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)--;
145 reiserfs_error(sb, "vs-7002", "no path to here");
149 set_de_item_location(de, path);
151 #ifdef CONFIG_REISERFS_CHECK
152 if (!is_direntry_le_ih(de->de_ih) ||
153 COMP_SHORT_KEYS(&de->de_ih->ih_key, key)) {
154 print_block(de->de_bh, 0, -1, -1);
155 reiserfs_panic(sb, "vs-7005", "found item %h is not directory "
156 "item or does not belong to the same directory "
157 "as key %K", de->de_ih, key);
159 #endif /* CONFIG_REISERFS_CHECK */
162 * binary search in directory item by third component of the
163 * key. sets de->de_entry_num of de
165 retval = bin_search_in_dir_item(de, cpu_key_k_offset(key));
166 path->pos_in_item = de->de_entry_num;
167 if (retval != NAME_NOT_FOUND) {
169 * ugly, but rename needs de_bh, de_deh, de_name,
170 * de_namelen, de_objectid set
172 set_de_name_and_namelen(de);
173 set_de_object_key(de);
178 /* Keyed 32-bit hash function using TEA in a Davis-Meyer function */
181 * The third component is hashed, and you can choose from more than
182 * one hash function. Per directory hashes are not yet implemented
183 * but are thought about. This function should be moved to hashes.c
184 * Jedi, please do so. -Hans
186 static __u32 get_third_component(struct super_block *s,
187 const char *name, int len)
191 if (!len || (len == 1 && name[0] == '.'))
193 if (len == 2 && name[0] == '.' && name[1] == '.')
194 return DOT_DOT_OFFSET;
196 res = REISERFS_SB(s)->s_hash_function(name, len);
198 /* take bits from 7-th to 30-th including both bounds */
199 res = GET_HASH_VALUE(res);
202 * needed to have no names before "." and ".." those have hash
203 * value == 0 and generation conters 1 and 2 accordingly
206 return res + MAX_GENERATION_NUMBER;
209 static int reiserfs_match(struct reiserfs_dir_entry *de,
210 const char *name, int namelen)
212 int retval = NAME_NOT_FOUND;
214 if ((namelen == de->de_namelen) &&
215 !memcmp(de->de_name, name, de->de_namelen))
217 (de_visible(de->de_deh + de->de_entry_num) ? NAME_FOUND :
218 NAME_FOUND_INVISIBLE);
223 /* de's de_bh, de_ih, de_deh, de_item_num, de_entry_num are set already */
225 /* used when hash collisions exist */
227 static int linear_search_in_dir_item(struct cpu_key *key,
228 struct reiserfs_dir_entry *de,
229 const char *name, int namelen)
231 struct reiserfs_de_head *deh = de->de_deh;
235 i = de->de_entry_num;
237 if (i == ih_entry_count(de->de_ih) ||
238 GET_HASH_VALUE(deh_offset(deh + i)) !=
239 GET_HASH_VALUE(cpu_key_k_offset(key))) {
243 RFALSE(de->de_deh != B_I_DEH(de->de_bh, de->de_ih),
244 "vs-7010: array of entry headers not found");
248 for (; i >= 0; i--, deh--) {
249 /* hash value does not match, no need to check whole name */
250 if (GET_HASH_VALUE(deh_offset(deh)) !=
251 GET_HASH_VALUE(cpu_key_k_offset(key))) {
252 return NAME_NOT_FOUND;
255 /* mark that this generation number is used */
256 if (de->de_gen_number_bit_string)
257 set_bit(GET_GENERATION_NUMBER(deh_offset(deh)),
258 de->de_gen_number_bit_string);
260 /* calculate pointer to name and namelen */
261 de->de_entry_num = i;
262 set_de_name_and_namelen(de);
265 * de's de_name, de_namelen, de_recordlen are set.
