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>
43 #include <linux/quotaops.h>
45 #define MLOG_MASK_PREFIX ML_NAMEI
46 #include <cluster/masklog.h>
54 #include "extent_map.h"
67 #include "buffer_head_io.h"
69 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
72 struct dentry *dentry,
74 struct buffer_head **new_fe_bh,
75 struct buffer_head *parent_fe_bh,
77 struct ocfs2_alloc_context *inode_ac);
79 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
80 struct inode **ret_orphan_dir,
83 struct buffer_head **de_bh);
85 static int ocfs2_orphan_add(struct ocfs2_super *osb,
88 struct ocfs2_dinode *fe,
90 struct buffer_head *de_bh,
91 struct inode *orphan_dir_inode);
93 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
98 /* An orphan dir name is an 8 byte value, printed as a hex string */
99 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
101 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
102 struct nameidata *nd)
106 struct inode *inode = NULL;
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_inode_lock(dir, NULL, 0);
123 if (status != -ENOENT)
125 ret = ERR_PTR(status);
129 status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
130 dentry->d_name.len, &blkno);
134 inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
136 ret = ERR_PTR(-EACCES);
141 /* Clear any orphaned state... If we were able to look up the
142 * inode from a directory, it certainly can't be orphaned. We
143 * might have the bad state from a node which intended to
144 * orphan this inode but crashed before it could commit the
146 spin_lock(&oi->ip_lock);
147 oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
148 spin_unlock(&oi->ip_lock);
151 dentry->d_op = &ocfs2_dentry_ops;
152 ret = d_splice_alias(inode, dentry);
156 * If d_splice_alias() finds a DCACHE_DISCONNECTED
157 * dentry, it will d_move() it on top of ourse. The
158 * return value will indicate this however, so in
159 * those cases, we switch them around for the locking
162 * NOTE: This dentry already has ->d_op set from
163 * ocfs2_get_parent() and ocfs2_get_dentry()
168 status = ocfs2_dentry_attach_lock(dentry, inode,
169 OCFS2_I(dir)->ip_blkno);
172 ret = ERR_PTR(status);
178 /* Don't drop the cluster lock until *after* the d_add --
179 * unlink on another node will message us to remove that
180 * dentry under this lock so otherwise we can race this with
181 * the downconvert thread and have a stale dentry. */
182 ocfs2_inode_unlock(dir, 0);
191 static struct inode *ocfs2_get_init_inode(struct inode *dir, int mode)
195 inode = new_inode(dir->i_sb);
197 mlog(ML_ERROR, "new_inode failed!\n");
201 /* populate as many fields early on as possible - many of
202 * these are used by the support functions here and in
208 inode->i_uid = current_fsuid();
209 if (dir->i_mode & S_ISGID) {
210 inode->i_gid = dir->i_gid;
214 inode->i_gid = current_fsgid();
215 inode->i_mode = mode;
220 static int ocfs2_mknod(struct inode *dir,
221 struct dentry *dentry,
226 struct buffer_head *parent_fe_bh = NULL;
227 handle_t *handle = NULL;
228 struct ocfs2_super *osb;
229 struct ocfs2_dinode *dirfe;
230 struct buffer_head *new_fe_bh = NULL;
231 struct buffer_head *de_bh = NULL;
232 struct inode *inode = NULL;
233 struct ocfs2_alloc_context *inode_ac = NULL;
234 struct ocfs2_alloc_context *data_ac = NULL;
235 struct ocfs2_alloc_context *xattr_ac = NULL;
236 int want_clusters = 0;
237 int xattr_credits = 0;
238 struct ocfs2_security_xattr_info si = {
241 int did_quota_inode = 0;
243 mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
244 (unsigned long)dev, dentry->d_name.len,
245 dentry->d_name.name);
247 /* get our super block */
248 osb = OCFS2_SB(dir->i_sb);
250 status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
252 if (status != -ENOENT)
257 if (S_ISDIR(mode) && (dir->i_nlink >= OCFS2_LINK_MAX)) {
262 dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
263 if (!dirfe->i_links_count) {
264 /* can't make a file in a deleted directory. */
269 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
274 /* get a spot inside the dir. */
275 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
277 dentry->d_name.len, &de_bh);
283 /* reserve an inode spot */
