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"
66 #include "buffer_head_io.h"
68 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
71 struct dentry *dentry,
73 struct buffer_head **new_fe_bh,
74 struct buffer_head *parent_fe_bh,
76 struct ocfs2_alloc_context *inode_ac);
78 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
79 struct inode **ret_orphan_dir,
82 struct buffer_head **de_bh);
84 static int ocfs2_orphan_add(struct ocfs2_super *osb,
87 struct ocfs2_dinode *fe,
89 struct buffer_head *de_bh,
90 struct inode *orphan_dir_inode);
92 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
97 /* An orphan dir name is an 8 byte value, printed as a hex string */
98 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
100 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
101 struct nameidata *nd)
105 struct inode *inode = NULL;
107 struct ocfs2_inode_info *oi;
109 mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
110 dentry->d_name.len, dentry->d_name.name);
112 if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
113 ret = ERR_PTR(-ENAMETOOLONG);
117 mlog(0, "find name %.*s in directory %llu\n", dentry->d_name.len,
118 dentry->d_name.name, (unsigned long long)OCFS2_I(dir)->ip_blkno);
120 status = ocfs2_inode_lock(dir, NULL, 0);
122 if (status != -ENOENT)
124 ret = ERR_PTR(status);
128 status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
129 dentry->d_name.len, &blkno);
133 inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
135 ret = ERR_PTR(-EACCES);
140 /* Clear any orphaned state... If we were able to look up the
141 * inode from a directory, it certainly can't be orphaned. We
142 * might have the bad state from a node which intended to
143 * orphan this inode but crashed before it could commit the
145 spin_lock(&oi->ip_lock);
146 oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
147 spin_unlock(&oi->ip_lock);
150 dentry->d_op = &ocfs2_dentry_ops;
151 ret = d_splice_alias(inode, dentry);
155 * If d_splice_alias() finds a DCACHE_DISCONNECTED
156 * dentry, it will d_move() it on top of ourse. The
157 * return value will indicate this however, so in
158 * those cases, we switch them around for the locking
161 * NOTE: This dentry already has ->d_op set from
162 * ocfs2_get_parent() and ocfs2_get_dentry()
167 status = ocfs2_dentry_attach_lock(dentry, inode,
168 OCFS2_I(dir)->ip_blkno);
171 ret = ERR_PTR(status);
177 /* Don't drop the cluster lock until *after* the d_add --
178 * unlink on another node will message us to remove that
179 * dentry under this lock so otherwise we can race this with
180 * the downconvert thread and have a stale dentry. */
181 ocfs2_inode_unlock(dir, 0);
190 static struct inode *ocfs2_get_init_inode(struct inode *dir, int mode)
194 inode = new_inode(dir->i_sb);
196 mlog(ML_ERROR, "new_inode failed!\n");
200 /* populate as many fields early on as possible - many of
201 * these are used by the support functions here and in
207 inode->i_uid = current_fsuid();
208 if (dir->i_mode & S_ISGID) {
209 inode->i_gid = dir->i_gid;
213 inode->i_gid = current_fsgid();
214 inode->i_mode = mode;
218 static int ocfs2_mknod(struct inode *dir,
219 struct dentry *dentry,
224 struct buffer_head *parent_fe_bh = NULL;
225 handle_t *handle = NULL;
226 struct ocfs2_super *osb;
227 struct ocfs2_dinode *dirfe;
228 struct buffer_head *new_fe_bh = NULL;
229 struct buffer_head *de_bh = NULL;
230 struct inode *inode = NULL;
231 struct ocfs2_alloc_context *inode_ac = NULL;
232 struct ocfs2_alloc_context *data_ac = NULL;
233 struct ocfs2_alloc_context *xattr_ac = NULL;
234 int want_clusters = 0;
235 int xattr_credits = 0;
236 struct ocfs2_security_xattr_info si = {
240 mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
241 (unsigned long)dev, dentry->d_name.len,
242 dentry->d_name.name);
244 /* get our super block */
245 osb = OCFS2_SB(dir->i_sb);
247 status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
249 if (status != -ENOENT)
254 if (S_ISDIR(mode) && (dir->i_nlink >= OCFS2_LINK_MAX)) {
259 dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
260 if (!dirfe->i_links_count) {
261 /* can't make a file in a deleted directory. */