269 reiserfs_match(de, name, namelen)) != NAME_NOT_FOUND) {
271 /* key of pointed object */
272 set_de_object_key(de);
274 store_de_entry_key(de);
276 /* retval can be NAME_FOUND or NAME_FOUND_INVISIBLE */
281 if (GET_GENERATION_NUMBER(le_ih_k_offset(de->de_ih)) == 0)
283 * we have reached left most entry in the node. In common we
284 * have to go to the left neighbor, but if generation counter
285 * is 0 already, we know for sure, that there is no name with
286 * the same hash value
289 * FIXME: this work correctly only because hash value can not
290 * be 0. Btw, in case of Yura's hash it is probably possible,
293 return NAME_NOT_FOUND;
295 RFALSE(de->de_item_num,
296 "vs-7015: two diritems of the same directory in one node?");
298 return GOTO_PREVIOUS_ITEM;
302 * may return NAME_FOUND, NAME_FOUND_INVISIBLE, NAME_NOT_FOUND
303 * FIXME: should add something like IOERROR
305 static int reiserfs_find_entry(struct inode *dir, const char *name, int namelen,
306 struct treepath *path_to_entry,
307 struct reiserfs_dir_entry *de)
309 struct cpu_key key_to_search;
312 if (namelen > REISERFS_MAX_NAME(dir->i_sb->s_blocksize))
313 return NAME_NOT_FOUND;
315 /* we will search for this key in the tree */
316 make_cpu_key(&key_to_search, dir,
317 get_third_component(dir->i_sb, name, namelen),
322 search_by_entry_key(dir->i_sb, &key_to_search,
324 if (retval == IO_ERROR) {
325 reiserfs_error(dir->i_sb, "zam-7001", "io error");
329 /* compare names for all entries having given hash value */
331 linear_search_in_dir_item(&key_to_search, de, name,
334 * there is no need to scan directory anymore.
335 * Given entry found or does not exist
337 if (retval != GOTO_PREVIOUS_ITEM) {
338 path_to_entry->pos_in_item = de->de_entry_num;
343 * there is left neighboring item of this directory
344 * and given entry can be there
346 set_cpu_key_k_offset(&key_to_search,
347 le_ih_k_offset(de->de_ih) - 1);
348 pathrelse(path_to_entry);
353 static struct dentry *reiserfs_lookup(struct inode *dir, struct dentry *dentry,
357 struct inode *inode = NULL;
358 struct reiserfs_dir_entry de;
359 INITIALIZE_PATH(path_to_entry);
361 if (REISERFS_MAX_NAME(dir->i_sb->s_blocksize) < dentry->d_name.len)
362 return ERR_PTR(-ENAMETOOLONG);
364 reiserfs_write_lock(dir->i_sb);
366 de.de_gen_number_bit_string = NULL;
368 reiserfs_find_entry(dir, dentry->d_name.name, dentry->d_name.len,
369 &path_to_entry, &de);
370 pathrelse(&path_to_entry);
371 if (retval == NAME_FOUND) {
372 inode = reiserfs_iget(dir->i_sb,
373 (struct cpu_key *)&de.de_dir_id);
374 if (!inode || IS_ERR(inode)) {
375 reiserfs_write_unlock(dir->i_sb);
376 return ERR_PTR(-EACCES);
380 * Propagate the private flag so we know we're
381 * in the priv tree. Also clear IOP_XATTR
382 * since we don't have xattrs on xattr files.
384 if (IS_PRIVATE(dir)) {
385 inode->i_flags |= S_PRIVATE;
386 inode->i_opflags &= ~IOP_XATTR;
389 reiserfs_write_unlock(dir->i_sb);
390 if (retval == IO_ERROR) {
391 return ERR_PTR(-EIO);
394 return d_splice_alias(inode, dentry);
398 * looks up the dentry of the parent directory for child.
399 * taken from ext2_get_parent
401 struct dentry *reiserfs_get_parent(struct dentry *child)
404 struct inode *inode = NULL;
405 struct reiserfs_dir_entry de;
406 INITIALIZE_PATH(path_to_entry);
407 struct inode *dir = d_inode(child);
409 if (dir->i_nlink == 0) {
410 return ERR_PTR(-ENOENT);
412 de.de_gen_number_bit_string = NULL;
414 reiserfs_write_lock(dir->i_sb);
415 retval = reiserfs_find_entry(dir, "..", 2, &path_to_entry, &de);
416 pathrelse(&path_to_entry);
417 if (retval != NAME_FOUND) {
418 reiserfs_write_unlock(dir->i_sb);
419 return ERR_PTR(-ENOENT);
421 inode = reiserfs_iget(dir->i_sb, (struct cpu_key *)&de.de_dir_id);
422 reiserfs_write_unlock(dir->i_sb);
424 return d_obtain_alias(inode);
427 /* add entry to the directory (entry can be hidden).