284 status = ocfs2_reserve_new_inode(osb, &inode_ac);
286 if (status != -ENOSPC)
291 inode = ocfs2_get_init_inode(dir, mode);
298 /* get security xattr */
299 status = ocfs2_init_security_get(inode, dir, &si);
301 if (status == -EOPNOTSUPP)
309 /* calculate meta data/clusters for setting security and acl xattr */
310 status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
312 &xattr_credits, &xattr_ac);
318 /* Reserve a cluster if creating an extent based directory. */
319 if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb))
322 status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
324 if (status != -ENOSPC)
329 handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb) +
331 if (IS_ERR(handle)) {
332 status = PTR_ERR(handle);
338 /* We don't use standard VFS wrapper because we don't want vfs_dq_init
340 if (sb_any_quota_active(osb->sb) &&
341 osb->sb->dq_op->alloc_inode(inode, 1) == NO_QUOTA) {
347 /* do the real work now. */
348 status = ocfs2_mknod_locked(osb, dir, inode, dentry, dev,
349 &new_fe_bh, parent_fe_bh, handle,
357 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
364 status = ocfs2_journal_access_di(handle, dir, parent_fe_bh,
365 OCFS2_JOURNAL_ACCESS_WRITE);
370 le16_add_cpu(&dirfe->i_links_count, 1);
371 status = ocfs2_journal_dirty(handle, parent_fe_bh);
379 status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
387 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
395 status = ocfs2_add_entry(handle, dentry, inode,
396 OCFS2_I(inode)->ip_blkno, parent_fe_bh,
403 status = ocfs2_dentry_attach_lock(dentry, inode,
404 OCFS2_I(dir)->ip_blkno);
410 insert_inode_hash(inode);
411 dentry->d_op = &ocfs2_dentry_ops;
412 d_instantiate(dentry, inode);
415 if (status < 0 && did_quota_inode)
416 vfs_dq_free_inode(inode);
418 ocfs2_commit_trans(osb, handle);
420 ocfs2_inode_unlock(dir, 1);
422 if (status == -ENOSPC)
423 mlog(0, "Disk is full\n");
427 brelse(parent_fe_bh);
431 if ((status < 0) && inode) {
437 ocfs2_free_alloc_context(inode_ac);
440 ocfs2_free_alloc_context(data_ac);
443 ocfs2_free_alloc_context(xattr_ac);
450 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
453 struct dentry *dentry,
455 struct buffer_head **new_fe_bh,
456 struct buffer_head *parent_fe_bh,
458 struct ocfs2_alloc_context *inode_ac)
461 struct ocfs2_dinode *fe = NULL;
462 struct ocfs2_extent_list *fel;
466 mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry,
467 inode->i_mode, (unsigned long)dev, dentry->d_name.len,
468 dentry->d_name.name);
472 status = ocfs2_claim_new_inode(osb, handle, inode_ac, &suballoc_bit,
479 /* populate as many fields early on as possible - many of
480 * these are used by the support functions here and in
482 inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
483 OCFS2_I(inode)->ip_blkno = fe_blkno;
484 spin_lock(&osb->osb_lock);
485 inode->i_generation = osb->s_next_generation++;
486 spin_unlock(&osb->osb_lock);
488 *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
494 ocfs2_set_new_buffer_uptodate(inode, *new_fe_bh);
496 status = ocfs2_journal_access_di(handle, inode, *new_fe_bh,
497 OCFS2_JOURNAL_ACCESS_CREATE);
503 fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
504 memset(fe, 0, osb->sb->s_blocksize);
506 fe->i_generation = cpu_to_le32(inode->i_generation);
507 fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
508 fe->i_blkno = cpu_to_le64(fe_blkno);
509 fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
510 fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
511 fe->i_uid = cpu_to_le32(inode->i_uid);
512 fe->i_gid = cpu_to_le32(inode->i_gid);
513 fe->i_mode = cpu_to_le16(inode->i_mode);
514 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
515 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
516 fe->i_links_count = cpu_to_le16(inode->i_nlink);
518 fe->i_last_eb_blk = 0;
519 strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
520 le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
521 fe->i_atime = fe->i_ctime = fe->i_mtime =
522 cpu_to_le64(CURRENT_TIME.tv_sec);
523 fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
524 cpu_to_le32(CURRENT_TIME.tv_nsec);
528 * If supported, directories start with inline data.