266 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
271 /* get a spot inside the dir. */
272 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
274 dentry->d_name.len, &de_bh);
280 /* reserve an inode spot */
281 status = ocfs2_reserve_new_inode(osb, &inode_ac);
283 if (status != -ENOSPC)
288 inode = ocfs2_get_init_inode(dir, mode);
295 /* get security xattr */
296 status = ocfs2_init_security_get(inode, dir, &si);
298 if (status == -EOPNOTSUPP)
306 /* calculate meta data/clusters for setting security and acl xattr */
307 status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
309 &xattr_credits, &xattr_ac);
315 /* Reserve a cluster if creating an extent based directory. */
316 if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb))
319 status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
321 if (status != -ENOSPC)
326 handle = ocfs2_start_trans(osb, OCFS2_MKNOD_CREDITS + xattr_credits);
327 if (IS_ERR(handle)) {
328 status = PTR_ERR(handle);
334 /* do the real work now. */
335 status = ocfs2_mknod_locked(osb, dir, inode, dentry, dev,
336 &new_fe_bh, parent_fe_bh, handle,
344 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
351 status = ocfs2_journal_access(handle, dir, parent_fe_bh,
352 OCFS2_JOURNAL_ACCESS_WRITE);
357 le16_add_cpu(&dirfe->i_links_count, 1);
358 status = ocfs2_journal_dirty(handle, parent_fe_bh);
366 status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
374 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
382 status = ocfs2_add_entry(handle, dentry, inode,
383 OCFS2_I(inode)->ip_blkno, parent_fe_bh,
390 status = ocfs2_dentry_attach_lock(dentry, inode,
391 OCFS2_I(dir)->ip_blkno);
397 insert_inode_hash(inode);
398 dentry->d_op = &ocfs2_dentry_ops;
399 d_instantiate(dentry, inode);
403 ocfs2_commit_trans(osb, handle);
405 ocfs2_inode_unlock(dir, 1);
407 if (status == -ENOSPC)
408 mlog(0, "Disk is full\n");
412 brelse(parent_fe_bh);
416 if ((status < 0) && inode) {
422 ocfs2_free_alloc_context(inode_ac);
425 ocfs2_free_alloc_context(data_ac);
428 ocfs2_free_alloc_context(xattr_ac);
435 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
438 struct dentry *dentry,
440 struct buffer_head **new_fe_bh,
441 struct buffer_head *parent_fe_bh,
443 struct ocfs2_alloc_context *inode_ac)
446 struct ocfs2_dinode *fe = NULL;
447 struct ocfs2_extent_list *fel;
451 mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry,
452 inode->i_mode, (unsigned long)dev, dentry->d_name.len,
453 dentry->d_name.name);
457 status = ocfs2_claim_new_inode(osb, handle, inode_ac, &suballoc_bit,
464 /* populate as many fields early on as possible - many of
465 * these are used by the support functions here and in
467 inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
468 OCFS2_I(inode)->ip_blkno = fe_blkno;
469 spin_lock(&osb->osb_lock);
470 inode->i_generation = osb->s_next_generation++;
471 spin_unlock(&osb->osb_lock);
473 *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
479 ocfs2_set_new_buffer_uptodate(inode, *new_fe_bh);
481 status = ocfs2_journal_access(handle, inode, *new_fe_bh,
482 OCFS2_JOURNAL_ACCESS_CREATE);
488 fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
489 memset(fe, 0, osb->sb->s_blocksize);
491 fe->i_generation = cpu_to_le32(inode->i_generation);
492 fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
493 fe->i_blkno = cpu_to_le64(fe_blkno);
494 fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
495 fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
496 fe->i_uid = cpu_to_le32(inode->i_uid);
497 fe->i_gid = cpu_to_le32(inode->i_gid);
498 fe->i_mode = cpu_to_le16(inode->i_mode);
499 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
500 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
501 fe->i_links_count = cpu_to_le16(inode->i_nlink);
503 fe->i_last_eb_blk = 0;
504 strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
505 le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
506 fe->i_atime = fe->i_ctime = fe->i_mtime =
507 cpu_to_le64(CURRENT_TIME.tv_sec);
508 fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
509 cpu_to_le32(CURRENT_TIME.tv_nsec);
513 * If supported, directories start with inline data.