429 insert definition of when hidden directories are used here -Hans
431 Does not mark dir inode dirty, do it after successesfull call to it */
433 static int reiserfs_add_entry(struct reiserfs_transaction_handle *th,
434 struct inode *dir, const char *name, int namelen,
435 struct inode *inode, int visible)
437 struct cpu_key entry_key;
438 struct reiserfs_de_head *deh;
439 INITIALIZE_PATH(path);
440 struct reiserfs_dir_entry de;
441 DECLARE_BITMAP(bit_string, MAX_GENERATION_NUMBER + 1);
445 * 48 bytes now and we avoid kmalloc if we
446 * create file with short name
448 char small_buf[32 + DEH_SIZE];
451 int buflen, paste_size;
454 BUG_ON(!th->t_trans_id);
456 /* cannot allow items to be added into a busy deleted directory */
460 if (namelen > REISERFS_MAX_NAME(dir->i_sb->s_blocksize))
461 return -ENAMETOOLONG;
463 /* each entry has unique key. compose it */
464 make_cpu_key(&entry_key, dir,
465 get_third_component(dir->i_sb, name, namelen),
468 /* get memory for composing the entry */
469 buflen = DEH_SIZE + ROUND_UP(namelen);
470 if (buflen > sizeof(small_buf)) {
471 buffer = kmalloc(buflen, GFP_NOFS);
478 (get_inode_sd_version(dir) ==
479 STAT_DATA_V1) ? (DEH_SIZE + namelen) : buflen;
482 * fill buffer : directory entry head, name[, dir objectid | ,
483 * stat data | ,stat data, dir objectid ]
485 deh = (struct reiserfs_de_head *)buffer;
486 deh->deh_location = 0; /* JDM Endian safe if 0 */
487 put_deh_offset(deh, cpu_key_k_offset(&entry_key));
488 deh->deh_state = 0; /* JDM Endian safe if 0 */
489 /* put key (ino analog) to de */
491 /* safe: k_dir_id is le */
492 deh->deh_dir_id = INODE_PKEY(inode)->k_dir_id;
493 /* safe: k_objectid is le */
494 deh->deh_objectid = INODE_PKEY(inode)->k_objectid;
497 memcpy((char *)(deh + 1), name, namelen);
498 /* padd by 0s to the 4 byte boundary */
499 padd_item((char *)(deh + 1), ROUND_UP(namelen), namelen);
502 * entry is ready to be pasted into tree, set 'visibility'
503 * and 'stat data in entry' attributes
505 mark_de_without_sd(deh);
506 visible ? mark_de_visible(deh) : mark_de_hidden(deh);
508 /* find the proper place for the new entry */
509 memset(bit_string, 0, sizeof(bit_string));
510 de.de_gen_number_bit_string = bit_string;
511 retval = reiserfs_find_entry(dir, name, namelen, &path, &de);
512 if (retval != NAME_NOT_FOUND) {
513 if (buffer != small_buf)
517 if (retval == IO_ERROR) {
521 if (retval != NAME_FOUND) {
522 reiserfs_error(dir->i_sb, "zam-7002",
523 "reiserfs_find_entry() returned "
524 "unexpected value (%d)", retval);
531 find_first_zero_bit(bit_string,
532 MAX_GENERATION_NUMBER + 1);
533 if (gen_number > MAX_GENERATION_NUMBER) {
534 /* there is no free generation number */
535 reiserfs_warning(dir->i_sb, "reiserfs-7010",
536 "Congratulations! we have got hash function "
538 if (buffer != small_buf)
543 /* adjust offset of directory enrty */
544 put_deh_offset(deh, SET_GENERATION_NUMBER(deh_offset(deh), gen_number));
545 set_cpu_key_k_offset(&entry_key, deh_offset(deh));
547 /* update max-hash-collisions counter in reiserfs_sb_info */
548 PROC_INFO_MAX(th->t_super, max_hash_collisions, gen_number);
550 /* we need to re-search for the insertion point */
551 if (gen_number != 0) {
552 if (search_by_entry_key(dir->i_sb, &entry_key, &path, &de) !=
554 reiserfs_warning(dir->i_sb, "vs-7032",
555 "entry with this key (%K) already "
556 "exists", &entry_key);
558 if (buffer != small_buf)
565 /* perform the insertion of the entry that we have prepared */
567 reiserfs_paste_into_item(th, &path, &entry_key, dir, buffer,
569 if (buffer != small_buf)
572 reiserfs_check_path(&path);
576 dir->i_size += paste_size;
577 dir->i_mtime = dir->i_ctime = current_time(dir);
578 if (!S_ISDIR(inode->i_mode) && visible)
579 /* reiserfs_mkdir or reiserfs_rename will do that by itself */
580 reiserfs_update_sd(th, dir);
582 reiserfs_check_path(&path);
587 * quota utility function, call if you've had to abort after calling
588 * new_inode_init, and have not called reiserfs_new_inode yet.
