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 #include <cluster/masklog.h>
53 #include "extent_map.h"
65 #include "ocfs2_trace.h"
67 #include "buffer_head_io.h"
69 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
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 ocfs2_dir_lookup_result *lookup);
84 static int ocfs2_orphan_add(struct ocfs2_super *osb,
87 struct buffer_head *fe_bh,
89 struct ocfs2_dir_lookup_result *lookup,
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,
105 struct inode *inode = NULL;
107 struct ocfs2_inode_info *oi;
109 trace_ocfs2_lookup(dir, dentry, dentry->d_name.len,
111 (unsigned long long)OCFS2_I(dir)->ip_blkno, 0);
113 if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
114 ret = ERR_PTR(-ENAMETOOLONG);
118 status = ocfs2_inode_lock_nested(dir, NULL, 0, OI_LS_PARENT);
120 if (status != -ENOENT)
122 ret = ERR_PTR(status);
126 status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
127 dentry->d_name.len, &blkno);
131 inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
133 ret = ERR_PTR(-EACCES);
138 /* Clear any orphaned state... If we were able to look up the
139 * inode from a directory, it certainly can't be orphaned. We
140 * might have the bad state from a node which intended to
141 * orphan this inode but crashed before it could commit the
143 spin_lock(&oi->ip_lock);
144 oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
145 spin_unlock(&oi->ip_lock);
148 ret = d_splice_alias(inode, dentry);
152 * If d_splice_alias() finds a DCACHE_DISCONNECTED
153 * dentry, it will d_move() it on top of ourse. The
154 * return value will indicate this however, so in
155 * those cases, we switch them around for the locking
158 * NOTE: This dentry already has ->d_op set from
159 * ocfs2_get_parent() and ocfs2_get_dentry()
164 status = ocfs2_dentry_attach_lock(dentry, inode,
165 OCFS2_I(dir)->ip_blkno);
168 ret = ERR_PTR(status);
172 ocfs2_dentry_attach_gen(dentry);
175 /* Don't drop the cluster lock until *after* the d_add --
176 * unlink on another node will message us to remove that
177 * dentry under this lock so otherwise we can race this with
178 * the downconvert thread and have a stale dentry. */
179 ocfs2_inode_unlock(dir, 0);
183 trace_ocfs2_lookup_ret(ret);
188 static struct inode *ocfs2_get_init_inode(struct inode *dir, umode_t mode)
192 inode = new_inode(dir->i_sb);
194 mlog(ML_ERROR, "new_inode failed!\n");
198 /* populate as many fields early on as possible - many of
199 * these are used by the support functions here and in
203 inode_init_owner(inode, dir, mode);
204 dquot_initialize(inode);
208 static int ocfs2_mknod(struct inode *dir,
209 struct dentry *dentry,
214 struct buffer_head *parent_fe_bh = NULL;
215 handle_t *handle = NULL;
216 struct ocfs2_super *osb;
217 struct ocfs2_dinode *dirfe;
218 struct buffer_head *new_fe_bh = NULL;
219 struct inode *inode = NULL;
220 struct ocfs2_alloc_context *inode_ac = NULL;
221 struct ocfs2_alloc_context *data_ac = NULL;
222 struct ocfs2_alloc_context *meta_ac = NULL;
223 int want_clusters = 0;
225 int xattr_credits = 0;
226 struct ocfs2_security_xattr_info si = {
229 int did_quota_inode = 0;
230 struct ocfs2_dir_lookup_result lookup = { NULL, };
232 int did_block_signals = 0;
233 struct posix_acl *default_acl = NULL, *acl = NULL;
235 trace_ocfs2_mknod(dir, dentry, dentry->d_name.len, dentry->d_name.name,
236 (unsigned long long)OCFS2_I(dir)->ip_blkno,
237 (unsigned long)dev, mode);
239 dquot_initialize(dir);
241 /* get our super block */
242 osb = OCFS2_SB(dir->i_sb);
244 status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
246 if (status != -ENOENT)
251 if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
256 dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
257 if (!ocfs2_read_links_count(dirfe)) {
258 /* can't make a file in a deleted directory. */
263 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
268 /* get a spot inside the dir. */
269 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
271 dentry->d_name.len, &lookup);
277 /* reserve an inode spot */
278 status = ocfs2_reserve_new_inode(osb, &inode_ac);
280 if (status != -ENOSPC)
285 inode = ocfs2_get_init_inode(dir, mode);
292 /* get security xattr */
293 status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
295 if (status == -EOPNOTSUPP)
303 /* calculate meta data/clusters for setting security and acl xattr */
304 status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
306 &xattr_credits, &want_meta);
312 /* Reserve a cluster if creating an extent based directory. */
313 if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
316 /* Dir indexing requires extra space as well */
317 if (ocfs2_supports_indexed_dirs(osb))
321 status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
323 if (status != -ENOSPC)
328 status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
330 if (status != -ENOSPC)
335 status = posix_acl_create(dir, &mode, &default_acl, &acl);
341 handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
344 if (IS_ERR(handle)) {
345 status = PTR_ERR(handle);
351 /* Starting to change things, restart is no longer possible. */
352 ocfs2_block_signals(&oldset);
353 did_block_signals = 1;
355 status = dquot_alloc_inode(inode);
360 /* do the real work now. */
361 status = ocfs2_mknod_locked(osb, dir, inode, dev,
362 &new_fe_bh, parent_fe_bh, handle,
370 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
371 new_fe_bh, data_ac, meta_ac);
377 status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
379 OCFS2_JOURNAL_ACCESS_WRITE);
384 ocfs2_add_links_count(dirfe, 1);
385 ocfs2_journal_dirty(handle, parent_fe_bh);
390 status = ocfs2_set_acl(handle, inode, new_fe_bh,
391 ACL_TYPE_DEFAULT, default_acl,
394 if (!status && acl) {
395 status = ocfs2_set_acl(handle, inode, new_fe_bh,
396 ACL_TYPE_ACCESS, acl,
406 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
415 * Do this before adding the entry to the directory. We add
416 * also set d_op after success so that ->d_iput() will cleanup
417 * the dentry lock even if ocfs2_add_entry() fails below.