515 if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
516 u16 feat = le16_to_cpu(fe->i_dyn_features);
518 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
520 fe->id2.i_data.id_count = cpu_to_le16(ocfs2_max_inline_data(osb->sb));
522 fel = &fe->id2.i_list;
523 fel->l_tree_depth = 0;
524 fel->l_next_free_rec = 0;
525 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
528 status = ocfs2_journal_dirty(handle, *new_fe_bh);
534 ocfs2_populate_inode(inode, fe, 1);
535 ocfs2_inode_set_new(osb, inode);
536 if (!ocfs2_mount_local(osb)) {
537 status = ocfs2_create_new_inode_locks(inode);
542 status = 0; /* error in ocfs2_create_new_inode_locks is not
557 static int ocfs2_mkdir(struct inode *dir,
558 struct dentry *dentry,
563 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
564 dentry->d_name.len, dentry->d_name.name);
565 ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
571 static int ocfs2_create(struct inode *dir,
572 struct dentry *dentry,
574 struct nameidata *nd)
578 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
579 dentry->d_name.len, dentry->d_name.name);
580 ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
586 static int ocfs2_link(struct dentry *old_dentry,
588 struct dentry *dentry)
591 struct inode *inode = old_dentry->d_inode;
593 struct buffer_head *fe_bh = NULL;
594 struct buffer_head *parent_fe_bh = NULL;
595 struct buffer_head *de_bh = NULL;
596 struct ocfs2_dinode *fe = NULL;
597 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
599 mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
600 old_dentry->d_name.len, old_dentry->d_name.name,
601 dentry->d_name.len, dentry->d_name.name);
603 if (S_ISDIR(inode->i_mode))
606 err = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
618 err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
623 err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
625 dentry->d_name.len, &de_bh);
631 err = ocfs2_inode_lock(inode, &fe_bh, 1);
638 fe = (struct ocfs2_dinode *) fe_bh->b_data;
639 if (le16_to_cpu(fe->i_links_count) >= OCFS2_LINK_MAX) {
641 goto out_unlock_inode;
644 handle = ocfs2_start_trans(osb, OCFS2_LINK_CREDITS);
645 if (IS_ERR(handle)) {
646 err = PTR_ERR(handle);
649 goto out_unlock_inode;
652 err = ocfs2_journal_access(handle, inode, fe_bh,
653 OCFS2_JOURNAL_ACCESS_WRITE);
660 inode->i_ctime = CURRENT_TIME;
661 fe->i_links_count = cpu_to_le16(inode->i_nlink);
662 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
663 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
665 err = ocfs2_journal_dirty(handle, fe_bh);
667 le16_add_cpu(&fe->i_links_count, -1);
673 err = ocfs2_add_entry(handle, dentry, inode,
674 OCFS2_I(inode)->ip_blkno,
675 parent_fe_bh, de_bh);
677 le16_add_cpu(&fe->i_links_count, -1);
683 err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
689 atomic_inc(&inode->i_count);
690 dentry->d_op = &ocfs2_dentry_ops;
691 d_instantiate(dentry, inode);
694 ocfs2_commit_trans(osb, handle);
696 ocfs2_inode_unlock(inode, 1);
699 ocfs2_inode_unlock(dir, 1);
703 brelse(parent_fe_bh);
711 * Takes and drops an exclusive lock on the given dentry. This will
712 * force other nodes to drop it.
714 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
718 ret = ocfs2_dentry_lock(dentry, 1);
722 ocfs2_dentry_unlock(dentry, 1);
727 static inline int inode_is_unlinkable(struct inode *inode)
729 if (S_ISDIR(inode->i_mode)) {
730 if (inode->i_nlink == 2)
735 if (inode->i_nlink == 1)
740 static int ocfs2_unlink(struct inode *dir,
741 struct dentry *dentry)
744 int child_locked = 0;
745 struct inode *inode = dentry->d_inode;
746 struct inode *orphan_dir = NULL;
747 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
749 struct ocfs2_dinode *fe = NULL;
750 struct buffer_head *fe_bh = NULL;
751 struct buffer_head *parent_node_bh = NULL;
752 handle_t *handle = NULL;
753 struct ocfs2_dir_entry *dirent = NULL;
754 struct buffer_head *dirent_bh = NULL;
755 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
756 struct buffer_head *orphan_entry_bh = NULL;
758 mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
759 dentry->d_name.len, dentry->d_name.name);
761 BUG_ON(dentry->d_parent->d_inode != dir);
763 mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
765 if (inode == osb->root_inode) {
766 mlog(0, "Cannot delete the root directory\n");
770 status = ocfs2_inode_lock(dir, &parent_node_bh, 1);
772 if (status != -ENOENT)
777 status = ocfs2_find_files_on_disk(dentry->d_name.name,
778 dentry->d_name.len, &blkno,
779 dir, &dirent_bh, &dirent);
781 if (status != -ENOENT)
786 if (OCFS2_I(inode)->ip_blkno != blkno) {
