1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* -*- mode: c; c-basic-offset: 8; -*-
3 * vim: noexpandtab sw=8 ts=8 sts=0:
7 * Create and rename file, directory, symlinks
9 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
11 * Portions of this code from linux/fs/ext3/dir.c
13 * Copyright (C) 1992, 1993, 1994, 1995
14 * Remy Card (card@masi.ibp.fr)
15 * Laboratoire MASI - Institut Blaise pascal
16 * Universite Pierre et Marie Curie (Paris VI)
20 * linux/fs/minix/dir.c
22 * Copyright (C) 1991, 1992 Linux Torvalds
26 #include <linux/types.h>
27 #include <linux/slab.h>
28 #include <linux/highmem.h>
29 #include <linux/quotaops.h>
30 #include <linux/iversion.h>
32 #include <cluster/masklog.h>
40 #include "extent_map.h"
52 #include "ocfs2_trace.h"
54 #include "buffer_head_io.h"
56 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
60 struct buffer_head **new_fe_bh,
61 struct buffer_head *parent_fe_bh,
63 struct ocfs2_alloc_context *inode_ac);
65 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
66 struct inode **ret_orphan_dir,
69 struct ocfs2_dir_lookup_result *lookup,
72 static int ocfs2_orphan_add(struct ocfs2_super *osb,
75 struct buffer_head *fe_bh,
77 struct ocfs2_dir_lookup_result *lookup,
78 struct inode *orphan_dir_inode,
81 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
86 static int ocfs2_double_lock(struct ocfs2_super *osb,
87 struct buffer_head **bh1,
89 struct buffer_head **bh2,
93 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2);
94 /* An orphan dir name is an 8 byte value, printed as a hex string */
95 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
97 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
102 struct inode *inode = NULL;
104 struct ocfs2_inode_info *oi;
106 trace_ocfs2_lookup(dir, dentry, dentry->d_name.len,
108 (unsigned long long)OCFS2_I(dir)->ip_blkno, 0);
110 if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
111 ret = ERR_PTR(-ENAMETOOLONG);
115 status = ocfs2_inode_lock_nested(dir, NULL, 0, OI_LS_PARENT);
117 if (status != -ENOENT)
119 ret = ERR_PTR(status);
123 status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
124 dentry->d_name.len, &blkno);
128 inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
130 ret = ERR_PTR(-EACCES);
135 /* Clear any orphaned state... If we were able to look up the
136 * inode from a directory, it certainly can't be orphaned. We
137 * might have the bad state from a node which intended to
138 * orphan this inode but crashed before it could commit the
140 spin_lock(&oi->ip_lock);
141 oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
142 spin_unlock(&oi->ip_lock);
145 ret = d_splice_alias(inode, dentry);
149 * If d_splice_alias() finds a DCACHE_DISCONNECTED
150 * dentry, it will d_move() it on top of ourse. The
151 * return value will indicate this however, so in
152 * those cases, we switch them around for the locking
155 * NOTE: This dentry already has ->d_op set from
156 * ocfs2_get_parent() and ocfs2_get_dentry()
158 if (!IS_ERR_OR_NULL(ret))
161 status = ocfs2_dentry_attach_lock(dentry, inode,
162 OCFS2_I(dir)->ip_blkno);
165 ret = ERR_PTR(status);
169 ocfs2_dentry_attach_gen(dentry);
172 /* Don't drop the cluster lock until *after* the d_add --
173 * unlink on another node will message us to remove that
174 * dentry under this lock so otherwise we can race this with
175 * the downconvert thread and have a stale dentry. */
176 ocfs2_inode_unlock(dir, 0);
180 trace_ocfs2_lookup_ret(ret);
185 static struct inode *ocfs2_get_init_inode(struct inode *dir, umode_t mode)
190 inode = new_inode(dir->i_sb);
192 mlog(ML_ERROR, "new_inode failed!\n");
193 return ERR_PTR(-ENOMEM);
196 /* populate as many fields early on as possible - many of
197 * these are used by the support functions here and in
201 inode_init_owner(inode, dir, mode);
202 status = dquot_initialize(inode);
204 return ERR_PTR(status);
209 static void ocfs2_cleanup_add_entry_failure(struct ocfs2_super *osb,
210 struct dentry *dentry, struct inode *inode)
212 struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
214 ocfs2_simple_drop_lockres(osb, &dl->dl_lockres);
215 ocfs2_lock_res_free(&dl->dl_lockres);
216 BUG_ON(dl->dl_count != 1);
217 spin_lock(&dentry_attach_lock);
218 dentry->d_fsdata = NULL;
219 spin_unlock(&dentry_attach_lock);
224 static int ocfs2_mknod(struct inode *dir,
225 struct dentry *dentry,
230 struct buffer_head *parent_fe_bh = NULL;
231 handle_t *handle = NULL;
232 struct ocfs2_super *osb;
233 struct ocfs2_dinode *dirfe;
234 struct buffer_head *new_fe_bh = NULL;
235 struct inode *inode = NULL;
236 struct ocfs2_alloc_context *inode_ac = NULL;
237 struct ocfs2_alloc_context *data_ac = NULL;
238 struct ocfs2_alloc_context *meta_ac = NULL;
239 int want_clusters = 0;
241 int xattr_credits = 0;
242 struct ocfs2_security_xattr_info si = {
245 int did_quota_inode = 0;
246 struct ocfs2_dir_lookup_result lookup = { NULL, };
248 int did_block_signals = 0;
249 struct ocfs2_dentry_lock *dl = NULL;
251 trace_ocfs2_mknod(dir, dentry, dentry->d_name.len, dentry->d_name.name,
252 (unsigned long long)OCFS2_I(dir)->ip_blkno,
253 (unsigned long)dev, mode);
255 status = dquot_initialize(dir);
261 /* get our super block */
262 osb = OCFS2_SB(dir->i_sb);
264 status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
266 if (status != -ENOENT)
271 if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
276 dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
277 if (!ocfs2_read_links_count(dirfe)) {
278 /* can't make a file in a deleted directory. */
283 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
288 /* get a spot inside the dir. */
289 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
291 dentry->d_name.len, &lookup);
297 /* reserve an inode spot */
298 status = ocfs2_reserve_new_inode(osb, &inode_ac);
300 if (status != -ENOSPC)
305 inode = ocfs2_get_init_inode(dir, mode);
307 status = PTR_ERR(inode);
313 /* get security xattr */
314 status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
316 if (status == -EOPNOTSUPP)
324 /* calculate meta data/clusters for setting security and acl xattr */
325 status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
327 &xattr_credits, &want_meta);
333 /* Reserve a cluster if creating an extent based directory. */
334 if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
337 /* Dir indexing requires extra space as well */
338 if (ocfs2_supports_indexed_dirs(osb))
342 status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
344 if (status != -ENOSPC)
349 status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
351 if (status != -ENOSPC)
356 handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
359 if (IS_ERR(handle)) {
360 status = PTR_ERR(handle);
366 /* Starting to change things, restart is no longer possible. */
367 ocfs2_block_signals(&oldset);
368 did_block_signals = 1;
370 status = dquot_alloc_inode(inode);
375 /* do the real work now. */
376 status = ocfs2_mknod_locked(osb, dir, inode, dev,
377 &new_fe_bh, parent_fe_bh, handle,
385 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
386 new_fe_bh, data_ac, meta_ac);
392 status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
394 OCFS2_JOURNAL_ACCESS_WRITE);
399 ocfs2_add_links_count(dirfe, 1);
400 ocfs2_journal_dirty(handle, parent_fe_bh);
404 status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
413 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
422 * Do this before adding the entry to the directory. We add
423 * also set d_op after success so that ->d_iput() will cleanup
424 * the dentry lock even if ocfs2_add_entry() fails below.