419 status = ocfs2_dentry_attach_lock(dentry, inode,
420 OCFS2_I(dir)->ip_blkno);
426 status = ocfs2_add_entry(handle, dentry, inode,
427 OCFS2_I(inode)->ip_blkno, parent_fe_bh,
434 insert_inode_hash(inode);
435 d_instantiate(dentry, inode);
439 posix_acl_release(default_acl);
441 posix_acl_release(acl);
442 if (status < 0 && did_quota_inode)
443 dquot_free_inode(inode);
445 ocfs2_commit_trans(osb, handle);
447 ocfs2_inode_unlock(dir, 1);
448 if (did_block_signals)
449 ocfs2_unblock_signals(&oldset);
452 brelse(parent_fe_bh);
455 ocfs2_free_dir_lookup_result(&lookup);
458 ocfs2_free_alloc_context(inode_ac);
461 ocfs2_free_alloc_context(data_ac);
464 ocfs2_free_alloc_context(meta_ac);
467 * We should call iput after the i_mutex of the bitmap been
468 * unlocked in ocfs2_free_alloc_context, or the
469 * ocfs2_delete_inode will mutex_lock again.
471 if ((status < 0) && inode) {
472 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
483 static int __ocfs2_mknod_locked(struct inode *dir,
486 struct buffer_head **new_fe_bh,
487 struct buffer_head *parent_fe_bh,
489 struct ocfs2_alloc_context *inode_ac,
490 u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
493 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
494 struct ocfs2_dinode *fe = NULL;
495 struct ocfs2_extent_list *fel;
497 struct ocfs2_inode_info *oi = OCFS2_I(inode);
501 /* populate as many fields early on as possible - many of
502 * these are used by the support functions here and in
504 inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
505 OCFS2_I(inode)->ip_blkno = fe_blkno;
506 spin_lock(&osb->osb_lock);
507 inode->i_generation = osb->s_next_generation++;
508 spin_unlock(&osb->osb_lock);
510 *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
516 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
518 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
520 OCFS2_JOURNAL_ACCESS_CREATE);
526 fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
527 memset(fe, 0, osb->sb->s_blocksize);
529 fe->i_generation = cpu_to_le32(inode->i_generation);
530 fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
531 fe->i_blkno = cpu_to_le64(fe_blkno);
532 fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
533 fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
534 fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
535 fe->i_uid = cpu_to_le32(i_uid_read(inode));
536 fe->i_gid = cpu_to_le32(i_gid_read(inode));
537 fe->i_mode = cpu_to_le16(inode->i_mode);
538 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
539 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
541 ocfs2_set_links_count(fe, inode->i_nlink);
543 fe->i_last_eb_blk = 0;
544 strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
545 fe->i_flags |= cpu_to_le32(OCFS2_VALID_FL);
546 fe->i_atime = fe->i_ctime = fe->i_mtime =
547 cpu_to_le64(CURRENT_TIME.tv_sec);
548 fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
549 cpu_to_le32(CURRENT_TIME.tv_nsec);
553 * If supported, directories start with inline data. If inline
554 * isn't supported, but indexing is, we start them as indexed.
556 feat = le16_to_cpu(fe->i_dyn_features);
557 if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
558 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
560 fe->id2.i_data.id_count = cpu_to_le16(
561 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
563 fel = &fe->id2.i_list;
564 fel->l_tree_depth = 0;
565 fel->l_next_free_rec = 0;
566 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
569 ocfs2_journal_dirty(handle, *new_fe_bh);
571 ocfs2_populate_inode(inode, fe, 1);
572 ocfs2_ci_set_new(osb, INODE_CACHE(inode));
573 if (!ocfs2_mount_local(osb)) {
574 status = ocfs2_create_new_inode_locks(inode);
579 oi->i_sync_tid = handle->h_transaction->t_tid;
580 oi->i_datasync_tid = handle->h_transaction->t_tid;
595 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
599 struct buffer_head **new_fe_bh,
600 struct buffer_head *parent_fe_bh,
602 struct ocfs2_alloc_context *inode_ac)
605 u64 suballoc_loc, fe_blkno = 0;
610 status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
611 inode_ac, &suballoc_loc,
612 &suballoc_bit, &fe_blkno);
618 return __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
619 parent_fe_bh, handle, inode_ac,
620 fe_blkno, suballoc_loc, suballoc_bit);
623 static int ocfs2_mkdir(struct inode *dir,
624 struct dentry *dentry,
629 trace_ocfs2_mkdir(dir, dentry, dentry->d_name.len, dentry->d_name.name,
630 OCFS2_I(dir)->ip_blkno, mode);
631 ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
638 static int ocfs2_create(struct inode *dir,
639 struct dentry *dentry,
645 trace_ocfs2_create(dir, dentry, dentry->d_name.len, dentry->d_name.name,
646 (unsigned long long)OCFS2_I(dir)->ip_blkno, mode);
647 ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
654 static int ocfs2_link(struct dentry *old_dentry,
656 struct dentry *dentry)
659 struct inode *inode = old_dentry->d_inode;
661 struct buffer_head *fe_bh = NULL;
662 struct buffer_head *parent_fe_bh = NULL;
663 struct ocfs2_dinode *fe = NULL;
664 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
665 struct ocfs2_dir_lookup_result lookup = { NULL, };
669 trace_ocfs2_link((unsigned long long)OCFS2_I(inode)->ip_blkno,
670 old_dentry->d_name.len, old_dentry->d_name.name,
671 dentry->d_name.len, dentry->d_name.name);
673 if (S_ISDIR(inode->i_mode))
676 dquot_initialize(dir);
678 err = ocfs2_inode_lock_nested(dir, &parent_fe_bh, 1, OI_LS_PARENT);
690 err = ocfs2_lookup_ino_from_name(dir, old_dentry->d_name.name,
691 old_dentry->d_name.len, &old_de_ino);
698 * Check whether another node removed the source inode while we
701 if (old_de_ino != OCFS2_I(inode)->ip_blkno) {
706 err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
711 err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
713 dentry->d_name.len, &lookup);
719 err = ocfs2_inode_lock(inode, &fe_bh, 1);
726 fe = (struct ocfs2_dinode *) fe_bh->b_data;
727 if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
729 goto out_unlock_inode;
732 handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
733 if (IS_ERR(handle)) {
734 err = PTR_ERR(handle);
737 goto out_unlock_inode;
740 /* Starting to change things, restart is no longer possible. */
741 ocfs2_block_signals(&oldset);
743 err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
744 OCFS2_JOURNAL_ACCESS_WRITE);
751 inode->i_ctime = CURRENT_TIME;
752 ocfs2_set_links_count(fe, inode->i_nlink);
753 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
754 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
755 ocfs2_journal_dirty(handle, fe_bh);
757 err = ocfs2_add_entry(handle, dentry, inode,
758 OCFS2_I(inode)->ip_blkno,
759 parent_fe_bh, &lookup);
761 ocfs2_add_links_count(fe, -1);
767 err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
774 d_instantiate(dentry, inode);
777 ocfs2_commit_trans(osb, handle);
778 ocfs2_unblock_signals(&oldset);
780 ocfs2_inode_unlock(inode, 1);
783 ocfs2_inode_unlock(dir, 1);
786 brelse(parent_fe_bh);
788 ocfs2_free_dir_lookup_result(&lookup);
797 * Takes and drops an exclusive lock on the given dentry. This will
798 * force other nodes to drop it.
