1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * Create and rename file, directory, symlinks
8 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
10 * Portions of this code from linux/fs/ext3/dir.c
12 * Copyright (C) 1992, 1993, 1994, 1995
13 * Remy Card (card@masi.ibp.fr)
14 * Laboratoire MASI - Institut Blaise pascal
15 * Universite Pierre et Marie Curie (Paris VI)
19 * linux/fs/minix/dir.c
21 * Copyright (C) 1991, 1992 Linux Torvalds
23 * This program is free software; you can redistribute it and/or
24 * modify it under the terms of the GNU General Public
25 * License as published by the Free Software Foundation; either
26 * version 2 of the License, or (at your option) any later version.
28 * This program is distributed in the hope that it will be useful,
29 * but WITHOUT ANY WARRANTY; without even the implied warranty of
30 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
31 * General Public License for more details.
33 * You should have received a copy of the GNU General Public
34 * License along with this program; if not, write to the
35 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
36 * Boston, MA 021110-1307, USA.
40 #include <linux/types.h>
41 #include <linux/slab.h>
42 #include <linux/highmem.h>
44 #define MLOG_MASK_PREFIX ML_NAMEI
45 #include <cluster/masklog.h>
53 #include "extent_map.h"
65 #include "buffer_head_io.h"
67 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
69 struct dentry *dentry, int mode,
71 struct buffer_head **new_fe_bh,
72 struct buffer_head *parent_fe_bh,
74 struct inode **ret_inode,
75 struct ocfs2_alloc_context *inode_ac);
77 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
78 struct inode **ret_orphan_dir,
81 struct buffer_head **de_bh);
83 static int ocfs2_orphan_add(struct ocfs2_super *osb,
86 struct ocfs2_dinode *fe,
88 struct buffer_head *de_bh,
89 struct inode *orphan_dir_inode);
91 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
96 /* An orphan dir name is an 8 byte value, printed as a hex string */
97 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
99 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
100 struct nameidata *nd)
104 struct buffer_head *dirent_bh = NULL;
105 struct inode *inode = NULL;
107 struct ocfs2_dir_entry *dirent;
108 struct ocfs2_inode_info *oi;
110 mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
111 dentry->d_name.len, dentry->d_name.name);
113 if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
114 ret = ERR_PTR(-ENAMETOOLONG);
118 mlog(0, "find name %.*s in directory %llu\n", dentry->d_name.len,
119 dentry->d_name.name, (unsigned long long)OCFS2_I(dir)->ip_blkno);
121 status = ocfs2_meta_lock(dir, NULL, 0);
123 if (status != -ENOENT)
125 ret = ERR_PTR(status);
129 status = ocfs2_find_files_on_disk(dentry->d_name.name,
130 dentry->d_name.len, &blkno,
131 dir, &dirent_bh, &dirent);
135 inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0);
137 ret = ERR_PTR(-EACCES);
142 /* Clear any orphaned state... If we were able to look up the
143 * inode from a directory, it certainly can't be orphaned. We
144 * might have the bad state from a node which intended to
145 * orphan this inode but crashed before it could commit the
147 spin_lock(&oi->ip_lock);
148 oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
149 spin_unlock(&oi->ip_lock);
152 dentry->d_op = &ocfs2_dentry_ops;
153 ret = d_splice_alias(inode, dentry);
157 * If d_splice_alias() finds a DCACHE_DISCONNECTED
158 * dentry, it will d_move() it on top of ourse. The
159 * return value will indicate this however, so in
160 * those cases, we switch them around for the locking
163 * NOTE: This dentry already has ->d_op set from
164 * ocfs2_get_parent() and ocfs2_get_dentry()
169 status = ocfs2_dentry_attach_lock(dentry, inode,
170 OCFS2_I(dir)->ip_blkno);
173 ret = ERR_PTR(status);
179 /* Don't drop the cluster lock until *after* the d_add --
180 * unlink on another node will message us to remove that