589 * This should only be called on inodes that do not have stat data
590 * inserted into the tree yet.
592 static int drop_new_inode(struct inode *inode)
595 make_bad_inode(inode);
596 inode->i_flags |= S_NOQUOTA;
602 * utility function that does setup for reiserfs_new_inode.
603 * dquot_initialize needs lots of credits so it's better to have it
604 * outside of a transaction, so we had to pull some bits of
605 * reiserfs_new_inode out into this func.
607 static int new_inode_init(struct inode *inode, struct inode *dir, umode_t mode)
610 * Make inode invalid - just in case we are going to drop it before
611 * the initialization happens
613 INODE_PKEY(inode)->k_objectid = 0;
616 * the quota init calls have to know who to charge the quota to, so
617 * we have to set uid and gid here
619 inode_init_owner(&init_user_ns, inode, dir, mode);
620 return dquot_initialize(inode);
623 static int reiserfs_create(struct user_namespace *mnt_userns, struct inode *dir,
624 struct dentry *dentry, umode_t mode, bool excl)
629 * We need blocks for transaction + (user+group)*(quotas
630 * for new inode + update of quota for directory owner)
633 JOURNAL_PER_BALANCE_CNT * 2 +
634 2 * (REISERFS_QUOTA_INIT_BLOCKS(dir->i_sb) +
635 REISERFS_QUOTA_TRANS_BLOCKS(dir->i_sb));
636 struct reiserfs_transaction_handle th;
637 struct reiserfs_security_handle security;
639 retval = dquot_initialize(dir);
643 if (!(inode = new_inode(dir->i_sb))) {
646 retval = new_inode_init(inode, dir, mode);
648 drop_new_inode(inode);
652 jbegin_count += reiserfs_cache_default_acl(dir);
653 retval = reiserfs_security_init(dir, inode, &dentry->d_name, &security);
655 drop_new_inode(inode);
658 jbegin_count += retval;
659 reiserfs_write_lock(dir->i_sb);
661 retval = journal_begin(&th, dir->i_sb, jbegin_count);
663 drop_new_inode(inode);
668 reiserfs_new_inode(&th, dir, mode, NULL, 0 /*i_size */ , dentry,
673 inode->i_op = &reiserfs_file_inode_operations;
674 inode->i_fop = &reiserfs_file_operations;
675 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
678 reiserfs_add_entry(&th, dir, dentry->d_name.name,
679 dentry->d_name.len, inode, 1 /*visible */ );
683 reiserfs_update_sd(&th, inode);
684 err = journal_end(&th);
687 unlock_new_inode(inode);
691 reiserfs_update_inode_transaction(inode);
692 reiserfs_update_inode_transaction(dir);
694 d_instantiate_new(dentry, inode);
695 retval = journal_end(&th);
698 reiserfs_write_unlock(dir->i_sb);
702 static int reiserfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
703 struct dentry *dentry, umode_t mode, dev_t rdev)
707 struct reiserfs_transaction_handle th;
708 struct reiserfs_security_handle security;
710 * We need blocks for transaction + (user+group)*(quotas
711 * for new inode + update of quota for directory owner)
714 JOURNAL_PER_BALANCE_CNT * 3 +
715 2 * (REISERFS_QUOTA_INIT_BLOCKS(dir->i_sb) +
716 REISERFS_QUOTA_TRANS_BLOCKS(dir->i_sb));
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 d_instantiate_new(dentry, inode);
778 retval = journal_end(&th);
781 reiserfs_write_unlock(dir->i_sb);
785 static int reiserfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
786 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 d_instantiate_new(dentry, inode);
878 retval = journal_end(&th);
880 reiserfs_write_unlock(dir->i_sb);
884 static inline int reiserfs_empty_dir(struct inode *inode)
887 * we can cheat because an old format dir cannot have
888 * EMPTY_DIR_SIZE, and a new format dir cannot have
889 * EMPTY_DIR_SIZE_V1. So, if the inode is either size,
890 * regardless of disk format version, the directory is empty.