800 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
804 ret = ocfs2_dentry_lock(dentry, 1);
808 ocfs2_dentry_unlock(dentry, 1);
813 static inline int ocfs2_inode_is_unlinkable(struct inode *inode)
815 if (S_ISDIR(inode->i_mode)) {
816 if (inode->i_nlink == 2)
821 if (inode->i_nlink == 1)
826 static int ocfs2_unlink(struct inode *dir,
827 struct dentry *dentry)
830 int child_locked = 0;
831 bool is_unlinkable = false;
832 struct inode *inode = dentry->d_inode;
833 struct inode *orphan_dir = NULL;
834 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
836 struct ocfs2_dinode *fe = NULL;
837 struct buffer_head *fe_bh = NULL;
838 struct buffer_head *parent_node_bh = NULL;
839 handle_t *handle = NULL;
840 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
841 struct ocfs2_dir_lookup_result lookup = { NULL, };
842 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
844 trace_ocfs2_unlink(dir, dentry, dentry->d_name.len,
846 (unsigned long long)OCFS2_I(dir)->ip_blkno,
847 (unsigned long long)OCFS2_I(inode)->ip_blkno);
849 dquot_initialize(dir);
851 BUG_ON(dentry->d_parent->d_inode != dir);
853 if (inode == osb->root_inode)
856 status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
859 if (status != -ENOENT)
864 status = ocfs2_find_files_on_disk(dentry->d_name.name,
865 dentry->d_name.len, &blkno, dir,
868 if (status != -ENOENT)
873 if (OCFS2_I(inode)->ip_blkno != blkno) {
876 trace_ocfs2_unlink_noent(
877 (unsigned long long)OCFS2_I(inode)->ip_blkno,
878 (unsigned long long)blkno,
879 OCFS2_I(inode)->ip_flags);
883 status = ocfs2_inode_lock(inode, &fe_bh, 1);
885 if (status != -ENOENT)
891 if (S_ISDIR(inode->i_mode)) {
892 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
898 status = ocfs2_remote_dentry_delete(dentry);
900 /* This remote delete should succeed under all normal
906 if (ocfs2_inode_is_unlinkable(inode)) {
907 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
908 OCFS2_I(inode)->ip_blkno,
909 orphan_name, &orphan_insert);
914 is_unlinkable = true;
917 handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
918 if (IS_ERR(handle)) {
919 status = PTR_ERR(handle);
925 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
926 OCFS2_JOURNAL_ACCESS_WRITE);
932 fe = (struct ocfs2_dinode *) fe_bh->b_data;
934 /* delete the name from the parent dir */
935 status = ocfs2_delete_entry(handle, dir, &lookup);
941 if (S_ISDIR(inode->i_mode))
944 ocfs2_set_links_count(fe, inode->i_nlink);
945 ocfs2_journal_dirty(handle, fe_bh);
947 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
948 if (S_ISDIR(inode->i_mode))
951 status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
954 if (S_ISDIR(inode->i_mode))
960 status = ocfs2_orphan_add(osb, handle, inode, fe_bh,
961 orphan_name, &orphan_insert, orphan_dir);
968 ocfs2_commit_trans(osb, handle);
971 ocfs2_inode_unlock(inode, 1);
973 ocfs2_inode_unlock(dir, 1);
976 /* This was locked for us in ocfs2_prepare_orphan_dir() */
977 ocfs2_inode_unlock(orphan_dir, 1);
978 mutex_unlock(&orphan_dir->i_mutex);
983 brelse(parent_node_bh);
985 ocfs2_free_dir_lookup_result(&orphan_insert);
986 ocfs2_free_dir_lookup_result(&lookup);
988 if (status && (status != -ENOTEMPTY) && (status != -ENOENT))
994 static int ocfs2_check_if_ancestor(struct ocfs2_super *osb,
995 u64 src_inode_no, u64 dest_inode_no)
998 u64 parent_inode_no = 0;
999 u64 child_inode_no = src_inode_no;
1000 struct inode *child_inode;
1002 #define MAX_LOOKUP_TIMES 32
1004 child_inode = ocfs2_iget(osb, child_inode_no, 0, 0);
1005 if (IS_ERR(child_inode)) {
1006 ret = PTR_ERR(child_inode);
1010 ret = ocfs2_inode_lock(child_inode, NULL, 0);
1018 ret = ocfs2_lookup_ino_from_name(child_inode, "..", 2,
1020 ocfs2_inode_unlock(child_inode, 0);
1027 if (parent_inode_no == dest_inode_no) {
1032 if (parent_inode_no == osb->root_inode->i_ino) {
1037 child_inode_no = parent_inode_no;
1039 if (++i >= MAX_LOOKUP_TIMES) {
1040 mlog(ML_NOTICE, "max lookup times reached, filesystem "
1041 "may have nested directories, "
1042 "src inode: %llu, dest inode: %llu.\n",
1043 (unsigned long long)src_inode_no,
1044 (unsigned long long)dest_inode_no);
1054 * The only place this should be used is rename!
1055 * if they have the same id, then the 1st one is the only one locked.