181 * dentry under this lock so otherwise we can race this with
182 * the vote thread and have a stale dentry. */
183 ocfs2_meta_unlock(dir, 0);
194 static int ocfs2_mknod(struct inode *dir,
195 struct dentry *dentry,
200 struct buffer_head *parent_fe_bh = NULL;
201 handle_t *handle = NULL;
202 struct ocfs2_super *osb;
203 struct ocfs2_dinode *dirfe;
204 struct buffer_head *new_fe_bh = NULL;
205 struct buffer_head *de_bh = NULL;
206 struct inode *inode = NULL;
207 struct ocfs2_alloc_context *inode_ac = NULL;
208 struct ocfs2_alloc_context *data_ac = NULL;
210 mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
211 (unsigned long)dev, dentry->d_name.len,
212 dentry->d_name.name);
214 /* get our super block */
215 osb = OCFS2_SB(dir->i_sb);
217 status = ocfs2_meta_lock(dir, &parent_fe_bh, 1);
219 if (status != -ENOENT)
224 if (S_ISDIR(mode) && (dir->i_nlink >= OCFS2_LINK_MAX)) {
229 dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
230 if (!dirfe->i_links_count) {
231 /* can't make a file in a deleted directory. */
236 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
241 /* get a spot inside the dir. */
242 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
244 dentry->d_name.len, &de_bh);
250 /* reserve an inode spot */
251 status = ocfs2_reserve_new_inode(osb, &inode_ac);
253 if (status != -ENOSPC)
258 /* are we making a directory? If so, reserve a cluster for his
261 status = ocfs2_reserve_clusters(osb, 1, &data_ac);
263 if (status != -ENOSPC)
269 handle = ocfs2_start_trans(osb, OCFS2_MKNOD_CREDITS);
270 if (IS_ERR(handle)) {
271 status = PTR_ERR(handle);
277 /* do the real work now. */
278 status = ocfs2_mknod_locked(osb, dir, dentry, mode, dev,
279 &new_fe_bh, parent_fe_bh, handle,
287 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
294 status = ocfs2_journal_access(handle, dir, parent_fe_bh,
295 OCFS2_JOURNAL_ACCESS_WRITE);
300 le16_add_cpu(&dirfe->i_links_count, 1);
301 status = ocfs2_journal_dirty(handle, parent_fe_bh);
309 status = ocfs2_add_entry(handle, dentry, inode,
310 OCFS2_I(inode)->ip_blkno, parent_fe_bh,
317 status = ocfs2_dentry_attach_lock(dentry, inode,
318 OCFS2_I(dir)->ip_blkno);
324 insert_inode_hash(inode);
325 dentry->d_op = &ocfs2_dentry_ops;
326 d_instantiate(dentry, inode);
330 ocfs2_commit_trans(osb, handle);
332 ocfs2_meta_unlock(dir, 1);
334 if (status == -ENOSPC)
335 mlog(0, "Disk is full\n");
344 brelse(parent_fe_bh);
346 if ((status < 0) && inode)
350 ocfs2_free_alloc_context(inode_ac);
353 ocfs2_free_alloc_context(data_ac);
360 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
362 struct dentry *dentry, int mode,
364 struct buffer_head **new_fe_bh,
365 struct buffer_head *parent_fe_bh,
367 struct inode **ret_inode,
368 struct ocfs2_alloc_context *inode_ac)
371 struct ocfs2_dinode *fe = NULL;
372 struct ocfs2_extent_list *fel;
375 struct inode *inode = NULL;
377 mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
378 (unsigned long)dev, dentry->d_name.len,
379 dentry->d_name.name);
384 status = ocfs2_claim_new_inode(osb, handle, inode_ac, &suballoc_bit,
391 inode = new_inode(dir->i_sb);
393 status = PTR_ERR(inode);
394 mlog(ML_ERROR, "new_inode failed!\n");
398 /* populate as many fields early on as possible - many of
399 * these are used by the support functions here and in
401 inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
402 OCFS2_I(inode)->ip_blkno = fe_blkno;
407 inode->i_mode = mode;
408 spin_lock(&osb->osb_lock);
409 inode->i_generation = osb->s_next_generation++;
410 spin_unlock(&osb->osb_lock);
412 *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
418 ocfs2_set_new_buffer_uptodate(inode, *new_fe_bh);
420 status = ocfs2_journal_access(handle, inode, *new_fe_bh,
421 OCFS2_JOURNAL_ACCESS_CREATE);
427 fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
428 memset(fe, 0, osb->sb->s_blocksize);
430 fe->i_generation = cpu_to_le32(inode->i_generation);
431 fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
