041 drop_new_inode (inode);
1042 reiserfs_kfree (name, item_len, parent_dir->i_sb);
1046 retval = reiserfs_new_inode (&th, parent_dir, mode, name, strlen (symname),
1048 reiserfs_kfree (name, item_len, parent_dir->i_sb);
1049 if (retval) { /* reiserfs_new_inode iputs for us */
1053 reiserfs_update_inode_transaction(inode) ;
1054 reiserfs_update_inode_transaction(parent_dir) ;
1056 inode->i_op = &reiserfs_symlink_inode_operations;
1057 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1059 // must be sure this inode is written with this transaction
1061 //reiserfs_update_sd (&th, inode, READ_BLOCKS);
1063 retval = reiserfs_add_entry (&th, parent_dir, dentry->d_name.name,
1064 dentry->d_name.len, inode, 1/*visible*/);
1068 reiserfs_update_sd (&th, inode);
1069 err = journal_end(&th, parent_dir->i_sb, jbegin_count) ;
1076 d_instantiate(dentry, inode);
1077 retval = journal_end(&th, parent_dir->i_sb, jbegin_count) ;
1079 reiserfs_write_unlock(parent_dir->i_sb);
1083 static int reiserfs_link (struct dentry * old_dentry, struct inode * dir, struct dentry * dentry)
1086 struct inode *inode = old_dentry->d_inode;
1087 struct reiserfs_transaction_handle th ;
1088 int jbegin_count = JOURNAL_PER_BALANCE_CNT * 3;
1090 reiserfs_write_lock(dir->i_sb);
1091 if (inode->i_nlink >= REISERFS_LINK_MAX) {
1092 //FIXME: sd_nlink is 32 bit for new files
1093 reiserfs_write_unlock(dir->i_sb);
1096 if (inode->i_nlink == 0) {
1097 reiserfs_write_unlock(dir->i_sb);
1101 /* inc before scheduling so reiserfs_unlink knows we are here */
1104 retval = journal_begin(&th, dir->i_sb, jbegin_count) ;
1107 reiserfs_write_unlock (dir->i_sb);
1111 /* create new entry */
1112 retval = reiserfs_add_entry (&th, dir, dentry->d_name.name, dentry->d_name.len,
1113 inode, 1/*visible*/);
1115 reiserfs_update_inode_transaction(inode) ;
1116 reiserfs_update_inode_transaction(dir) ;
1121 err = journal_end(&th, dir->i_sb, jbegin_count) ;
1122 reiserfs_write_unlock(dir->i_sb);
1123 return err ? err : retval;
1126 inode->i_ctime = CURRENT_TIME;
1127 reiserfs_update_sd (&th, inode);
1129 atomic_inc(&inode->i_count) ;
1130 d_instantiate(dentry, inode);
1131 retval = journal_end(&th, dir->i_sb, jbegin_count) ;
1132 reiserfs_write_unlock(dir->i_sb);
1137 // de contains information pointing to an entry which
1138 static int de_still_valid (const char * name, int len, struct reiserfs_dir_entry * de)
1140 struct reiserfs_dir_entry tmp = *de;
1142 // recalculate pointer to name and name length
1143 set_de_name_and_namelen (&tmp);
1144 // FIXME: could check more
1145 if (tmp.de_namelen != len || memcmp (name, de->de_name, len))
1151 static int entry_points_to_object (const char * name, int len, struct reiserfs_dir_entry * de, struct inode * inode)
1153 if (!de_still_valid (name, len, de))
1157 if (!de_visible (de->de_deh + de->de_entry_num))
1158 reiserfs_panic (NULL, "vs-7042: entry_points_to_object: entry must be visible");
1159 return (de->de_objectid == inode->i_ino) ? 1 : 0;
1162 /* this must be added hidden entry */
1163 if (de_visible (de->de_deh + de->de_entry_num))
1164 reiserfs_panic (NULL, "vs-7043: entry_points_to_object: entry must be visible");
1170 /* sets key of objectid the entry has to point to */
1171 static void set_ino_in_dir_entry (struct reiserfs_dir_entry * de, struct reiserfs_key * key)
1173 /* JDM These operations are endian safe - both are le */
1174 de->de_deh[de->de_entry_num].deh_dir_id = key->k_dir_id;
1175 de->de_deh[de->de_entry_num].deh_objectid = key->k_objectid;
1180 * process, that is going to call fix_nodes/do_balance must hold only
1181 * one path. If it holds 2 or more, it can get into endless waiting in
1182 * get_empty_nodes or its clones
1184 static int reiserfs_rename (struct inode * old_dir, struct dentry *old_dentry,
1185 struct inode * new_dir, struct dentry *new_dentry)
1188 INITIALIZE_PATH (old_entry_path);
1189 INITIALIZE_PATH (new_entry_path);