1057 static int ocfs2_double_lock(struct ocfs2_super *osb,
1058 struct buffer_head **bh1,
1059 struct inode *inode1,
1060 struct buffer_head **bh2,
1061 struct inode *inode2)
1064 int inode1_is_ancestor, inode2_is_ancestor;
1065 struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
1066 struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
1067 struct buffer_head **tmpbh;
1068 struct inode *tmpinode;
1070 trace_ocfs2_double_lock((unsigned long long)oi1->ip_blkno,
1071 (unsigned long long)oi2->ip_blkno);
1078 /* we always want to lock the one with the lower lockid first.
1079 * and if they are nested, we lock ancestor first */
1080 if (oi1->ip_blkno != oi2->ip_blkno) {
1081 inode1_is_ancestor = ocfs2_check_if_ancestor(osb, oi2->ip_blkno,
1083 if (inode1_is_ancestor < 0) {
1084 status = inode1_is_ancestor;
1088 inode2_is_ancestor = ocfs2_check_if_ancestor(osb, oi1->ip_blkno,
1090 if (inode2_is_ancestor < 0) {
1091 status = inode2_is_ancestor;
1095 if ((inode1_is_ancestor == 1) ||
1096 (oi1->ip_blkno < oi2->ip_blkno &&
1097 inode2_is_ancestor == 0)) {
1098 /* switch id1 and id2 around */
1108 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
1111 if (status != -ENOENT)
1118 status = ocfs2_inode_lock_nested(inode1, bh1, 1, OI_LS_RENAME2);
1121 * An error return must mean that no cluster locks
1122 * were held on function exit.
1124 if (oi1->ip_blkno != oi2->ip_blkno) {
1125 ocfs2_inode_unlock(inode2, 1);
1130 if (status != -ENOENT)
1134 trace_ocfs2_double_lock_end(
1135 (unsigned long long)OCFS2_I(inode1)->ip_blkno,
1136 (unsigned long long)OCFS2_I(inode2)->ip_blkno);
1144 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1146 ocfs2_inode_unlock(inode1, 1);
1148 if (inode1 != inode2)
1149 ocfs2_inode_unlock(inode2, 1);
1152 static int ocfs2_rename(struct inode *old_dir,
1153 struct dentry *old_dentry,
1154 struct inode *new_dir,
1155 struct dentry *new_dentry)
1157 int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1158 int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1159 struct inode *old_inode = old_dentry->d_inode;
1160 struct inode *new_inode = new_dentry->d_inode;
1161 struct inode *orphan_dir = NULL;
1162 struct ocfs2_dinode *newfe = NULL;
1163 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1164 struct buffer_head *newfe_bh = NULL;
1165 struct buffer_head *old_inode_bh = NULL;
1166 struct ocfs2_super *osb = NULL;
1167 u64 newfe_blkno, old_de_ino;
1168 handle_t *handle = NULL;
1169 struct buffer_head *old_dir_bh = NULL;
1170 struct buffer_head *new_dir_bh = NULL;
1171 u32 old_dir_nlink = old_dir->i_nlink;
1172 struct ocfs2_dinode *old_di;
1173 struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1174 struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1175 struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1176 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1177 struct ocfs2_dir_lookup_result target_insert = { NULL, };
1178 bool should_add_orphan = false;
1180 /* At some point it might be nice to break this function up a
1183 trace_ocfs2_rename(old_dir, old_dentry, new_dir, new_dentry,
1184 old_dentry->d_name.len, old_dentry->d_name.name,
1185 new_dentry->d_name.len, new_dentry->d_name.name);
1187 dquot_initialize(old_dir);
1188 dquot_initialize(new_dir);
1190 osb = OCFS2_SB(old_dir->i_sb);
1193 if (!igrab(new_inode))
1197 /* Assume a directory hierarchy thusly:
1200 * a,b,c, and d are all directories.
1202 * from cwd of 'a' on both nodes:
1206 * And that's why, just like the VFS, we need a file system
1208 if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1209 status = ocfs2_rename_lock(osb);
1216 /* here we cannot guarantee the inodes haven't just been
1217 * changed, so check if they are nested again */
1218 status = ocfs2_check_if_ancestor(osb, new_dir->i_ino,
1223 } else if (status == 1) {
1225 trace_ocfs2_rename_not_permitted(
1226 (unsigned long long)old_inode->i_ino,
1227 (unsigned long long)new_dir->i_ino);
1232 /* if old and new are the same, this'll just do one lock. */
1233 status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1234 &new_dir_bh, new_dir);
1241 /* make sure both dirs have bhs
1242 * get an extra ref on old_dir_bh if old==new */
1245 new_dir_bh = old_dir_bh;
1248 mlog(ML_ERROR, "no old_dir_bh!\n");
1255 * Aside from allowing a meta data update, the locking here
1256 * also ensures that the downconvert thread on other nodes
1257 * won't have to concurrently downconvert the inode and the
1260 status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1263 if (status != -ENOENT)
1267 old_child_locked = 1;
1269 status = ocfs2_remote_dentry_delete(old_dentry);
1275 if (S_ISDIR(old_inode->i_mode)) {
1276 u64 old_inode_parent;
1279 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1281 &old_inode_dot_dot_res);
1287 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1292 if (!new_inode && new_dir != old_dir &&
1293 new_dir->i_nlink >= ocfs2_link_max(osb)) {
1299 status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1300 old_dentry->d_name.len,
1308 * Check for inode number is _not_ due to possible IO errors.
1309 * We might rmdir the source, keep it as pwd of some process
1310 * and merrily kill the link to whatever was created under the
1311 * same name. Goodbye sticky bit ;-<
1313 if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1318 /* check if the target already exists (in which case we need
1320 status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1321 new_dentry->d_name.len,
1322 &newfe_blkno, new_dir,
1323 &target_lookup_res);
1324 /* The only error we allow here is -ENOENT because the new
1325 * file not existing is perfectly valid. */
1326 if ((status < 0) && (status != -ENOENT)) {
1327 /* If we cannot find the file specified we should just */
1328 /* return the error... */
1335 if (!target_exists && new_inode) {
1337 * Target was unlinked by another node while we were
1338 * waiting to get to ocfs2_rename(). There isn't
1339 * anything we can do here to help the situation, so
1340 * bubble up the appropriate error.