530 if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
531 u16 feat = le16_to_cpu(fe->i_dyn_features);
533 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
535 fe->id2.i_data.id_count = cpu_to_le16(
536 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
538 fel = &fe->id2.i_list;
539 fel->l_tree_depth = 0;
540 fel->l_next_free_rec = 0;
541 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
544 status = ocfs2_journal_dirty(handle, *new_fe_bh);
550 ocfs2_populate_inode(inode, fe, 1);
551 ocfs2_inode_set_new(osb, inode);
552 if (!ocfs2_mount_local(osb)) {
553 status = ocfs2_create_new_inode_locks(inode);
558 status = 0; /* error in ocfs2_create_new_inode_locks is not
573 static int ocfs2_mkdir(struct inode *dir,
574 struct dentry *dentry,
579 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
580 dentry->d_name.len, dentry->d_name.name);
581 ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
587 static int ocfs2_create(struct inode *dir,
588 struct dentry *dentry,
590 struct nameidata *nd)
594 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
595 dentry->d_name.len, dentry->d_name.name);
596 ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
602 static int ocfs2_link(struct dentry *old_dentry,
604 struct dentry *dentry)
607 struct inode *inode = old_dentry->d_inode;
609 struct buffer_head *fe_bh = NULL;
610 struct buffer_head *parent_fe_bh = NULL;
611 struct buffer_head *de_bh = NULL;
612 struct ocfs2_dinode *fe = NULL;
613 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
615 mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
616 old_dentry->d_name.len, old_dentry->d_name.name,
617 dentry->d_name.len, dentry->d_name.name);
619 if (S_ISDIR(inode->i_mode))
622 err = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
634 err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
639 err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
641 dentry->d_name.len, &de_bh);
647 err = ocfs2_inode_lock(inode, &fe_bh, 1);
654 fe = (struct ocfs2_dinode *) fe_bh->b_data;
655 if (le16_to_cpu(fe->i_links_count) >= OCFS2_LINK_MAX) {
657 goto out_unlock_inode;
660 handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
661 if (IS_ERR(handle)) {
662 err = PTR_ERR(handle);
665 goto out_unlock_inode;
668 err = ocfs2_journal_access_di(handle, inode, fe_bh,
669 OCFS2_JOURNAL_ACCESS_WRITE);
676 inode->i_ctime = CURRENT_TIME;
677 fe->i_links_count = cpu_to_le16(inode->i_nlink);
678 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
679 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
681 err = ocfs2_journal_dirty(handle, fe_bh);
683 le16_add_cpu(&fe->i_links_count, -1);
689 err = ocfs2_add_entry(handle, dentry, inode,
690 OCFS2_I(inode)->ip_blkno,
691 parent_fe_bh, de_bh);
693 le16_add_cpu(&fe->i_links_count, -1);
699 err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
705 atomic_inc(&inode->i_count);
706 dentry->d_op = &ocfs2_dentry_ops;
707 d_instantiate(dentry, inode);
710 ocfs2_commit_trans(osb, handle);
712 ocfs2_inode_unlock(inode, 1);
715 ocfs2_inode_unlock(dir, 1);
719 brelse(parent_fe_bh);
727 * Takes and drops an exclusive lock on the given dentry. This will
728 * force other nodes to drop it.
730 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
734 ret = ocfs2_dentry_lock(dentry, 1);
738 ocfs2_dentry_unlock(dentry, 1);
743 static inline int inode_is_unlinkable(struct inode *inode)
745 if (S_ISDIR(inode->i_mode)) {
746 if (inode->i_nlink == 2)
751 if (inode->i_nlink == 1)
756 static int ocfs2_unlink(struct inode *dir,
757 struct dentry *dentry)
760 int child_locked = 0;
761 struct inode *inode = dentry->d_inode;
762 struct inode *orphan_dir = NULL;
763 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
765 struct ocfs2_dinode *fe = NULL;
766 struct buffer_head *fe_bh = NULL;
767 struct buffer_head *parent_node_bh = NULL;
768 handle_t *handle = NULL;
769 struct ocfs2_dir_entry *dirent = NULL;
770 struct buffer_head *dirent_bh = NULL;
771 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
772 struct buffer_head *orphan_entry_bh = NULL;
774 mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
775 dentry->d_name.len, dentry->d_name.name);
777 BUG_ON(dentry->d_parent->d_inode != dir);
779 mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
781 if (inode == osb->root_inode) {
782 mlog(0, "Cannot delete the root directory\n");
786 status = ocfs2_inode_lock(dir, &parent_node_bh, 1);
788 if (status != -ENOENT)
793 status = ocfs2_find_files_on_disk(dentry->d_name.name,
794 dentry->d_name.len, &blkno,
795 dir, &dirent_bh, &dirent);
797 if (status != -ENOENT)
802 if (OCFS2_I(inode)->ip_blkno != blkno) {