432 fe->i_blkno = cpu_to_le64(fe_blkno);
433 fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
434 fe->i_suballoc_slot = cpu_to_le16(osb->slot_num);
435 fe->i_uid = cpu_to_le32(current->fsuid);
436 if (dir->i_mode & S_ISGID) {
437 fe->i_gid = cpu_to_le32(dir->i_gid);
441 fe->i_gid = cpu_to_le32(current->fsgid);
442 fe->i_mode = cpu_to_le16(mode);
443 if (S_ISCHR(mode) || S_ISBLK(mode))
444 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
446 fe->i_links_count = cpu_to_le16(inode->i_nlink);
448 fe->i_last_eb_blk = 0;
449 strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
450 le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
451 fe->i_atime = fe->i_ctime = fe->i_mtime =
452 cpu_to_le64(CURRENT_TIME.tv_sec);
453 fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
454 cpu_to_le32(CURRENT_TIME.tv_nsec);
457 fel = &fe->id2.i_list;
458 fel->l_tree_depth = 0;
459 fel->l_next_free_rec = 0;
460 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
462 status = ocfs2_journal_dirty(handle, *new_fe_bh);
468 if (ocfs2_populate_inode(inode, fe, 1) < 0) {
469 mlog(ML_ERROR, "populate inode failed! bh->b_blocknr=%llu, "
470 "i_blkno=%llu, i_ino=%lu\n",
471 (unsigned long long)(*new_fe_bh)->b_blocknr,
472 (unsigned long long)le64_to_cpu(fe->i_blkno),
477 ocfs2_inode_set_new(osb, inode);
478 if (!ocfs2_mount_local(osb)) {
479 status = ocfs2_create_new_inode_locks(inode);
484 status = 0; /* error in ocfs2_create_new_inode_locks is not
502 static int ocfs2_mkdir(struct inode *dir,
503 struct dentry *dentry,
508 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
509 dentry->d_name.len, dentry->d_name.name);
510 ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
516 static int ocfs2_create(struct inode *dir,
517 struct dentry *dentry,
519 struct nameidata *nd)
523 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
524 dentry->d_name.len, dentry->d_name.name);
525 ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
531 static int ocfs2_link(struct dentry *old_dentry,
533 struct dentry *dentry)
536 struct inode *inode = old_dentry->d_inode;
538 struct buffer_head *fe_bh = NULL;
539 struct buffer_head *parent_fe_bh = NULL;
540 struct buffer_head *de_bh = NULL;
541 struct ocfs2_dinode *fe = NULL;
542 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
544 mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
545 old_dentry->d_name.len, old_dentry->d_name.name,
546 dentry->d_name.len, dentry->d_name.name);
548 if (S_ISDIR(inode->i_mode))
551 err = ocfs2_meta_lock(dir, &parent_fe_bh, 1);
563 err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
568 err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
570 dentry->d_name.len, &de_bh);
576 err = ocfs2_meta_lock(inode, &fe_bh, 1);
583 fe = (struct ocfs2_dinode *) fe_bh->b_data;
584 if (le16_to_cpu(fe->i_links_count) >= OCFS2_LINK_MAX) {
586 goto out_unlock_inode;
589 handle = ocfs2_start_trans(osb, OCFS2_LINK_CREDITS);
590 if (IS_ERR(handle)) {
591 err = PTR_ERR(handle);
594 goto out_unlock_inode;
597 err = ocfs2_journal_access(handle, inode, fe_bh,
598 OCFS2_JOURNAL_ACCESS_WRITE);
605 inode->i_ctime = CURRENT_TIME;
606 fe->i_links_count = cpu_to_le16(inode->i_nlink);
607 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
608 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
610 err = ocfs2_journal_dirty(handle, fe_bh);
612 le16_add_cpu(&fe->i_links_count, -1);
618 err = ocfs2_add_entry(handle, dentry, inode,
619 OCFS2_I(inode)->ip_blkno,
620 parent_fe_bh, de_bh);
622 le16_add_cpu(&fe->i_links_count, -1);
628 err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
634 atomic_inc(&inode->i_count);
635 dentry->d_op = &ocfs2_dentry_ops;
636 d_instantiate(dentry, inode);
639 ocfs2_commit_trans(osb, handle);
641 ocfs2_meta_unlock(inode, 1);
644 ocfs2_meta_unlock(dir, 1);
651 brelse(parent_fe_bh);
659 * Takes and drops an exclusive lock on the given dentry. This will
660 * force other nodes to drop it.