892 if (inode->i_size != EMPTY_DIR_SIZE &&
893 inode->i_size != EMPTY_DIR_SIZE_V1) {
899 static int reiserfs_rmdir(struct inode *dir, struct dentry *dentry)
903 struct reiserfs_transaction_handle th;
905 INITIALIZE_PATH(path);
906 struct reiserfs_dir_entry de;
909 * we will be doing 2 balancings and update 2 stat data, we
910 * change quotas of the owner of the directory and of the owner
911 * of the parent directory. The quota structure is possibly
912 * deleted only on last iput => outside of this transaction
915 JOURNAL_PER_BALANCE_CNT * 2 + 2 +
916 4 * REISERFS_QUOTA_TRANS_BLOCKS(dir->i_sb);
918 retval = dquot_initialize(dir);
922 reiserfs_write_lock(dir->i_sb);
923 retval = journal_begin(&th, dir->i_sb, jbegin_count);
927 de.de_gen_number_bit_string = NULL;
929 reiserfs_find_entry(dir, dentry->d_name.name, dentry->d_name.len,
930 &path, &de)) == NAME_NOT_FOUND) {
933 } else if (retval == IO_ERROR) {
938 inode = d_inode(dentry);
940 reiserfs_update_inode_transaction(inode);
941 reiserfs_update_inode_transaction(dir);
943 if (de.de_objectid != inode->i_ino) {
945 * FIXME: compare key of an object and a key found in the entry
950 if (!reiserfs_empty_dir(inode)) {
955 /* cut entry from dir directory */
956 retval = reiserfs_cut_from_item(&th, &path, &de.de_entry_key,
957 dir, NULL, /* page */
958 0 /*new file size - not used here */ );
962 if (inode->i_nlink != 2 && inode->i_nlink != 1)
963 reiserfs_error(inode->i_sb, "reiserfs-7040",
964 "empty directory has nlink != 2 (%d)",
968 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(dir);
969 reiserfs_update_sd(&th, inode);
971 DEC_DIR_INODE_NLINK(dir)
972 dir->i_size -= (DEH_SIZE + de.de_entrylen);
973 reiserfs_update_sd(&th, dir);
975 /* prevent empty directory from getting lost */
976 add_save_link(&th, inode, 0 /* not truncate */ );
978 retval = journal_end(&th);
979 reiserfs_check_path(&path);
981 reiserfs_write_unlock(dir->i_sb);
986 * we must release path, because we did not call
987 * reiserfs_cut_from_item, or reiserfs_cut_from_item does not
988 * release path if operation was not complete
991 err = journal_end(&th);
992 reiserfs_write_unlock(dir->i_sb);
993 return err ? err : retval;
996 static int reiserfs_unlink(struct inode *dir, struct dentry *dentry)
1000 struct reiserfs_dir_entry de;
1001 INITIALIZE_PATH(path);
1002 struct reiserfs_transaction_handle th;
1004 unsigned long savelink;
1006 retval = dquot_initialize(dir);
1010 inode = d_inode(dentry);
1013 * in this transaction we can be doing at max two balancings and
1014 * update two stat datas, we change quotas of the owner of the
1015 * directory and of the owner of the parent directory. The quota
1016 * structure is possibly deleted only on iput => outside of
1020 JOURNAL_PER_BALANCE_CNT * 2 + 2 +
1021 4 * REISERFS_QUOTA_TRANS_BLOCKS(dir->i_sb);
1023 reiserfs_write_lock(dir->i_sb);
1024 retval = journal_begin(&th, dir->i_sb, jbegin_count);
1028 de.de_gen_number_bit_string = NULL;
1030 reiserfs_find_entry(dir, dentry->d_name.name, dentry->d_name.len,
1031 &path, &de)) == NAME_NOT_FOUND) {
1034 } else if (retval == IO_ERROR) {
1039 reiserfs_update_inode_transaction(inode);
1040 reiserfs_update_inode_transaction(dir);
1042 if (de.de_objectid != inode->i_ino) {
1044 * FIXME: compare key of an object and a key found in the entry
1050 if (!inode->i_nlink) {
1051 reiserfs_warning(inode->i_sb, "reiserfs-7042",
1052 "deleting nonexistent file (%lu), %d",
1053 inode->i_ino, inode->i_nlink);
1054 set_nlink(inode, 1);
1060 * we schedule before doing the add_save_link call, save the link
1061 * count so we don't race
1063 savelink = inode->i_nlink;
1066 reiserfs_cut_from_item(&th, &path, &de.de_entry_key, dir, NULL,
1072 inode->i_ctime = current_time(inode);
1073 reiserfs_update_sd(&th, inode);
1075 dir->i_size -= (de.de_entrylen + DEH_SIZE);