426 status = ocfs2_dentry_attach_lock(dentry, inode,
427 OCFS2_I(dir)->ip_blkno);
433 dl = dentry->d_fsdata;
435 status = ocfs2_add_entry(handle, dentry, inode,
436 OCFS2_I(inode)->ip_blkno, parent_fe_bh,
443 insert_inode_hash(inode);
444 d_instantiate(dentry, inode);
447 if (status < 0 && did_quota_inode)
448 dquot_free_inode(inode);
450 ocfs2_commit_trans(osb, handle);
452 ocfs2_inode_unlock(dir, 1);
453 if (did_block_signals)
454 ocfs2_unblock_signals(&oldset);
457 brelse(parent_fe_bh);
460 ocfs2_free_dir_lookup_result(&lookup);
463 ocfs2_free_alloc_context(inode_ac);
466 ocfs2_free_alloc_context(data_ac);
469 ocfs2_free_alloc_context(meta_ac);
472 * We should call iput after the i_mutex of the bitmap been
473 * unlocked in ocfs2_free_alloc_context, or the
474 * ocfs2_delete_inode will mutex_lock again.
476 if ((status < 0) && inode) {
478 ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
480 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
491 static int __ocfs2_mknod_locked(struct inode *dir,
494 struct buffer_head **new_fe_bh,
495 struct buffer_head *parent_fe_bh,
497 struct ocfs2_alloc_context *inode_ac,
498 u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
501 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
502 struct ocfs2_dinode *fe = NULL;
503 struct ocfs2_extent_list *fel;
505 struct ocfs2_inode_info *oi = OCFS2_I(inode);
506 struct timespec64 ts;
510 /* populate as many fields early on as possible - many of
511 * these are used by the support functions here and in
513 inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
514 oi->ip_blkno = fe_blkno;
515 spin_lock(&osb->osb_lock);
516 inode->i_generation = osb->s_next_generation++;
517 spin_unlock(&osb->osb_lock);
519 *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
525 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
527 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
529 OCFS2_JOURNAL_ACCESS_CREATE);
535 fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
536 memset(fe, 0, osb->sb->s_blocksize);
538 fe->i_generation = cpu_to_le32(inode->i_generation);
539 fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
540 fe->i_blkno = cpu_to_le64(fe_blkno);
541 fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
542 fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
543 fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
544 fe->i_uid = cpu_to_le32(i_uid_read(inode));
545 fe->i_gid = cpu_to_le32(i_gid_read(inode));
546 fe->i_mode = cpu_to_le16(inode->i_mode);
547 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
548 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
550 ocfs2_set_links_count(fe, inode->i_nlink);
552 fe->i_last_eb_blk = 0;
553 strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
554 fe->i_flags |= cpu_to_le32(OCFS2_VALID_FL);
555 ktime_get_real_ts64(&ts);
556 fe->i_atime = fe->i_ctime = fe->i_mtime =
557 cpu_to_le64(ts.tv_sec);
558 fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
559 cpu_to_le32(ts.tv_nsec);
563 * If supported, directories start with inline data. If inline
564 * isn't supported, but indexing is, we start them as indexed.
566 feat = le16_to_cpu(fe->i_dyn_features);
567 if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
568 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
570 fe->id2.i_data.id_count = cpu_to_le16(
571 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
573 fel = &fe->id2.i_list;
574 fel->l_tree_depth = 0;
575 fel->l_next_free_rec = 0;
576 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
579 ocfs2_journal_dirty(handle, *new_fe_bh);
581 ocfs2_populate_inode(inode, fe, 1);
582 ocfs2_ci_set_new(osb, INODE_CACHE(inode));
583 if (!ocfs2_mount_local(osb)) {
584 status = ocfs2_create_new_inode_locks(inode);
589 ocfs2_update_inode_fsync_trans(handle, inode, 1);
604 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
608 struct buffer_head **new_fe_bh,
609 struct buffer_head *parent_fe_bh,
611 struct ocfs2_alloc_context *inode_ac)
614 u64 suballoc_loc, fe_blkno = 0;
619 status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
620 inode_ac, &suballoc_loc,
621 &suballoc_bit, &fe_blkno);
627 status = __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
628 parent_fe_bh, handle, inode_ac,
629 fe_blkno, suballoc_loc, suballoc_bit);
631 u64 bg_blkno = ocfs2_which_suballoc_group(fe_blkno, suballoc_bit);
632 int tmp = ocfs2_free_suballoc_bits(handle, inode_ac->ac_inode,
633 inode_ac->ac_bh, suballoc_bit, bg_blkno, 1);
641 static int ocfs2_mkdir(struct inode *dir,
642 struct dentry *dentry,
647 trace_ocfs2_mkdir(dir, dentry, dentry->d_name.len, dentry->d_name.name,
648 OCFS2_I(dir)->ip_blkno, mode);
649 ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
656 static int ocfs2_create(struct inode *dir,
657 struct dentry *dentry,
663 trace_ocfs2_create(dir, dentry, dentry->d_name.len, dentry->d_name.name,
664 (unsigned long long)OCFS2_I(dir)->ip_blkno, mode);
665 ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
672 static int ocfs2_link(struct dentry *old_dentry,
674 struct dentry *dentry)
677 struct inode *inode = d_inode(old_dentry);
678 struct inode *old_dir = d_inode(old_dentry->d_parent);
680 struct buffer_head *fe_bh = NULL;
681 struct buffer_head *old_dir_bh = NULL;
682 struct buffer_head *parent_fe_bh = NULL;
683 struct ocfs2_dinode *fe = NULL;
684 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
685 struct ocfs2_dir_lookup_result lookup = { NULL, };
689 trace_ocfs2_link((unsigned long long)OCFS2_I(inode)->ip_blkno,
690 old_dentry->d_name.len, old_dentry->d_name.name,
691 dentry->d_name.len, dentry->d_name.name);
693 if (S_ISDIR(inode->i_mode))
696 err = dquot_initialize(dir);
702 err = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
703 &parent_fe_bh, dir, 0);
710 /* make sure both dirs have bhs
711 * get an extra ref on old_dir_bh if old==new */
714 parent_fe_bh = old_dir_bh;
715 get_bh(parent_fe_bh);
717 mlog(ML_ERROR, "%s: no old_dir_bh!\n", osb->uuid_str);
728 err = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
729 old_dentry->d_name.len, &old_de_ino);
736 * Check whether another node removed the source inode while we
739 if (old_de_ino != OCFS2_I(inode)->ip_blkno) {
744 err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
749 err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
751 dentry->d_name.len, &lookup);
757 err = ocfs2_inode_lock(inode, &fe_bh, 1);
764 fe = (struct ocfs2_dinode *) fe_bh->b_data;
765 if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
767 goto out_unlock_inode;
770 handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
771 if (IS_ERR(handle)) {
772 err = PTR_ERR(handle);
775 goto out_unlock_inode;
778 /* Starting to change things, restart is no longer possible. */
779 ocfs2_block_signals(&oldset);
781 err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
782 OCFS2_JOURNAL_ACCESS_WRITE);
789 inode->i_ctime = current_time(inode);
790 ocfs2_set_links_count(fe, inode->i_nlink);
791 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
792 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
793 ocfs2_journal_dirty(handle, fe_bh);
795 err = ocfs2_add_entry(handle, dentry, inode,
796 OCFS2_I(inode)->ip_blkno,
797 parent_fe_bh, &lookup);
799 ocfs2_add_links_count(fe, -1);
805 err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
812 d_instantiate(dentry, inode);
815 ocfs2_commit_trans(osb, handle);
816 ocfs2_unblock_signals(&oldset);
818 ocfs2_inode_unlock(inode, 1);
821 ocfs2_double_unlock(old_dir, dir);
824 brelse(parent_fe_bh);
827 ocfs2_free_dir_lookup_result(&lookup);
836 * Takes and drops an exclusive lock on the given dentry. This will
837 * force other nodes to drop it.
