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/time.h>
15 #include <linux/bitops.h>
16 #include <linux/slab.h>
20 #include <linux/quotaops.h>
22 #define INC_DIR_INODE_NLINK(i) if (i->i_nlink != 1) { inc_nlink(i); if (i->i_nlink >= REISERFS_LINK_MAX) set_nlink(i, 1); }
23 #define DEC_DIR_INODE_NLINK(i) if (i->i_nlink != 1) drop_nlink(i);
26 * directory item contains array of entry headers. This performs
27 * 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 = ih_entry_count(ih) - 1;
38 for (j = (rbound + lbound) / 2; lbound <= rbound;
39 j = (rbound + lbound) / 2) {
40 if (off < deh_offset(deh + j)) {
44 if (off > deh_offset(deh + j)) {
48 /* this is not name found, but matched third key component */
53 de->de_entry_num = lbound;
54 return NAME_NOT_FOUND;
58 * comment? maybe something like set de to point to what the path points to?
60 static inline void set_de_item_location(struct reiserfs_dir_entry *de,
61 struct treepath *path)
63 de->de_bh = get_last_bh(path);
64 de->de_ih = tp_item_head(path);
65 de->de_deh = B_I_DEH(de->de_bh, de->de_ih);
66 de->de_item_num = PATH_LAST_POSITION(path);
70 * de_bh, de_ih, de_deh (points to first element of array), de_item_num is set
72 inline void set_de_name_and_namelen(struct reiserfs_dir_entry *de)
74 struct reiserfs_de_head *deh = de->de_deh + de->de_entry_num;
76 BUG_ON(de->de_entry_num >= ih_entry_count(de->de_ih));
78 de->de_entrylen = entry_length(de->de_bh, de->de_ih, de->de_entry_num);
79 de->de_namelen = de->de_entrylen - (de_with_sd(deh) ? SD_SIZE : 0);
80 de->de_name = ih_item_body(de->de_bh, de->de_ih) + deh_location(deh);
81 if (de->de_name[de->de_namelen - 1] == 0)
82 de->de_namelen = strlen(de->de_name);
85 /* what entry points to */
86 static inline void set_de_object_key(struct reiserfs_dir_entry *de)
88 BUG_ON(de->de_entry_num >= ih_entry_count(de->de_ih));
89 de->de_dir_id = deh_dir_id(&de->de_deh[de->de_entry_num]);
90 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 BUG_ON(de->de_entry_num >= ih_entry_count(de->de_ih));
99 /* store key of the found entry */
100 de->de_entry_key.version = KEY_FORMAT_3_5;
101 de->de_entry_key.on_disk_key.k_dir_id =
102 le32_to_cpu(de->de_ih->ih_key.k_dir_id);
103 de->de_entry_key.on_disk_key.k_objectid =
104 le32_to_cpu(de->de_ih->ih_key.k_objectid);
105 set_cpu_key_k_offset(&de->de_entry_key, deh_offset(deh));
106 set_cpu_key_k_type(&de->de_entry_key, TYPE_DIRENTRY);
110 * We assign a key to each directory item, and place multiple entries in a
111 * single directory item. A directory item has a key equal to the key of
112 * the first directory entry in it.
114 * This function first calls search_by_key, then, if item whose first entry
115 * matches is not found it looks for the entry inside directory item found
116 * by search_by_key. Fills the path to the entry, and to the entry position
119 /* The function is NOT SCHEDULE-SAFE! */
120 int search_by_entry_key(struct super_block *sb, const struct cpu_key *key,
121 struct treepath *path, struct reiserfs_dir_entry *de)
125 retval = search_item(sb, key, path);
128 if (!PATH_LAST_POSITION(path)) {
129 reiserfs_error(sb, "vs-7000", "search_by_key "
130 "returned item position == 0");
134 PATH_LAST_POSITION(path)--;
144 reiserfs_error(sb, "vs-7002", "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", "found item %h is not directory "
155 "item or does not belong to the same directory "
156 "as key %K", de->de_ih, key);
158 #endif /* CONFIG_REISERFS_CHECK */
161 * binary search in directory item by third component of the
162 * key. sets de->de_entry_num of de
164 retval = bin_search_in_dir_item(de, cpu_key_k_offset(key));
165 path->pos_in_item = de->de_entry_num;
166 if (retval != NAME_NOT_FOUND) {
168 * ugly, but rename needs de_bh, de_deh, de_name,
169 * de_namelen, de_objectid set
171 set_de_name_and_namelen(de);
172 set_de_object_key(de);
177 /* Keyed 32-bit hash function using TEA in a Davis-Meyer function */
180 * The third component is hashed, and you can choose from more than
181 * one hash function. Per directory hashes are not yet implemented
182 * but are thought about. This function should be moved to hashes.c
183 * Jedi, please do so. -Hans
185 static __u32 get_third_component(struct super_block *s,
186 const char *name, int len)
190 if (!len || (len == 1 && name[0] == '.'))
192 if (len == 2 && name[0] == '.' && name[1] == '.')