662 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
666 ret = ocfs2_dentry_lock(dentry, 1);
670 ocfs2_dentry_unlock(dentry, 1);
675 static inline int inode_is_unlinkable(struct inode *inode)
677 if (S_ISDIR(inode->i_mode)) {
678 if (inode->i_nlink == 2)
683 if (inode->i_nlink == 1)
688 static int ocfs2_unlink(struct inode *dir,
689 struct dentry *dentry)
692 int child_locked = 0;
693 struct inode *inode = dentry->d_inode;
694 struct inode *orphan_dir = NULL;
695 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
697 struct ocfs2_dinode *fe = NULL;
698 struct buffer_head *fe_bh = NULL;
699 struct buffer_head *parent_node_bh = NULL;
700 handle_t *handle = NULL;
701 struct ocfs2_dir_entry *dirent = NULL;
702 struct buffer_head *dirent_bh = NULL;
703 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
704 struct buffer_head *orphan_entry_bh = NULL;
706 mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
707 dentry->d_name.len, dentry->d_name.name);
709 BUG_ON(dentry->d_parent->d_inode != dir);
711 mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
713 if (inode == osb->root_inode) {
714 mlog(0, "Cannot delete the root directory\n");
718 status = ocfs2_meta_lock(dir, &parent_node_bh, 1);
720 if (status != -ENOENT)
725 status = ocfs2_find_files_on_disk(dentry->d_name.name,
726 dentry->d_name.len, &blkno,
727 dir, &dirent_bh, &dirent);
729 if (status != -ENOENT)
734 if (OCFS2_I(inode)->ip_blkno != blkno) {
737 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
738 (unsigned long long)OCFS2_I(inode)->ip_blkno,
739 (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
743 status = ocfs2_meta_lock(inode, &fe_bh, 1);
745 if (status != -ENOENT)
751 if (S_ISDIR(inode->i_mode)) {
752 if (!ocfs2_empty_dir(inode)) {
755 } else if (inode->i_nlink != 2) {
761 status = ocfs2_remote_dentry_delete(dentry);
763 /* This vote should succeed under all normal
769 if (inode_is_unlinkable(inode)) {
770 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir, inode,
779 handle = ocfs2_start_trans(osb, OCFS2_UNLINK_CREDITS);
780 if (IS_ERR(handle)) {
781 status = PTR_ERR(handle);
787 status = ocfs2_journal_access(handle, inode, fe_bh,
788 OCFS2_JOURNAL_ACCESS_WRITE);
794 fe = (struct ocfs2_dinode *) fe_bh->b_data;
796 if (inode_is_unlinkable(inode)) {
797 status = ocfs2_orphan_add(osb, handle, inode, fe, orphan_name,
798 orphan_entry_bh, orphan_dir);
805 /* delete the name from the parent dir */
806 status = ocfs2_delete_entry(handle, dir, dirent, dirent_bh);
812 if (S_ISDIR(inode->i_mode))
815 fe->i_links_count = cpu_to_le16(inode->i_nlink);
817 status = ocfs2_journal_dirty(handle, fe_bh);
823 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
824 if (S_ISDIR(inode->i_mode))
827 status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
830 if (S_ISDIR(inode->i_mode))
836 ocfs2_commit_trans(osb, handle);
839 ocfs2_meta_unlock(inode, 1);
841 ocfs2_meta_unlock(dir, 1);
844 /* This was locked for us in ocfs2_prepare_orphan_dir() */
845 ocfs2_meta_unlock(orphan_dir, 1);
846 mutex_unlock(&orphan_dir->i_mutex);
857 brelse(parent_node_bh);
860 brelse(orphan_entry_bh);
868 * The only place this should be used is rename!
869 * if they have the same id, then the 1st one is the only one locked.
871 static int ocfs2_double_lock(struct ocfs2_super *osb,
872 struct buffer_head **bh1,
873 struct inode *inode1,
874 struct buffer_head **bh2,
875 struct inode *inode2)
878 struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
879 struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
880 struct buffer_head **tmpbh;
881 struct inode *tmpinode;
883 mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
884 (unsigned long long)oi1->ip_blkno,
885 (unsigned long long)oi2->ip_blkno);
892 /* we always want to lock the one with the lower lockid first. */
893 if (oi1->ip_blkno != oi2->ip_blkno) {
894 if (oi1->ip_blkno < oi2->ip_blkno) {
895 /* switch id1 and id2 around */
896 mlog(0, "switching them around...\n");
906 status = ocfs2_meta_lock(inode2, bh2, 1);
908 if (status != -ENOENT)
915 status = ocfs2_meta_lock(inode1, bh1, 1);
918 * An error return must mean that no cluster locks
919 * were held on function exit.
921 if (oi1->ip_blkno != oi2->ip_blkno)
922 ocfs2_meta_unlock(inode2, 1);
924 if (status != -ENOENT)
933 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
935 ocfs2_meta_unlock(inode1, 1);
937 if (inode1 != inode2)
938 ocfs2_meta_unlock(inode2, 1);
941 #define PARENT_INO(buffer) \
942 ((struct ocfs2_dir_entry *) \
944 le16_to_cpu(((struct ocfs2_dir_entry *)buffer)->rec_len)))->inode
946 static int ocfs2_rename(struct inode *old_dir,
947 struct dentry *old_dentry,
948 struct inode *new_dir,
949 struct dentry *new_dentry)
951 int status = 0, rename_lock = 0, parents_locked = 0;
952 int old_child_locked = 0, new_child_locked = 0;
953 struct inode *old_inode = old_dentry->d_inode;
954 struct inode *new_inode = new_dentry->d_inode;
955 struct inode *orphan_dir = NULL;
956 struct ocfs2_dinode *newfe = NULL;
957 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
958 struct buffer_head *orphan_entry_bh = NULL;
959 struct buffer_head *newfe_bh = NULL;
960 struct buffer_head *old_inode_bh = NULL;
961 struct buffer_head *insert_entry_bh = NULL;
962 struct ocfs2_super *osb = NULL;
964 handle_t *handle = NULL;
965 struct buffer_head *old_dir_bh = NULL;
966 struct buffer_head *new_dir_bh = NULL;
967 struct ocfs2_dir_entry *old_de = NULL, *new_de = NULL; // dirent for old_dentry
969 struct buffer_head *new_de_bh = NULL, *old_de_bh = NULL; // bhs for above
970 struct buffer_head *old_inode_de_bh = NULL; // if old_dentry is a dir,
971 // this is the 1st dirent bh
972 nlink_t old_dir_nlink = old_dir->i_nlink;
973 struct ocfs2_dinode *old_di;
975 /* At some point it might be nice to break this function up a
978 mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
979 old_dir, old_dentry, new_dir, new_dentry,
980 old_dentry->d_name.len, old_dentry->d_name.name,
981 new_dentry->d_name.len, new_dentry->d_name.name);
983 osb = OCFS2_SB(old_dir->i_sb);
986 if (!igrab(new_inode))
990 /* Assume a directory hierarchy thusly:
993 * a,b,c, and d are all directories.