1190 INITIALIZE_PATH (dot_dot_entry_path);
1191 struct item_head new_entry_ih, old_entry_ih, dot_dot_ih ;
1192 struct reiserfs_dir_entry old_de, new_de, dot_dot_de;
1193 struct inode * old_inode, * new_dentry_inode;
1194 struct reiserfs_transaction_handle th ;
1196 umode_t old_inode_mode;
1197 unsigned long savelink = 1;
1198 struct timespec ctime;
1200 /* three balancings: (1) old name removal, (2) new name insertion
1201 and (3) maybe "save" link insertion
1202 stat data updates: (1) old directory,
1203 (2) new directory and (3) maybe old object stat data (when it is
1204 directory) and (4) maybe stat data of object to which new entry
1205 pointed initially and (5) maybe block containing ".." of
1206 renamed directory */
1207 jbegin_count = JOURNAL_PER_BALANCE_CNT * 3 + 5;
1209 old_inode = old_dentry->d_inode;
1210 new_dentry_inode = new_dentry->d_inode;
1212 // make sure, that oldname still exists and points to an object we
1213 // are going to rename
1214 old_de.de_gen_number_bit_string = NULL;
1215 reiserfs_write_lock(old_dir->i_sb);
1216 retval = reiserfs_find_entry (old_dir, old_dentry->d_name.name, old_dentry->d_name.len,
1217 &old_entry_path, &old_de);
1218 pathrelse (&old_entry_path);
1219 if (retval == IO_ERROR) {
1220 reiserfs_write_unlock(old_dir->i_sb);
1224 if (retval != NAME_FOUND || old_de.de_objectid != old_inode->i_ino) {
1225 reiserfs_write_unlock(old_dir->i_sb);
1229 old_inode_mode = old_inode->i_mode;
1230 if (S_ISDIR(old_inode_mode)) {
1231 // make sure, that directory being renamed has correct ".."
1232 // and that its new parent directory has not too many links
1235 if (new_dentry_inode) {
1236 if (!reiserfs_empty_dir(new_dentry_inode)) {
1237 reiserfs_write_unlock(old_dir->i_sb);
1242 /* directory is renamed, its parent directory will be changed,
1243 ** so find ".." entry
1245 dot_dot_de.de_gen_number_bit_string = NULL;
1246 retval = reiserfs_find_entry (old_inode, "..", 2, &dot_dot_entry_path, &dot_dot_de);
1247 pathrelse (&dot_dot_entry_path);
1248 if (retval != NAME_FOUND) {
1249 reiserfs_write_unlock(old_dir->i_sb);
1253 /* inode number of .. must equal old_dir->i_ino */
1254 if (dot_dot_de.de_objectid != old_dir->i_ino) {
1255 reiserfs_write_unlock(old_dir->i_sb);
1260 retval = journal_begin(&th, old_dir->i_sb, jbegin_count) ;
1262 reiserfs_write_unlock (old_dir->i_sb);
1267 /* add new entry (or find the existing one) */
1268 retval = reiserfs_add_entry (&th, new_dir, new_dentry->d_name.name, new_dentry->d_name.len,
1270 if (retval == -EEXIST) {
1271 if (!new_dentry_inode) {
1272 reiserfs_panic (old_dir->i_sb,
1273 "vs-7050: new entry is found, new inode == 0\n");
1275 } else if (retval) {
1276 int err = journal_end(&th, old_dir->i_sb, jbegin_count) ;
1277 reiserfs_write_unlock(old_dir->i_sb);
1278 return err ? err : retval;
1281 reiserfs_update_inode_transaction(old_dir) ;
1282 reiserfs_update_inode_transaction(new_dir) ;
1284 /* this makes it so an fsync on an open fd for the old name will
1285 ** commit the rename operation
1287 reiserfs_update_inode_transaction(old_inode) ;
1289 if (new_dentry_inode)
1290 reiserfs_update_inode_transaction(new_dentry_inode) ;
1293 // look for old name using corresponding entry key (found by reiserfs_find_entry)
1294 if (search_by_entry_key (new_dir->i_sb, &old_de.de_entry_key, &old_entry_path, &old_de) != NAME_FOUND)
1297 copy_item_head(&old_entry_ih, get_ih(&old_entry_path)) ;
1299 reiserfs_prepare_for_journal(old_inode->i_sb, old_de.de_bh, 1) ;
1301 // look for new name by reiserfs_find_entry
1302 new_de.de_gen_number_bit_string = NULL;
1303 retval = reiserfs_find_entry (new_dir, new_dentry->d_name.name, new_dentry->d_name.len,
1304 &new_entry_path, &new_de);
1305 // reiserfs_add_entry should not return IO_ERROR, because it is called with essentially same parameters from
1306 // reiserfs_add_entry above, and we'll catch any i/o errors before we get here.