193 return DOT_DOT_OFFSET;
195 res = REISERFS_SB(s)->s_hash_function(name, len);
197 /* take bits from 7-th to 30-th including both bounds */
198 res = GET_HASH_VALUE(res);
201 * needed to have no names before "." and ".." those have hash
202 * value == 0 and generation conters 1 and 2 accordingly
205 return res + MAX_GENERATION_NUMBER;
208 static int reiserfs_match(struct reiserfs_dir_entry *de,
209 const char *name, int namelen)
211 int retval = NAME_NOT_FOUND;
213 if ((namelen == de->de_namelen) &&
214 !memcmp(de->de_name, name, de->de_namelen))
216 (de_visible(de->de_deh + de->de_entry_num) ? NAME_FOUND :
217 NAME_FOUND_INVISIBLE);
222 /* de's de_bh, de_ih, de_deh, de_item_num, de_entry_num are set already */
224 /* used when hash collisions exist */
226 static int linear_search_in_dir_item(struct cpu_key *key,
227 struct reiserfs_dir_entry *de,
228 const char *name, int namelen)
230 struct reiserfs_de_head *deh = de->de_deh;
234 i = de->de_entry_num;
236 if (i == ih_entry_count(de->de_ih) ||
237 GET_HASH_VALUE(deh_offset(deh + i)) !=
238 GET_HASH_VALUE(cpu_key_k_offset(key))) {
242 RFALSE(de->de_deh != B_I_DEH(de->de_bh, de->de_ih),
243 "vs-7010: array of entry headers not found");
247 for (; i >= 0; i--, deh--) {
248 /* hash value does not match, no need to check whole name */
249 if (GET_HASH_VALUE(deh_offset(deh)) !=
250 GET_HASH_VALUE(cpu_key_k_offset(key))) {
251 return NAME_NOT_FOUND;
254 /* mark that this generation number is used */
255 if (de->de_gen_number_bit_string)
256 set_bit(GET_GENERATION_NUMBER(deh_offset(deh)),
257 de->de_gen_number_bit_string);
259 /* calculate pointer to name and namelen */
260 de->de_entry_num = i;
261 set_de_name_and_namelen(de);
264 * de's de_name, de_namelen, de_recordlen are set.
268 reiserfs_match(de, name, namelen)) != NAME_NOT_FOUND) {
270 /* key of pointed object */
271 set_de_object_key(de);
273 store_de_entry_key(de);
275 /* retval can be NAME_FOUND or NAME_FOUND_INVISIBLE */
280 if (GET_GENERATION_NUMBER(le_ih_k_offset(de->de_ih)) == 0)
282 * we have reached left most entry in the node. In common we
283 * have to go to the left neighbor, but if generation counter
284 * is 0 already, we know for sure, that there is no name with
285 * the same hash value
288 * FIXME: this work correctly only because hash value can not
289 * be 0. Btw, in case of Yura's hash it is probably possible,
292 return NAME_NOT_FOUND;
294 RFALSE(de->de_item_num,
295 "vs-7015: two diritems of the same directory in one node?");
297 return GOTO_PREVIOUS_ITEM;
301 * may return NAME_FOUND, NAME_FOUND_INVISIBLE, NAME_NOT_FOUND
302 * FIXME: should add something like IOERROR
304 static int reiserfs_find_entry(struct inode *dir, const char *name, int namelen,
305 struct treepath *path_to_entry,
306 struct reiserfs_dir_entry *de)
308 struct cpu_key key_to_search;
311 if (namelen > REISERFS_MAX_NAME(dir->i_sb->s_blocksize))
312 return NAME_NOT_FOUND;
314 /* we will search for this key in the tree */
315 make_cpu_key(&key_to_search, dir,
316 get_third_component(dir->i_sb, name, namelen),
321 search_by_entry_key(dir->i_sb, &key_to_search,
323 if (retval == IO_ERROR) {
324 reiserfs_error(dir->i_sb, "zam-7001", "io error");
328 /* compare names for all entries having given hash value */
330 linear_search_in_dir_item(&key_to_search, de, name,
333 * there is no need to scan directory anymore.
334 * Given entry found or does not exist
336 if (retval != GOTO_PREVIOUS_ITEM) {
337 path_to_entry->pos_in_item = de->de_entry_num;
342 * there is left neighboring item of this directory
343 * and given entry can be there
345 set_cpu_key_k_offset(&key_to_search,
346 le_ih_k_offset(de->de_ih) - 1);
347 pathrelse(path_to_entry);
352 static struct dentry *reiserfs_lookup(struct inode *dir, struct dentry *dentry,
356 struct inode *inode = NULL;
357 struct reiserfs_dir_entry de;
358 INITIALIZE_PATH(path_to_entry);
360 if (REISERFS_MAX_NAME(dir->i_sb->s_blocksize) < dentry->d_name.len)
361 return ERR_PTR(-ENAMETOOLONG);
363 reiserfs_write_lock(dir->i_sb);
365 de.de_gen_number_bit_string = NULL;
367 reiserfs_find_entry(dir, dentry->d_name.name, dentry->d_name.len,
368 &path_to_entry, &de);
369 pathrelse(&path_to_entry);
370 if (retval == NAME_FOUND) {
371 inode = reiserfs_iget(dir->i_sb,
372 (struct cpu_key *)&de.de_dir_id);
373 if (!inode || IS_ERR(inode)) {
374 reiserfs_write_unlock(dir->i_sb);
375 return ERR_PTR(-EACCES);
379 * Propagate the private flag so we know we're
383 inode->i_flags |= S_PRIVATE;
385 reiserfs_write_unlock(dir->i_sb);
386 if (retval == IO_ERROR) {
387 return ERR_PTR(-EIO);
390 return d_splice_alias(inode, dentry);
394 * looks up the dentry of the parent directory for child.
395 * taken from ext2_get_parent
397 struct dentry *reiserfs_get_parent(struct dentry *child)
400 struct inode *inode = NULL;
401 struct reiserfs_dir_entry de;
402 INITIALIZE_PATH(path_to_entry);
403 struct inode *dir = d_inode(child);
405 if (dir->i_nlink == 0) {
406 return ERR_PTR(-ENOENT);
408 de.de_gen_number_bit_string = NULL;
410 reiserfs_write_lock(dir->i_sb);
411 retval = reiserfs_find_entry(dir, "..", 2, &path_to_entry, &de);
412 pathrelse(&path_to_entry);
413 if (retval != NAME_FOUND) {
414 reiserfs_write_unlock(dir->i_sb);
415 return ERR_PTR(-ENOENT);
417 inode = reiserfs_iget(dir->i_sb, (struct cpu_key *)&de.de_dir_id);
418 reiserfs_write_unlock(dir->i_sb);
420 return d_obtain_alias(inode);
423 /* add entry to the directory (entry can be hidden).