995 * from cwd of 'a' on both nodes:
999 * And that's why, just like the VFS, we need a file system
1001 if (old_dentry != new_dentry) {
1002 status = ocfs2_rename_lock(osb);
1010 /* if old and new are the same, this'll just do one lock. */
1011 status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1012 &new_dir_bh, new_dir);
1019 /* make sure both dirs have bhs
1020 * get an extra ref on old_dir_bh if old==new */
1023 new_dir_bh = old_dir_bh;
1026 mlog(ML_ERROR, "no old_dir_bh!\n");
1033 * Aside from allowing a meta data update, the locking here
1034 * also ensures that the vote thread on other nodes won't have
1035 * to concurrently downconvert the inode and the dentry locks.
1037 status = ocfs2_meta_lock(old_inode, &old_inode_bh, 1);
1039 if (status != -ENOENT)
1043 old_child_locked = 1;
1045 status = ocfs2_remote_dentry_delete(old_dentry);
1051 if (S_ISDIR(old_inode->i_mode)) {
1053 old_inode_de_bh = ocfs2_bread(old_inode, 0, &status, 0);
1054 if (!old_inode_de_bh)
1058 if (le64_to_cpu(PARENT_INO(old_inode_de_bh->b_data)) !=
1059 OCFS2_I(old_dir)->ip_blkno)
1062 if (!new_inode && new_dir!=old_dir &&
1063 new_dir->i_nlink >= OCFS2_LINK_MAX)
1068 old_de_bh = ocfs2_find_entry(old_dentry->d_name.name,
1069 old_dentry->d_name.len,
1075 * Check for inode number is _not_ due to possible IO errors.
1076 * We might rmdir the source, keep it as pwd of some process
1077 * and merrily kill the link to whatever was created under the
1078 * same name. Goodbye sticky bit ;-<
1080 if (le64_to_cpu(old_de->inode) != OCFS2_I(old_inode)->ip_blkno)
1083 /* check if the target already exists (in which case we need
1085 status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1086 new_dentry->d_name.len,
1087 &newfe_blkno, new_dir, &new_de_bh,
1089 /* The only error we allow here is -ENOENT because the new
1090 * file not existing is perfectly valid. */
1091 if ((status < 0) && (status != -ENOENT)) {
1092 /* If we cannot find the file specified we should just */
1093 /* return the error... */
1098 if (!new_de && new_inode)
1099 mlog(ML_ERROR, "inode %lu does not exist in it's parent "
1100 "directory!", new_inode->i_ino);
1102 /* In case we need to overwrite an existing file, we blow it
1105 /* VFS didn't think there existed an inode here, but
1106 * someone else in the cluster must have raced our
1107 * rename to create one. Today we error cleanly, in
1108 * the future we should consider calling iget to build
1109 * a new struct inode for this entry. */
1113 mlog(0, "We found an inode for name %.*s but VFS "
1114 "didn't give us one.\n", new_dentry->d_name.len,
1115 new_dentry->d_name.name);
1119 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1122 mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1123 (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1124 (unsigned long long)newfe_blkno,
1125 OCFS2_I(new_inode)->ip_flags);
1129 status = ocfs2_meta_lock(new_inode, &newfe_bh, 1);
1131 if (status != -ENOENT)
1135 new_child_locked = 1;
1137 status = ocfs2_remote_dentry_delete(new_dentry);
1143 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1145 mlog(0, "aha rename over existing... new_de=%p new_blkno=%llu "
1146 "newfebh=%p bhblocknr=%llu\n", new_de,
1147 (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1148 (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1150 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1151 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1161 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1163 status = ocfs2_check_dir_for_entry(new_dir,
1164 new_dentry->d_name.name,
1165 new_dentry->d_name.len);
1169 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1170 new_dentry->d_name.name,
1171 new_dentry->d_name.len,
1179 handle = ocfs2_start_trans(osb, OCFS2_RENAME_CREDITS);
1180 if (IS_ERR(handle)) {
1181 status = PTR_ERR(handle);