839 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
843 ret = ocfs2_dentry_lock(dentry, 1);
847 ocfs2_dentry_unlock(dentry, 1);
852 static inline int ocfs2_inode_is_unlinkable(struct inode *inode)
854 if (S_ISDIR(inode->i_mode)) {
855 if (inode->i_nlink == 2)
860 if (inode->i_nlink == 1)
865 static int ocfs2_unlink(struct inode *dir,
866 struct dentry *dentry)
869 int child_locked = 0;
870 bool is_unlinkable = false;
871 struct inode *inode = d_inode(dentry);
872 struct inode *orphan_dir = NULL;
873 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
875 struct ocfs2_dinode *fe = NULL;
876 struct buffer_head *fe_bh = NULL;
877 struct buffer_head *parent_node_bh = NULL;
878 handle_t *handle = NULL;
879 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
880 struct ocfs2_dir_lookup_result lookup = { NULL, };
881 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
883 trace_ocfs2_unlink(dir, dentry, dentry->d_name.len,
885 (unsigned long long)OCFS2_I(dir)->ip_blkno,
886 (unsigned long long)OCFS2_I(inode)->ip_blkno);
888 status = dquot_initialize(dir);
894 BUG_ON(d_inode(dentry->d_parent) != dir);
896 if (inode == osb->root_inode)
899 status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
902 if (status != -ENOENT)
907 status = ocfs2_find_files_on_disk(dentry->d_name.name,
908 dentry->d_name.len, &blkno, dir,
911 if (status != -ENOENT)
916 if (OCFS2_I(inode)->ip_blkno != blkno) {
919 trace_ocfs2_unlink_noent(
920 (unsigned long long)OCFS2_I(inode)->ip_blkno,
921 (unsigned long long)blkno,
922 OCFS2_I(inode)->ip_flags);
926 status = ocfs2_inode_lock(inode, &fe_bh, 1);
928 if (status != -ENOENT)
934 if (S_ISDIR(inode->i_mode)) {
935 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
941 status = ocfs2_remote_dentry_delete(dentry);
943 /* This remote delete should succeed under all normal
949 if (ocfs2_inode_is_unlinkable(inode)) {
950 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
951 OCFS2_I(inode)->ip_blkno,
952 orphan_name, &orphan_insert,
958 is_unlinkable = true;
961 handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
962 if (IS_ERR(handle)) {
963 status = PTR_ERR(handle);
969 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
970 OCFS2_JOURNAL_ACCESS_WRITE);
976 fe = (struct ocfs2_dinode *) fe_bh->b_data;
978 /* delete the name from the parent dir */
979 status = ocfs2_delete_entry(handle, dir, &lookup);
985 if (S_ISDIR(inode->i_mode))
988 ocfs2_set_links_count(fe, inode->i_nlink);
989 ocfs2_journal_dirty(handle, fe_bh);
991 dir->i_ctime = dir->i_mtime = current_time(dir);
992 if (S_ISDIR(inode->i_mode))
995 status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
998 if (S_ISDIR(inode->i_mode))
1003 if (is_unlinkable) {
1004 status = ocfs2_orphan_add(osb, handle, inode, fe_bh,
1005 orphan_name, &orphan_insert, orphan_dir, false);
1012 ocfs2_commit_trans(osb, handle);
1015 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1016 ocfs2_inode_unlock(orphan_dir, 1);
1017 inode_unlock(orphan_dir);
1022 ocfs2_inode_unlock(inode, 1);
1024 ocfs2_inode_unlock(dir, 1);
1027 brelse(parent_node_bh);
1029 ocfs2_free_dir_lookup_result(&orphan_insert);
1030 ocfs2_free_dir_lookup_result(&lookup);
1032 if (status && (status != -ENOTEMPTY) && (status != -ENOENT))
1038 static int ocfs2_check_if_ancestor(struct ocfs2_super *osb,
1039 u64 src_inode_no, u64 dest_inode_no)
1042 u64 parent_inode_no = 0;
1043 u64 child_inode_no = src_inode_no;
1044 struct inode *child_inode;
1046 #define MAX_LOOKUP_TIMES 32
1048 child_inode = ocfs2_iget(osb, child_inode_no, 0, 0);
1049 if (IS_ERR(child_inode)) {
1050 ret = PTR_ERR(child_inode);
1054 ret = ocfs2_inode_lock(child_inode, NULL, 0);
1062 ret = ocfs2_lookup_ino_from_name(child_inode, "..", 2,
1064 ocfs2_inode_unlock(child_inode, 0);
1071 if (parent_inode_no == dest_inode_no) {
1076 if (parent_inode_no == osb->root_inode->i_ino) {
1081 child_inode_no = parent_inode_no;
1083 if (++i >= MAX_LOOKUP_TIMES) {
1084 mlog(ML_NOTICE, "max lookup times reached, filesystem "
1085 "may have nested directories, "
1086 "src inode: %llu, dest inode: %llu.\n",
1087 (unsigned long long)src_inode_no,
1088 (unsigned long long)dest_inode_no);
1098 * The only place this should be used is rename and link!
1099 * if they have the same id, then the 1st one is the only one locked.
1101 static int ocfs2_double_lock(struct ocfs2_super *osb,
1102 struct buffer_head **bh1,
1103 struct inode *inode1,
1104 struct buffer_head **bh2,
1105 struct inode *inode2,
1109 int inode1_is_ancestor, inode2_is_ancestor;
1110 struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
1111 struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
1113 trace_ocfs2_double_lock((unsigned long long)oi1->ip_blkno,
1114 (unsigned long long)oi2->ip_blkno);
1121 /* we always want to lock the one with the lower lockid first.
