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,
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,
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_nested(dir, NULL, 0, OI_LS_PARENT);
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 d_set_d_op(dentry, &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);
175 ocfs2_dentry_attach_gen(dentry);
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_init_owner(inode, dir, mode);
209 dquot_initialize(inode);
213 static int ocfs2_mknod(struct inode *dir,
214 struct dentry *dentry,
219 struct buffer_head *parent_fe_bh = NULL;
220 handle_t *handle = NULL;
221 struct ocfs2_super *osb;
222 struct ocfs2_dinode *dirfe;
223 struct buffer_head *new_fe_bh = NULL;
224 struct inode *inode = NULL;
225 struct ocfs2_alloc_context *inode_ac = NULL;
226 struct ocfs2_alloc_context *data_ac = NULL;
227 struct ocfs2_alloc_context *meta_ac = NULL;
228 int want_clusters = 0;
230 int xattr_credits = 0;
231 struct ocfs2_security_xattr_info si = {
234 int did_quota_inode = 0;
235 struct ocfs2_dir_lookup_result lookup = { NULL, };
237 int did_block_signals = 0;
239 mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
240 (unsigned long)dev, dentry->d_name.len,
241 dentry->d_name.name);
243 dquot_initialize(dir);
245 /* get our super block */
246 osb = OCFS2_SB(dir->i_sb);
248 status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
250 if (status != -ENOENT)
255 if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
260 dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
261 if (!ocfs2_read_links_count(dirfe)) {
262 /* can't make a file in a deleted directory. */
267 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
272 /* get a spot inside the dir. */
273 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
275 dentry->d_name.len, &lookup);
281 /* reserve an inode spot */
282 status = ocfs2_reserve_new_inode(osb, &inode_ac);
284 if (status != -ENOSPC)
289 inode = ocfs2_get_init_inode(dir, mode);
296 /* get security xattr */
297 status = ocfs2_init_security_get(inode, dir, &si);
299 if (status == -EOPNOTSUPP)
307 /* calculate meta data/clusters for setting security and acl xattr */
308 status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
310 &xattr_credits, &want_meta);
316 /* Reserve a cluster if creating an extent based directory. */
317 if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
320 /* Dir indexing requires extra space as well */
321 if (ocfs2_supports_indexed_dirs(osb))
325 status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
327 if (status != -ENOSPC)
332 status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
334 if (status != -ENOSPC)
339 handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
342 if (IS_ERR(handle)) {
343 status = PTR_ERR(handle);
349 /* Starting to change things, restart is no longer possible. */
350 ocfs2_block_signals(&oldset);
351 did_block_signals = 1;
353 status = dquot_alloc_inode(inode);
358 mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry,
359 inode->i_mode, (unsigned long)dev, dentry->d_name.len,
360 dentry->d_name.name);
362 /* do the real work now. */
363 status = ocfs2_mknod_locked(osb, dir, inode, dev,
364 &new_fe_bh, parent_fe_bh, handle,
372 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
373 new_fe_bh, data_ac, meta_ac);
379 status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
381 OCFS2_JOURNAL_ACCESS_WRITE);
386 ocfs2_add_links_count(dirfe, 1);
387 ocfs2_journal_dirty(handle, parent_fe_bh);
391 status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
399 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
408 * Do this before adding the entry to the directory. We add
409 * also set d_op after success so that ->d_iput() will cleanup
410 * the dentry lock even if ocfs2_add_entry() fails below.
412 status = ocfs2_dentry_attach_lock(dentry, inode,
413 OCFS2_I(dir)->ip_blkno);
418 d_set_d_op(dentry, &ocfs2_dentry_ops);
420 status = ocfs2_add_entry(handle, dentry, inode,
421 OCFS2_I(inode)->ip_blkno, parent_fe_bh,
428 insert_inode_hash(inode);
429 d_instantiate(dentry, inode);
432 if (status < 0 && did_quota_inode)
433 dquot_free_inode(inode);
435 ocfs2_commit_trans(osb, handle);
437 ocfs2_inode_unlock(dir, 1);
438 if (did_block_signals)
439 ocfs2_unblock_signals(&oldset);
441 if (status == -ENOSPC)
442 mlog(0, "Disk is full\n");
445 brelse(parent_fe_bh);
449 ocfs2_free_dir_lookup_result(&lookup);
452 ocfs2_free_alloc_context(inode_ac);
455 ocfs2_free_alloc_context(data_ac);
458 ocfs2_free_alloc_context(meta_ac);
461 * We should call iput after the i_mutex of the bitmap been
462 * unlocked in ocfs2_free_alloc_context, or the
463 * ocfs2_delete_inode will mutex_lock again.