789 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
790 (unsigned long long)OCFS2_I(inode)->ip_blkno,
791 (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
795 status = ocfs2_inode_lock(inode, &fe_bh, 1);
797 if (status != -ENOENT)
803 if (S_ISDIR(inode->i_mode)) {
804 if (!ocfs2_empty_dir(inode)) {
807 } else if (inode->i_nlink != 2) {
813 status = ocfs2_remote_dentry_delete(dentry);
815 /* This remote delete should succeed under all normal
821 if (inode_is_unlinkable(inode)) {
822 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir, inode,
831 handle = ocfs2_start_trans(osb, OCFS2_UNLINK_CREDITS);
832 if (IS_ERR(handle)) {
833 status = PTR_ERR(handle);
839 status = ocfs2_journal_access(handle, inode, fe_bh,
840 OCFS2_JOURNAL_ACCESS_WRITE);
846 fe = (struct ocfs2_dinode *) fe_bh->b_data;
848 if (inode_is_unlinkable(inode)) {
849 status = ocfs2_orphan_add(osb, handle, inode, fe, orphan_name,
850 orphan_entry_bh, orphan_dir);
857 /* delete the name from the parent dir */
858 status = ocfs2_delete_entry(handle, dir, dirent, dirent_bh);
864 if (S_ISDIR(inode->i_mode))
867 fe->i_links_count = cpu_to_le16(inode->i_nlink);
869 status = ocfs2_journal_dirty(handle, fe_bh);
875 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
876 if (S_ISDIR(inode->i_mode))
879 status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
882 if (S_ISDIR(inode->i_mode))
888 ocfs2_commit_trans(osb, handle);
891 ocfs2_inode_unlock(inode, 1);
893 ocfs2_inode_unlock(dir, 1);
896 /* This was locked for us in ocfs2_prepare_orphan_dir() */
897 ocfs2_inode_unlock(orphan_dir, 1);
898 mutex_unlock(&orphan_dir->i_mutex);
904 brelse(parent_node_bh);
905 brelse(orphan_entry_bh);
913 * The only place this should be used is rename!
914 * if they have the same id, then the 1st one is the only one locked.
916 static int ocfs2_double_lock(struct ocfs2_super *osb,
917 struct buffer_head **bh1,
918 struct inode *inode1,
919 struct buffer_head **bh2,
920 struct inode *inode2)
923 struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
924 struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
925 struct buffer_head **tmpbh;
926 struct inode *tmpinode;
928 mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
929 (unsigned long long)oi1->ip_blkno,
930 (unsigned long long)oi2->ip_blkno);
937 /* we always want to lock the one with the lower lockid first. */
938 if (oi1->ip_blkno != oi2->ip_blkno) {
939 if (oi1->ip_blkno < oi2->ip_blkno) {
940 /* switch id1 and id2 around */
941 mlog(0, "switching them around...\n");
951 status = ocfs2_inode_lock(inode2, bh2, 1);
953 if (status != -ENOENT)
960 status = ocfs2_inode_lock(inode1, bh1, 1);
963 * An error return must mean that no cluster locks
964 * were held on function exit.
966 if (oi1->ip_blkno != oi2->ip_blkno)
967 ocfs2_inode_unlock(inode2, 1);
969 if (status != -ENOENT)
978 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
980 ocfs2_inode_unlock(inode1, 1);
982 if (inode1 != inode2)
983 ocfs2_inode_unlock(inode2, 1);
986 static int ocfs2_rename(struct inode *old_dir,
987 struct dentry *old_dentry,
988 struct inode *new_dir,
989 struct dentry *new_dentry)
991 int status = 0, rename_lock = 0, parents_locked = 0;
992 int old_child_locked = 0, new_child_locked = 0;
993 struct inode *old_inode = old_dentry->d_inode;
994 struct inode *new_inode = new_dentry->d_inode;
995 struct inode *orphan_dir = NULL;
996 struct ocfs2_dinode *newfe = NULL;
997 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
998 struct buffer_head *orphan_entry_bh = NULL;
999 struct buffer_head *newfe_bh = NULL;
1000 struct buffer_head *old_inode_bh = NULL;
1001 struct buffer_head *insert_entry_bh = NULL;
1002 struct ocfs2_super *osb = NULL;
1003 u64 newfe_blkno, old_de_ino;
1004 handle_t *handle = NULL;
1005 struct buffer_head *old_dir_bh = NULL;
1006 struct buffer_head *new_dir_bh = NULL;
1007 struct ocfs2_dir_entry *old_inode_dot_dot_de = NULL, *old_de = NULL,
1009 struct buffer_head *new_de_bh = NULL, *old_de_bh = NULL; // bhs for above
1010 struct buffer_head *old_inode_de_bh = NULL; // if old_dentry is a dir,
1011 // this is the 1st dirent bh
1012 nlink_t old_dir_nlink = old_dir->i_nlink;
1013 struct ocfs2_dinode *old_di;
1015 /* At some point it might be nice to break this function up a
1018 mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1019 old_dir, old_dentry, new_dir, new_dentry,
1020 old_dentry->d_name.len, old_dentry->d_name.name,
1021 new_dentry->d_name.len, new_dentry->d_name.name);
1023 osb = OCFS2_SB(old_dir->i_sb);
1026 if (!igrab(new_inode))
1030 /* Assume a directory hierarchy thusly:
1033 * a,b,c, and d are all directories.