1122 * and if they are nested, we lock ancestor first */
1123 if (oi1->ip_blkno != oi2->ip_blkno) {
1124 inode1_is_ancestor = ocfs2_check_if_ancestor(osb, oi2->ip_blkno,
1126 if (inode1_is_ancestor < 0) {
1127 status = inode1_is_ancestor;
1131 inode2_is_ancestor = ocfs2_check_if_ancestor(osb, oi1->ip_blkno,
1133 if (inode2_is_ancestor < 0) {
1134 status = inode2_is_ancestor;
1138 if ((inode1_is_ancestor == 1) ||
1139 (oi1->ip_blkno < oi2->ip_blkno &&
1140 inode2_is_ancestor == 0)) {
1141 /* switch id1 and id2 around */
1143 swap(inode2, inode1);
1146 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
1147 rename == 1 ? OI_LS_RENAME1 : OI_LS_PARENT);
1149 if (status != -ENOENT)
1156 status = ocfs2_inode_lock_nested(inode1, bh1, 1,
1157 rename == 1 ? OI_LS_RENAME2 : OI_LS_PARENT);
1160 * An error return must mean that no cluster locks
1161 * were held on function exit.
1163 if (oi1->ip_blkno != oi2->ip_blkno) {
1164 ocfs2_inode_unlock(inode2, 1);
1169 if (status != -ENOENT)
1173 trace_ocfs2_double_lock_end(
1174 (unsigned long long)oi1->ip_blkno,
1175 (unsigned long long)oi2->ip_blkno);
1183 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1185 ocfs2_inode_unlock(inode1, 1);
1187 if (inode1 != inode2)
1188 ocfs2_inode_unlock(inode2, 1);
1191 static int ocfs2_rename(struct inode *old_dir,
1192 struct dentry *old_dentry,
1193 struct inode *new_dir,
1194 struct dentry *new_dentry,
1197 int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1198 int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1199 struct inode *old_inode = d_inode(old_dentry);
1200 struct inode *new_inode = d_inode(new_dentry);
1201 struct inode *orphan_dir = NULL;
1202 struct ocfs2_dinode *newfe = NULL;
1203 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1204 struct buffer_head *newfe_bh = NULL;
1205 struct buffer_head *old_inode_bh = NULL;
1206 struct ocfs2_super *osb = NULL;
1207 u64 newfe_blkno, old_de_ino;
1208 handle_t *handle = NULL;
1209 struct buffer_head *old_dir_bh = NULL;
1210 struct buffer_head *new_dir_bh = NULL;
1211 u32 old_dir_nlink = old_dir->i_nlink;
1212 struct ocfs2_dinode *old_di;
1213 struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1214 struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1215 struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1216 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1217 struct ocfs2_dir_lookup_result target_insert = { NULL, };
1218 bool should_add_orphan = false;
1223 /* At some point it might be nice to break this function up a
1226 trace_ocfs2_rename(old_dir, old_dentry, new_dir, new_dentry,
1227 old_dentry->d_name.len, old_dentry->d_name.name,
1228 new_dentry->d_name.len, new_dentry->d_name.name);
1230 status = dquot_initialize(old_dir);
1235 status = dquot_initialize(new_dir);
1241 osb = OCFS2_SB(old_dir->i_sb);
1244 if (!igrab(new_inode))
1248 /* Assume a directory hierarchy thusly:
1251 * a,b,c, and d are all directories.
1253 * from cwd of 'a' on both nodes:
1257 * And that's why, just like the VFS, we need a file system
1259 if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1260 status = ocfs2_rename_lock(osb);
1267 /* here we cannot guarantee the inodes haven't just been
1268 * changed, so check if they are nested again */
1269 status = ocfs2_check_if_ancestor(osb, new_dir->i_ino,
1274 } else if (status == 1) {
1276 trace_ocfs2_rename_not_permitted(
1277 (unsigned long long)old_inode->i_ino,
1278 (unsigned long long)new_dir->i_ino);
1283 /* if old and new are the same, this'll just do one lock. */
1284 status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1285 &new_dir_bh, new_dir, 1);
1292 if (!new_dir->i_nlink) {
1297 /* make sure both dirs have bhs
1298 * get an extra ref on old_dir_bh if old==new */
1301 new_dir_bh = old_dir_bh;
1304 mlog(ML_ERROR, "no old_dir_bh!\n");
1311 * Aside from allowing a meta data update, the locking here
1312 * also ensures that the downconvert thread on other nodes
1313 * won't have to concurrently downconvert the inode and the
1316 status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1319 if (status != -ENOENT)
1323 old_child_locked = 1;
1325 status = ocfs2_remote_dentry_delete(old_dentry);
1331 if (S_ISDIR(old_inode->i_mode)) {
1332 u64 old_inode_parent;
1335 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1337 &old_inode_dot_dot_res);
1343 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1348 if (!new_inode && new_dir != old_dir &&
1349 new_dir->i_nlink >= ocfs2_link_max(osb)) {
1355 status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1356 old_dentry->d_name.len,
1364 * Check for inode number is _not_ due to possible IO errors.
1365 * We might rmdir the source, keep it as pwd of some process
1366 * and merrily kill the link to whatever was created under the
1367 * same name. Goodbye sticky bit ;-<
1369 if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1374 /* check if the target already exists (in which case we need
1376 status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1377 new_dentry->d_name.len,
1378 &newfe_blkno, new_dir,
1379 &target_lookup_res);
1380 /* The only error we allow here is -ENOENT because the new
1381 * file not existing is perfectly valid. */
1382 if ((status < 0) && (status != -ENOENT)) {
1383 /* If we cannot find the file specified we should just */
1384 /* return the error... */
1391 if (!target_exists && new_inode) {
1393 * Target was unlinked by another node while we were
1394 * waiting to get to ocfs2_rename(). There isn't
1395 * anything we can do here to help the situation, so
1396 * bubble up the appropriate error.