465 if ((status < 0) && inode) {
466 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
476 static int __ocfs2_mknod_locked(struct inode *dir,
479 struct buffer_head **new_fe_bh,
480 struct buffer_head *parent_fe_bh,
482 struct ocfs2_alloc_context *inode_ac,
483 u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
486 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
487 struct ocfs2_dinode *fe = NULL;
488 struct ocfs2_extent_list *fel;
493 /* populate as many fields early on as possible - many of
494 * these are used by the support functions here and in
496 inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
497 OCFS2_I(inode)->ip_blkno = fe_blkno;
498 spin_lock(&osb->osb_lock);
499 inode->i_generation = osb->s_next_generation++;
500 spin_unlock(&osb->osb_lock);
502 *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
508 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
510 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
512 OCFS2_JOURNAL_ACCESS_CREATE);
518 fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
519 memset(fe, 0, osb->sb->s_blocksize);
521 fe->i_generation = cpu_to_le32(inode->i_generation);
522 fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
523 fe->i_blkno = cpu_to_le64(fe_blkno);
524 fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
525 fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
526 fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
527 fe->i_uid = cpu_to_le32(inode->i_uid);
528 fe->i_gid = cpu_to_le32(inode->i_gid);
529 fe->i_mode = cpu_to_le16(inode->i_mode);
530 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
531 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
533 ocfs2_set_links_count(fe, inode->i_nlink);
535 fe->i_last_eb_blk = 0;
536 strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
537 le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
538 fe->i_atime = fe->i_ctime = fe->i_mtime =
539 cpu_to_le64(CURRENT_TIME.tv_sec);
540 fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
541 cpu_to_le32(CURRENT_TIME.tv_nsec);
545 * If supported, directories start with inline data. If inline
546 * isn't supported, but indexing is, we start them as indexed.
548 feat = le16_to_cpu(fe->i_dyn_features);
549 if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
550 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
552 fe->id2.i_data.id_count = cpu_to_le16(
553 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
555 fel = &fe->id2.i_list;
556 fel->l_tree_depth = 0;
557 fel->l_next_free_rec = 0;
558 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
561 ocfs2_journal_dirty(handle, *new_fe_bh);
563 ocfs2_populate_inode(inode, fe, 1);
564 ocfs2_ci_set_new(osb, INODE_CACHE(inode));
565 if (!ocfs2_mount_local(osb)) {
566 status = ocfs2_create_new_inode_locks(inode);
571 status = 0; /* error in ocfs2_create_new_inode_locks is not
586 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
590 struct buffer_head **new_fe_bh,
591 struct buffer_head *parent_fe_bh,
593 struct ocfs2_alloc_context *inode_ac)
596 u64 suballoc_loc, fe_blkno = 0;
601 status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
602 inode_ac, &suballoc_loc,
603 &suballoc_bit, &fe_blkno);
609 return __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
610 parent_fe_bh, handle, inode_ac,
611 fe_blkno, suballoc_loc, suballoc_bit);
614 static int ocfs2_mkdir(struct inode *dir,
615 struct dentry *dentry,
620 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
621 dentry->d_name.len, dentry->d_name.name);
622 ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
628 static int ocfs2_create(struct inode *dir,
629 struct dentry *dentry,
631 struct nameidata *nd)
635 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
636 dentry->d_name.len, dentry->d_name.name);
637 ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
643 static int ocfs2_link(struct dentry *old_dentry,
645 struct dentry *dentry)
648 struct inode *inode = old_dentry->d_inode;
650 struct buffer_head *fe_bh = NULL;
651 struct buffer_head *parent_fe_bh = NULL;
652 struct ocfs2_dinode *fe = NULL;
653 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
654 struct ocfs2_dir_lookup_result lookup = { NULL, };
657 mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
658 old_dentry->d_name.len, old_dentry->d_name.name,
659 dentry->d_name.len, dentry->d_name.name);
661 if (S_ISDIR(inode->i_mode))
664 dquot_initialize(dir);
666 err = ocfs2_inode_lock_nested(dir, &parent_fe_bh, 1, OI_LS_PARENT);
678 err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
683 err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
685 dentry->d_name.len, &lookup);
691 err = ocfs2_inode_lock(inode, &fe_bh, 1);
698 fe = (struct ocfs2_dinode *) fe_bh->b_data;
699 if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
701 goto out_unlock_inode;
704 handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
705 if (IS_ERR(handle)) {
706 err = PTR_ERR(handle);
709 goto out_unlock_inode;
712 /* Starting to change things, restart is no longer possible. */
713 ocfs2_block_signals(&oldset);
715 err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
716 OCFS2_JOURNAL_ACCESS_WRITE);
723 inode->i_ctime = CURRENT_TIME;
724 ocfs2_set_links_count(fe, inode->i_nlink);
725 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
726 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
727 ocfs2_journal_dirty(handle, fe_bh);
729 err = ocfs2_add_entry(handle, dentry, inode,
730 OCFS2_I(inode)->ip_blkno,
731 parent_fe_bh, &lookup);
733 ocfs2_add_links_count(fe, -1);
739 err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
746 d_set_d_op(dentry, &ocfs2_dentry_ops);
747 d_instantiate(dentry, inode);
750 ocfs2_commit_trans(osb, handle);
751 ocfs2_unblock_signals(&oldset);
753 ocfs2_inode_unlock(inode, 1);
756 ocfs2_inode_unlock(dir, 1);
759 brelse(parent_fe_bh);
761 ocfs2_free_dir_lookup_result(&lookup);
769 * Takes and drops an exclusive lock on the given dentry. This will
770 * force other nodes to drop it.