1307 if (retval != NAME_FOUND_INVISIBLE && retval != NAME_FOUND)
1310 copy_item_head(&new_entry_ih, get_ih(&new_entry_path)) ;
1312 reiserfs_prepare_for_journal(old_inode->i_sb, new_de.de_bh, 1) ;
1314 if (S_ISDIR(old_inode->i_mode)) {
1315 if (search_by_entry_key (new_dir->i_sb, &dot_dot_de.de_entry_key, &dot_dot_entry_path, &dot_dot_de) != NAME_FOUND)
1317 copy_item_head(&dot_dot_ih, get_ih(&dot_dot_entry_path)) ;
1318 // node containing ".." gets into transaction
1319 reiserfs_prepare_for_journal(old_inode->i_sb, dot_dot_de.de_bh, 1) ;
1321 /* we should check seals here, not do
1322 this stuff, yes? Then, having
1323 gathered everything into RAM we
1324 should lock the buffers, yes? -Hans */
1325 /* probably. our rename needs to hold more
1326 ** than one path at once. The seals would
1327 ** have to be written to deal with multi-path
1330 /* sanity checking before doing the rename - avoid races many
1331 ** of the above checks could have scheduled. We have to be
1332 ** sure our items haven't been shifted by another process.
1334 if (item_moved(&new_entry_ih, &new_entry_path) ||
1335 !entry_points_to_object(new_dentry->d_name.name,
1336 new_dentry->d_name.len,
1337 &new_de, new_dentry_inode) ||
1338 item_moved(&old_entry_ih, &old_entry_path) ||
1339 !entry_points_to_object (old_dentry->d_name.name,
1340 old_dentry->d_name.len,
1341 &old_de, old_inode)) {
1342 reiserfs_restore_prepared_buffer (old_inode->i_sb, new_de.de_bh);
1343 reiserfs_restore_prepared_buffer (old_inode->i_sb, old_de.de_bh);
1344 if (S_ISDIR(old_inode_mode))
1345 reiserfs_restore_prepared_buffer (old_inode->i_sb, dot_dot_de.de_bh);
1348 if (S_ISDIR(old_inode_mode)) {
1349 if ( item_moved(&dot_dot_ih, &dot_dot_entry_path) ||
1350 !entry_points_to_object ( "..", 2, &dot_dot_de, old_dir) ) {
1351 reiserfs_restore_prepared_buffer (old_inode->i_sb, old_de.de_bh);
1352 reiserfs_restore_prepared_buffer (old_inode->i_sb, new_de.de_bh);
1353 reiserfs_restore_prepared_buffer (old_inode->i_sb, dot_dot_de.de_bh);
1358 RFALSE( S_ISDIR(old_inode_mode) &&
1359 !buffer_journal_prepared(dot_dot_de.de_bh), "" );
1364 /* ok, all the changes can be done in one fell swoop when we
1365 have claimed all the buffers needed.*/
1367 mark_de_visible (new_de.de_deh + new_de.de_entry_num);
1368 set_ino_in_dir_entry (&new_de, INODE_PKEY (old_inode));
1369 journal_mark_dirty (&th, old_dir->i_sb, new_de.de_bh);
1371 mark_de_hidden (old_de.de_deh + old_de.de_entry_num);
1372 journal_mark_dirty (&th, old_dir->i_sb, old_de.de_bh);
1373 ctime = CURRENT_TIME;
1374 old_dir->i_ctime = old_dir->i_mtime = ctime;
1375 new_dir->i_ctime = new_dir->i_mtime = ctime;
1376 /* thanks to Alex Adriaanse <alex_a@caltech.edu> for patch which adds ctime update of
1378 old_inode->i_ctime = ctime;
1380 if (new_dentry_inode) {
1381 // adjust link number of the victim
1382 if (S_ISDIR(new_dentry_inode->i_mode)) {
1383 new_dentry_inode->i_nlink = 0;
1385 new_dentry_inode->i_nlink--;
1387 new_dentry_inode->i_ctime = ctime;