1402 /* In case we need to overwrite an existing file, we blow it
1404 if (target_exists) {
1405 /* VFS didn't think there existed an inode here, but
1406 * someone else in the cluster must have raced our
1407 * rename to create one. Today we error cleanly, in
1408 * the future we should consider calling iget to build
1409 * a new struct inode for this entry. */
1413 trace_ocfs2_rename_target_exists(new_dentry->d_name.len,
1414 new_dentry->d_name.name);
1418 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1421 trace_ocfs2_rename_disagree(
1422 (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1423 (unsigned long long)newfe_blkno,
1424 OCFS2_I(new_inode)->ip_flags);
1428 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1430 if (status != -ENOENT)
1434 new_child_locked = 1;
1436 status = ocfs2_remote_dentry_delete(new_dentry);
1442 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1444 trace_ocfs2_rename_over_existing(
1445 (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1446 (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1448 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1449 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1450 OCFS2_I(new_inode)->ip_blkno,
1451 orphan_name, &orphan_insert,
1457 should_add_orphan = true;
1460 BUG_ON(d_inode(new_dentry->d_parent) != new_dir);
1462 status = ocfs2_check_dir_for_entry(new_dir,
1463 new_dentry->d_name.name,
1464 new_dentry->d_name.len);
1468 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1469 new_dentry->d_name.name,
1470 new_dentry->d_name.len,
1478 handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1479 if (IS_ERR(handle)) {
1480 status = PTR_ERR(handle);
1486 if (target_exists) {
1487 if (S_ISDIR(new_inode->i_mode)) {
1488 if (new_inode->i_nlink != 2 ||
1489 !ocfs2_empty_dir(new_inode)) {
1490 status = -ENOTEMPTY;
1494 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1496 OCFS2_JOURNAL_ACCESS_WRITE);
1502 /* change the dirent to point to the correct inode */
1503 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1509 inode_inc_iversion(new_dir);
1511 if (S_ISDIR(new_inode->i_mode))
1512 ocfs2_set_links_count(newfe, 0);
1514 ocfs2_add_links_count(newfe, -1);
1515 ocfs2_journal_dirty(handle, newfe_bh);
1516 if (should_add_orphan) {
1517 status = ocfs2_orphan_add(osb, handle, new_inode,
1518 newfe_bh, orphan_name,
1519 &orphan_insert, orphan_dir, false);
1526 /* if the name was not found in new_dir, add it now */
1527 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1528 OCFS2_I(old_inode)->ip_blkno,
1529 new_dir_bh, &target_insert);
1532 old_inode->i_ctime = current_time(old_inode);
1533 mark_inode_dirty(old_inode);
1535 status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1537 OCFS2_JOURNAL_ACCESS_WRITE);
1539 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1541 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1542 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1543 ocfs2_journal_dirty(handle, old_inode_bh);
1548 * Now that the name has been added to new_dir, remove the old name.
1550 * We don't keep any directory entry context around until now
1551 * because the insert might have changed the type of directory
1552 * we're dealing with.
1554 status = ocfs2_find_entry(old_dentry->d_name.name,
1555 old_dentry->d_name.len, old_dir,
1558 if (!is_journal_aborted(osb->journal->j_journal)) {
1559 ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1561 new_dentry->d_name.len, new_dentry->d_name.name,
1562 old_dentry->d_name.len, old_dentry->d_name.name);
1567 status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1570 if (!is_journal_aborted(osb->journal->j_journal)) {
1571 ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1573 new_dentry->d_name.len, new_dentry->d_name.name,
1574 old_dentry->d_name.len, old_dentry->d_name.name);
1580 drop_nlink(new_inode);
1581 new_inode->i_ctime = current_time(new_inode);
1583 old_dir->i_ctime = old_dir->i_mtime = current_time(old_dir);
1585 if (update_dot_dot) {
1586 status = ocfs2_update_entry(old_inode, handle,
1587 &old_inode_dot_dot_res, new_dir);
1588 drop_nlink(old_dir);
1590 drop_nlink(new_inode);
1593 mark_inode_dirty(new_dir);
1596 mark_inode_dirty(old_dir);
1597 ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1599 mark_inode_dirty(new_inode);
1600 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1603 if (old_dir != new_dir) {
1604 /* Keep the same times on both directories.*/
1605 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1608 * This will also pick up the i_nlink change from the
1611 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1614 if (old_dir_nlink != old_dir->i_nlink) {
1616 mlog(ML_ERROR, "need to change nlink for old dir "
1617 "%llu from %d to %d but bh is NULL!\n",
1618 (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1619 (int)old_dir_nlink, old_dir->i_nlink);
1621 struct ocfs2_dinode *fe;
1622 status = ocfs2_journal_access_di(handle,
1623 INODE_CACHE(old_dir),
1625 OCFS2_JOURNAL_ACCESS_WRITE);
1626 fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1627 ocfs2_set_links_count(fe, old_dir->i_nlink);
1628 ocfs2_journal_dirty(handle, old_dir_bh);
1631 ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1635 ocfs2_commit_trans(osb, handle);
1638 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1639 ocfs2_inode_unlock(orphan_dir, 1);
1640 inode_unlock(orphan_dir);
1644 if (new_child_locked)
1645 ocfs2_inode_unlock(new_inode, 1);
1647 if (old_child_locked)
1648 ocfs2_inode_unlock(old_inode, 1);
1651 ocfs2_double_unlock(old_dir, new_dir);
1654 ocfs2_rename_unlock(osb);
1657 sync_mapping_buffers(old_inode->i_mapping);
1661 ocfs2_free_dir_lookup_result(&target_lookup_res);
1662 ocfs2_free_dir_lookup_result(&old_entry_lookup);
1663 ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1664 ocfs2_free_dir_lookup_result(&orphan_insert);
1665 ocfs2_free_dir_lookup_result(&target_insert);
1668 brelse(old_inode_bh);
1679 * we expect i_size = strlen(symname). Copy symname into the file
1680 * data, including the null terminator.
1682 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1684 struct inode *inode,
1685 const char *symname)
1687 struct buffer_head **bhs = NULL;
1689 struct super_block *sb = osb->sb;
1690 u64 p_blkno, p_blocks;
1691 int virtual, blocks, status, i, bytes_left;
1693 bytes_left = i_size_read(inode) + 1;
1694 /* we can't trust i_blocks because we're actually going to
1695 * write i_size + 1 bytes. */
1696 blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1698 trace_ocfs2_create_symlink_data((unsigned long long)inode->i_blocks,
1699 i_size_read(inode), blocks);
1701 /* Sanity check -- make sure we're going to fit. */
1703 ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1709 bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1716 status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1723 /* links can never be larger than one cluster so we know this
1724 * is all going to be contiguous, but do a sanity check
1726 if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1733 while(bytes_left > 0) {
1734 c = &symname[virtual * sb->s_blocksize];
1736 bhs[virtual] = sb_getblk(sb, p_blkno);
1737 if (!bhs[virtual]) {
1742 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1745 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1747 OCFS2_JOURNAL_ACCESS_CREATE);
1753 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1755 memcpy(bhs[virtual]->b_data, c,
1756 (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1759 ocfs2_journal_dirty(handle, bhs[virtual]);
1763 bytes_left -= sb->s_blocksize;
1770 for(i = 0; i < blocks; i++)
1780 static int ocfs2_symlink(struct inode *dir,
1781 struct dentry *dentry,
1782 const char *symname)
1784 int status, l, credits;
1786 struct ocfs2_super *osb = NULL;
1787 struct inode *inode = NULL;
1788 struct super_block *sb;
1789 struct buffer_head *new_fe_bh = NULL;
1790 struct buffer_head *parent_fe_bh = NULL;