772 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
776 ret = ocfs2_dentry_lock(dentry, 1);
780 ocfs2_dentry_unlock(dentry, 1);
785 static inline int inode_is_unlinkable(struct inode *inode)
787 if (S_ISDIR(inode->i_mode)) {
788 if (inode->i_nlink == 2)
793 if (inode->i_nlink == 1)
798 static int ocfs2_unlink(struct inode *dir,
799 struct dentry *dentry)
802 int child_locked = 0;
803 struct inode *inode = dentry->d_inode;
804 struct inode *orphan_dir = NULL;
805 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
807 struct ocfs2_dinode *fe = NULL;
808 struct buffer_head *fe_bh = NULL;
809 struct buffer_head *parent_node_bh = NULL;
810 handle_t *handle = NULL;
811 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
812 struct ocfs2_dir_lookup_result lookup = { NULL, };
813 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
815 mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
816 dentry->d_name.len, dentry->d_name.name);
818 dquot_initialize(dir);
820 BUG_ON(dentry->d_parent->d_inode != dir);
822 mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
824 if (inode == osb->root_inode) {
825 mlog(0, "Cannot delete the root directory\n");
829 status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
832 if (status != -ENOENT)
837 status = ocfs2_find_files_on_disk(dentry->d_name.name,
838 dentry->d_name.len, &blkno, dir,
841 if (status != -ENOENT)
846 if (OCFS2_I(inode)->ip_blkno != blkno) {
849 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
850 (unsigned long long)OCFS2_I(inode)->ip_blkno,
851 (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
855 status = ocfs2_inode_lock(inode, &fe_bh, 1);
857 if (status != -ENOENT)
863 if (S_ISDIR(inode->i_mode)) {
864 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
870 status = ocfs2_remote_dentry_delete(dentry);
872 /* This remote delete should succeed under all normal
878 if (inode_is_unlinkable(inode)) {
879 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
880 OCFS2_I(inode)->ip_blkno,
881 orphan_name, &orphan_insert);
888 handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
889 if (IS_ERR(handle)) {
890 status = PTR_ERR(handle);
896 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
897 OCFS2_JOURNAL_ACCESS_WRITE);
903 fe = (struct ocfs2_dinode *) fe_bh->b_data;
905 if (inode_is_unlinkable(inode)) {
906 status = ocfs2_orphan_add(osb, handle, inode, fe_bh, orphan_name,
907 &orphan_insert, orphan_dir);
914 /* delete the name from the parent dir */
915 status = ocfs2_delete_entry(handle, dir, &lookup);
921 if (S_ISDIR(inode->i_mode))
924 ocfs2_set_links_count(fe, inode->i_nlink);
925 ocfs2_journal_dirty(handle, fe_bh);
927 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
928 if (S_ISDIR(inode->i_mode))
931 status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
934 if (S_ISDIR(inode->i_mode))
940 ocfs2_commit_trans(osb, handle);
943 ocfs2_inode_unlock(inode, 1);
945 ocfs2_inode_unlock(dir, 1);
948 /* This was locked for us in ocfs2_prepare_orphan_dir() */
949 ocfs2_inode_unlock(orphan_dir, 1);
950 mutex_unlock(&orphan_dir->i_mutex);
955 brelse(parent_node_bh);
957 ocfs2_free_dir_lookup_result(&orphan_insert);
958 ocfs2_free_dir_lookup_result(&lookup);
966 * The only place this should be used is rename!
967 * if they have the same id, then the 1st one is the only one locked.
969 static int ocfs2_double_lock(struct ocfs2_super *osb,
970 struct buffer_head **bh1,
971 struct inode *inode1,
972 struct buffer_head **bh2,
973 struct inode *inode2)
976 struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
977 struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
978 struct buffer_head **tmpbh;
979 struct inode *tmpinode;
981 mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
982 (unsigned long long)oi1->ip_blkno,
983 (unsigned long long)oi2->ip_blkno);
990 /* we always want to lock the one with the lower lockid first. */
991 if (oi1->ip_blkno != oi2->ip_blkno) {
992 if (oi1->ip_blkno < oi2->ip_blkno) {
993 /* switch id1 and id2 around */
994 mlog(0, "switching them around...\n");
1004 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
1007 if (status != -ENOENT)
1014 status = ocfs2_inode_lock_nested(inode1, bh1, 1, OI_LS_RENAME2);
1017 * An error return must mean that no cluster locks
1018 * were held on function exit.
1020 if (oi1->ip_blkno != oi2->ip_blkno) {
1021 ocfs2_inode_unlock(inode2, 1);
1026 if (status != -ENOENT)
1035 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1037 ocfs2_inode_unlock(inode1, 1);
1039 if (inode1 != inode2)
1040 ocfs2_inode_unlock(inode2, 1);
1043 static int ocfs2_rename(struct inode *old_dir,
1044 struct dentry *old_dentry,
1045 struct inode *new_dir,
1046 struct dentry *new_dentry)
1048 int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1049 int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1050 struct inode *old_inode = old_dentry->d_inode;
1051 struct inode *new_inode = new_dentry->d_inode;
1052 struct inode *orphan_dir = NULL;
1053 struct ocfs2_dinode *newfe = NULL;
1054 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1055 struct buffer_head *newfe_bh = NULL;
1056 struct buffer_head *old_inode_bh = NULL;
1057 struct ocfs2_super *osb = NULL;
1058 u64 newfe_blkno, old_de_ino;
1059 handle_t *handle = NULL;
1060 struct buffer_head *old_dir_bh = NULL;
1061 struct buffer_head *new_dir_bh = NULL;
1062 nlink_t old_dir_nlink = old_dir->i_nlink;
1063 struct ocfs2_dinode *old_di;
1064 struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1065 struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1066 struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1067 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1068 struct ocfs2_dir_lookup_result target_insert = { NULL, };
1070 /* At some point it might be nice to break this function up a
1073 mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1074 old_dir, old_dentry, new_dir, new_dentry,
1075 old_dentry->d_name.len, old_dentry->d_name.name,
1076 new_dentry->d_name.len, new_dentry->d_name.name);
1078 dquot_initialize(old_dir);
1079 dquot_initialize(new_dir);
1081 osb = OCFS2_SB(old_dir->i_sb);
1084 if (!igrab(new_inode))
1088 /* Assume a directory hierarchy thusly:
1091 * a,b,c, and d are all directories.