1076 dir->i_ctime = dir->i_mtime = current_time(dir);
1077 reiserfs_update_sd(&th, dir);
1080 /* prevent file from getting lost */
1081 add_save_link(&th, inode, 0 /* not truncate */ );
1083 retval = journal_end(&th);
1084 reiserfs_check_path(&path);
1085 reiserfs_write_unlock(dir->i_sb);
1090 err = journal_end(&th);
1091 reiserfs_check_path(&path);
1095 reiserfs_write_unlock(dir->i_sb);
1099 static int reiserfs_symlink(struct user_namespace *mnt_userns,
1100 struct inode *parent_dir, struct dentry *dentry,
1101 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_nohighmem(inode);
1177 inode->i_mapping->a_ops = &reiserfs_address_space_operations;
1179 retval = reiserfs_add_entry(&th, parent_dir, dentry->d_name.name,
1180 dentry->d_name.len, inode, 1 /*visible */ );
1184 reiserfs_update_sd(&th, inode);
1185 err = journal_end(&th);
1188 unlock_new_inode(inode);
1193 d_instantiate_new(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(inode);
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 user_namespace *mnt_userns,
1311 struct inode *old_dir, struct dentry *old_dentry,
1312 struct inode *new_dir, struct dentry *new_dentry,
1316 INITIALIZE_PATH(old_entry_path);
1317 INITIALIZE_PATH(new_entry_path);
1318 INITIALIZE_PATH(dot_dot_entry_path);
1319 struct item_head new_entry_ih, old_entry_ih, dot_dot_ih;
1320 struct reiserfs_dir_entry old_de, new_de, dot_dot_de;
1321 struct inode *old_inode, *new_dentry_inode;
1322 struct reiserfs_transaction_handle th;
1324 umode_t old_inode_mode;
1325 unsigned long savelink = 1;
1326 struct timespec64 ctime;
1328 if (flags & ~RENAME_NOREPLACE)
1332 * three balancings: (1) old name removal, (2) new name insertion
1333 * and (3) maybe "save" link insertion
1334 * stat data updates: (1) old directory,
1335 * (2) new directory and (3) maybe old object stat data (when it is
1336 * directory) and (4) maybe stat data of object to which new entry
1337 * pointed initially and (5) maybe block containing ".." of
1339 * quota updates: two parent directories
1342 JOURNAL_PER_BALANCE_CNT * 3 + 5 +
1343 4 * REISERFS_QUOTA_TRANS_BLOCKS(old_dir->i_sb);
1345 retval = dquot_initialize(old_dir);
1348 retval = dquot_initialize(new_dir);
1352 old_inode = d_inode(old_dentry);
1353 new_dentry_inode = d_inode(new_dentry);
1356 * make sure that oldname still exists and points to an object we
1357 * are going to rename
1359 old_de.de_gen_number_bit_string = NULL;
1360 reiserfs_write_lock(old_dir->i_sb);
1362 reiserfs_find_entry(old_dir, old_dentry->d_name.name,
1363 old_dentry->d_name.len, &old_entry_path,
1365 pathrelse(&old_entry_path);
1366 if (retval == IO_ERROR) {
1367 reiserfs_write_unlock(old_dir->i_sb);
1371 if (retval != NAME_FOUND || old_de.de_objectid != old_inode->i_ino) {
1372 reiserfs_write_unlock(old_dir->i_sb);
1376 old_inode_mode = old_inode->i_mode;
1377 if (S_ISDIR(old_inode_mode)) {
1379 * make sure that directory being renamed has correct ".."
1380 * and that its new parent directory has not too many links
1383 if (new_dentry_inode) {
1384 if (!reiserfs_empty_dir(new_dentry_inode)) {
1385 reiserfs_write_unlock(old_dir->i_sb);
1391 * directory is renamed, its parent directory will be changed,
1392 * so find ".." entry
1394 dot_dot_de.de_gen_number_bit_string = NULL;
1396 reiserfs_find_entry(old_inode, "..", 2, &dot_dot_entry_path,
1398 pathrelse(&dot_dot_entry_path);
1399 if (retval != NAME_FOUND) {
1400 reiserfs_write_unlock(old_dir->i_sb);
1404 /* inode number of .. must equal old_dir->i_ino */
1405 if (dot_dot_de.de_objectid != old_dir->i_ino) {
1406 reiserfs_write_unlock(old_dir->i_sb);
1411 retval = journal_begin(&th, old_dir->i_sb, jbegin_count);
1413 reiserfs_write_unlock(old_dir->i_sb);
1417 /* add new entry (or find the existing one) */
1419 reiserfs_add_entry(&th, new_dir, new_dentry->d_name.name,