1188 if (S_ISDIR(new_inode->i_mode)) {
1189 if (!ocfs2_empty_dir(new_inode) ||
1190 new_inode->i_nlink != 2) {
1191 status = -ENOTEMPTY;
1195 status = ocfs2_journal_access(handle, new_inode, newfe_bh,
1196 OCFS2_JOURNAL_ACCESS_WRITE);
1202 if (S_ISDIR(new_inode->i_mode) ||
1203 (newfe->i_links_count == cpu_to_le16(1))){
1204 status = ocfs2_orphan_add(osb, handle, new_inode,
1206 orphan_entry_bh, orphan_dir);
1213 /* change the dirent to point to the correct inode */
1214 status = ocfs2_journal_access(handle, new_dir, new_de_bh,
1215 OCFS2_JOURNAL_ACCESS_WRITE);
1220 new_de->inode = cpu_to_le64(OCFS2_I(old_inode)->ip_blkno);
1221 new_de->file_type = old_de->file_type;
1222 new_dir->i_version++;
1223 status = ocfs2_journal_dirty(handle, new_de_bh);
1229 if (S_ISDIR(new_inode->i_mode))
1230 newfe->i_links_count = 0;
1232 le16_add_cpu(&newfe->i_links_count, -1);
1234 status = ocfs2_journal_dirty(handle, newfe_bh);
1240 /* if the name was not found in new_dir, add it now */
1241 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1242 OCFS2_I(old_inode)->ip_blkno,
1243 new_dir_bh, insert_entry_bh);
1246 old_inode->i_ctime = CURRENT_TIME;
1247 mark_inode_dirty(old_inode);
1249 status = ocfs2_journal_access(handle, old_inode, old_inode_bh,
1250 OCFS2_JOURNAL_ACCESS_WRITE);
1252 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1254 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1255 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1257 status = ocfs2_journal_dirty(handle, old_inode_bh);
1263 /* now that the name has been added to new_dir, remove the old name */
1264 status = ocfs2_delete_entry(handle, old_dir, old_de, old_de_bh);
1271 new_inode->i_nlink--;
1272 new_inode->i_ctime = CURRENT_TIME;
1274 old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1275 if (old_inode_de_bh) {
1276 status = ocfs2_journal_access(handle, old_inode,
1278 OCFS2_JOURNAL_ACCESS_WRITE);
1279 PARENT_INO(old_inode_de_bh->b_data) =
1280 cpu_to_le64(OCFS2_I(new_dir)->ip_blkno);
1281 status = ocfs2_journal_dirty(handle, old_inode_de_bh);
1284 new_inode->i_nlink--;
1287 mark_inode_dirty(new_dir);
1290 mark_inode_dirty(old_dir);
1291 ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1293 mark_inode_dirty(new_inode);
1294 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1297 if (old_dir != new_dir) {
1298 /* Keep the same times on both directories.*/
1299 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1302 * This will also pick up the i_nlink change from the
1305 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1308 if (old_dir_nlink != old_dir->i_nlink) {
1310 mlog(ML_ERROR, "need to change nlink for old dir "
1311 "%llu from %d to %d but bh is NULL!\n",
1312 (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1313 (int)old_dir_nlink, old_dir->i_nlink);
1315 struct ocfs2_dinode *fe;
1316 status = ocfs2_journal_access(handle, old_dir,
1318 OCFS2_JOURNAL_ACCESS_WRITE);
1319 fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1320 fe->i_links_count = cpu_to_le16(old_dir->i_nlink);
1321 status = ocfs2_journal_dirty(handle, old_dir_bh);
1325 ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1329 ocfs2_rename_unlock(osb);
1332 ocfs2_commit_trans(osb, handle);
1335 ocfs2_double_unlock(old_dir, new_dir);
1337 if (old_child_locked)
1338 ocfs2_meta_unlock(old_inode, 1);
1340 if (new_child_locked)
1341 ocfs2_meta_unlock(new_inode, 1);
1344 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1345 ocfs2_meta_unlock(orphan_dir, 1);
1346 mutex_unlock(&orphan_dir->i_mutex);
1351 sync_mapping_buffers(old_inode->i_mapping);
1358 brelse(old_inode_bh);
1367 if (old_inode_de_bh)
1368 brelse(old_inode_de_bh);
1369 if (orphan_entry_bh)
1370 brelse(orphan_entry_bh);
1371 if (insert_entry_bh)
1372 brelse(insert_entry_bh);
1380 * we expect i_size = strlen(symname). Copy symname into the file
1381 * data, including the null terminator.