1093 * from cwd of 'a' on both nodes:
1097 * And that's why, just like the VFS, we need a file system
1099 if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1100 status = ocfs2_rename_lock(osb);
1108 /* if old and new are the same, this'll just do one lock. */
1109 status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1110 &new_dir_bh, new_dir);
1117 /* make sure both dirs have bhs
1118 * get an extra ref on old_dir_bh if old==new */
1121 new_dir_bh = old_dir_bh;
1124 mlog(ML_ERROR, "no old_dir_bh!\n");
1131 * Aside from allowing a meta data update, the locking here
1132 * also ensures that the downconvert thread on other nodes
1133 * won't have to concurrently downconvert the inode and the
1136 status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1139 if (status != -ENOENT)
1143 old_child_locked = 1;
1145 status = ocfs2_remote_dentry_delete(old_dentry);
1151 if (S_ISDIR(old_inode->i_mode)) {
1152 u64 old_inode_parent;
1155 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1157 &old_inode_dot_dot_res);
1163 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1168 if (!new_inode && new_dir != old_dir &&
1169 new_dir->i_nlink >= ocfs2_link_max(osb)) {
1175 status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1176 old_dentry->d_name.len,
1184 * Check for inode number is _not_ due to possible IO errors.
1185 * We might rmdir the source, keep it as pwd of some process
1186 * and merrily kill the link to whatever was created under the
1187 * same name. Goodbye sticky bit ;-<
1189 if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1194 /* check if the target already exists (in which case we need
1196 status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1197 new_dentry->d_name.len,
1198 &newfe_blkno, new_dir,
1199 &target_lookup_res);
1200 /* The only error we allow here is -ENOENT because the new
1201 * file not existing is perfectly valid. */
1202 if ((status < 0) && (status != -ENOENT)) {
1203 /* If we cannot find the file specified we should just */
1204 /* return the error... */
1211 if (!target_exists && new_inode) {
1213 * Target was unlinked by another node while we were
1214 * waiting to get to ocfs2_rename(). There isn't
1215 * anything we can do here to help the situation, so
1216 * bubble up the appropriate error.
1222 /* In case we need to overwrite an existing file, we blow it
1224 if (target_exists) {
1225 /* VFS didn't think there existed an inode here, but
1226 * someone else in the cluster must have raced our
1227 * rename to create one. Today we error cleanly, in
1228 * the future we should consider calling iget to build
1229 * a new struct inode for this entry. */
1233 mlog(0, "We found an inode for name %.*s but VFS "
1234 "didn't give us one.\n", new_dentry->d_name.len,
1235 new_dentry->d_name.name);
1239 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1242 mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1243 (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1244 (unsigned long long)newfe_blkno,
1245 OCFS2_I(new_inode)->ip_flags);
1249 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1251 if (status != -ENOENT)
1255 new_child_locked = 1;
1257 status = ocfs2_remote_dentry_delete(new_dentry);
1263 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1265 mlog(0, "aha rename over existing... new_blkno=%llu "
1266 "newfebh=%p bhblocknr=%llu\n",
1267 (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1268 (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1270 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1271 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1272 OCFS2_I(new_inode)->ip_blkno,
1273 orphan_name, &orphan_insert);
1280 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1282 status = ocfs2_check_dir_for_entry(new_dir,
1283 new_dentry->d_name.name,
1284 new_dentry->d_name.len);
1288 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1289 new_dentry->d_name.name,
1290 new_dentry->d_name.len,
1298 handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1299 if (IS_ERR(handle)) {
1300 status = PTR_ERR(handle);
1306 if (target_exists) {
1307 if (S_ISDIR(new_inode->i_mode)) {
1308 if (new_inode->i_nlink != 2 ||
1309 !ocfs2_empty_dir(new_inode)) {
1310 status = -ENOTEMPTY;
1314 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1316 OCFS2_JOURNAL_ACCESS_WRITE);
1322 if (S_ISDIR(new_inode->i_mode) ||
1323 (ocfs2_read_links_count(newfe) == 1)) {
1324 status = ocfs2_orphan_add(osb, handle, new_inode,
1325 newfe_bh, orphan_name,
1326 &orphan_insert, orphan_dir);
1333 /* change the dirent to point to the correct inode */
1334 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1340 new_dir->i_version++;
1342 if (S_ISDIR(new_inode->i_mode))
1343 ocfs2_set_links_count(newfe, 0);
1345 ocfs2_add_links_count(newfe, -1);
1346 ocfs2_journal_dirty(handle, newfe_bh);
1348 /* if the name was not found in new_dir, add it now */
1349 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1350 OCFS2_I(old_inode)->ip_blkno,
1351 new_dir_bh, &target_insert);
1354 old_inode->i_ctime = CURRENT_TIME;
1355 mark_inode_dirty(old_inode);
1357 status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1359 OCFS2_JOURNAL_ACCESS_WRITE);
1361 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1363 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1364 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1365 ocfs2_journal_dirty(handle, old_inode_bh);
1370 * Now that the name has been added to new_dir, remove the old name.
1372 * We don't keep any directory entry context around until now
1373 * because the insert might have changed the type of directory
1374 * we're dealing with.