1346 /* In case we need to overwrite an existing file, we blow it
1348 if (target_exists) {
1349 /* VFS didn't think there existed an inode here, but
1350 * someone else in the cluster must have raced our
1351 * rename to create one. Today we error cleanly, in
1352 * the future we should consider calling iget to build
1353 * a new struct inode for this entry. */
1357 trace_ocfs2_rename_target_exists(new_dentry->d_name.len,
1358 new_dentry->d_name.name);
1362 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1365 trace_ocfs2_rename_disagree(
1366 (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1367 (unsigned long long)newfe_blkno,
1368 OCFS2_I(new_inode)->ip_flags);
1372 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1374 if (status != -ENOENT)
1378 new_child_locked = 1;
1380 status = ocfs2_remote_dentry_delete(new_dentry);
1386 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1388 trace_ocfs2_rename_over_existing(
1389 (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1390 (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1392 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1393 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1394 OCFS2_I(new_inode)->ip_blkno,
1395 orphan_name, &orphan_insert);
1400 should_add_orphan = true;
1403 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1405 status = ocfs2_check_dir_for_entry(new_dir,
1406 new_dentry->d_name.name,
1407 new_dentry->d_name.len);
1411 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1412 new_dentry->d_name.name,
1413 new_dentry->d_name.len,
1421 handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1422 if (IS_ERR(handle)) {
1423 status = PTR_ERR(handle);
1429 if (target_exists) {
1430 if (S_ISDIR(new_inode->i_mode)) {
1431 if (new_inode->i_nlink != 2 ||
1432 !ocfs2_empty_dir(new_inode)) {
1433 status = -ENOTEMPTY;
1437 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1439 OCFS2_JOURNAL_ACCESS_WRITE);
1445 /* change the dirent to point to the correct inode */
1446 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1452 new_dir->i_version++;
1454 if (S_ISDIR(new_inode->i_mode))
1455 ocfs2_set_links_count(newfe, 0);
1457 ocfs2_add_links_count(newfe, -1);
1458 ocfs2_journal_dirty(handle, newfe_bh);
1459 if (should_add_orphan) {
1460 status = ocfs2_orphan_add(osb, handle, new_inode,
1461 newfe_bh, orphan_name,
1462 &orphan_insert, orphan_dir);
1469 /* if the name was not found in new_dir, add it now */
1470 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1471 OCFS2_I(old_inode)->ip_blkno,
1472 new_dir_bh, &target_insert);
1475 old_inode->i_ctime = CURRENT_TIME;
1476 mark_inode_dirty(old_inode);
1478 status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1480 OCFS2_JOURNAL_ACCESS_WRITE);
1482 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1484 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1485 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1486 ocfs2_journal_dirty(handle, old_inode_bh);
1491 * Now that the name has been added to new_dir, remove the old name.
1493 * We don't keep any directory entry context around until now
1494 * because the insert might have changed the type of directory
1495 * we're dealing with.
1497 status = ocfs2_find_entry(old_dentry->d_name.name,
1498 old_dentry->d_name.len, old_dir,
1503 status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1510 drop_nlink(new_inode);
1511 new_inode->i_ctime = CURRENT_TIME;
1513 old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1515 if (update_dot_dot) {
1516 status = ocfs2_update_entry(old_inode, handle,
1517 &old_inode_dot_dot_res, new_dir);
1518 drop_nlink(old_dir);
1520 drop_nlink(new_inode);
1523 mark_inode_dirty(new_dir);
1526 mark_inode_dirty(old_dir);
1527 ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1529 mark_inode_dirty(new_inode);
1530 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1533 if (old_dir != new_dir) {
1534 /* Keep the same times on both directories.*/
1535 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1538 * This will also pick up the i_nlink change from the
1541 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1544 if (old_dir_nlink != old_dir->i_nlink) {
1546 mlog(ML_ERROR, "need to change nlink for old dir "
1547 "%llu from %d to %d but bh is NULL!\n",
1548 (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1549 (int)old_dir_nlink, old_dir->i_nlink);
1551 struct ocfs2_dinode *fe;
1552 status = ocfs2_journal_access_di(handle,
1553 INODE_CACHE(old_dir),
1555 OCFS2_JOURNAL_ACCESS_WRITE);
1556 fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1557 ocfs2_set_links_count(fe, old_dir->i_nlink);
1558 ocfs2_journal_dirty(handle, old_dir_bh);
1561 ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1565 ocfs2_rename_unlock(osb);
1568 ocfs2_commit_trans(osb, handle);
1571 ocfs2_double_unlock(old_dir, new_dir);
1573 if (old_child_locked)
1574 ocfs2_inode_unlock(old_inode, 1);
1576 if (new_child_locked)
1577 ocfs2_inode_unlock(new_inode, 1);
1580 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1581 ocfs2_inode_unlock(orphan_dir, 1);
1582 mutex_unlock(&orphan_dir->i_mutex);
1587 sync_mapping_buffers(old_inode->i_mapping);
1592 ocfs2_free_dir_lookup_result(&target_lookup_res);
1593 ocfs2_free_dir_lookup_result(&old_entry_lookup);
1594 ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1595 ocfs2_free_dir_lookup_result(&orphan_insert);
1596 ocfs2_free_dir_lookup_result(&target_insert);
1599 brelse(old_inode_bh);
1610 * we expect i_size = strlen(symname). Copy symname into the file
1611 * data, including the null terminator.