805 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
806 (unsigned long long)OCFS2_I(inode)->ip_blkno,
807 (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
811 status = ocfs2_inode_lock(inode, &fe_bh, 1);
813 if (status != -ENOENT)
819 if (S_ISDIR(inode->i_mode)) {
820 if (!ocfs2_empty_dir(inode)) {
823 } else if (inode->i_nlink != 2) {
829 status = ocfs2_remote_dentry_delete(dentry);
831 /* This remote delete should succeed under all normal
837 if (inode_is_unlinkable(inode)) {
838 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir, inode,
847 handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
848 if (IS_ERR(handle)) {
849 status = PTR_ERR(handle);
855 status = ocfs2_journal_access_di(handle, inode, fe_bh,
856 OCFS2_JOURNAL_ACCESS_WRITE);
862 fe = (struct ocfs2_dinode *) fe_bh->b_data;
864 if (inode_is_unlinkable(inode)) {
865 status = ocfs2_orphan_add(osb, handle, inode, fe, orphan_name,
866 orphan_entry_bh, orphan_dir);
873 /* delete the name from the parent dir */
874 status = ocfs2_delete_entry(handle, dir, dirent, dirent_bh);
880 if (S_ISDIR(inode->i_mode))
883 fe->i_links_count = cpu_to_le16(inode->i_nlink);
885 status = ocfs2_journal_dirty(handle, fe_bh);
891 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
892 if (S_ISDIR(inode->i_mode))
895 status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
898 if (S_ISDIR(inode->i_mode))
904 ocfs2_commit_trans(osb, handle);
907 ocfs2_inode_unlock(inode, 1);
909 ocfs2_inode_unlock(dir, 1);
912 /* This was locked for us in ocfs2_prepare_orphan_dir() */
913 ocfs2_inode_unlock(orphan_dir, 1);
914 mutex_unlock(&orphan_dir->i_mutex);
920 brelse(parent_node_bh);
921 brelse(orphan_entry_bh);
929 * The only place this should be used is rename!
930 * if they have the same id, then the 1st one is the only one locked.
932 static int ocfs2_double_lock(struct ocfs2_super *osb,
933 struct buffer_head **bh1,
934 struct inode *inode1,
935 struct buffer_head **bh2,
936 struct inode *inode2)
939 struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
940 struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
941 struct buffer_head **tmpbh;
942 struct inode *tmpinode;
944 mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
945 (unsigned long long)oi1->ip_blkno,
946 (unsigned long long)oi2->ip_blkno);
953 /* we always want to lock the one with the lower lockid first. */
954 if (oi1->ip_blkno != oi2->ip_blkno) {
955 if (oi1->ip_blkno < oi2->ip_blkno) {
956 /* switch id1 and id2 around */
957 mlog(0, "switching them around...\n");
967 status = ocfs2_inode_lock(inode2, bh2, 1);
969 if (status != -ENOENT)
976 status = ocfs2_inode_lock(inode1, bh1, 1);
979 * An error return must mean that no cluster locks
980 * were held on function exit.
982 if (oi1->ip_blkno != oi2->ip_blkno)
983 ocfs2_inode_unlock(inode2, 1);
985 if (status != -ENOENT)
994 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
996 ocfs2_inode_unlock(inode1, 1);
998 if (inode1 != inode2)
999 ocfs2_inode_unlock(inode2, 1);
1002 static int ocfs2_rename(struct inode *old_dir,
1003 struct dentry *old_dentry,
1004 struct inode *new_dir,
1005 struct dentry *new_dentry)
1007 int status = 0, rename_lock = 0, parents_locked = 0;
1008 int old_child_locked = 0, new_child_locked = 0;
1009 struct inode *old_inode = old_dentry->d_inode;
1010 struct inode *new_inode = new_dentry->d_inode;
1011 struct inode *orphan_dir = NULL;
1012 struct ocfs2_dinode *newfe = NULL;
1013 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1014 struct buffer_head *orphan_entry_bh = NULL;
1015 struct buffer_head *newfe_bh = NULL;
1016 struct buffer_head *old_inode_bh = NULL;
1017 struct buffer_head *insert_entry_bh = NULL;
1018 struct ocfs2_super *osb = NULL;
1019 u64 newfe_blkno, old_de_ino;
1020 handle_t *handle = NULL;
1021 struct buffer_head *old_dir_bh = NULL;
1022 struct buffer_head *new_dir_bh = NULL;
1023 struct ocfs2_dir_entry *old_inode_dot_dot_de = NULL, *old_de = NULL,
1025 struct buffer_head *new_de_bh = NULL, *old_de_bh = NULL; // bhs for above
1026 struct buffer_head *old_inode_de_bh = NULL; // if old_dentry is a dir,
1027 // this is the 1st dirent bh
1028 nlink_t old_dir_nlink = old_dir->i_nlink;
1029 struct ocfs2_dinode *old_di;
1031 /* At some point it might be nice to break this function up a
1034 mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1035 old_dir, old_dentry, new_dir, new_dentry,
1036 old_dentry->d_name.len, old_dentry->d_name.name,
1037 new_dentry->d_name.len, new_dentry->d_name.name);
1039 osb = OCFS2_SB(old_dir->i_sb);
1042 if (!igrab(new_inode))
1046 /* Assume a directory hierarchy thusly:
1049 * a,b,c, and d are all directories.