425 insert definition of when hidden directories are used here -Hans
427 Does not mark dir inode dirty, do it after successesfull call to it */
429 static int reiserfs_add_entry(struct reiserfs_transaction_handle *th,
430 struct inode *dir, const char *name, int namelen,
431 struct inode *inode, int visible)
433 struct cpu_key entry_key;
434 struct reiserfs_de_head *deh;
435 INITIALIZE_PATH(path);
436 struct reiserfs_dir_entry de;
437 DECLARE_BITMAP(bit_string, MAX_GENERATION_NUMBER + 1);
441 * 48 bytes now and we avoid kmalloc if we
442 * create file with short name
444 char small_buf[32 + DEH_SIZE];
447 int buflen, paste_size;
450 BUG_ON(!th->t_trans_id);
452 /* cannot allow items to be added into a busy deleted directory */
456 if (namelen > REISERFS_MAX_NAME(dir->i_sb->s_blocksize))
457 return -ENAMETOOLONG;
459 /* each entry has unique key. compose it */
460 make_cpu_key(&entry_key, dir,
461 get_third_component(dir->i_sb, name, namelen),
464 /* get memory for composing the entry */
465 buflen = DEH_SIZE + ROUND_UP(namelen);
466 if (buflen > sizeof(small_buf)) {
467 buffer = kmalloc(buflen, GFP_NOFS);
474 (get_inode_sd_version(dir) ==
475 STAT_DATA_V1) ? (DEH_SIZE + namelen) : buflen;
478 * fill buffer : directory entry head, name[, dir objectid | ,
479 * stat data | ,stat data, dir objectid ]
481 deh = (struct reiserfs_de_head *)buffer;
482 deh->deh_location = 0; /* JDM Endian safe if 0 */
483 put_deh_offset(deh, cpu_key_k_offset(&entry_key));
484 deh->deh_state = 0; /* JDM Endian safe if 0 */
485 /* put key (ino analog) to de */
487 /* safe: k_dir_id is le */
488 deh->deh_dir_id = INODE_PKEY(inode)->k_dir_id;
489 /* safe: k_objectid is le */
490 deh->deh_objectid = INODE_PKEY(inode)->k_objectid;
493 memcpy((char *)(deh + 1), name, namelen);
494 /* padd by 0s to the 4 byte boundary */
495 padd_item((char *)(deh + 1), ROUND_UP(namelen), namelen);
498 * entry is ready to be pasted into tree, set 'visibility'
499 * and 'stat data in entry' attributes
501 mark_de_without_sd(deh);
502 visible ? mark_de_visible(deh) : mark_de_hidden(deh);
504 /* find the proper place for the new entry */
505 memset(bit_string, 0, sizeof(bit_string));
506 de.de_gen_number_bit_string = bit_string;
507 retval = reiserfs_find_entry(dir, name, namelen, &path, &de);
508 if (retval != NAME_NOT_FOUND) {
509 if (buffer != small_buf)
513 if (retval == IO_ERROR) {
517 if (retval != NAME_FOUND) {
518 reiserfs_error(dir->i_sb, "zam-7002",
519 "reiserfs_find_entry() returned "
520 "unexpected value (%d)", retval);
527 find_first_zero_bit(bit_string,
528 MAX_GENERATION_NUMBER + 1);
529 if (gen_number > MAX_GENERATION_NUMBER) {
530 /* there is no free generation number */
531 reiserfs_warning(dir->i_sb, "reiserfs-7010",
532 "Congratulations! we have got hash function "
534 if (buffer != small_buf)
539 /* adjust offset of directory enrty */
540 put_deh_offset(deh, SET_GENERATION_NUMBER(deh_offset(deh), gen_number));
541 set_cpu_key_k_offset(&entry_key, deh_offset(deh));
543 /* update max-hash-collisions counter in reiserfs_sb_info */
544 PROC_INFO_MAX(th->t_super, max_hash_collisions, gen_number);
546 /* we need to re-search for the insertion point */
547 if (gen_number != 0) {
548 if (search_by_entry_key(dir->i_sb, &entry_key, &path, &de) !=
550 reiserfs_warning(dir->i_sb, "vs-7032",
551 "entry with this key (%K) already "
552 "exists", &entry_key);
554 if (buffer != small_buf)
561 /* perform the insertion of the entry that we have prepared */
563 reiserfs_paste_into_item(th, &path, &entry_key, dir, buffer,
565 if (buffer != small_buf)
568 reiserfs_check_path(&path);
572 dir->i_size += paste_size;
573 dir->i_mtime = dir->i_ctime = CURRENT_TIME_SEC;
574 if (!S_ISDIR(inode->i_mode) && visible)
575 /* reiserfs_mkdir or reiserfs_rename will do that by itself */
576 reiserfs_update_sd(th, dir);
578 reiserfs_check_path(&path);
583 * quota utility function, call if you've had to abort after calling
584 * new_inode_init, and have not called reiserfs_new_inode yet.
585 * This should only be called on inodes that do not have stat data
586 * inserted into the tree yet.
588 static int drop_new_inode(struct inode *inode)
591 make_bad_inode(inode);
592 inode->i_flags |= S_NOQUOTA;
598 * utility function that does setup for reiserfs_new_inode.
599 * dquot_initialize needs lots of credits so it's better to have it
600 * outside of a transaction, so we had to pull some bits of
601 * reiserfs_new_inode out into this func.