1791 struct ocfs2_dinode *fe = NULL;
1792 struct ocfs2_dinode *dirfe;
1793 handle_t *handle = NULL;
1794 struct ocfs2_alloc_context *inode_ac = NULL;
1795 struct ocfs2_alloc_context *data_ac = NULL;
1796 struct ocfs2_alloc_context *xattr_ac = NULL;
1797 int want_clusters = 0;
1798 int xattr_credits = 0;
1799 struct ocfs2_security_xattr_info si = {
1802 int did_quota = 0, did_quota_inode = 0;
1803 struct ocfs2_dir_lookup_result lookup = { NULL, };
1805 int did_block_signals = 0;
1806 struct ocfs2_dentry_lock *dl = NULL;
1808 trace_ocfs2_symlink_begin(dir, dentry, symname,
1809 dentry->d_name.len, dentry->d_name.name);
1811 status = dquot_initialize(dir);
1820 l = strlen(symname) + 1;
1822 credits = ocfs2_calc_symlink_credits(sb);
1824 /* lock the parent directory */
1825 status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1827 if (status != -ENOENT)
1832 dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1833 if (!ocfs2_read_links_count(dirfe)) {
1834 /* can't make a file in a deleted directory. */
1839 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1840 dentry->d_name.len);
1844 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1845 dentry->d_name.name,
1846 dentry->d_name.len, &lookup);
1852 status = ocfs2_reserve_new_inode(osb, &inode_ac);
1854 if (status != -ENOSPC)
1859 inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1860 if (IS_ERR(inode)) {
1861 status = PTR_ERR(inode);
1867 /* get security xattr */
1868 status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
1870 if (status == -EOPNOTSUPP)
1878 /* calculate meta data/clusters for setting security xattr */
1880 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1881 &xattr_credits, &xattr_ac);
1888 /* don't reserve bitmap space for fast symlinks. */
1889 if (l > ocfs2_fast_symlink_chars(sb))
1892 status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1894 if (status != -ENOSPC)
1899 handle = ocfs2_start_trans(osb, credits + xattr_credits);
1900 if (IS_ERR(handle)) {
1901 status = PTR_ERR(handle);
1907 /* Starting to change things, restart is no longer possible. */
1908 ocfs2_block_signals(&oldset);
1909 did_block_signals = 1;
1911 status = dquot_alloc_inode(inode);
1914 did_quota_inode = 1;
1916 trace_ocfs2_symlink_create(dir, dentry, dentry->d_name.len,
1917 dentry->d_name.name,
1918 (unsigned long long)OCFS2_I(dir)->ip_blkno,
1921 status = ocfs2_mknod_locked(osb, dir, inode,
1922 0, &new_fe_bh, parent_fe_bh, handle,
1929 fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1932 inode->i_op = &ocfs2_symlink_inode_operations;
1933 inode_nohighmem(inode);
1934 if (l > ocfs2_fast_symlink_chars(sb)) {
1937 status = dquot_alloc_space_nodirty(inode,
1938 ocfs2_clusters_to_bytes(osb->sb, 1));
1942 inode->i_mapping->a_ops = &ocfs2_aops;
1943 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1945 handle, data_ac, NULL,
1948 if (status != -ENOSPC && status != -EINTR) {
1950 "Failed to extend file to %llu\n",
1951 (unsigned long long)newsize);
1957 i_size_write(inode, newsize);
1958 inode->i_blocks = ocfs2_inode_sector_count(inode);
1960 inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
1961 memcpy((char *) fe->id2.i_symlink, symname, l);
1962 i_size_write(inode, newsize);
1963 inode->i_blocks = 0;
1966 status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1972 if (!ocfs2_inode_is_fast_symlink(inode)) {
1973 status = ocfs2_create_symlink_data(osb, handle, inode,
1982 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1991 * Do this before adding the entry to the directory. We add
1992 * also set d_op after success so that ->d_iput() will cleanup
1993 * the dentry lock even if ocfs2_add_entry() fails below.
1995 status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
2001 dl = dentry->d_fsdata;
2003 status = ocfs2_add_entry(handle, dentry, inode,
2004 le64_to_cpu(fe->i_blkno), parent_fe_bh,
2011 insert_inode_hash(inode);
2012 d_instantiate(dentry, inode);
2014 if (status < 0 && did_quota)
2015 dquot_free_space_nodirty(inode,
2016 ocfs2_clusters_to_bytes(osb->sb, 1));
2017 if (status < 0 && did_quota_inode)
2018 dquot_free_inode(inode);
2020 ocfs2_commit_trans(osb, handle);
2022 ocfs2_inode_unlock(dir, 1);
2023 if (did_block_signals)
2024 ocfs2_unblock_signals(&oldset);
2027 brelse(parent_fe_bh);
2029 ocfs2_free_dir_lookup_result(&lookup);
2031 ocfs2_free_alloc_context(inode_ac);
2033 ocfs2_free_alloc_context(data_ac);
2035 ocfs2_free_alloc_context(xattr_ac);
2036 if ((status < 0) && inode) {
2038 ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
2040 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
2051 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2053 int status, namelen;
2055 namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2065 if (namelen != OCFS2_ORPHAN_NAMELEN) {
2071 trace_ocfs2_blkno_stringify(blkno, name, namelen);
2080 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
2081 struct inode **ret_orphan_dir,
2082 struct buffer_head **ret_orphan_dir_bh)
2084 struct inode *orphan_dir_inode;
2085 struct buffer_head *orphan_dir_bh = NULL;
2088 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2089 ORPHAN_DIR_SYSTEM_INODE,
2091 if (!orphan_dir_inode) {
2097 inode_lock(orphan_dir_inode);
2099 ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2101 inode_unlock(orphan_dir_inode);
2102 iput(orphan_dir_inode);
2108 *ret_orphan_dir = orphan_dir_inode;
2109 *ret_orphan_dir_bh = orphan_dir_bh;
2114 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
2115 struct buffer_head *orphan_dir_bh,
2118 struct ocfs2_dir_lookup_result *lookup,
2122 struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
2124 (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2125 OCFS2_ORPHAN_NAMELEN;
2128 ret = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2129 OCFS2_DIO_ORPHAN_PREFIX);
2130 if (ret != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2136 ret = ocfs2_blkno_stringify(blkno,
2137 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2139 ret = ocfs2_blkno_stringify(blkno, name);
2145 ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2146 orphan_dir_bh, name,
2157 * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
2158 * insertion of an orphan.
2159 * @osb: ocfs2 file system
2160 * @ret_orphan_dir: Orphan dir inode - returned locked!
2161 * @blkno: Actual block number of the inode to be inserted into orphan dir.
2162 * @lookup: dir lookup result, to be passed back into functions like
2165 * Returns zero on success and the ret_orphan_dir, name and lookup
2166 * fields will be populated.
2168 * Returns non-zero on failure.
2170 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2171 struct inode **ret_orphan_dir,
2174 struct ocfs2_dir_lookup_result *lookup,
2177 struct inode *orphan_dir_inode = NULL;
2178 struct buffer_head *orphan_dir_bh = NULL;
2181 ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
2188 ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
2189 blkno, name, lookup, dio);
2195 *ret_orphan_dir = orphan_dir_inode;
2198 brelse(orphan_dir_bh);
2201 ocfs2_inode_unlock(orphan_dir_inode, 1);
2202 inode_unlock(orphan_dir_inode);
2203 iput(orphan_dir_inode);
2211 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2213 struct inode *inode,
2214 struct buffer_head *fe_bh,
2216 struct ocfs2_dir_lookup_result *lookup,
2217 struct inode *orphan_dir_inode,
2220 struct buffer_head *orphan_dir_bh = NULL;
2222 struct ocfs2_dinode *orphan_fe;
2223 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
2225 (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2226 OCFS2_ORPHAN_NAMELEN;
2228 trace_ocfs2_orphan_add_begin(
2229 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2231 status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2237 status = ocfs2_journal_access_di(handle,
2238 INODE_CACHE(orphan_dir_inode),
2240 OCFS2_JOURNAL_ACCESS_WRITE);
2247 * We're going to journal the change of i_flags and i_orphaned_slot.