1376 status = ocfs2_find_entry(old_dentry->d_name.name,
1377 old_dentry->d_name.len, old_dir,
1382 status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1389 new_inode->i_nlink--;
1390 new_inode->i_ctime = CURRENT_TIME;
1392 old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1394 if (update_dot_dot) {
1395 status = ocfs2_update_entry(old_inode, handle,
1396 &old_inode_dot_dot_res, new_dir);
1399 new_inode->i_nlink--;
1402 mark_inode_dirty(new_dir);
1405 mark_inode_dirty(old_dir);
1406 ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1408 mark_inode_dirty(new_inode);
1409 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1412 if (old_dir != new_dir) {
1413 /* Keep the same times on both directories.*/
1414 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1417 * This will also pick up the i_nlink change from the
1420 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1423 if (old_dir_nlink != old_dir->i_nlink) {
1425 mlog(ML_ERROR, "need to change nlink for old dir "
1426 "%llu from %d to %d but bh is NULL!\n",
1427 (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1428 (int)old_dir_nlink, old_dir->i_nlink);
1430 struct ocfs2_dinode *fe;
1431 status = ocfs2_journal_access_di(handle,
1432 INODE_CACHE(old_dir),
1434 OCFS2_JOURNAL_ACCESS_WRITE);
1435 fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1436 ocfs2_set_links_count(fe, old_dir->i_nlink);
1437 ocfs2_journal_dirty(handle, old_dir_bh);
1440 ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1444 ocfs2_rename_unlock(osb);
1447 ocfs2_commit_trans(osb, handle);
1450 ocfs2_double_unlock(old_dir, new_dir);
1452 if (old_child_locked)
1453 ocfs2_inode_unlock(old_inode, 1);
1455 if (new_child_locked)
1456 ocfs2_inode_unlock(new_inode, 1);
1459 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1460 ocfs2_inode_unlock(orphan_dir, 1);
1461 mutex_unlock(&orphan_dir->i_mutex);
1466 sync_mapping_buffers(old_inode->i_mapping);
1471 ocfs2_free_dir_lookup_result(&target_lookup_res);
1472 ocfs2_free_dir_lookup_result(&old_entry_lookup);
1473 ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1474 ocfs2_free_dir_lookup_result(&orphan_insert);
1475 ocfs2_free_dir_lookup_result(&target_insert);
1478 brelse(old_inode_bh);
1488 * we expect i_size = strlen(symname). Copy symname into the file
1489 * data, including the null terminator.
1491 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1493 struct inode *inode,
1494 const char *symname)
1496 struct buffer_head **bhs = NULL;
1498 struct super_block *sb = osb->sb;
1499 u64 p_blkno, p_blocks;
1500 int virtual, blocks, status, i, bytes_left;
1502 bytes_left = i_size_read(inode) + 1;
1503 /* we can't trust i_blocks because we're actually going to
1504 * write i_size + 1 bytes. */
1505 blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1507 mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1508 (unsigned long long)inode->i_blocks,
1509 i_size_read(inode), blocks);
1511 /* Sanity check -- make sure we're going to fit. */
1513 ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1519 bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1526 status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1533 /* links can never be larger than one cluster so we know this
1534 * is all going to be contiguous, but do a sanity check
1536 if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1543 while(bytes_left > 0) {
1544 c = &symname[virtual * sb->s_blocksize];
1546 bhs[virtual] = sb_getblk(sb, p_blkno);
1547 if (!bhs[virtual]) {
1552 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1555 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1557 OCFS2_JOURNAL_ACCESS_CREATE);
1563 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1565 memcpy(bhs[virtual]->b_data, c,
1566 (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1569 ocfs2_journal_dirty(handle, bhs[virtual]);
1573 bytes_left -= sb->s_blocksize;
1580 for(i = 0; i < blocks; i++)
1589 static int ocfs2_symlink(struct inode *dir,
1590 struct dentry *dentry,
1591 const char *symname)
1593 int status, l, credits;
1595 struct ocfs2_super *osb = NULL;
1596 struct inode *inode = NULL;
1597 struct super_block *sb;
1598 struct buffer_head *new_fe_bh = NULL;
1599 struct buffer_head *parent_fe_bh = NULL;
1600 struct ocfs2_dinode *fe = NULL;
1601 struct ocfs2_dinode *dirfe;
1602 handle_t *handle = NULL;
1603 struct ocfs2_alloc_context *inode_ac = NULL;
1604 struct ocfs2_alloc_context *data_ac = NULL;
1605 struct ocfs2_alloc_context *xattr_ac = NULL;
1606 int want_clusters = 0;
1607 int xattr_credits = 0;
1608 struct ocfs2_security_xattr_info si = {
1611 int did_quota = 0, did_quota_inode = 0;
1612 struct ocfs2_dir_lookup_result lookup = { NULL, };
1614 int did_block_signals = 0;
1616 mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1617 dentry, symname, dentry->d_name.len, dentry->d_name.name);
1619 dquot_initialize(dir);
1624 l = strlen(symname) + 1;
1626 credits = ocfs2_calc_symlink_credits(sb);
1628 /* lock the parent directory */
1629 status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1631 if (status != -ENOENT)
1636 dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1637 if (!ocfs2_read_links_count(dirfe)) {