603 static int new_inode_init(struct inode *inode, struct inode *dir, umode_t mode)
606 * Make inode invalid - just in case we are going to drop it before
607 * the initialization happens
609 INODE_PKEY(inode)->k_objectid = 0;
612 * the quota init calls have to know who to charge the quota to, so
613 * we have to set uid and gid here
615 inode_init_owner(inode, dir, mode);
616 return dquot_initialize(inode);
619 static int reiserfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
625 * We need blocks for transaction + (user+group)*(quotas
626 * for new inode + update of quota for directory owner)
629 JOURNAL_PER_BALANCE_CNT * 2 +
630 2 * (REISERFS_QUOTA_INIT_BLOCKS(dir->i_sb) +
631 REISERFS_QUOTA_TRANS_BLOCKS(dir->i_sb));
632 struct reiserfs_transaction_handle th;
633 struct reiserfs_security_handle security;
635 retval = dquot_initialize(dir);
639 if (!(inode = new_inode(dir->i_sb))) {
642 retval = new_inode_init(inode, dir, mode);
644 drop_new_inode(inode);
648 jbegin_count += reiserfs_cache_default_acl(dir);
649 retval = reiserfs_security_init(dir, inode, &dentry->d_name, &security);
651 drop_new_inode(inode);
654 jbegin_count += retval;
655 reiserfs_write_lock(dir->i_sb);
657 retval = journal_begin(&th, dir->i_sb, jbegin_count);
659 drop_new_inode(inode);
664 reiserfs_new_inode(&th, dir, mode, NULL, 0 /*i_size */ , dentry,
669 inode->i_op = &reiserfs_file_inode_operations;
670 inode->i_fop = &reiserfs_file_operations;
671 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
674 reiserfs_add_entry(&th, dir, dentry->d_name.name,
675 dentry->d_name.len, inode, 1 /*visible */ );
679 reiserfs_update_sd(&th, inode);
680 err = journal_end(&th);
683 unlock_new_inode(inode);
687 reiserfs_update_inode_transaction(inode);
688 reiserfs_update_inode_transaction(dir);
690 unlock_new_inode(inode);
691 d_instantiate(dentry, inode);
692 retval = journal_end(&th);
695 reiserfs_write_unlock(dir->i_sb);
699 static int reiserfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
704 struct reiserfs_transaction_handle th;
705 struct reiserfs_security_handle security;
707 * We need blocks for transaction + (user+group)*(quotas
708 * for new inode + update of quota for directory owner)
711 JOURNAL_PER_BALANCE_CNT * 3 +
712 2 * (REISERFS_QUOTA_INIT_BLOCKS(dir->i_sb) +
713 REISERFS_QUOTA_TRANS_BLOCKS(dir->i_sb));
715 if (!new_valid_dev(rdev))
718 retval = dquot_initialize(dir);
722 if (!(inode = new_inode(dir->i_sb))) {
725 retval = new_inode_init(inode, dir, mode);
727 drop_new_inode(inode);
731 jbegin_count += reiserfs_cache_default_acl(dir);
732 retval = reiserfs_security_init(dir, inode, &dentry->d_name, &security);
734 drop_new_inode(inode);
737 jbegin_count += retval;
738 reiserfs_write_lock(dir->i_sb);
740 retval = journal_begin(&th, dir->i_sb, jbegin_count);
742 drop_new_inode(inode);
747 reiserfs_new_inode(&th, dir, mode, NULL, 0 /*i_size */ , dentry,
753 inode->i_op = &reiserfs_special_inode_operations;
754 init_special_inode(inode, inode->i_mode, rdev);
756 /* FIXME: needed for block and char devices only */
757 reiserfs_update_sd(&th, inode);
759 reiserfs_update_inode_transaction(inode);
760 reiserfs_update_inode_transaction(dir);
763 reiserfs_add_entry(&th, dir, dentry->d_name.name,
764 dentry->d_name.len, inode, 1 /*visible */ );
768 reiserfs_update_sd(&th, inode);
769 err = journal_end(&th);
772 unlock_new_inode(inode);
777 unlock_new_inode(inode);
778 d_instantiate(dentry, inode);
779 retval = journal_end(&th);
782 reiserfs_write_unlock(dir->i_sb);
786 static int reiserfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
790 struct reiserfs_transaction_handle th;
791 struct reiserfs_security_handle security;
793 * We need blocks for transaction + (user+group)*(quotas
794 * for new inode + update of quota for directory owner)
797 JOURNAL_PER_BALANCE_CNT * 3 +
798 2 * (REISERFS_QUOTA_INIT_BLOCKS(dir->i_sb) +
799 REISERFS_QUOTA_TRANS_BLOCKS(dir->i_sb));
801 retval = dquot_initialize(dir);
805 #ifdef DISPLACE_NEW_PACKING_LOCALITIES
807 * set flag that new packing locality created and new blocks
808 * for the content of that directory are not displaced yet
810 REISERFS_I(dir)->new_packing_locality = 1;
812 mode = S_IFDIR | mode;
813 if (!(inode = new_inode(dir->i_sb))) {
816 retval = new_inode_init(inode, dir, mode);
818 drop_new_inode(inode);
822 jbegin_count += reiserfs_cache_default_acl(dir);
823 retval = reiserfs_security_init(dir, inode, &dentry->d_name, &security);
825 drop_new_inode(inode);
828 jbegin_count += retval;
829 reiserfs_write_lock(dir->i_sb);
831 retval = journal_begin(&th, dir->i_sb, jbegin_count);
833 drop_new_inode(inode);
838 * inc the link count now, so another writer doesn't overflow
839 * it while we sleep later on.
841 INC_DIR_INODE_NLINK(dir)
843 retval = reiserfs_new_inode(&th, dir, mode, NULL /*symlink */ ,
844 old_format_only(dir->i_sb) ?
845 EMPTY_DIR_SIZE_V1 : EMPTY_DIR_SIZE,
846 dentry, inode, &security);
848 DEC_DIR_INODE_NLINK(dir)
852 reiserfs_update_inode_transaction(inode);
853 reiserfs_update_inode_transaction(dir);
855 inode->i_op = &reiserfs_dir_inode_operations;
856 inode->i_fop = &reiserfs_dir_operations;
858 /* note, _this_ add_entry will not update dir's stat data */
860 reiserfs_add_entry(&th, dir, dentry->d_name.name,
861 dentry->d_name.len, inode, 1 /*visible */ );
865 DEC_DIR_INODE_NLINK(dir);
866 reiserfs_update_sd(&th, inode);
867 err = journal_end(&th);
870 unlock_new_inode(inode);
874 /* the above add_entry did not update dir's stat data */
875 reiserfs_update_sd(&th, dir);
877 unlock_new_inode(inode);
878 d_instantiate(dentry, inode);
879 retval = journal_end(&th);
881 reiserfs_write_unlock(dir->i_sb);
885 static inline int reiserfs_empty_dir(struct inode *inode)
888 * we can cheat because an old format dir cannot have
889 * EMPTY_DIR_SIZE, and a new format dir cannot have
890 * EMPTY_DIR_SIZE_V1. So, if the inode is either size,
891 * regardless of disk format version, the directory is empty.