2248 * It's safe anyway, though some callers may duplicate the journaling.
2249 * Journaling within the func just make the logic look more
2252 status = ocfs2_journal_access_di(handle,
2255 OCFS2_JOURNAL_ACCESS_WRITE);
2261 /* we're a cluster, and nlink can change on disk from
2262 * underneath us... */
2263 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2264 if (S_ISDIR(inode->i_mode))
2265 ocfs2_add_links_count(orphan_fe, 1);
2266 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2267 ocfs2_journal_dirty(handle, orphan_dir_bh);
2269 status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2271 OCFS2_I(inode)->ip_blkno,
2272 orphan_dir_bh, lookup);
2279 /* Update flag OCFS2_DIO_ORPHANED_FL and record the orphan
2282 fe->i_flags |= cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2283 fe->i_dio_orphaned_slot = cpu_to_le16(osb->slot_num);
2285 fe->i_flags |= cpu_to_le32(OCFS2_ORPHANED_FL);
2286 OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2288 /* Record which orphan dir our inode now resides
2289 * in. delete_inode will use this to determine which orphan
2291 fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2294 ocfs2_journal_dirty(handle, fe_bh);
2296 trace_ocfs2_orphan_add_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
2301 if (S_ISDIR(inode->i_mode))
2302 ocfs2_add_links_count(orphan_fe, -1);
2303 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2307 brelse(orphan_dir_bh);
2312 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2313 int ocfs2_orphan_del(struct ocfs2_super *osb,
2315 struct inode *orphan_dir_inode,
2316 struct inode *inode,
2317 struct buffer_head *orphan_dir_bh,
2320 char name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2321 struct ocfs2_dinode *orphan_fe;
2323 struct ocfs2_dir_lookup_result lookup = { NULL, };
2326 status = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2327 OCFS2_DIO_ORPHAN_PREFIX);
2328 if (status != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2334 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno,
2335 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2337 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2343 trace_ocfs2_orphan_del(
2344 (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2345 name, strlen(name));
2347 status = ocfs2_journal_access_di(handle,
2348 INODE_CACHE(orphan_dir_inode),
2350 OCFS2_JOURNAL_ACCESS_WRITE);
2356 /* find it's spot in the orphan directory */
2357 status = ocfs2_find_entry(name, strlen(name), orphan_dir_inode,
2364 /* remove it from the orphan directory */
2365 status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2371 /* do the i_nlink dance! :) */
2372 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2373 if (S_ISDIR(inode->i_mode))
2374 ocfs2_add_links_count(orphan_fe, -1);
2375 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2376 ocfs2_journal_dirty(handle, orphan_dir_bh);
2379 ocfs2_free_dir_lookup_result(&lookup);
2387 * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to receive a newly
2388 * allocated file. This is different from the typical 'add to orphan dir'
2389 * operation in that the inode does not yet exist. This is a problem because
2390 * the orphan dir stringifies the inode block number to come up with it's
2391 * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2392 * problem. This function works around it by calling deeper into the orphan
2393 * and suballoc code than other callers. Use this only by necessity.
2394 * @dir: The directory which this inode will ultimately wind up under - not the
2396 * @dir_bh: buffer_head the @dir inode block
2397 * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2398 * with the string to be used for orphan dirent. Pass back to the orphan dir
2400 * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2402 * @ret_di_blkno: block number where the new inode will be allocated.
2403 * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2404 * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2406 * Returns zero on success and the ret_orphan_dir, name and lookup
2407 * fields will be populated.
2409 * Returns non-zero on failure.
2411 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2412 struct buffer_head *dir_bh,
2414 struct inode **ret_orphan_dir,
2416 struct ocfs2_dir_lookup_result *orphan_insert,
2417 struct ocfs2_alloc_context **ret_inode_ac)
2421 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2422 struct inode *orphan_dir = NULL;
2423 struct buffer_head *orphan_dir_bh = NULL;
2424 struct ocfs2_alloc_context *inode_ac = NULL;
2426 ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2432 /* reserve an inode spot */
2433 ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2440 ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2447 ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2448 di_blkno, orphan_name, orphan_insert,
2457 *ret_orphan_dir = orphan_dir;
2458 *ret_di_blkno = di_blkno;
2459 *ret_inode_ac = inode_ac;
2461 * orphan_name and orphan_insert are already up to
2462 * date via prepare_orphan_dir
2465 /* Unroll reserve_new_inode* */
2467 ocfs2_free_alloc_context(inode_ac);
2469 /* Unroll orphan dir locking */
2470 inode_unlock(orphan_dir);
2471 ocfs2_inode_unlock(orphan_dir, 1);
2475 brelse(orphan_dir_bh);
2480 int ocfs2_create_inode_in_orphan(struct inode *dir,
2482 struct inode **new_inode)
2484 int status, did_quota_inode = 0;
2485 struct inode *inode = NULL;
2486 struct inode *orphan_dir = NULL;
2487 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2488 handle_t *handle = NULL;
2489 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2490 struct buffer_head *parent_di_bh = NULL;
2491 struct buffer_head *new_di_bh = NULL;
2492 struct ocfs2_alloc_context *inode_ac = NULL;
2493 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2494 u64 uninitialized_var(di_blkno), suballoc_loc;
2497 status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2499 if (status != -ENOENT)
2504 status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2505 orphan_name, &orphan_dir,
2506 &di_blkno, &orphan_insert, &inode_ac);
2508 if (status != -ENOSPC)
2513 inode = ocfs2_get_init_inode(dir, mode);
2514 if (IS_ERR(inode)) {
2515 status = PTR_ERR(inode);
2521 handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2522 if (IS_ERR(handle)) {
2523 status = PTR_ERR(handle);
2529 status = dquot_alloc_inode(inode);
2532 did_quota_inode = 1;
2534 status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2536 &suballoc_bit, di_blkno);
2543 /* do the real work now. */
2544 status = __ocfs2_mknod_locked(dir, inode,
2545 0, &new_di_bh, parent_di_bh, handle,
2546 inode_ac, di_blkno, suballoc_loc,
2553 status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2554 &orphan_insert, orphan_dir, false);
2560 /* get open lock so that only nodes can't remove it from orphan dir. */
2561 status = ocfs2_open_lock(inode);
2565 insert_inode_hash(inode);
2567 if (status < 0 && did_quota_inode)
2568 dquot_free_inode(inode);
2570 ocfs2_commit_trans(osb, handle);
2573 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2574 ocfs2_inode_unlock(orphan_dir, 1);
2575 inode_unlock(orphan_dir);
2579 if ((status < 0) && inode) {
2585 ocfs2_free_alloc_context(inode_ac);
2592 ocfs2_free_dir_lookup_result(&orphan_insert);
2594 ocfs2_inode_unlock(dir, 1);
2595 brelse(parent_di_bh);
2599 int ocfs2_add_inode_to_orphan(struct ocfs2_super *osb,
2600 struct inode *inode)
2602 char orphan_name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2603 struct inode *orphan_dir_inode = NULL;
2604 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2605 struct buffer_head *di_bh = NULL;
2607 handle_t *handle = NULL;
2608 struct ocfs2_dinode *di = NULL;
2610 status = ocfs2_inode_lock(inode, &di_bh, 1);
2616 di = (struct ocfs2_dinode *) di_bh->b_data;
2618 * Another append dio crashed?