1638 /* can't make a file in a deleted directory. */
1643 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1644 dentry->d_name.len);
1648 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1649 dentry->d_name.name,
1650 dentry->d_name.len, &lookup);
1656 status = ocfs2_reserve_new_inode(osb, &inode_ac);
1658 if (status != -ENOSPC)
1663 inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1670 /* get security xattr */
1671 status = ocfs2_init_security_get(inode, dir, &si);
1673 if (status == -EOPNOTSUPP)
1681 /* calculate meta data/clusters for setting security xattr */
1683 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1684 &xattr_credits, &xattr_ac);
1691 /* don't reserve bitmap space for fast symlinks. */
1692 if (l > ocfs2_fast_symlink_chars(sb))
1695 status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1697 if (status != -ENOSPC)
1702 handle = ocfs2_start_trans(osb, credits + xattr_credits);
1703 if (IS_ERR(handle)) {
1704 status = PTR_ERR(handle);
1710 /* Starting to change things, restart is no longer possible. */
1711 ocfs2_block_signals(&oldset);
1712 did_block_signals = 1;
1714 status = dquot_alloc_inode(inode);
1717 did_quota_inode = 1;
1719 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry,
1720 inode->i_mode, dentry->d_name.len,
1721 dentry->d_name.name);
1723 status = ocfs2_mknod_locked(osb, dir, inode,
1724 0, &new_fe_bh, parent_fe_bh, handle,
1731 fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1734 if (l > ocfs2_fast_symlink_chars(sb)) {
1737 inode->i_op = &ocfs2_symlink_inode_operations;
1738 status = dquot_alloc_space_nodirty(inode,
1739 ocfs2_clusters_to_bytes(osb->sb, 1));
1743 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1745 handle, data_ac, NULL,
1748 if (status != -ENOSPC && status != -EINTR) {
1750 "Failed to extend file to %llu\n",
1751 (unsigned long long)newsize);
1757 i_size_write(inode, newsize);
1758 inode->i_blocks = ocfs2_inode_sector_count(inode);
1760 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1761 memcpy((char *) fe->id2.i_symlink, symname, l);
1762 i_size_write(inode, newsize);
1763 inode->i_blocks = 0;
1766 status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1772 if (!ocfs2_inode_is_fast_symlink(inode)) {
1773 status = ocfs2_create_symlink_data(osb, handle, inode,
1782 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1791 * Do this before adding the entry to the directory. We add
1792 * also set d_op after success so that ->d_iput() will cleanup
1793 * the dentry lock even if ocfs2_add_entry() fails below.
1795 status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1800 d_set_d_op(dentry, &ocfs2_dentry_ops);
1802 status = ocfs2_add_entry(handle, dentry, inode,
1803 le64_to_cpu(fe->i_blkno), parent_fe_bh,
1810 insert_inode_hash(inode);
1811 d_instantiate(dentry, inode);
1813 if (status < 0 && did_quota)
1814 dquot_free_space_nodirty(inode,
1815 ocfs2_clusters_to_bytes(osb->sb, 1));
1816 if (status < 0 && did_quota_inode)
1817 dquot_free_inode(inode);
1819 ocfs2_commit_trans(osb, handle);
1821 ocfs2_inode_unlock(dir, 1);
1822 if (did_block_signals)
1823 ocfs2_unblock_signals(&oldset);
1826 brelse(parent_fe_bh);
1829 ocfs2_free_dir_lookup_result(&lookup);
1831 ocfs2_free_alloc_context(inode_ac);
1833 ocfs2_free_alloc_context(data_ac);
1835 ocfs2_free_alloc_context(xattr_ac);
1836 if ((status < 0) && inode) {
1837 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
1847 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1849 int status, namelen;
1853 namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1863 if (namelen != OCFS2_ORPHAN_NAMELEN) {
1869 mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
1878 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
1879 struct inode **ret_orphan_dir,
1880 struct buffer_head **ret_orphan_dir_bh)
1882 struct inode *orphan_dir_inode;
1883 struct buffer_head *orphan_dir_bh = NULL;
1886 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1887 ORPHAN_DIR_SYSTEM_INODE,
1889 if (!orphan_dir_inode) {
1895 mutex_lock(&orphan_dir_inode->i_mutex);
1897 ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1899 mutex_unlock(&orphan_dir_inode->i_mutex);
1900 iput(orphan_dir_inode);
1906 *ret_orphan_dir = orphan_dir_inode;
1907 *ret_orphan_dir_bh = orphan_dir_bh;
1912 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
1913 struct buffer_head *orphan_dir_bh,
1916 struct ocfs2_dir_lookup_result *lookup)
1919 struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
1921 ret = ocfs2_blkno_stringify(blkno, name);
1927 ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1928 orphan_dir_bh, name,
1929 OCFS2_ORPHAN_NAMELEN, lookup);
1939 * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
1940 * insertion of an orphan.
1941 * @osb: ocfs2 file system
1942 * @ret_orphan_dir: Orphan dir inode - returned locked!
1943 * @blkno: Actual block number of the inode to be inserted into orphan dir.
1944 * @lookup: dir lookup result, to be passed back into functions like
1947 * Returns zero on success and the ret_orphan_dir, name and lookup
1948 * fields will be populated.
1950 * Returns non-zero on failure.