893 if (inode->i_size != EMPTY_DIR_SIZE &&
894 inode->i_size != EMPTY_DIR_SIZE_V1) {
900 static int reiserfs_rmdir(struct inode *dir, struct dentry *dentry)
904 struct reiserfs_transaction_handle th;
906 INITIALIZE_PATH(path);
907 struct reiserfs_dir_entry de;
910 * we will be doing 2 balancings and update 2 stat data, we
911 * change quotas of the owner of the directory and of the owner
912 * of the parent directory. The quota structure is possibly
913 * deleted only on last iput => outside of this transaction
916 JOURNAL_PER_BALANCE_CNT * 2 + 2 +
917 4 * REISERFS_QUOTA_TRANS_BLOCKS(dir->i_sb);
919 retval = dquot_initialize(dir);
923 reiserfs_write_lock(dir->i_sb);
924 retval = journal_begin(&th, dir->i_sb, jbegin_count);
928 de.de_gen_number_bit_string = NULL;
930 reiserfs_find_entry(dir, dentry->d_name.name, dentry->d_name.len,
931 &path, &de)) == NAME_NOT_FOUND) {
934 } else if (retval == IO_ERROR) {
939 inode = d_inode(dentry);
941 reiserfs_update_inode_transaction(inode);
942 reiserfs_update_inode_transaction(dir);
944 if (de.de_objectid != inode->i_ino) {
946 * FIXME: compare key of an object and a key found in the entry
951 if (!reiserfs_empty_dir(inode)) {
956 /* cut entry from dir directory */
957 retval = reiserfs_cut_from_item(&th, &path, &de.de_entry_key,
958 dir, NULL, /* page */
959 0 /*new file size - not used here */ );
963 if (inode->i_nlink != 2 && inode->i_nlink != 1)
964 reiserfs_error(inode->i_sb, "reiserfs-7040",
965 "empty directory has nlink != 2 (%d)",
969 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
970 reiserfs_update_sd(&th, inode);
972 DEC_DIR_INODE_NLINK(dir)
973 dir->i_size -= (DEH_SIZE + de.de_entrylen);
974 reiserfs_update_sd(&th, dir);
976 /* prevent empty directory from getting lost */
977 add_save_link(&th, inode, 0 /* not truncate */ );
979 retval = journal_end(&th);
980 reiserfs_check_path(&path);
982 reiserfs_write_unlock(dir->i_sb);
987 * we must release path, because we did not call
988 * reiserfs_cut_from_item, or reiserfs_cut_from_item does not
989 * release path if operation was not complete
992 err = journal_end(&th);
993 reiserfs_write_unlock(dir->i_sb);
994 return err ? err : retval;
997 static int reiserfs_unlink(struct inode *dir, struct dentry *dentry)
1000 struct inode *inode;
1001 struct reiserfs_dir_entry de;
1002 INITIALIZE_PATH(path);
1003 struct reiserfs_transaction_handle th;
1005 unsigned long savelink;
1007 retval = dquot_initialize(dir);
1011 inode = d_inode(dentry);
1014 * in this transaction we can be doing at max two balancings and
1015 * update two stat datas, we change quotas of the owner of the
1016 * directory and of the owner of the parent directory. The quota
1017 * structure is possibly deleted only on iput => outside of
1021 JOURNAL_PER_BALANCE_CNT * 2 + 2 +
1022 4 * REISERFS_QUOTA_TRANS_BLOCKS(dir->i_sb);
1024 reiserfs_write_lock(dir->i_sb);
1025 retval = journal_begin(&th, dir->i_sb, jbegin_count);
1029 de.de_gen_number_bit_string = NULL;
1031 reiserfs_find_entry(dir, dentry->d_name.name, dentry->d_name.len,
1032 &path, &de)) == NAME_NOT_FOUND) {
1035 } else if (retval == IO_ERROR) {
1040 reiserfs_update_inode_transaction(inode);
1041 reiserfs_update_inode_transaction(dir);
1043 if (de.de_objectid != inode->i_ino) {
1045 * FIXME: compare key of an object and a key found in the entry
1051 if (!inode->i_nlink) {
1052 reiserfs_warning(inode->i_sb, "reiserfs-7042",
1053 "deleting nonexistent file (%lu), %d",
1054 inode->i_ino, inode->i_nlink);
1055 set_nlink(inode, 1);
1061 * we schedule before doing the add_save_link call, save the link
1062 * count so we don't race
1064 savelink = inode->i_nlink;
1067 reiserfs_cut_from_item(&th, &path, &de.de_entry_key, dir, NULL,
1073 inode->i_ctime = CURRENT_TIME_SEC;
1074 reiserfs_update_sd(&th, inode);
1076 dir->i_size -= (de.de_entrylen + DEH_SIZE);
1077 dir->i_ctime = dir->i_mtime = CURRENT_TIME_SEC;
1078 reiserfs_update_sd(&th, dir);
1081 /* prevent file from getting lost */
1082 add_save_link(&th, inode, 0 /* not truncate */ );
1084 retval = journal_end(&th);
1085 reiserfs_check_path(&path);
1086 reiserfs_write_unlock(dir->i_sb);
1091 err = journal_end(&th);
1092 reiserfs_check_path(&path);
1096 reiserfs_write_unlock(dir->i_sb);
1100 static int reiserfs_symlink(struct inode *parent_dir,
1101 struct dentry *dentry, const char *symname)
1104 struct inode *inode;
1107 struct reiserfs_transaction_handle th;
1108 struct reiserfs_security_handle security;
1109 int mode = S_IFLNK | S_IRWXUGO;
1111 * We need blocks for transaction + (user+group)*(quotas for
1112 * new inode + update of quota for directory owner)
1115 JOURNAL_PER_BALANCE_CNT * 3 +
1116 2 * (REISERFS_QUOTA_INIT_BLOCKS(parent_dir->i_sb) +
1117 REISERFS_QUOTA_TRANS_BLOCKS(parent_dir->i_sb));
1119 retval = dquot_initialize(parent_dir);
1123 if (!(inode = new_inode(parent_dir->i_sb))) {
1126 retval = new_inode_init(inode, parent_dir, mode);
1128 drop_new_inode(inode);
1132 retval = reiserfs_security_init(parent_dir, inode, &dentry->d_name,
1135 drop_new_inode(inode);
1138 jbegin_count += retval;
1140 reiserfs_write_lock(parent_dir->i_sb);
1141 item_len = ROUND_UP(strlen(symname));
1142 if (item_len > MAX_DIRECT_ITEM_LEN(parent_dir->i_sb->s_blocksize)) {
1143 retval = -ENAMETOOLONG;
1144 drop_new_inode(inode);
1148 name = kmalloc(item_len, GFP_NOFS);
1150 drop_new_inode(inode);
1154 memcpy(name, symname, strlen(symname));