1051 * from cwd of 'a' on both nodes:
1055 * And that's why, just like the VFS, we need a file system
1057 if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1058 status = ocfs2_rename_lock(osb);
1066 /* if old and new are the same, this'll just do one lock. */
1067 status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1068 &new_dir_bh, new_dir);
1075 /* make sure both dirs have bhs
1076 * get an extra ref on old_dir_bh if old==new */
1079 new_dir_bh = old_dir_bh;
1082 mlog(ML_ERROR, "no old_dir_bh!\n");
1089 * Aside from allowing a meta data update, the locking here
1090 * also ensures that the downconvert thread on other nodes
1091 * won't have to concurrently downconvert the inode and the
1094 status = ocfs2_inode_lock(old_inode, &old_inode_bh, 1);
1096 if (status != -ENOENT)
1100 old_child_locked = 1;
1102 status = ocfs2_remote_dentry_delete(old_dentry);
1108 if (S_ISDIR(old_inode->i_mode)) {
1109 u64 old_inode_parent;
1111 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1112 old_inode, &old_inode_de_bh,
1113 &old_inode_dot_dot_de);
1119 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1124 if (!new_inode && new_dir != old_dir &&
1125 new_dir->i_nlink >= OCFS2_LINK_MAX) {
1131 status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1132 old_dentry->d_name.len,
1140 * Check for inode number is _not_ due to possible IO errors.
1141 * We might rmdir the source, keep it as pwd of some process
1142 * and merrily kill the link to whatever was created under the
1143 * same name. Goodbye sticky bit ;-<
1145 if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1150 /* check if the target already exists (in which case we need
1152 status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1153 new_dentry->d_name.len,
1154 &newfe_blkno, new_dir, &new_de_bh,
1156 /* The only error we allow here is -ENOENT because the new
1157 * file not existing is perfectly valid. */
1158 if ((status < 0) && (status != -ENOENT)) {
1159 /* If we cannot find the file specified we should just */
1160 /* return the error... */
1165 if (!new_de && new_inode) {
1167 * Target was unlinked by another node while we were
1168 * waiting to get to ocfs2_rename(). There isn't
1169 * anything we can do here to help the situation, so
1170 * bubble up the appropriate error.
1176 /* In case we need to overwrite an existing file, we blow it
1179 /* VFS didn't think there existed an inode here, but
1180 * someone else in the cluster must have raced our
1181 * rename to create one. Today we error cleanly, in
1182 * the future we should consider calling iget to build
1183 * a new struct inode for this entry. */
1187 mlog(0, "We found an inode for name %.*s but VFS "
1188 "didn't give us one.\n", new_dentry->d_name.len,
1189 new_dentry->d_name.name);
1193 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1196 mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1197 (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1198 (unsigned long long)newfe_blkno,
1199 OCFS2_I(new_inode)->ip_flags);
1203 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1205 if (status != -ENOENT)
1209 new_child_locked = 1;
1211 status = ocfs2_remote_dentry_delete(new_dentry);
1217 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1219 mlog(0, "aha rename over existing... new_de=%p new_blkno=%llu "
1220 "newfebh=%p bhblocknr=%llu\n", new_de,
1221 (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1222 (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1224 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1225 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1235 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1237 status = ocfs2_check_dir_for_entry(new_dir,
1238 new_dentry->d_name.name,
1239 new_dentry->d_name.len);
1243 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1244 new_dentry->d_name.name,
1245 new_dentry->d_name.len,
1253 handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1254 if (IS_ERR(handle)) {
1255 status = PTR_ERR(handle);
1262 if (S_ISDIR(new_inode->i_mode)) {
1263 if (!ocfs2_empty_dir(new_inode) ||
1264 new_inode->i_nlink != 2) {
1265 status = -ENOTEMPTY;
1269 status = ocfs2_journal_access_di(handle, new_inode, newfe_bh,
1270 OCFS2_JOURNAL_ACCESS_WRITE);
1276 if (S_ISDIR(new_inode->i_mode) ||
1277 (newfe->i_links_count == cpu_to_le16(1))){
1278 status = ocfs2_orphan_add(osb, handle, new_inode,
1280 orphan_entry_bh, orphan_dir);
1287 /* change the dirent to point to the correct inode */
1288 status = ocfs2_update_entry(new_dir, handle, new_de_bh,
1294 new_dir->i_version++;
1296 if (S_ISDIR(new_inode->i_mode))
1297 newfe->i_links_count = 0;
1299 le16_add_cpu(&newfe->i_links_count, -1);
1301 status = ocfs2_journal_dirty(handle, newfe_bh);
1307 /* if the name was not found in new_dir, add it now */
1308 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1309 OCFS2_I(old_inode)->ip_blkno,
1310 new_dir_bh, insert_entry_bh);
1313 old_inode->i_ctime = CURRENT_TIME;
1314 mark_inode_dirty(old_inode);
1316 status = ocfs2_journal_access_di(handle, old_inode, old_inode_bh,
1317 OCFS2_JOURNAL_ACCESS_WRITE);
1319 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1321 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1322 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1324 status = ocfs2_journal_dirty(handle, old_inode_bh);
1331 * Now that the name has been added to new_dir, remove the old name.