1613 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1615 struct inode *inode,
1616 const char *symname)
1618 struct buffer_head **bhs = NULL;
1620 struct super_block *sb = osb->sb;
1621 u64 p_blkno, p_blocks;
1622 int virtual, blocks, status, i, bytes_left;
1624 bytes_left = i_size_read(inode) + 1;
1625 /* we can't trust i_blocks because we're actually going to
1626 * write i_size + 1 bytes. */
1627 blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1629 trace_ocfs2_create_symlink_data((unsigned long long)inode->i_blocks,
1630 i_size_read(inode), blocks);
1632 /* Sanity check -- make sure we're going to fit. */
1634 ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1640 bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1647 status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1654 /* links can never be larger than one cluster so we know this
1655 * is all going to be contiguous, but do a sanity check
1657 if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1664 while(bytes_left > 0) {
1665 c = &symname[virtual * sb->s_blocksize];
1667 bhs[virtual] = sb_getblk(sb, p_blkno);
1668 if (!bhs[virtual]) {
1673 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1676 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1678 OCFS2_JOURNAL_ACCESS_CREATE);
1684 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1686 memcpy(bhs[virtual]->b_data, c,
1687 (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1690 ocfs2_journal_dirty(handle, bhs[virtual]);
1694 bytes_left -= sb->s_blocksize;
1701 for(i = 0; i < blocks; i++)
1711 static int ocfs2_symlink(struct inode *dir,
1712 struct dentry *dentry,
1713 const char *symname)
1715 int status, l, credits;
1717 struct ocfs2_super *osb = NULL;
1718 struct inode *inode = NULL;
1719 struct super_block *sb;
1720 struct buffer_head *new_fe_bh = NULL;
1721 struct buffer_head *parent_fe_bh = NULL;
1722 struct ocfs2_dinode *fe = NULL;
1723 struct ocfs2_dinode *dirfe;
1724 handle_t *handle = NULL;
1725 struct ocfs2_alloc_context *inode_ac = NULL;
1726 struct ocfs2_alloc_context *data_ac = NULL;
1727 struct ocfs2_alloc_context *xattr_ac = NULL;
1728 int want_clusters = 0;
1729 int xattr_credits = 0;
1730 struct ocfs2_security_xattr_info si = {
1733 int did_quota = 0, did_quota_inode = 0;
1734 struct ocfs2_dir_lookup_result lookup = { NULL, };
1736 int did_block_signals = 0;
1738 trace_ocfs2_symlink_begin(dir, dentry, symname,
1739 dentry->d_name.len, dentry->d_name.name);
1741 dquot_initialize(dir);
1746 l = strlen(symname) + 1;
1748 credits = ocfs2_calc_symlink_credits(sb);
1750 /* lock the parent directory */
1751 status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1753 if (status != -ENOENT)
1758 dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1759 if (!ocfs2_read_links_count(dirfe)) {
1760 /* can't make a file in a deleted directory. */
1765 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1766 dentry->d_name.len);
1770 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1771 dentry->d_name.name,
1772 dentry->d_name.len, &lookup);
1778 status = ocfs2_reserve_new_inode(osb, &inode_ac);
1780 if (status != -ENOSPC)
1785 inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1792 /* get security xattr */
1793 status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
1795 if (status == -EOPNOTSUPP)
1803 /* calculate meta data/clusters for setting security xattr */
1805 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1806 &xattr_credits, &xattr_ac);
1813 /* don't reserve bitmap space for fast symlinks. */
1814 if (l > ocfs2_fast_symlink_chars(sb))
1817 status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1819 if (status != -ENOSPC)
1824 handle = ocfs2_start_trans(osb, credits + xattr_credits);
1825 if (IS_ERR(handle)) {
1826 status = PTR_ERR(handle);
1832 /* Starting to change things, restart is no longer possible. */
1833 ocfs2_block_signals(&oldset);
1834 did_block_signals = 1;
1836 status = dquot_alloc_inode(inode);
1839 did_quota_inode = 1;
1841 trace_ocfs2_symlink_create(dir, dentry, dentry->d_name.len,
1842 dentry->d_name.name,
1843 (unsigned long long)OCFS2_I(dir)->ip_blkno,
1846 status = ocfs2_mknod_locked(osb, dir, inode,
1847 0, &new_fe_bh, parent_fe_bh, handle,
1854 fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1857 inode->i_op = &ocfs2_symlink_inode_operations;
1858 if (l > ocfs2_fast_symlink_chars(sb)) {
1861 status = dquot_alloc_space_nodirty(inode,
1862 ocfs2_clusters_to_bytes(osb->sb, 1));
1866 inode->i_mapping->a_ops = &ocfs2_aops;
1867 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1869 handle, data_ac, NULL,
1872 if (status != -ENOSPC && status != -EINTR) {
1874 "Failed to extend file to %llu\n",
1875 (unsigned long long)newsize);
1881 i_size_write(inode, newsize);
1882 inode->i_blocks = ocfs2_inode_sector_count(inode);
1884 inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
1885 memcpy((char *) fe->id2.i_symlink, symname, l);
1886 i_size_write(inode, newsize);
1887 inode->i_blocks = 0;
1890 status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1896 if (!ocfs2_inode_is_fast_symlink(inode)) {
1897 status = ocfs2_create_symlink_data(osb, handle, inode,
1906 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1915 * Do this before adding the entry to the directory. We add
1916 * also set d_op after success so that ->d_iput() will cleanup
1917 * the dentry lock even if ocfs2_add_entry() fails below.
1919 status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1925 status = ocfs2_add_entry(handle, dentry, inode,
1926 le64_to_cpu(fe->i_blkno), parent_fe_bh,
1933 insert_inode_hash(inode);
1934 d_instantiate(dentry, inode);
1936 if (status < 0 && did_quota)
1937 dquot_free_space_nodirty(inode,
1938 ocfs2_clusters_to_bytes(osb->sb, 1));
1939 if (status < 0 && did_quota_inode)
1940 dquot_free_inode(inode);
1942 ocfs2_commit_trans(osb, handle);
1944 ocfs2_inode_unlock(dir, 1);
1945 if (did_block_signals)
1946 ocfs2_unblock_signals(&oldset);
1949 brelse(parent_fe_bh);
1951 ocfs2_free_dir_lookup_result(&lookup);
1953 ocfs2_free_alloc_context(inode_ac);
1955 ocfs2_free_alloc_context(data_ac);
1957 ocfs2_free_alloc_context(xattr_ac);
1958 if ((status < 0) && inode) {
1959 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
1970 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1972 int status, namelen;
1974 namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1984 if (namelen != OCFS2_ORPHAN_NAMELEN) {
1990 trace_ocfs2_blkno_stringify(blkno, name, namelen);
1999 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
2000 struct inode **ret_orphan_dir,
2001 struct buffer_head **ret_orphan_dir_bh)
2003 struct inode *orphan_dir_inode;
2004 struct buffer_head *orphan_dir_bh = NULL;
2007 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2008 ORPHAN_DIR_SYSTEM_INODE,
2010 if (!orphan_dir_inode) {
2016 mutex_lock(&orphan_dir_inode->i_mutex);
2018 ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2020 mutex_unlock(&orphan_dir_inode->i_mutex);
2021 iput(orphan_dir_inode);
2027 *ret_orphan_dir = orphan_dir_inode;
2028 *ret_orphan_dir_bh = orphan_dir_bh;
2033 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
2034 struct buffer_head *orphan_dir_bh,
2037 struct ocfs2_dir_lookup_result *lookup)
2040 struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
2042 ret = ocfs2_blkno_stringify(blkno, name);
2048 ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2049 orphan_dir_bh, name,
2050 OCFS2_ORPHAN_NAMELEN, lookup);
2060 * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
2061 * insertion of an orphan.