1155 padd_item(name, item_len, strlen(symname));
1157 retval = journal_begin(&th, parent_dir->i_sb, jbegin_count);
1159 drop_new_inode(inode);
1165 reiserfs_new_inode(&th, parent_dir, mode, name, strlen(symname),
1166 dentry, inode, &security);
1168 if (retval) { /* reiserfs_new_inode iputs for us */
1172 reiserfs_update_inode_transaction(inode);
1173 reiserfs_update_inode_transaction(parent_dir);
1175 inode->i_op = &reiserfs_symlink_inode_operations;
1176 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1178 retval = reiserfs_add_entry(&th, parent_dir, dentry->d_name.name,
1179 dentry->d_name.len, inode, 1 /*visible */ );
1183 reiserfs_update_sd(&th, inode);
1184 err = journal_end(&th);
1187 unlock_new_inode(inode);
1192 unlock_new_inode(inode);
1193 d_instantiate(dentry, inode);
1194 retval = journal_end(&th);
1196 reiserfs_write_unlock(parent_dir->i_sb);
1200 static int reiserfs_link(struct dentry *old_dentry, struct inode *dir,
1201 struct dentry *dentry)
1204 struct inode *inode = d_inode(old_dentry);
1205 struct reiserfs_transaction_handle th;
1207 * We need blocks for transaction + update of quotas for
1208 * the owners of the directory
1211 JOURNAL_PER_BALANCE_CNT * 3 +
1212 2 * REISERFS_QUOTA_TRANS_BLOCKS(dir->i_sb);
1214 retval = dquot_initialize(dir);
1218 reiserfs_write_lock(dir->i_sb);
1219 if (inode->i_nlink >= REISERFS_LINK_MAX) {
1220 /* FIXME: sd_nlink is 32 bit for new files */
1221 reiserfs_write_unlock(dir->i_sb);
1225 /* inc before scheduling so reiserfs_unlink knows we are here */
1228 retval = journal_begin(&th, dir->i_sb, jbegin_count);
1231 reiserfs_write_unlock(dir->i_sb);
1235 /* create new entry */
1237 reiserfs_add_entry(&th, dir, dentry->d_name.name,
1238 dentry->d_name.len, inode, 1 /*visible */ );
1240 reiserfs_update_inode_transaction(inode);
1241 reiserfs_update_inode_transaction(dir);
1246 err = journal_end(&th);
1247 reiserfs_write_unlock(dir->i_sb);
1248 return err ? err : retval;
1251 inode->i_ctime = CURRENT_TIME_SEC;
1252 reiserfs_update_sd(&th, inode);
1255 d_instantiate(dentry, inode);
1256 retval = journal_end(&th);
1257 reiserfs_write_unlock(dir->i_sb);
1261 /* de contains information pointing to an entry which */
1262 static int de_still_valid(const char *name, int len,
1263 struct reiserfs_dir_entry *de)
1265 struct reiserfs_dir_entry tmp = *de;
1267 /* recalculate pointer to name and name length */
1268 set_de_name_and_namelen(&tmp);
1269 /* FIXME: could check more */
1270 if (tmp.de_namelen != len || memcmp(name, de->de_name, len))
1275 static int entry_points_to_object(const char *name, int len,
1276 struct reiserfs_dir_entry *de,
1277 struct inode *inode)
1279 if (!de_still_valid(name, len, de))
1283 if (!de_visible(de->de_deh + de->de_entry_num))
1284 reiserfs_panic(inode->i_sb, "vs-7042",
1285 "entry must be visible");
1286 return (de->de_objectid == inode->i_ino) ? 1 : 0;
1289 /* this must be added hidden entry */
1290 if (de_visible(de->de_deh + de->de_entry_num))
1291 reiserfs_panic(NULL, "vs-7043", "entry must be visible");
1296 /* sets key of objectid the entry has to point to */
1297 static void set_ino_in_dir_entry(struct reiserfs_dir_entry *de,
1298 struct reiserfs_key *key)
1300 /* JDM These operations are endian safe - both are le */
1301 de->de_deh[de->de_entry_num].deh_dir_id = key->k_dir_id;
1302 de->de_deh[de->de_entry_num].deh_objectid = key->k_objectid;
1306 * process, that is going to call fix_nodes/do_balance must hold only
1307 * one path. If it holds 2 or more, it can get into endless waiting in
1308 * get_empty_nodes or its clones
1310 static int reiserfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1311 struct inode *new_dir, struct dentry *new_dentry)
1314 INITIALIZE_PATH(old_entry_path);
1315 INITIALIZE_PATH(new_entry_path);
1316 INITIALIZE_PATH(dot_dot_entry_path);
1317 struct item_head new_entry_ih, old_entry_ih, dot_dot_ih;
1318 struct reiserfs_dir_entry old_de, new_de, dot_dot_de;
1319 struct inode *old_inode, *new_dentry_inode;
1320 struct reiserfs_transaction_handle th;
1322 umode_t old_inode_mode;
1323 unsigned long savelink = 1;
1324 struct timespec ctime;
1327 * three balancings: (1) old name removal, (2) new name insertion
1328 * and (3) maybe "save" link insertion
1329 * stat data updates: (1) old directory,
1330 * (2) new directory and (3) maybe old object stat data (when it is
1331 * directory) and (4) maybe stat data of object to which new entry
1332 * pointed initially and (5) maybe block containing ".." of
1334 * quota updates: two parent directories
1337 JOURNAL_PER_BALANCE_CNT * 3 + 5 +
1338 4 * REISERFS_QUOTA_TRANS_BLOCKS(old_dir->i_sb);
1340 retval = dquot_initialize(old_dir);
1343 retval = dquot_initialize(new_dir);
1347 old_inode = d_inode(old_dentry);
1348 new_dentry_inode = d_inode(new_dentry);
1351 * make sure that oldname still exists and points to an object we
1352 * are going to rename
1354 old_de.de_gen_number_bit_string = NULL;
1355 reiserfs_write_lock(old_dir->i_sb);
1357 reiserfs_find_entry(old_dir, old_dentry->d_name.name,
1358 old_dentry->d_name.len, &old_entry_path,
1360 pathrelse(&old_entry_path);
1361 if (retval == IO_ERROR) {
1362 reiserfs_write_unlock(old_dir->i_sb);
1366 if (retval != NAME_FOUND || old_de.de_objectid != old_inode->i_ino) {
1367 reiserfs_write_unlock(old_dir->i_sb);
1371 old_inode_mode = old_inode->i_mode;
1372 if (S_ISDIR(old_inode_mode)) {
1374 * make sure that directory being renamed has correct ".."