1035 * from cwd of 'a' on both nodes:
1039 * And that's why, just like the VFS, we need a file system
1041 if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1042 status = ocfs2_rename_lock(osb);
1050 /* if old and new are the same, this'll just do one lock. */
1051 status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1052 &new_dir_bh, new_dir);
1059 /* make sure both dirs have bhs
1060 * get an extra ref on old_dir_bh if old==new */
1063 new_dir_bh = old_dir_bh;
1066 mlog(ML_ERROR, "no old_dir_bh!\n");
1073 * Aside from allowing a meta data update, the locking here
1074 * also ensures that the downconvert thread on other nodes
1075 * won't have to concurrently downconvert the inode and the
1078 status = ocfs2_inode_lock(old_inode, &old_inode_bh, 1);
1080 if (status != -ENOENT)
1084 old_child_locked = 1;
1086 status = ocfs2_remote_dentry_delete(old_dentry);
1092 if (S_ISDIR(old_inode->i_mode)) {
1093 u64 old_inode_parent;
1095 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1096 old_inode, &old_inode_de_bh,
1097 &old_inode_dot_dot_de);
1103 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1108 if (!new_inode && new_dir != old_dir &&
1109 new_dir->i_nlink >= OCFS2_LINK_MAX) {
1115 status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1116 old_dentry->d_name.len,
1124 * Check for inode number is _not_ due to possible IO errors.
1125 * We might rmdir the source, keep it as pwd of some process
1126 * and merrily kill the link to whatever was created under the
1127 * same name. Goodbye sticky bit ;-<
1129 if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1134 /* check if the target already exists (in which case we need
1136 status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1137 new_dentry->d_name.len,
1138 &newfe_blkno, new_dir, &new_de_bh,
1140 /* The only error we allow here is -ENOENT because the new
1141 * file not existing is perfectly valid. */
1142 if ((status < 0) && (status != -ENOENT)) {
1143 /* If we cannot find the file specified we should just */
1144 /* return the error... */
1149 if (!new_de && new_inode) {
1151 * Target was unlinked by another node while we were
1152 * waiting to get to ocfs2_rename(). There isn't
1153 * anything we can do here to help the situation, so
1154 * bubble up the appropriate error.
1160 /* In case we need to overwrite an existing file, we blow it
1163 /* VFS didn't think there existed an inode here, but
1164 * someone else in the cluster must have raced our
1165 * rename to create one. Today we error cleanly, in
1166 * the future we should consider calling iget to build
1167 * a new struct inode for this entry. */
1171 mlog(0, "We found an inode for name %.*s but VFS "
1172 "didn't give us one.\n", new_dentry->d_name.len,
1173 new_dentry->d_name.name);
1177 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1180 mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1181 (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1182 (unsigned long long)newfe_blkno,
1183 OCFS2_I(new_inode)->ip_flags);
1187 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1189 if (status != -ENOENT)
1193 new_child_locked = 1;
1195 status = ocfs2_remote_dentry_delete(new_dentry);
1201 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1203 mlog(0, "aha rename over existing... new_de=%p new_blkno=%llu "
1204 "newfebh=%p bhblocknr=%llu\n", new_de,
1205 (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1206 (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1208 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1209 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1219 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1221 status = ocfs2_check_dir_for_entry(new_dir,
1222 new_dentry->d_name.name,
1223 new_dentry->d_name.len);
1227 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1228 new_dentry->d_name.name,