2062 * @osb: ocfs2 file system
2063 * @ret_orphan_dir: Orphan dir inode - returned locked!
2064 * @blkno: Actual block number of the inode to be inserted into orphan dir.
2065 * @lookup: dir lookup result, to be passed back into functions like
2068 * Returns zero on success and the ret_orphan_dir, name and lookup
2069 * fields will be populated.
2071 * Returns non-zero on failure.
2073 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2074 struct inode **ret_orphan_dir,
2077 struct ocfs2_dir_lookup_result *lookup)
2079 struct inode *orphan_dir_inode = NULL;
2080 struct buffer_head *orphan_dir_bh = NULL;
2083 ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
2090 ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
2091 blkno, name, lookup);
2097 *ret_orphan_dir = orphan_dir_inode;
2100 brelse(orphan_dir_bh);
2103 ocfs2_inode_unlock(orphan_dir_inode, 1);
2104 mutex_unlock(&orphan_dir_inode->i_mutex);
2105 iput(orphan_dir_inode);
2113 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2115 struct inode *inode,
2116 struct buffer_head *fe_bh,
2118 struct ocfs2_dir_lookup_result *lookup,
2119 struct inode *orphan_dir_inode)
2121 struct buffer_head *orphan_dir_bh = NULL;
2123 struct ocfs2_dinode *orphan_fe;
2124 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
2126 trace_ocfs2_orphan_add_begin(
2127 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2129 status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2135 status = ocfs2_journal_access_di(handle,
2136 INODE_CACHE(orphan_dir_inode),
2138 OCFS2_JOURNAL_ACCESS_WRITE);
2145 * We're going to journal the change of i_flags and i_orphaned_slot.
2146 * It's safe anyway, though some callers may duplicate the journaling.
2147 * Journaling within the func just make the logic look more
2150 status = ocfs2_journal_access_di(handle,
2153 OCFS2_JOURNAL_ACCESS_WRITE);
2159 /* we're a cluster, and nlink can change on disk from
2160 * underneath us... */
2161 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2162 if (S_ISDIR(inode->i_mode))
2163 ocfs2_add_links_count(orphan_fe, 1);
2164 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2165 ocfs2_journal_dirty(handle, orphan_dir_bh);
2167 status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2168 OCFS2_ORPHAN_NAMELEN, inode,
2169 OCFS2_I(inode)->ip_blkno,
2170 orphan_dir_bh, lookup);
2176 fe->i_flags |= cpu_to_le32(OCFS2_ORPHANED_FL);
2177 OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2179 /* Record which orphan dir our inode now resides
2180 * in. delete_inode will use this to determine which orphan
2182 fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2184 ocfs2_journal_dirty(handle, fe_bh);
2186 trace_ocfs2_orphan_add_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
2191 if (S_ISDIR(inode->i_mode))
2192 ocfs2_add_links_count(orphan_fe, -1);
2193 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2197 brelse(orphan_dir_bh);
2202 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2203 int ocfs2_orphan_del(struct ocfs2_super *osb,
2205 struct inode *orphan_dir_inode,
2206 struct inode *inode,
2207 struct buffer_head *orphan_dir_bh)
2209 char name[OCFS2_ORPHAN_NAMELEN + 1];
2210 struct ocfs2_dinode *orphan_fe;
2212 struct ocfs2_dir_lookup_result lookup = { NULL, };
2214 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2220 trace_ocfs2_orphan_del(
2221 (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2222 name, OCFS2_ORPHAN_NAMELEN);
2224 /* find it's spot in the orphan directory */
2225 status = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN, orphan_dir_inode,
2232 /* remove it from the orphan directory */
2233 status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2239 status = ocfs2_journal_access_di(handle,
2240 INODE_CACHE(orphan_dir_inode),
2242 OCFS2_JOURNAL_ACCESS_WRITE);
2248 /* do the i_nlink dance! :) */
2249 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2250 if (S_ISDIR(inode->i_mode))
2251 ocfs2_add_links_count(orphan_fe, -1);
2252 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2253 ocfs2_journal_dirty(handle, orphan_dir_bh);
2256 ocfs2_free_dir_lookup_result(&lookup);
2264 * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to receive a newly
2265 * allocated file. This is different from the typical 'add to orphan dir'
2266 * operation in that the inode does not yet exist. This is a problem because
2267 * the orphan dir stringifies the inode block number to come up with it's
2268 * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2269 * problem. This function works around it by calling deeper into the orphan
2270 * and suballoc code than other callers. Use this only by necessity.
2271 * @dir: The directory which this inode will ultimately wind up under - not the
2273 * @dir_bh: buffer_head the @dir inode block
2274 * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2275 * with the string to be used for orphan dirent. Pass back to the orphan dir
2277 * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2279 * @ret_di_blkno: block number where the new inode will be allocated.
2280 * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2281 * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2283 * Returns zero on success and the ret_orphan_dir, name and lookup
2284 * fields will be populated.
2286 * Returns non-zero on failure.