1375 * and that its new parent directory has not too many links
1378 if (new_dentry_inode) {
1379 if (!reiserfs_empty_dir(new_dentry_inode)) {
1380 reiserfs_write_unlock(old_dir->i_sb);
1386 * directory is renamed, its parent directory will be changed,
1387 * so find ".." entry
1389 dot_dot_de.de_gen_number_bit_string = NULL;
1391 reiserfs_find_entry(old_inode, "..", 2, &dot_dot_entry_path,
1393 pathrelse(&dot_dot_entry_path);
1394 if (retval != NAME_FOUND) {
1395 reiserfs_write_unlock(old_dir->i_sb);
1399 /* inode number of .. must equal old_dir->i_ino */
1400 if (dot_dot_de.de_objectid != old_dir->i_ino) {
1401 reiserfs_write_unlock(old_dir->i_sb);
1406 retval = journal_begin(&th, old_dir->i_sb, jbegin_count);
1408 reiserfs_write_unlock(old_dir->i_sb);
1412 /* add new entry (or find the existing one) */
1414 reiserfs_add_entry(&th, new_dir, new_dentry->d_name.name,
1415 new_dentry->d_name.len, old_inode, 0);
1416 if (retval == -EEXIST) {
1417 if (!new_dentry_inode) {
1418 reiserfs_panic(old_dir->i_sb, "vs-7050",
1419 "new entry is found, new inode == 0");
1421 } else if (retval) {
1422 int err = journal_end(&th);
1423 reiserfs_write_unlock(old_dir->i_sb);
1424 return err ? err : retval;
1427 reiserfs_update_inode_transaction(old_dir);
1428 reiserfs_update_inode_transaction(new_dir);
1431 * this makes it so an fsync on an open fd for the old name will
1432 * commit the rename operation
1434 reiserfs_update_inode_transaction(old_inode);
1436 if (new_dentry_inode)
1437 reiserfs_update_inode_transaction(new_dentry_inode);
1441 * look for old name using corresponding entry key
1442 * (found by reiserfs_find_entry)
1445 search_by_entry_key(new_dir->i_sb, &old_de.de_entry_key,
1447 &old_de)) != NAME_FOUND) {
1448 pathrelse(&old_entry_path);
1450 reiserfs_write_unlock(old_dir->i_sb);
1454 copy_item_head(&old_entry_ih, tp_item_head(&old_entry_path));
1456 reiserfs_prepare_for_journal(old_inode->i_sb, old_de.de_bh, 1);
1458 /* look for new name by reiserfs_find_entry */
1459 new_de.de_gen_number_bit_string = NULL;
1461 reiserfs_find_entry(new_dir, new_dentry->d_name.name,
1462 new_dentry->d_name.len, &new_entry_path,
1465 * reiserfs_add_entry should not return IO_ERROR,
1466 * because it is called with essentially same parameters from
1467 * reiserfs_add_entry above, and we'll catch any i/o errors
1468 * before we get here.
1470 if (retval != NAME_FOUND_INVISIBLE && retval != NAME_FOUND) {
1471 pathrelse(&new_entry_path);
1472 pathrelse(&old_entry_path);
1474 reiserfs_write_unlock(old_dir->i_sb);
1478 copy_item_head(&new_entry_ih, tp_item_head(&new_entry_path));
1480 reiserfs_prepare_for_journal(old_inode->i_sb, new_de.de_bh, 1);
1482 if (S_ISDIR(old_inode->i_mode)) {
1484 search_by_entry_key(new_dir->i_sb,
1485 &dot_dot_de.de_entry_key,
1486 &dot_dot_entry_path,
1487 &dot_dot_de)) != NAME_FOUND) {
1488 pathrelse(&dot_dot_entry_path);
1489 pathrelse(&new_entry_path);
1490 pathrelse(&old_entry_path);
1492 reiserfs_write_unlock(old_dir->i_sb);
1495 copy_item_head(&dot_dot_ih,
1496 tp_item_head(&dot_dot_entry_path));
1497 /* node containing ".." gets into transaction */
1498 reiserfs_prepare_for_journal(old_inode->i_sb,
1499 dot_dot_de.de_bh, 1);
1502 * we should check seals here, not do
1503 * this stuff, yes? Then, having
1504 * gathered everything into RAM we
1505 * should lock the buffers, yes? -Hans
1508 * probably. our rename needs to hold more
1509 * than one path at once. The seals would
1510 * have to be written to deal with multi-path
1514 * sanity checking before doing the rename - avoid races many
1515 * of the above checks could have scheduled. We have to be
1516 * sure our items haven't been shifted by another process.