1229 new_dentry->d_name.len,
1237 handle = ocfs2_start_trans(osb, OCFS2_RENAME_CREDITS);
1238 if (IS_ERR(handle)) {
1239 status = PTR_ERR(handle);
1246 if (S_ISDIR(new_inode->i_mode)) {
1247 if (!ocfs2_empty_dir(new_inode) ||
1248 new_inode->i_nlink != 2) {
1249 status = -ENOTEMPTY;
1253 status = ocfs2_journal_access(handle, new_inode, newfe_bh,
1254 OCFS2_JOURNAL_ACCESS_WRITE);
1260 if (S_ISDIR(new_inode->i_mode) ||
1261 (newfe->i_links_count == cpu_to_le16(1))){
1262 status = ocfs2_orphan_add(osb, handle, new_inode,
1264 orphan_entry_bh, orphan_dir);
1271 /* change the dirent to point to the correct inode */
1272 status = ocfs2_update_entry(new_dir, handle, new_de_bh,
1278 new_dir->i_version++;
1280 if (S_ISDIR(new_inode->i_mode))
1281 newfe->i_links_count = 0;
1283 le16_add_cpu(&newfe->i_links_count, -1);
1285 status = ocfs2_journal_dirty(handle, newfe_bh);
1291 /* if the name was not found in new_dir, add it now */
1292 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1293 OCFS2_I(old_inode)->ip_blkno,
1294 new_dir_bh, insert_entry_bh);
1297 old_inode->i_ctime = CURRENT_TIME;
1298 mark_inode_dirty(old_inode);
1300 status = ocfs2_journal_access(handle, old_inode, old_inode_bh,
1301 OCFS2_JOURNAL_ACCESS_WRITE);
1303 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1305 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1306 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1308 status = ocfs2_journal_dirty(handle, old_inode_bh);
1315 * Now that the name has been added to new_dir, remove the old name.
1317 * We don't keep any directory entry context around until now
1318 * because the insert might have changed the type of directory
1319 * we're dealing with.
1321 old_de_bh = ocfs2_find_entry(old_dentry->d_name.name,
1322 old_dentry->d_name.len,
1329 status = ocfs2_delete_entry(handle, old_dir, old_de, old_de_bh);
1336 new_inode->i_nlink--;
1337 new_inode->i_ctime = CURRENT_TIME;
1339 old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1340 if (old_inode_de_bh) {
1341 status = ocfs2_update_entry(old_inode, handle, old_inode_de_bh,
1342 old_inode_dot_dot_de, new_dir);
1345 new_inode->i_nlink--;
1348 mark_inode_dirty(new_dir);
1351 mark_inode_dirty(old_dir);
1352 ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1354 mark_inode_dirty(new_inode);
1355 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1358 if (old_dir != new_dir) {
1359 /* Keep the same times on both directories.*/
1360 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1363 * This will also pick up the i_nlink change from the
1366 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1369 if (old_dir_nlink != old_dir->i_nlink) {
1371 mlog(ML_ERROR, "need to change nlink for old dir "
1372 "%llu from %d to %d but bh is NULL!\n",
1373 (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1374 (int)old_dir_nlink, old_dir->i_nlink);
1376 struct ocfs2_dinode *fe;
1377 status = ocfs2_journal_access(handle, old_dir,
1379 OCFS2_JOURNAL_ACCESS_WRITE);
1380 fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1381 fe->i_links_count = cpu_to_le16(old_dir->i_nlink);
1382 status = ocfs2_journal_dirty(handle, old_dir_bh);
1386 ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1390 ocfs2_rename_unlock(osb);
1393 ocfs2_commit_trans(osb, handle);
1396 ocfs2_double_unlock(old_dir, new_dir);
1398 if (old_child_locked)
1399 ocfs2_inode_unlock(old_inode, 1);
1401 if (new_child_locked)
1402 ocfs2_inode_unlock(new_inode, 1);
1405 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1406 ocfs2_inode_unlock(orphan_dir, 1);
1407 mutex_unlock(&orphan_dir->i_mutex);
1412 sync_mapping_buffers(old_inode->i_mapping);
1417 brelse(old_inode_bh);
1422 brelse(old_inode_de_bh);
1423 brelse(orphan_entry_bh);
1424 brelse(insert_entry_bh);
1432 * we expect i_size = strlen(symname). Copy symname into the file
1433 * data, including the null terminator.