1383 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1385 struct inode *inode,
1386 const char *symname)
1388 struct buffer_head **bhs = NULL;
1390 struct super_block *sb = osb->sb;
1391 u64 p_blkno, p_blocks;
1392 int virtual, blocks, status, i, bytes_left;
1394 bytes_left = i_size_read(inode) + 1;
1395 /* we can't trust i_blocks because we're actually going to
1396 * write i_size + 1 bytes. */
1397 blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1399 mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1400 (unsigned long long)inode->i_blocks,
1401 i_size_read(inode), blocks);
1403 /* Sanity check -- make sure we're going to fit. */
1405 ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1411 bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1418 status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1425 /* links can never be larger than one cluster so we know this
1426 * is all going to be contiguous, but do a sanity check
1428 if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1435 while(bytes_left > 0) {
1436 c = &symname[virtual * sb->s_blocksize];
1438 bhs[virtual] = sb_getblk(sb, p_blkno);
1439 if (!bhs[virtual]) {
1444 ocfs2_set_new_buffer_uptodate(inode, bhs[virtual]);
1446 status = ocfs2_journal_access(handle, inode, bhs[virtual],
1447 OCFS2_JOURNAL_ACCESS_CREATE);
1453 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1455 memcpy(bhs[virtual]->b_data, c,
1456 (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1459 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1467 bytes_left -= sb->s_blocksize;
1474 for(i = 0; i < blocks; i++)
1484 static int ocfs2_symlink(struct inode *dir,
1485 struct dentry *dentry,
1486 const char *symname)
1488 int status, l, credits;
1490 struct ocfs2_super *osb = NULL;
1491 struct inode *inode = NULL;
1492 struct super_block *sb;
1493 struct buffer_head *new_fe_bh = NULL;
1494 struct buffer_head *de_bh = NULL;
1495 struct buffer_head *parent_fe_bh = NULL;
1496 struct ocfs2_dinode *fe = NULL;
1497 struct ocfs2_dinode *dirfe;
1498 handle_t *handle = NULL;
1499 struct ocfs2_alloc_context *inode_ac = NULL;
1500 struct ocfs2_alloc_context *data_ac = NULL;
1502 mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1503 dentry, symname, dentry->d_name.len, dentry->d_name.name);
1508 l = strlen(symname) + 1;
1510 credits = ocfs2_calc_symlink_credits(sb);
1512 /* lock the parent directory */
1513 status = ocfs2_meta_lock(dir, &parent_fe_bh, 1);
1515 if (status != -ENOENT)
1520 dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1521 if (!dirfe->i_links_count) {
1522 /* can't make a file in a deleted directory. */
1527 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1528 dentry->d_name.len);
1532 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1533 dentry->d_name.name,
1534 dentry->d_name.len, &de_bh);
1540 status = ocfs2_reserve_new_inode(osb, &inode_ac);
1542 if (status != -ENOSPC)
1547 /* don't reserve bitmap space for fast symlinks. */
1548 if (l > ocfs2_fast_symlink_chars(sb)) {
1549 status = ocfs2_reserve_clusters(osb, 1, &data_ac);
1551 if (status != -ENOSPC)
1557 handle = ocfs2_start_trans(osb, credits);
1558 if (IS_ERR(handle)) {
1559 status = PTR_ERR(handle);
1565 status = ocfs2_mknod_locked(osb, dir, dentry,
1566 S_IFLNK | S_IRWXUGO, 0,
1567 &new_fe_bh, parent_fe_bh, handle,
1574 fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1577 if (l > ocfs2_fast_symlink_chars(sb)) {
1580 inode->i_op = &ocfs2_symlink_inode_operations;
1581 status = ocfs2_do_extend_allocation(osb, inode, &offset, 1, 0,
1583 handle, data_ac, NULL,
1586 if (status != -ENOSPC && status != -EINTR) {
1588 "Failed to extend file to %llu\n",
1589 (unsigned long long)newsize);
1595 i_size_write(inode, newsize);
1596 inode->i_blocks = ocfs2_inode_sector_count(inode);
1598 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1599 memcpy((char *) fe->id2.i_symlink, symname, l);
1600 i_size_write(inode, newsize);
1601 inode->i_blocks = 0;
1604 status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1610 if (!ocfs2_inode_is_fast_symlink(inode)) {
1611 status = ocfs2_create_symlink_data(osb, handle, inode,
1619 status = ocfs2_add_entry(handle, dentry, inode,
1620 le64_to_cpu(fe->i_blkno), parent_fe_bh,
1627 status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1633 insert_inode_hash(inode);
1634 dentry->d_op = &ocfs2_dentry_ops;
1635 d_instantiate(dentry, inode);
1638 ocfs2_commit_trans(osb, handle);