2619 * If so, manually recover it first.
2621 if (unlikely(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL))) {
2622 status = ocfs2_truncate_file(inode, di_bh, i_size_read(inode));
2624 if (status != -ENOSPC)
2626 goto bail_unlock_inode;
2629 status = ocfs2_del_inode_from_orphan(osb, inode, di_bh, 0, 0);
2632 goto bail_unlock_inode;
2636 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir_inode,
2637 OCFS2_I(inode)->ip_blkno,
2643 goto bail_unlock_inode;
2646 handle = ocfs2_start_trans(osb,
2647 OCFS2_INODE_ADD_TO_ORPHAN_CREDITS);
2648 if (IS_ERR(handle)) {
2649 status = PTR_ERR(handle);
2650 goto bail_unlock_orphan;
2653 status = ocfs2_orphan_add(osb, handle, inode, di_bh, orphan_name,
2654 &orphan_insert, orphan_dir_inode, true);
2658 ocfs2_commit_trans(osb, handle);
2661 ocfs2_inode_unlock(orphan_dir_inode, 1);
2662 inode_unlock(orphan_dir_inode);
2663 iput(orphan_dir_inode);
2665 ocfs2_free_dir_lookup_result(&orphan_insert);
2668 ocfs2_inode_unlock(inode, 1);
2675 int ocfs2_del_inode_from_orphan(struct ocfs2_super *osb,
2676 struct inode *inode, struct buffer_head *di_bh,
2677 int update_isize, loff_t end)
2679 struct inode *orphan_dir_inode = NULL;
2680 struct buffer_head *orphan_dir_bh = NULL;
2681 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2682 handle_t *handle = NULL;
2685 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2686 ORPHAN_DIR_SYSTEM_INODE,
2687 le16_to_cpu(di->i_dio_orphaned_slot));
2688 if (!orphan_dir_inode) {
2694 inode_lock(orphan_dir_inode);
2695 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2697 inode_unlock(orphan_dir_inode);
2698 iput(orphan_dir_inode);
2703 handle = ocfs2_start_trans(osb,
2704 OCFS2_INODE_DEL_FROM_ORPHAN_CREDITS);
2705 if (IS_ERR(handle)) {
2706 status = PTR_ERR(handle);
2707 goto bail_unlock_orphan;
2710 BUG_ON(!(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)));
2712 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode,
2713 inode, orphan_dir_bh, true);
2719 status = ocfs2_journal_access_di(handle,
2722 OCFS2_JOURNAL_ACCESS_WRITE);
2728 di->i_flags &= ~cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2729 di->i_dio_orphaned_slot = 0;
2732 status = ocfs2_set_inode_size(handle, inode, di_bh, end);
2736 ocfs2_journal_dirty(handle, di_bh);
2739 ocfs2_commit_trans(osb, handle);
2742 ocfs2_inode_unlock(orphan_dir_inode, 1);
2743 inode_unlock(orphan_dir_inode);
2744 brelse(orphan_dir_bh);
2745 iput(orphan_dir_inode);
2751 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2752 struct inode *inode,
2753 struct dentry *dentry)
2756 struct buffer_head *parent_di_bh = NULL;
2757 handle_t *handle = NULL;
2758 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2759 struct ocfs2_dinode *dir_di, *di;
2760 struct inode *orphan_dir_inode = NULL;
2761 struct buffer_head *orphan_dir_bh = NULL;
2762 struct buffer_head *di_bh = NULL;
2763 struct ocfs2_dir_lookup_result lookup = { NULL, };
2765 trace_ocfs2_mv_orphaned_inode_to_new(dir, dentry,
2766 dentry->d_name.len, dentry->d_name.name,
2767 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2768 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2770 status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2772 if (status != -ENOENT)
2777 dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2778 if (!dir_di->i_links_count) {
2779 /* can't make a file in a deleted directory. */
2784 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2785 dentry->d_name.len);
2789 /* get a spot inside the dir. */
2790 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2791 dentry->d_name.name,
2792 dentry->d_name.len, &lookup);
2798 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2799 ORPHAN_DIR_SYSTEM_INODE,
2801 if (!orphan_dir_inode) {
2807 inode_lock(orphan_dir_inode);
2809 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2812 inode_unlock(orphan_dir_inode);
2813 iput(orphan_dir_inode);
2817 status = ocfs2_read_inode_block(inode, &di_bh);
2823 handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2824 if (IS_ERR(handle)) {
2825 status = PTR_ERR(handle);
2831 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2832 di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2838 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2839 orphan_dir_bh, false);
2845 di = (struct ocfs2_dinode *)di_bh->b_data;
2846 di->i_flags &= ~cpu_to_le32(OCFS2_ORPHANED_FL);
2847 di->i_orphaned_slot = 0;
2848 set_nlink(inode, 1);
2849 ocfs2_set_links_count(di, inode->i_nlink);
2850 ocfs2_update_inode_fsync_trans(handle, inode, 1);
2851 ocfs2_journal_dirty(handle, di_bh);
2853 status = ocfs2_add_entry(handle, dentry, inode,
2854 OCFS2_I(inode)->ip_blkno, parent_di_bh,
2861 status = ocfs2_dentry_attach_lock(dentry, inode,
2862 OCFS2_I(dir)->ip_blkno);
2868 d_instantiate(dentry, inode);
2871 ocfs2_commit_trans(osb, handle);
2873 ocfs2_inode_unlock(orphan_dir_inode, 1);
2874 inode_unlock(orphan_dir_inode);
2875 iput(orphan_dir_inode);
2878 ocfs2_inode_unlock(dir, 1);
2881 brelse(parent_di_bh);
2882 brelse(orphan_dir_bh);
2884 ocfs2_free_dir_lookup_result(&lookup);
2892 const struct inode_operations ocfs2_dir_iops = {
2893 .create = ocfs2_create,
2894 .lookup = ocfs2_lookup,
2896 .unlink = ocfs2_unlink,
2897 .rmdir = ocfs2_unlink,
2898 .symlink = ocfs2_symlink,
2899 .mkdir = ocfs2_mkdir,
2900 .mknod = ocfs2_mknod,
2901 .rename = ocfs2_rename,
2902 .setattr = ocfs2_setattr,
2903 .getattr = ocfs2_getattr,
2904 .permission = ocfs2_permission,
2905 .listxattr = ocfs2_listxattr,
2906 .fiemap = ocfs2_fiemap,
2907 .get_acl = ocfs2_iop_get_acl,
2908 .set_acl = ocfs2_iop_set_acl,