1952 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1953 struct inode **ret_orphan_dir,
1956 struct ocfs2_dir_lookup_result *lookup)
1958 struct inode *orphan_dir_inode = NULL;
1959 struct buffer_head *orphan_dir_bh = NULL;
1962 ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
1969 ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
1970 blkno, name, lookup);
1976 *ret_orphan_dir = orphan_dir_inode;
1979 brelse(orphan_dir_bh);
1982 ocfs2_inode_unlock(orphan_dir_inode, 1);
1983 mutex_unlock(&orphan_dir_inode->i_mutex);
1984 iput(orphan_dir_inode);
1991 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1993 struct inode *inode,
1994 struct buffer_head *fe_bh,
1996 struct ocfs2_dir_lookup_result *lookup,
1997 struct inode *orphan_dir_inode)
1999 struct buffer_head *orphan_dir_bh = NULL;
2001 struct ocfs2_dinode *orphan_fe;
2002 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
2004 mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
2006 status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2012 status = ocfs2_journal_access_di(handle,
2013 INODE_CACHE(orphan_dir_inode),
2015 OCFS2_JOURNAL_ACCESS_WRITE);
2021 /* we're a cluster, and nlink can change on disk from
2022 * underneath us... */
2023 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2024 if (S_ISDIR(inode->i_mode))
2025 ocfs2_add_links_count(orphan_fe, 1);
2026 orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
2027 ocfs2_journal_dirty(handle, orphan_dir_bh);
2029 status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2030 OCFS2_ORPHAN_NAMELEN, inode,
2031 OCFS2_I(inode)->ip_blkno,
2032 orphan_dir_bh, lookup);
2039 * We're going to journal the change of i_flags and i_orphaned_slot.
2040 * It's safe anyway, though some callers may duplicate the journaling.
2041 * Journaling within the func just make the logic look more
2044 status = ocfs2_journal_access_di(handle,
2047 OCFS2_JOURNAL_ACCESS_WRITE);
2053 le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
2054 OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2056 /* Record which orphan dir our inode now resides
2057 * in. delete_inode will use this to determine which orphan
2059 fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2061 ocfs2_journal_dirty(handle, fe_bh);
2063 mlog(0, "Inode %llu orphaned in slot %d\n",
2064 (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
2067 brelse(orphan_dir_bh);
2073 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2074 int ocfs2_orphan_del(struct ocfs2_super *osb,
2076 struct inode *orphan_dir_inode,
2077 struct inode *inode,
2078 struct buffer_head *orphan_dir_bh)
2080 char name[OCFS2_ORPHAN_NAMELEN + 1];
2081 struct ocfs2_dinode *orphan_fe;
2083 struct ocfs2_dir_lookup_result lookup = { NULL, };
2087 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2093 mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
2094 name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2095 OCFS2_ORPHAN_NAMELEN);
2097 /* find it's spot in the orphan directory */
2098 status = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN, orphan_dir_inode,
2105 /* remove it from the orphan directory */
2106 status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2112 status = ocfs2_journal_access_di(handle,
2113 INODE_CACHE(orphan_dir_inode),
2115 OCFS2_JOURNAL_ACCESS_WRITE);
2121 /* do the i_nlink dance! :) */
2122 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2123 if (S_ISDIR(inode->i_mode))
2124 ocfs2_add_links_count(orphan_fe, -1);
2125 orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
2126 ocfs2_journal_dirty(handle, orphan_dir_bh);
2129 ocfs2_free_dir_lookup_result(&lookup);
2136 * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to recieve a newly
2137 * allocated file. This is different from the typical 'add to orphan dir'
2138 * operation in that the inode does not yet exist. This is a problem because
2139 * the orphan dir stringifies the inode block number to come up with it's
2140 * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2141 * problem. This function works around it by calling deeper into the orphan
2142 * and suballoc code than other callers. Use this only by necessity.
2143 * @dir: The directory which this inode will ultimately wind up under - not the
2145 * @dir_bh: buffer_head the @dir inode block
2146 * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2147 * with the string to be used for orphan dirent. Pass back to the orphan dir
2149 * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2151 * @ret_di_blkno: block number where the new inode will be allocated.
2152 * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2153 * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2155 * Returns zero on success and the ret_orphan_dir, name and lookup
2156 * fields will be populated.
2158 * Returns non-zero on failure.