1435 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1437 struct inode *inode,
1438 const char *symname)
1440 struct buffer_head **bhs = NULL;
1442 struct super_block *sb = osb->sb;
1443 u64 p_blkno, p_blocks;
1444 int virtual, blocks, status, i, bytes_left;
1446 bytes_left = i_size_read(inode) + 1;
1447 /* we can't trust i_blocks because we're actually going to
1448 * write i_size + 1 bytes. */
1449 blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1451 mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1452 (unsigned long long)inode->i_blocks,
1453 i_size_read(inode), blocks);
1455 /* Sanity check -- make sure we're going to fit. */
1457 ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1463 bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1470 status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1477 /* links can never be larger than one cluster so we know this
1478 * is all going to be contiguous, but do a sanity check
1480 if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1487 while(bytes_left > 0) {
1488 c = &symname[virtual * sb->s_blocksize];
1490 bhs[virtual] = sb_getblk(sb, p_blkno);
1491 if (!bhs[virtual]) {
1496 ocfs2_set_new_buffer_uptodate(inode, bhs[virtual]);
1498 status = ocfs2_journal_access(handle, inode, bhs[virtual],
1499 OCFS2_JOURNAL_ACCESS_CREATE);
1505 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1507 memcpy(bhs[virtual]->b_data, c,
1508 (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1511 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1519 bytes_left -= sb->s_blocksize;
1526 for(i = 0; i < blocks; i++)
1535 static int ocfs2_symlink(struct inode *dir,
1536 struct dentry *dentry,
1537 const char *symname)
1539 int status, l, credits;
1541 struct ocfs2_super *osb = NULL;
1542 struct inode *inode = NULL;
1543 struct super_block *sb;
1544 struct buffer_head *new_fe_bh = NULL;
1545 struct buffer_head *de_bh = NULL;
1546 struct buffer_head *parent_fe_bh = NULL;
1547 struct ocfs2_dinode *fe = NULL;
1548 struct ocfs2_dinode *dirfe;
1549 handle_t *handle = NULL;
1550 struct ocfs2_alloc_context *inode_ac = NULL;
1551 struct ocfs2_alloc_context *data_ac = NULL;
1552 struct ocfs2_alloc_context *xattr_ac = NULL;
1553 int want_clusters = 0;
1554 int xattr_credits = 0;
1555 struct ocfs2_security_xattr_info si = {
1559 mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1560 dentry, symname, dentry->d_name.len, dentry->d_name.name);
1565 l = strlen(symname) + 1;
1567 credits = ocfs2_calc_symlink_credits(sb);
1569 /* lock the parent directory */
1570 status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1572 if (status != -ENOENT)
1577 dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1578 if (!dirfe->i_links_count) {
1579 /* can't make a file in a deleted directory. */
1584 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1585 dentry->d_name.len);
1589 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1590 dentry->d_name.name,
1591 dentry->d_name.len, &de_bh);
1597 status = ocfs2_reserve_new_inode(osb, &inode_ac);
1599 if (status != -ENOSPC)
1604 inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1611 /* get security xattr */
1612 status = ocfs2_init_security_get(inode, dir, &si);
1614 if (status == -EOPNOTSUPP)
1622 /* calculate meta data/clusters for setting security xattr */
1624 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1625 &xattr_credits, &xattr_ac);
1632 /* don't reserve bitmap space for fast symlinks. */
1633 if (l > ocfs2_fast_symlink_chars(sb))
1636 status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1638 if (status != -ENOSPC)
1643 handle = ocfs2_start_trans(osb, credits + xattr_credits);
1644 if (IS_ERR(handle)) {
1645 status = PTR_ERR(handle);
1651 status = ocfs2_mknod_locked(osb, dir, inode, dentry,
1652 0, &new_fe_bh, parent_fe_bh, handle,
1659 fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1662 if (l > ocfs2_fast_symlink_chars(sb)) {
1665 inode->i_op = &ocfs2_symlink_inode_operations;
1666 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1668 handle, data_ac, NULL,
1671 if (status != -ENOSPC && status != -EINTR) {
1673 "Failed to extend file to %llu\n",
1674 (unsigned long long)newsize);
1680 i_size_write(inode, newsize);
1681 inode->i_blocks = ocfs2_inode_sector_count(inode);
1683 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1684 memcpy((char *) fe->id2.i_symlink, symname, l);
1685 i_size_write(inode, newsize);
1686 inode->i_blocks = 0;
1689 status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1695 if (!ocfs2_inode_is_fast_symlink(inode)) {
1696 status = ocfs2_create_symlink_data(osb, handle, inode,
1705 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1713 status = ocfs2_add_entry(handle, dentry, inode,
1714 le64_to_cpu(fe->i_blkno), parent_fe_bh,