1640 ocfs2_meta_unlock(dir, 1);
1645 brelse(parent_fe_bh);
1649 ocfs2_free_alloc_context(inode_ac);
1651 ocfs2_free_alloc_context(data_ac);
1652 if ((status < 0) && inode)
1660 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1662 int status, namelen;
1666 namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1676 if (namelen != OCFS2_ORPHAN_NAMELEN) {
1682 mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
1691 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1692 struct inode **ret_orphan_dir,
1693 struct inode *inode,
1695 struct buffer_head **de_bh)
1697 struct inode *orphan_dir_inode;
1698 struct buffer_head *orphan_dir_bh = NULL;
1701 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1707 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1708 ORPHAN_DIR_SYSTEM_INODE,
1710 if (!orphan_dir_inode) {
1716 mutex_lock(&orphan_dir_inode->i_mutex);
1718 status = ocfs2_meta_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1724 status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1725 orphan_dir_bh, name,
1726 OCFS2_ORPHAN_NAMELEN, de_bh);
1728 ocfs2_meta_unlock(orphan_dir_inode, 1);
1734 *ret_orphan_dir = orphan_dir_inode;
1738 mutex_unlock(&orphan_dir_inode->i_mutex);
1739 iput(orphan_dir_inode);
1743 brelse(orphan_dir_bh);
1749 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1751 struct inode *inode,
1752 struct ocfs2_dinode *fe,
1754 struct buffer_head *de_bh,
1755 struct inode *orphan_dir_inode)
1757 struct buffer_head *orphan_dir_bh = NULL;
1759 struct ocfs2_dinode *orphan_fe;
1761 mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
1763 status = ocfs2_read_block(osb,
1764 OCFS2_I(orphan_dir_inode)->ip_blkno,
1765 &orphan_dir_bh, OCFS2_BH_CACHED,
1772 status = ocfs2_journal_access(handle, orphan_dir_inode, orphan_dir_bh,
1773 OCFS2_JOURNAL_ACCESS_WRITE);
1779 /* we're a cluster, and nlink can change on disk from
1780 * underneath us... */
1781 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1782 if (S_ISDIR(inode->i_mode))
1783 le16_add_cpu(&orphan_fe->i_links_count, 1);
1784 orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
1786 status = ocfs2_journal_dirty(handle, orphan_dir_bh);
1792 status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
1793 OCFS2_ORPHAN_NAMELEN, inode,
1794 OCFS2_I(inode)->ip_blkno,
1795 orphan_dir_bh, de_bh);
1801 le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
1803 /* Record which orphan dir our inode now resides
1804 * in. delete_inode will use this to determine which orphan
1806 fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
1808 mlog(0, "Inode %llu orphaned in slot %d\n",
1809 (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
1813 brelse(orphan_dir_bh);
1819 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
1820 int ocfs2_orphan_del(struct ocfs2_super *osb,
1822 struct inode *orphan_dir_inode,
1823 struct inode *inode,
1824 struct buffer_head *orphan_dir_bh)
1826 char name[OCFS2_ORPHAN_NAMELEN + 1];
1827 struct ocfs2_dinode *orphan_fe;
1829 struct buffer_head *target_de_bh = NULL;
1830 struct ocfs2_dir_entry *target_de = NULL;
1834 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1840 mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
1841 name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
1842 OCFS2_ORPHAN_NAMELEN);
1844 /* find it's spot in the orphan directory */
1845 target_de_bh = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN,
1846 orphan_dir_inode, &target_de);
1847 if (!target_de_bh) {
1853 /* remove it from the orphan directory */
1854 status = ocfs2_delete_entry(handle, orphan_dir_inode, target_de,
1861 status = ocfs2_journal_access(handle,orphan_dir_inode, orphan_dir_bh,
1862 OCFS2_JOURNAL_ACCESS_WRITE);
1868 /* do the i_nlink dance! :) */
1869 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1870 if (S_ISDIR(inode->i_mode))
1871 le16_add_cpu(&orphan_fe->i_links_count, -1);
1872 orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
1874 status = ocfs2_journal_dirty(handle, orphan_dir_bh);
1882 brelse(target_de_bh);
1888 const struct inode_operations ocfs2_dir_iops = {
1889 .create = ocfs2_create,
1890 .lookup = ocfs2_lookup,
1892 .unlink = ocfs2_unlink,
1893 .rmdir = ocfs2_unlink,
1894 .symlink = ocfs2_symlink,
1895 .mkdir = ocfs2_mkdir,
1896 .mknod = ocfs2_mknod,
1897 .rename = ocfs2_rename,
1898 .setattr = ocfs2_setattr,
1899 .getattr = ocfs2_getattr,
1900 .permission = ocfs2_permission,