2160 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2161 struct buffer_head *dir_bh,
2163 struct inode **ret_orphan_dir,
2165 struct ocfs2_dir_lookup_result *orphan_insert,
2166 struct ocfs2_alloc_context **ret_inode_ac)
2170 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2171 struct inode *orphan_dir = NULL;
2172 struct buffer_head *orphan_dir_bh = NULL;
2173 struct ocfs2_alloc_context *inode_ac = NULL;
2175 ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2181 /* reserve an inode spot */
2182 ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2189 ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2196 ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2197 di_blkno, orphan_name, orphan_insert);
2205 *ret_orphan_dir = orphan_dir;
2206 *ret_di_blkno = di_blkno;
2207 *ret_inode_ac = inode_ac;
2209 * orphan_name and orphan_insert are already up to
2210 * date via prepare_orphan_dir
2213 /* Unroll reserve_new_inode* */
2215 ocfs2_free_alloc_context(inode_ac);
2217 /* Unroll orphan dir locking */
2218 mutex_unlock(&orphan_dir->i_mutex);
2219 ocfs2_inode_unlock(orphan_dir, 1);
2223 brelse(orphan_dir_bh);
2228 int ocfs2_create_inode_in_orphan(struct inode *dir,
2230 struct inode **new_inode)
2232 int status, did_quota_inode = 0;
2233 struct inode *inode = NULL;
2234 struct inode *orphan_dir = NULL;
2235 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2236 struct ocfs2_dinode *di = NULL;
2237 handle_t *handle = NULL;
2238 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2239 struct buffer_head *parent_di_bh = NULL;
2240 struct buffer_head *new_di_bh = NULL;
2241 struct ocfs2_alloc_context *inode_ac = NULL;
2242 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2243 u64 uninitialized_var(di_blkno), suballoc_loc;
2246 status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2248 if (status != -ENOENT)
2253 status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2254 orphan_name, &orphan_dir,
2255 &di_blkno, &orphan_insert, &inode_ac);
2257 if (status != -ENOSPC)
2262 inode = ocfs2_get_init_inode(dir, mode);
2269 handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2270 if (IS_ERR(handle)) {
2271 status = PTR_ERR(handle);
2277 status = dquot_alloc_inode(inode);
2280 did_quota_inode = 1;
2282 status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2284 &suballoc_bit, di_blkno);
2291 /* do the real work now. */
2292 status = __ocfs2_mknod_locked(dir, inode,
2293 0, &new_di_bh, parent_di_bh, handle,
2294 inode_ac, di_blkno, suballoc_loc,
2301 di = (struct ocfs2_dinode *)new_di_bh->b_data;
2302 status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2303 &orphan_insert, orphan_dir);
2309 /* get open lock so that only nodes can't remove it from orphan dir. */
2310 status = ocfs2_open_lock(inode);
2314 insert_inode_hash(inode);
2316 if (status < 0 && did_quota_inode)
2317 dquot_free_inode(inode);
2319 ocfs2_commit_trans(osb, handle);
2322 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2323 ocfs2_inode_unlock(orphan_dir, 1);
2324 mutex_unlock(&orphan_dir->i_mutex);
2328 if (status == -ENOSPC)
2329 mlog(0, "Disk is full\n");
2331 if ((status < 0) && inode) {
2337 ocfs2_free_alloc_context(inode_ac);
2344 ocfs2_free_dir_lookup_result(&orphan_insert);
2346 ocfs2_inode_unlock(dir, 1);
2347 brelse(parent_di_bh);
2351 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2352 struct inode *inode,
2353 struct dentry *dentry)
2356 struct buffer_head *parent_di_bh = NULL;
2357 handle_t *handle = NULL;
2358 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2359 struct ocfs2_dinode *dir_di, *di;
2360 struct inode *orphan_dir_inode = NULL;
2361 struct buffer_head *orphan_dir_bh = NULL;
2362 struct buffer_head *di_bh = NULL;
2363 struct ocfs2_dir_lookup_result lookup = { NULL, };
2365 mlog_entry("(0x%p, 0x%p, %.*s')\n", dir, dentry,
2366 dentry->d_name.len, dentry->d_name.name);
2368 status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2370 if (status != -ENOENT)
2375 dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2376 if (!dir_di->i_links_count) {
2377 /* can't make a file in a deleted directory. */
2382 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2383 dentry->d_name.len);
2387 /* get a spot inside the dir. */
2388 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2389 dentry->d_name.name,
2390 dentry->d_name.len, &lookup);
2396 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2397 ORPHAN_DIR_SYSTEM_INODE,
2399 if (!orphan_dir_inode) {
2405 mutex_lock(&orphan_dir_inode->i_mutex);
2407 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2410 mutex_unlock(&orphan_dir_inode->i_mutex);
2411 iput(orphan_dir_inode);
2415 status = ocfs2_read_inode_block(inode, &di_bh);
2421 handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2422 if (IS_ERR(handle)) {
2423 status = PTR_ERR(handle);
2429 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2430 di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2436 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2443 di = (struct ocfs2_dinode *)di_bh->b_data;
2444 le32_add_cpu(&di->i_flags, -OCFS2_ORPHANED_FL);
2445 di->i_orphaned_slot = 0;
2447 ocfs2_set_links_count(di, inode->i_nlink);
2448 ocfs2_journal_dirty(handle, di_bh);
2450 status = ocfs2_add_entry(handle, dentry, inode,
2451 OCFS2_I(inode)->ip_blkno, parent_di_bh,
2458 status = ocfs2_dentry_attach_lock(dentry, inode,
2459 OCFS2_I(dir)->ip_blkno);
2465 d_set_d_op(dentry, &ocfs2_dentry_ops);
2466 d_instantiate(dentry, inode);
2469 ocfs2_commit_trans(osb, handle);
2471 ocfs2_inode_unlock(orphan_dir_inode, 1);
2472 mutex_unlock(&orphan_dir_inode->i_mutex);
2473 iput(orphan_dir_inode);
2476 ocfs2_inode_unlock(dir, 1);
2479 brelse(parent_di_bh);
2480 brelse(orphan_dir_bh);
2482 ocfs2_free_dir_lookup_result(&lookup);
2489 const struct inode_operations ocfs2_dir_iops = {
2490 .create = ocfs2_create,
2491 .lookup = ocfs2_lookup,
2493 .unlink = ocfs2_unlink,
2494 .rmdir = ocfs2_unlink,
2495 .symlink = ocfs2_symlink,
2496 .mkdir = ocfs2_mkdir,
2497 .mknod = ocfs2_mknod,
2498 .rename = ocfs2_rename,
2499 .setattr = ocfs2_setattr,
2500 .getattr = ocfs2_getattr,
2501 .permission = ocfs2_permission,
2502 .setxattr = generic_setxattr,
2503 .getxattr = generic_getxattr,
2504 .listxattr = ocfs2_listxattr,
2505 .removexattr = generic_removexattr,
2506 .fiemap = ocfs2_fiemap,