1721 status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1727 insert_inode_hash(inode);
1728 dentry->d_op = &ocfs2_dentry_ops;
1729 d_instantiate(dentry, inode);
1732 ocfs2_commit_trans(osb, handle);
1734 ocfs2_inode_unlock(dir, 1);
1737 brelse(parent_fe_bh);
1742 ocfs2_free_alloc_context(inode_ac);
1744 ocfs2_free_alloc_context(data_ac);
1746 ocfs2_free_alloc_context(xattr_ac);
1747 if ((status < 0) && inode) {
1757 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1759 int status, namelen;
1763 namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1773 if (namelen != OCFS2_ORPHAN_NAMELEN) {
1779 mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
1788 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1789 struct inode **ret_orphan_dir,
1790 struct inode *inode,
1792 struct buffer_head **de_bh)
1794 struct inode *orphan_dir_inode;
1795 struct buffer_head *orphan_dir_bh = NULL;
1798 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1804 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1805 ORPHAN_DIR_SYSTEM_INODE,
1807 if (!orphan_dir_inode) {
1813 mutex_lock(&orphan_dir_inode->i_mutex);
1815 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1821 status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1822 orphan_dir_bh, name,
1823 OCFS2_ORPHAN_NAMELEN, de_bh);
1825 ocfs2_inode_unlock(orphan_dir_inode, 1);
1831 *ret_orphan_dir = orphan_dir_inode;
1835 mutex_unlock(&orphan_dir_inode->i_mutex);
1836 iput(orphan_dir_inode);
1839 brelse(orphan_dir_bh);
1845 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1847 struct inode *inode,
1848 struct ocfs2_dinode *fe,
1850 struct buffer_head *de_bh,
1851 struct inode *orphan_dir_inode)
1853 struct buffer_head *orphan_dir_bh = NULL;
1855 struct ocfs2_dinode *orphan_fe;
1857 mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
1859 status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
1865 status = ocfs2_journal_access(handle, orphan_dir_inode, orphan_dir_bh,
1866 OCFS2_JOURNAL_ACCESS_WRITE);
1872 /* we're a cluster, and nlink can change on disk from
1873 * underneath us... */
1874 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1875 if (S_ISDIR(inode->i_mode))
1876 le16_add_cpu(&orphan_fe->i_links_count, 1);
1877 orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
1879 status = ocfs2_journal_dirty(handle, orphan_dir_bh);
1885 status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
1886 OCFS2_ORPHAN_NAMELEN, inode,
1887 OCFS2_I(inode)->ip_blkno,
1888 orphan_dir_bh, de_bh);
1894 le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
1896 /* Record which orphan dir our inode now resides
1897 * in. delete_inode will use this to determine which orphan
1899 fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
1901 mlog(0, "Inode %llu orphaned in slot %d\n",
1902 (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
1905 brelse(orphan_dir_bh);
1911 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
1912 int ocfs2_orphan_del(struct ocfs2_super *osb,
1914 struct inode *orphan_dir_inode,
1915 struct inode *inode,
1916 struct buffer_head *orphan_dir_bh)
1918 char name[OCFS2_ORPHAN_NAMELEN + 1];
1919 struct ocfs2_dinode *orphan_fe;
1921 struct buffer_head *target_de_bh = NULL;
1922 struct ocfs2_dir_entry *target_de = NULL;
1926 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1932 mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
1933 name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
1934 OCFS2_ORPHAN_NAMELEN);
1936 /* find it's spot in the orphan directory */
1937 target_de_bh = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN,
1938 orphan_dir_inode, &target_de);
1939 if (!target_de_bh) {
1945 /* remove it from the orphan directory */
1946 status = ocfs2_delete_entry(handle, orphan_dir_inode, target_de,
1953 status = ocfs2_journal_access(handle,orphan_dir_inode, orphan_dir_bh,
1954 OCFS2_JOURNAL_ACCESS_WRITE);
1960 /* do the i_nlink dance! :) */
1961 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1962 if (S_ISDIR(inode->i_mode))
1963 le16_add_cpu(&orphan_fe->i_links_count, -1);
1964 orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
1966 status = ocfs2_journal_dirty(handle, orphan_dir_bh);
1973 brelse(target_de_bh);
1979 const struct inode_operations ocfs2_dir_iops = {
1980 .create = ocfs2_create,
1981 .lookup = ocfs2_lookup,
1983 .unlink = ocfs2_unlink,
1984 .rmdir = ocfs2_unlink,
1985 .symlink = ocfs2_symlink,
1986 .mkdir = ocfs2_mkdir,
1987 .mknod = ocfs2_mknod,
1988 .rename = ocfs2_rename,
1989 .setattr = ocfs2_setattr,
1990 .getattr = ocfs2_getattr,
1991 .permission = ocfs2_permission,
1992 .setxattr = generic_setxattr,
1993 .getxattr = generic_getxattr,
1994 .listxattr = ocfs2_listxattr,
1995 .removexattr = generic_removexattr,