1333 * We don't keep any directory entry context around until now
1334 * because the insert might have changed the type of directory
1335 * we're dealing with.
1337 old_de_bh = ocfs2_find_entry(old_dentry->d_name.name,
1338 old_dentry->d_name.len,
1345 status = ocfs2_delete_entry(handle, old_dir, old_de, old_de_bh);
1352 new_inode->i_nlink--;
1353 new_inode->i_ctime = CURRENT_TIME;
1355 old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1356 if (old_inode_de_bh) {
1357 status = ocfs2_update_entry(old_inode, handle, old_inode_de_bh,
1358 old_inode_dot_dot_de, new_dir);
1361 new_inode->i_nlink--;
1364 mark_inode_dirty(new_dir);
1367 mark_inode_dirty(old_dir);
1368 ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1370 mark_inode_dirty(new_inode);
1371 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1374 if (old_dir != new_dir) {
1375 /* Keep the same times on both directories.*/
1376 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1379 * This will also pick up the i_nlink change from the
1382 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1385 if (old_dir_nlink != old_dir->i_nlink) {
1387 mlog(ML_ERROR, "need to change nlink for old dir "
1388 "%llu from %d to %d but bh is NULL!\n",
1389 (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1390 (int)old_dir_nlink, old_dir->i_nlink);
1392 struct ocfs2_dinode *fe;
1393 status = ocfs2_journal_access_di(handle, old_dir,
1395 OCFS2_JOURNAL_ACCESS_WRITE);
1396 fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1397 fe->i_links_count = cpu_to_le16(old_dir->i_nlink);
1398 status = ocfs2_journal_dirty(handle, old_dir_bh);
1402 ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1406 ocfs2_rename_unlock(osb);
1409 ocfs2_commit_trans(osb, handle);
1412 ocfs2_double_unlock(old_dir, new_dir);
1414 if (old_child_locked)
1415 ocfs2_inode_unlock(old_inode, 1);
1417 if (new_child_locked)
1418 ocfs2_inode_unlock(new_inode, 1);
1421 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1422 ocfs2_inode_unlock(orphan_dir, 1);
1423 mutex_unlock(&orphan_dir->i_mutex);
1428 sync_mapping_buffers(old_inode->i_mapping);
1433 brelse(old_inode_bh);
1438 brelse(old_inode_de_bh);
1439 brelse(orphan_entry_bh);
1440 brelse(insert_entry_bh);
1448 * we expect i_size = strlen(symname). Copy symname into the file
1449 * data, including the null terminator.