1420 new_dentry->d_name.len, old_inode, 0);
1421 if (retval == -EEXIST) {
1422 if (!new_dentry_inode) {
1423 reiserfs_panic(old_dir->i_sb, "vs-7050",
1424 "new entry is found, new inode == 0");
1426 } else if (retval) {
1427 int err = journal_end(&th);
1428 reiserfs_write_unlock(old_dir->i_sb);
1429 return err ? err : retval;
1432 reiserfs_update_inode_transaction(old_dir);
1433 reiserfs_update_inode_transaction(new_dir);
1436 * this makes it so an fsync on an open fd for the old name will
1437 * commit the rename operation
1439 reiserfs_update_inode_transaction(old_inode);
1441 if (new_dentry_inode)
1442 reiserfs_update_inode_transaction(new_dentry_inode);
1446 * look for old name using corresponding entry key
1447 * (found by reiserfs_find_entry)
1450 search_by_entry_key(new_dir->i_sb, &old_de.de_entry_key,
1452 &old_de)) != NAME_FOUND) {
1453 pathrelse(&old_entry_path);
1455 reiserfs_write_unlock(old_dir->i_sb);
1459 copy_item_head(&old_entry_ih, tp_item_head(&old_entry_path));
1461 reiserfs_prepare_for_journal(old_inode->i_sb, old_de.de_bh, 1);
1463 /* look for new name by reiserfs_find_entry */
1464 new_de.de_gen_number_bit_string = NULL;
1466 reiserfs_find_entry(new_dir, new_dentry->d_name.name,
1467 new_dentry->d_name.len, &new_entry_path,
1470 * reiserfs_add_entry should not return IO_ERROR,
1471 * because it is called with essentially same parameters from
1472 * reiserfs_add_entry above, and we'll catch any i/o errors
1473 * before we get here.
1475 if (retval != NAME_FOUND_INVISIBLE && retval != NAME_FOUND) {
1476 pathrelse(&new_entry_path);
1477 pathrelse(&old_entry_path);
1479 reiserfs_write_unlock(old_dir->i_sb);
1483 copy_item_head(&new_entry_ih, tp_item_head(&new_entry_path));
1485 reiserfs_prepare_for_journal(old_inode->i_sb, new_de.de_bh, 1);
1487 if (S_ISDIR(old_inode->i_mode)) {
1489 search_by_entry_key(new_dir->i_sb,
1490 &dot_dot_de.de_entry_key,
1491 &dot_dot_entry_path,
1492 &dot_dot_de)) != NAME_FOUND) {
1493 pathrelse(&dot_dot_entry_path);
1494 pathrelse(&new_entry_path);
1495 pathrelse(&old_entry_path);
1497 reiserfs_write_unlock(old_dir->i_sb);
1500 copy_item_head(&dot_dot_ih,
1501 tp_item_head(&dot_dot_entry_path));
1502 /* node containing ".." gets into transaction */
1503 reiserfs_prepare_for_journal(old_inode->i_sb,
1504 dot_dot_de.de_bh, 1);
1507 * we should check seals here, not do
1508 * this stuff, yes? Then, having
1509 * gathered everything into RAM we
1510 * should lock the buffers, yes? -Hans
1513 * probably. our rename needs to hold more
1514 * than one path at once. The seals would
1515 * have to be written to deal with multi-path
1519 * sanity checking before doing the rename - avoid races many
1520 * of the above checks could have scheduled. We have to be
1521 * sure our items haven't been shifted by another process.
1523 if (item_moved(&new_entry_ih, &new_entry_path) ||
1524 !entry_points_to_object(new_dentry->d_name.name,
1525 new_dentry->d_name.len,
1526 &new_de, new_dentry_inode) ||
1527 item_moved(&old_entry_ih, &old_entry_path) ||
1528 !entry_points_to_object(old_dentry->d_name.name,
1529 old_dentry->d_name.len,
1530 &old_de, old_inode)) {
1531 reiserfs_restore_prepared_buffer(old_inode->i_sb,
1533 reiserfs_restore_prepared_buffer(old_inode->i_sb,
1535 if (S_ISDIR(old_inode_mode))
1536 reiserfs_restore_prepared_buffer(old_inode->
1542 if (S_ISDIR(old_inode_mode)) {
1543 if (item_moved(&dot_dot_ih, &dot_dot_entry_path) ||
1544 !entry_points_to_object("..", 2, &dot_dot_de,
1546 reiserfs_restore_prepared_buffer(old_inode->
1549 reiserfs_restore_prepared_buffer(old_inode->
1552 reiserfs_restore_prepared_buffer(old_inode->
1560 RFALSE(S_ISDIR(old_inode_mode) &&
1561 !buffer_journal_prepared(dot_dot_de.de_bh), "");
1567 * ok, all the changes can be done in one fell swoop when we
1568 * have claimed all the buffers needed.