1518 if (item_moved(&new_entry_ih, &new_entry_path) ||
1519 !entry_points_to_object(new_dentry->d_name.name,
1520 new_dentry->d_name.len,
1521 &new_de, new_dentry_inode) ||
1522 item_moved(&old_entry_ih, &old_entry_path) ||
1523 !entry_points_to_object(old_dentry->d_name.name,
1524 old_dentry->d_name.len,
1525 &old_de, old_inode)) {
1526 reiserfs_restore_prepared_buffer(old_inode->i_sb,
1528 reiserfs_restore_prepared_buffer(old_inode->i_sb,
1530 if (S_ISDIR(old_inode_mode))
1531 reiserfs_restore_prepared_buffer(old_inode->
1537 if (S_ISDIR(old_inode_mode)) {
1538 if (item_moved(&dot_dot_ih, &dot_dot_entry_path) ||
1539 !entry_points_to_object("..", 2, &dot_dot_de,
1541 reiserfs_restore_prepared_buffer(old_inode->
1544 reiserfs_restore_prepared_buffer(old_inode->
1547 reiserfs_restore_prepared_buffer(old_inode->
1555 RFALSE(S_ISDIR(old_inode_mode) &&
1556 !buffer_journal_prepared(dot_dot_de.de_bh), "");
1562 * ok, all the changes can be done in one fell swoop when we
1563 * have claimed all the buffers needed.
1566 mark_de_visible(new_de.de_deh + new_de.de_entry_num);
1567 set_ino_in_dir_entry(&new_de, INODE_PKEY(old_inode));
1568 journal_mark_dirty(&th, new_de.de_bh);
1570 mark_de_hidden(old_de.de_deh + old_de.de_entry_num);
1571 journal_mark_dirty(&th, old_de.de_bh);
1572 ctime = CURRENT_TIME_SEC;
1573 old_dir->i_ctime = old_dir->i_mtime = ctime;
1574 new_dir->i_ctime = new_dir->i_mtime = ctime;
1576 * thanks to Alex Adriaanse <alex_a@caltech.edu> for patch
1577 * which adds ctime update of renamed object
1579 old_inode->i_ctime = ctime;
1581 if (new_dentry_inode) {
1582 /* adjust link number of the victim */
1583 if (S_ISDIR(new_dentry_inode->i_mode)) {
1584 clear_nlink(new_dentry_inode);
1586 drop_nlink(new_dentry_inode);
1588 new_dentry_inode->i_ctime = ctime;
1589 savelink = new_dentry_inode->i_nlink;
1592 if (S_ISDIR(old_inode_mode)) {
1593 /* adjust ".." of renamed directory */
1594 set_ino_in_dir_entry(&dot_dot_de, INODE_PKEY(new_dir));
1595 journal_mark_dirty(&th, dot_dot_de.de_bh);
1598 * there (in new_dir) was no directory, so it got new link
1599 * (".." of renamed directory)
1601 if (!new_dentry_inode)
1602 INC_DIR_INODE_NLINK(new_dir);
1604 /* old directory lost one link - ".. " of renamed directory */
1605 DEC_DIR_INODE_NLINK(old_dir);
1608 * looks like in 2.3.99pre3 brelse is atomic.
1609 * so we can use pathrelse
1611 pathrelse(&new_entry_path);
1612 pathrelse(&dot_dot_entry_path);
1615 * FIXME: this reiserfs_cut_from_item's return value may screw up
1616 * anybody, but it will panic if will not be able to find the
1617 * entry. This needs one more clean up
1619 if (reiserfs_cut_from_item
1620 (&th, &old_entry_path, &old_de.de_entry_key, old_dir, NULL,
1622 reiserfs_error(old_dir->i_sb, "vs-7060",
1623 "couldn't not cut old name. Fsck later?");
1625 old_dir->i_size -= DEH_SIZE + old_de.de_entrylen;
1627 reiserfs_update_sd(&th, old_dir);
1628 reiserfs_update_sd(&th, new_dir);
1629 reiserfs_update_sd(&th, old_inode);
1631 if (new_dentry_inode) {
1633 add_save_link(&th, new_dentry_inode,
1634 0 /* not truncate */ );
1635 reiserfs_update_sd(&th, new_dentry_inode);
1638 retval = journal_end(&th);
1639 reiserfs_write_unlock(old_dir->i_sb);
1643 /* directories can handle most operations... */
1644 const struct inode_operations reiserfs_dir_inode_operations = {
1645 .create = reiserfs_create,
1646 .lookup = reiserfs_lookup,
1647 .link = reiserfs_link,
1648 .unlink = reiserfs_unlink,
1649 .symlink = reiserfs_symlink,
1650 .mkdir = reiserfs_mkdir,
1651 .rmdir = reiserfs_rmdir,
1652 .mknod = reiserfs_mknod,
1653 .rename = reiserfs_rename,
1654 .setattr = reiserfs_setattr,
1655 .setxattr = reiserfs_setxattr,
1656 .getxattr = reiserfs_getxattr,
1657 .listxattr = reiserfs_listxattr,
1658 .removexattr = reiserfs_removexattr,
1659 .permission = reiserfs_permission,
1660 .get_acl = reiserfs_get_acl,
1661 .set_acl = reiserfs_set_acl,
1665 * symlink operations.. same as page_symlink_inode_operations, with xattr
1668 const struct inode_operations reiserfs_symlink_inode_operations = {
1669 .readlink = generic_readlink,
1670 .follow_link = page_follow_link_light,
1671 .put_link = page_put_link,
1672 .setattr = reiserfs_setattr,
1673 .setxattr = reiserfs_setxattr,
1674 .getxattr = reiserfs_getxattr,
1675 .listxattr = reiserfs_listxattr,
1676 .removexattr = reiserfs_removexattr,
1677 .permission = reiserfs_permission,
1681 * special file operations.. just xattr/acl stuff
1683 const struct inode_operations reiserfs_special_inode_operations = {
1684 .setattr = reiserfs_setattr,
1685 .setxattr = reiserfs_setxattr,
1686 .getxattr = reiserfs_getxattr,
1687 .listxattr = reiserfs_listxattr,
1688 .removexattr = reiserfs_removexattr,
1689 .permission = reiserfs_permission,
1690 .get_acl = reiserfs_get_acl,
1691 .set_acl = reiserfs_set_acl,