1451 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1453 struct inode *inode,
1454 const char *symname)
1456 struct buffer_head **bhs = NULL;
1458 struct super_block *sb = osb->sb;
1459 u64 p_blkno, p_blocks;
1460 int virtual, blocks, status, i, bytes_left;
1462 bytes_left = i_size_read(inode) + 1;
1463 /* we can't trust i_blocks because we're actually going to
1464 * write i_size + 1 bytes. */
1465 blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1467 mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1468 (unsigned long long)inode->i_blocks,
1469 i_size_read(inode), blocks);
1471 /* Sanity check -- make sure we're going to fit. */
1473 ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1479 bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1486 status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1493 /* links can never be larger than one cluster so we know this
1494 * is all going to be contiguous, but do a sanity check
1496 if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1503 while(bytes_left > 0) {
1504 c = &symname[virtual * sb->s_blocksize];
1506 bhs[virtual] = sb_getblk(sb, p_blkno);
1507 if (!bhs[virtual]) {
1512 ocfs2_set_new_buffer_uptodate(inode, bhs[virtual]);
1514 status = ocfs2_journal_access(handle, inode, bhs[virtual],
1515 OCFS2_JOURNAL_ACCESS_CREATE);
1521 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1523 memcpy(bhs[virtual]->b_data, c,
1524 (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1527 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1535 bytes_left -= sb->s_blocksize;
1542 for(i = 0; i < blocks; i++)
1551 static int ocfs2_symlink(struct inode *dir,
1552 struct dentry *dentry,
1553 const char *symname)
1555 int status, l, credits;
1557 struct ocfs2_super *osb = NULL;
1558 struct inode *inode = NULL;
1559 struct super_block *sb;
1560 struct buffer_head *new_fe_bh = NULL;
1561 struct buffer_head *de_bh = NULL;
1562 struct buffer_head *parent_fe_bh = NULL;
1563 struct ocfs2_dinode *fe = NULL;
1564 struct ocfs2_dinode *dirfe;
1565 handle_t *handle = NULL;
1566 struct ocfs2_alloc_context *inode_ac = NULL;
1567 struct ocfs2_alloc_context *data_ac = NULL;
1568 struct ocfs2_alloc_context *xattr_ac = NULL;
1569 int want_clusters = 0;
1570 int xattr_credits = 0;
1571 struct ocfs2_security_xattr_info si = {
1574 int did_quota = 0, did_quota_inode = 0;
1576 mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1577 dentry, symname, dentry->d_name.len, dentry->d_name.name);
1582 l = strlen(symname) + 1;
1584 credits = ocfs2_calc_symlink_credits(sb);
1586 /* lock the parent directory */
1587 status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1589 if (status != -ENOENT)
1594 dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1595 if (!dirfe->i_links_count) {
1596 /* can't make a file in a deleted directory. */
1601 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1602 dentry->d_name.len);
1606 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1607 dentry->d_name.name,
1608 dentry->d_name.len, &de_bh);
1614 status = ocfs2_reserve_new_inode(osb, &inode_ac);
1616 if (status != -ENOSPC)
1621 inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1628 /* get security xattr */
1629 status = ocfs2_init_security_get(inode, dir, &si);
1631 if (status == -EOPNOTSUPP)
1639 /* calculate meta data/clusters for setting security xattr */
1641 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1642 &xattr_credits, &xattr_ac);
1649 /* don't reserve bitmap space for fast symlinks. */
1650 if (l > ocfs2_fast_symlink_chars(sb))
1653 status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1655 if (status != -ENOSPC)
1660 handle = ocfs2_start_trans(osb, credits + xattr_credits);
1661 if (IS_ERR(handle)) {
1662 status = PTR_ERR(handle);
1668 /* We don't use standard VFS wrapper because we don't want vfs_dq_init
1670 if (sb_any_quota_active(osb->sb) &&
1671 osb->sb->dq_op->alloc_inode(inode, 1) == NO_QUOTA) {
1675 did_quota_inode = 1;
1677 status = ocfs2_mknod_locked(osb, dir, inode, dentry,
1678 0, &new_fe_bh, parent_fe_bh, handle,
1685 fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1688 if (l > ocfs2_fast_symlink_chars(sb)) {
1691 inode->i_op = &ocfs2_symlink_inode_operations;
1692 if (vfs_dq_alloc_space_nodirty(inode,
1693 ocfs2_clusters_to_bytes(osb->sb, 1))) {
1698 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1700 handle, data_ac, NULL,
1703 if (status != -ENOSPC && status != -EINTR) {
1705 "Failed to extend file to %llu\n",
1706 (unsigned long long)newsize);
1712 i_size_write(inode, newsize);
1713 inode->i_blocks = ocfs2_inode_sector_count(inode);
1715 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1716 memcpy((char *) fe->id2.i_symlink, symname, l);
1717 i_size_write(inode, newsize);
1718 inode->i_blocks = 0;
1721 status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1727 if (!ocfs2_inode_is_fast_symlink(inode)) {
1728 status = ocfs2_create_symlink_data(osb, handle, inode,
1737 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1745 status = ocfs2_add_entry(handle, dentry, inode,