1571 mark_de_visible(new_de.de_deh + new_de.de_entry_num);
1572 set_ino_in_dir_entry(&new_de, INODE_PKEY(old_inode));
1573 journal_mark_dirty(&th, new_de.de_bh);
1575 mark_de_hidden(old_de.de_deh + old_de.de_entry_num);
1576 journal_mark_dirty(&th, old_de.de_bh);
1577 ctime = current_time(old_dir);
1578 old_dir->i_ctime = old_dir->i_mtime = ctime;
1579 new_dir->i_ctime = new_dir->i_mtime = ctime;
1582 * which adds ctime update of renamed object
1584 old_inode->i_ctime = ctime;
1586 if (new_dentry_inode) {
1587 /* adjust link number of the victim */
1588 if (S_ISDIR(new_dentry_inode->i_mode)) {
1589 clear_nlink(new_dentry_inode);
1591 drop_nlink(new_dentry_inode);
1593 new_dentry_inode->i_ctime = ctime;
1594 savelink = new_dentry_inode->i_nlink;
1597 if (S_ISDIR(old_inode_mode)) {
1598 /* adjust ".." of renamed directory */
1599 set_ino_in_dir_entry(&dot_dot_de, INODE_PKEY(new_dir));
1600 journal_mark_dirty(&th, dot_dot_de.de_bh);
1603 * there (in new_dir) was no directory, so it got new link
1604 * (".." of renamed directory)
1606 if (!new_dentry_inode)
1607 INC_DIR_INODE_NLINK(new_dir);
1609 /* old directory lost one link - ".. " of renamed directory */
1610 DEC_DIR_INODE_NLINK(old_dir);
1613 * looks like in 2.3.99pre3 brelse is atomic.
1614 * so we can use pathrelse
1616 pathrelse(&new_entry_path);
1617 pathrelse(&dot_dot_entry_path);
1620 * FIXME: this reiserfs_cut_from_item's return value may screw up
1621 * anybody, but it will panic if will not be able to find the
1622 * entry. This needs one more clean up
1624 if (reiserfs_cut_from_item
1625 (&th, &old_entry_path, &old_de.de_entry_key, old_dir, NULL,
1627 reiserfs_error(old_dir->i_sb, "vs-7060",
1628 "couldn't not cut old name. Fsck later?");
1630 old_dir->i_size -= DEH_SIZE + old_de.de_entrylen;
1632 reiserfs_update_sd(&th, old_dir);
1633 reiserfs_update_sd(&th, new_dir);
1634 reiserfs_update_sd(&th, old_inode);
1636 if (new_dentry_inode) {
1638 add_save_link(&th, new_dentry_inode,
1639 0 /* not truncate */ );
1640 reiserfs_update_sd(&th, new_dentry_inode);
1643 retval = journal_end(&th);
1644 reiserfs_write_unlock(old_dir->i_sb);
1648 /* directories can handle most operations... */
1649 const struct inode_operations reiserfs_dir_inode_operations = {
1650 .create = reiserfs_create,
1651 .lookup = reiserfs_lookup,
1652 .link = reiserfs_link,
1653 .unlink = reiserfs_unlink,
1654 .symlink = reiserfs_symlink,
1655 .mkdir = reiserfs_mkdir,
1656 .rmdir = reiserfs_rmdir,
1657 .mknod = reiserfs_mknod,
1658 .rename = reiserfs_rename,
1659 .setattr = reiserfs_setattr,
1660 .listxattr = reiserfs_listxattr,
1661 .permission = reiserfs_permission,
1662 .get_acl = reiserfs_get_acl,
1663 .set_acl = reiserfs_set_acl,
1664 .fileattr_get = reiserfs_fileattr_get,
1665 .fileattr_set = reiserfs_fileattr_set,
1669 * symlink operations.. same as page_symlink_inode_operations, with xattr
1672 const struct inode_operations reiserfs_symlink_inode_operations = {
1673 .get_link = page_get_link,
1674 .setattr = reiserfs_setattr,
1675 .listxattr = reiserfs_listxattr,
1676 .permission = reiserfs_permission,
1680 * special file operations.. just xattr/acl stuff
1682 const struct inode_operations reiserfs_special_inode_operations = {
1683 .setattr = reiserfs_setattr,
1684 .listxattr = reiserfs_listxattr,
1685 .permission = reiserfs_permission,
1686 .get_acl = reiserfs_get_acl,
1687 .set_acl = reiserfs_set_acl,