2288 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2289 struct buffer_head *dir_bh,
2291 struct inode **ret_orphan_dir,
2293 struct ocfs2_dir_lookup_result *orphan_insert,
2294 struct ocfs2_alloc_context **ret_inode_ac)
2298 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2299 struct inode *orphan_dir = NULL;
2300 struct buffer_head *orphan_dir_bh = NULL;
2301 struct ocfs2_alloc_context *inode_ac = NULL;
2303 ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2309 /* reserve an inode spot */
2310 ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2317 ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2324 ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2325 di_blkno, orphan_name, orphan_insert);
2333 *ret_orphan_dir = orphan_dir;
2334 *ret_di_blkno = di_blkno;
2335 *ret_inode_ac = inode_ac;
2337 * orphan_name and orphan_insert are already up to
2338 * date via prepare_orphan_dir
2341 /* Unroll reserve_new_inode* */
2343 ocfs2_free_alloc_context(inode_ac);
2345 /* Unroll orphan dir locking */
2346 mutex_unlock(&orphan_dir->i_mutex);
2347 ocfs2_inode_unlock(orphan_dir, 1);
2351 brelse(orphan_dir_bh);
2356 int ocfs2_create_inode_in_orphan(struct inode *dir,
2358 struct inode **new_inode)
2360 int status, did_quota_inode = 0;
2361 struct inode *inode = NULL;
2362 struct inode *orphan_dir = NULL;
2363 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2364 struct ocfs2_dinode *di = NULL;
2365 handle_t *handle = NULL;
2366 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2367 struct buffer_head *parent_di_bh = NULL;
2368 struct buffer_head *new_di_bh = NULL;
2369 struct ocfs2_alloc_context *inode_ac = NULL;
2370 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2371 u64 uninitialized_var(di_blkno), suballoc_loc;
2374 status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2376 if (status != -ENOENT)
2381 status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2382 orphan_name, &orphan_dir,
2383 &di_blkno, &orphan_insert, &inode_ac);
2385 if (status != -ENOSPC)
2390 inode = ocfs2_get_init_inode(dir, mode);
2397 handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2398 if (IS_ERR(handle)) {
2399 status = PTR_ERR(handle);
2405 status = dquot_alloc_inode(inode);
2408 did_quota_inode = 1;
2410 status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2412 &suballoc_bit, di_blkno);
2419 /* do the real work now. */
2420 status = __ocfs2_mknod_locked(dir, inode,
2421 0, &new_di_bh, parent_di_bh, handle,
2422 inode_ac, di_blkno, suballoc_loc,
2429 di = (struct ocfs2_dinode *)new_di_bh->b_data;
2430 status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2431 &orphan_insert, orphan_dir);
2437 /* get open lock so that only nodes can't remove it from orphan dir. */
2438 status = ocfs2_open_lock(inode);
2442 insert_inode_hash(inode);
2444 if (status < 0 && did_quota_inode)
2445 dquot_free_inode(inode);
2447 ocfs2_commit_trans(osb, handle);
2450 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2451 ocfs2_inode_unlock(orphan_dir, 1);
2452 mutex_unlock(&orphan_dir->i_mutex);
2456 if ((status < 0) && inode) {
2462 ocfs2_free_alloc_context(inode_ac);
2469 ocfs2_free_dir_lookup_result(&orphan_insert);
2471 ocfs2_inode_unlock(dir, 1);
2472 brelse(parent_di_bh);
2476 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2477 struct inode *inode,
2478 struct dentry *dentry)
2481 struct buffer_head *parent_di_bh = NULL;
2482 handle_t *handle = NULL;
2483 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2484 struct ocfs2_dinode *dir_di, *di;
2485 struct inode *orphan_dir_inode = NULL;
2486 struct buffer_head *orphan_dir_bh = NULL;
2487 struct buffer_head *di_bh = NULL;
2488 struct ocfs2_dir_lookup_result lookup = { NULL, };
2490 trace_ocfs2_mv_orphaned_inode_to_new(dir, dentry,
2491 dentry->d_name.len, dentry->d_name.name,
2492 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2493 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2495 status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2497 if (status != -ENOENT)
2502 dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2503 if (!dir_di->i_links_count) {
2504 /* can't make a file in a deleted directory. */
2509 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2510 dentry->d_name.len);
2514 /* get a spot inside the dir. */
2515 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2516 dentry->d_name.name,
2517 dentry->d_name.len, &lookup);
2523 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2524 ORPHAN_DIR_SYSTEM_INODE,
2526 if (!orphan_dir_inode) {
2532 mutex_lock(&orphan_dir_inode->i_mutex);
2534 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2537 mutex_unlock(&orphan_dir_inode->i_mutex);
2538 iput(orphan_dir_inode);
2542 status = ocfs2_read_inode_block(inode, &di_bh);
2548 handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2549 if (IS_ERR(handle)) {
2550 status = PTR_ERR(handle);
2556 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2557 di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2563 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2570 di = (struct ocfs2_dinode *)di_bh->b_data;
2571 di->i_flags &= ~cpu_to_le32(OCFS2_ORPHANED_FL);
2572 di->i_orphaned_slot = 0;
2573 set_nlink(inode, 1);
2574 ocfs2_set_links_count(di, inode->i_nlink);
2575 ocfs2_update_inode_fsync_trans(handle, inode, 1);
2576 ocfs2_journal_dirty(handle, di_bh);
2578 status = ocfs2_add_entry(handle, dentry, inode,
2579 OCFS2_I(inode)->ip_blkno, parent_di_bh,
2586 status = ocfs2_dentry_attach_lock(dentry, inode,
2587 OCFS2_I(dir)->ip_blkno);
2593 d_instantiate(dentry, inode);
2596 ocfs2_commit_trans(osb, handle);
2598 ocfs2_inode_unlock(orphan_dir_inode, 1);
2599 mutex_unlock(&orphan_dir_inode->i_mutex);
2600 iput(orphan_dir_inode);
2603 ocfs2_inode_unlock(dir, 1);
2606 brelse(parent_di_bh);
2607 brelse(orphan_dir_bh);
2609 ocfs2_free_dir_lookup_result(&lookup);
2617 const struct inode_operations ocfs2_dir_iops = {
2618 .create = ocfs2_create,
2619 .lookup = ocfs2_lookup,
2621 .unlink = ocfs2_unlink,
2622 .rmdir = ocfs2_unlink,
2623 .symlink = ocfs2_symlink,
2624 .mkdir = ocfs2_mkdir,
2625 .mknod = ocfs2_mknod,
2626 .rename = ocfs2_rename,
2627 .setattr = ocfs2_setattr,
2628 .getattr = ocfs2_getattr,
2629 .permission = ocfs2_permission,
2630 .setxattr = generic_setxattr,
2631 .getxattr = generic_getxattr,
2632 .listxattr = ocfs2_listxattr,
2633 .removexattr = generic_removexattr,
2634 .fiemap = ocfs2_fiemap,
2635 .get_acl = ocfs2_iop_get_acl,
2636 .set_acl = ocfs2_iop_set_acl,