1388 savelink = new_dentry_inode->i_nlink;
1391 if (S_ISDIR(old_inode_mode)) {
1392 // adjust ".." of renamed directory
1393 set_ino_in_dir_entry (&dot_dot_de, INODE_PKEY (new_dir));
1394 journal_mark_dirty (&th, new_dir->i_sb, dot_dot_de.de_bh);
1396 if (!new_dentry_inode)
1397 /* there (in new_dir) was no directory, so it got new link
1398 (".." of renamed directory) */
1399 INC_DIR_INODE_NLINK(new_dir);
1401 /* old directory lost one link - ".. " of renamed directory */
1402 DEC_DIR_INODE_NLINK(old_dir);
1405 // looks like in 2.3.99pre3 brelse is atomic. so we can use pathrelse
1406 pathrelse (&new_entry_path);
1407 pathrelse (&dot_dot_entry_path);
1409 // FIXME: this reiserfs_cut_from_item's return value may screw up
1410 // anybody, but it will panic if will not be able to find the
1411 // entry. This needs one more clean up
1412 if (reiserfs_cut_from_item (&th, &old_entry_path, &(old_de.de_entry_key), old_dir, NULL, 0) < 0)
1413 reiserfs_warning (old_dir->i_sb, "vs-7060: reiserfs_rename: couldn't not cut old name. Fsck later?");
1415 old_dir->i_size -= DEH_SIZE + old_de.de_entrylen;
1417 reiserfs_update_sd (&th, old_dir);
1418 reiserfs_update_sd (&th, new_dir);
1419 reiserfs_update_sd (&th, old_inode);
1421 if (new_dentry_inode) {
1423 add_save_link (&th, new_dentry_inode, 0/* not truncate */);
1424 reiserfs_update_sd (&th, new_dentry_inode);
1427 retval = journal_end(&th, old_dir->i_sb, jbegin_count) ;
1428 reiserfs_write_unlock(old_dir->i_sb);
1433 * directories can handle most operations...
1435 struct inode_operations reiserfs_dir_inode_operations = {
1436 //&reiserfs_dir_operations, /* default_file_ops */
1437 .create = reiserfs_create,
1438 .lookup = reiserfs_lookup,
1439 .link = reiserfs_link,
1440 .unlink = reiserfs_unlink,
1441 .symlink = reiserfs_symlink,
1442 .mkdir = reiserfs_mkdir,
1443 .rmdir = reiserfs_rmdir,
1444 .mknod = reiserfs_mknod,
1445 .rename = reiserfs_rename,
1446 .setattr = reiserfs_setattr,
1447 .setxattr = reiserfs_setxattr,
1448 .getxattr = reiserfs_getxattr,
1449 .listxattr = reiserfs_listxattr,
1450 .removexattr = reiserfs_removexattr,
1451 .permission = reiserfs_permission,
1455 * symlink operations.. same as page_symlink_inode_operations, with xattr
1458 struct inode_operations reiserfs_symlink_inode_operations = {
1459 .readlink = generic_readlink,
1460 .follow_link = page_follow_link_light,
1461 .put_link = page_put_link,
1462 .setattr = reiserfs_setattr,
1463 .setxattr = reiserfs_setxattr,
1464 .getxattr = reiserfs_getxattr,
1465 .listxattr = reiserfs_listxattr,
1466 .removexattr = reiserfs_removexattr,
1467 .permission = reiserfs_permission,
1473 * special file operations.. just xattr/acl stuff
1475 struct inode_operations reiserfs_special_inode_operations = {
1476 .setattr = reiserfs_setattr,
1477 .setxattr = reiserfs_setxattr,
1478 .getxattr = reiserfs_getxattr,
1479 .listxattr = reiserfs_listxattr,
1480 .removexattr = reiserfs_removexattr,
1481 .permission = reiserfs_permission,