1746 le64_to_cpu(fe->i_blkno), parent_fe_bh,
1753 status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1759 insert_inode_hash(inode);
1760 dentry->d_op = &ocfs2_dentry_ops;
1761 d_instantiate(dentry, inode);
1763 if (status < 0 && did_quota)
1764 vfs_dq_free_space_nodirty(inode,
1765 ocfs2_clusters_to_bytes(osb->sb, 1));
1766 if (status < 0 && did_quota_inode)
1767 vfs_dq_free_inode(inode);
1769 ocfs2_commit_trans(osb, handle);
1771 ocfs2_inode_unlock(dir, 1);
1774 brelse(parent_fe_bh);
1779 ocfs2_free_alloc_context(inode_ac);
1781 ocfs2_free_alloc_context(data_ac);
1783 ocfs2_free_alloc_context(xattr_ac);
1784 if ((status < 0) && inode) {
1794 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1796 int status, namelen;
1800 namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1810 if (namelen != OCFS2_ORPHAN_NAMELEN) {
1816 mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
1825 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1826 struct inode **ret_orphan_dir,
1827 struct inode *inode,
1829 struct buffer_head **de_bh)
1831 struct inode *orphan_dir_inode;
1832 struct buffer_head *orphan_dir_bh = NULL;
1835 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1841 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1842 ORPHAN_DIR_SYSTEM_INODE,
1844 if (!orphan_dir_inode) {
1850 mutex_lock(&orphan_dir_inode->i_mutex);
1852 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1858 status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1859 orphan_dir_bh, name,
1860 OCFS2_ORPHAN_NAMELEN, de_bh);
1862 ocfs2_inode_unlock(orphan_dir_inode, 1);
1868 *ret_orphan_dir = orphan_dir_inode;
1872 mutex_unlock(&orphan_dir_inode->i_mutex);
1873 iput(orphan_dir_inode);
1876 brelse(orphan_dir_bh);
1882 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1884 struct inode *inode,
1885 struct ocfs2_dinode *fe,
1887 struct buffer_head *de_bh,
1888 struct inode *orphan_dir_inode)
1890 struct buffer_head *orphan_dir_bh = NULL;
1892 struct ocfs2_dinode *orphan_fe;
1894 mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
1896 status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
1902 status = ocfs2_journal_access_di(handle, orphan_dir_inode, orphan_dir_bh,
1903 OCFS2_JOURNAL_ACCESS_WRITE);
1909 /* we're a cluster, and nlink can change on disk from
1910 * underneath us... */
1911 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1912 if (S_ISDIR(inode->i_mode))
1913 le16_add_cpu(&orphan_fe->i_links_count, 1);
1914 orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
1916 status = ocfs2_journal_dirty(handle, orphan_dir_bh);
1922 status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
1923 OCFS2_ORPHAN_NAMELEN, inode,
1924 OCFS2_I(inode)->ip_blkno,
1925 orphan_dir_bh, de_bh);
1931 le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
1933 /* Record which orphan dir our inode now resides
1934 * in. delete_inode will use this to determine which orphan
1936 fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
1938 mlog(0, "Inode %llu orphaned in slot %d\n",
1939 (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
1942 brelse(orphan_dir_bh);
1948 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
1949 int ocfs2_orphan_del(struct ocfs2_super *osb,
1951 struct inode *orphan_dir_inode,
1952 struct inode *inode,
1953 struct buffer_head *orphan_dir_bh)
1955 char name[OCFS2_ORPHAN_NAMELEN + 1];
1956 struct ocfs2_dinode *orphan_fe;
1958 struct buffer_head *target_de_bh = NULL;
1959 struct ocfs2_dir_entry *target_de = NULL;
1963 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1969 mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
1970 name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
1971 OCFS2_ORPHAN_NAMELEN);
1973 /* find it's spot in the orphan directory */
1974 target_de_bh = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN,
1975 orphan_dir_inode, &target_de);
1976 if (!target_de_bh) {
1982 /* remove it from the orphan directory */
1983 status = ocfs2_delete_entry(handle, orphan_dir_inode, target_de,
1990 status = ocfs2_journal_access_di(handle,orphan_dir_inode, orphan_dir_bh,
1991 OCFS2_JOURNAL_ACCESS_WRITE);
1997 /* do the i_nlink dance! :) */
1998 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1999 if (S_ISDIR(inode->i_mode))
2000 le16_add_cpu(&orphan_fe->i_links_count, -1);
2001 orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
2003 status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2010 brelse(target_de_bh);
2016 const struct inode_operations ocfs2_dir_iops = {
2017 .create = ocfs2_create,
2018 .lookup = ocfs2_lookup,
2020 .unlink = ocfs2_unlink,
2021 .rmdir = ocfs2_unlink,
2022 .symlink = ocfs2_symlink,
2023 .mkdir = ocfs2_mkdir,
2024 .mknod = ocfs2_mknod,
2025 .rename = ocfs2_rename,
2026 .setattr = ocfs2_setattr,
2027 .getattr = ocfs2_getattr,
2028 .permission = ocfs2_permission,
2029 .setxattr = generic_setxattr,
2030 .getxattr = generic_getxattr,
2031 .listxattr = ocfs2_listxattr,
2032 .removexattr = generic_removexattr,