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 ocfs2_dinode *fe = NULL;
235 struct buffer_head *new_fe_bh = NULL;
236 struct inode *inode = NULL;
237 struct ocfs2_alloc_context *inode_ac = NULL;
238 struct ocfs2_alloc_context *data_ac = NULL;
239 struct ocfs2_alloc_context *meta_ac = NULL;
240 int want_clusters = 0;
242 int xattr_credits = 0;
243 struct ocfs2_security_xattr_info si = {
246 int did_quota_inode = 0;
247 struct ocfs2_dir_lookup_result lookup = { NULL, };
249 int did_block_signals = 0;
250 struct ocfs2_dentry_lock *dl = NULL;
252 trace_ocfs2_mknod(dir, dentry, dentry->d_name.len, dentry->d_name.name,
253 (unsigned long long)OCFS2_I(dir)->ip_blkno,
254 (unsigned long)dev, mode);
256 status = dquot_initialize(dir);
262 /* get our super block */
263 osb = OCFS2_SB(dir->i_sb);
265 status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
267 if (status != -ENOENT)
272 if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
277 dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
278 if (!ocfs2_read_links_count(dirfe)) {
279 /* can't make a file in a deleted directory. */
284 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
289 /* get a spot inside the dir. */
290 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
292 dentry->d_name.len, &lookup);
298 /* reserve an inode spot */
299 status = ocfs2_reserve_new_inode(osb, &inode_ac);
301 if (status != -ENOSPC)
306 inode = ocfs2_get_init_inode(dir, mode);
308 status = PTR_ERR(inode);
314 /* get security xattr */
315 status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
317 if (status == -EOPNOTSUPP)
325 /* calculate meta data/clusters for setting security and acl xattr */
326 status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
328 &xattr_credits, &want_meta);
334 /* Reserve a cluster if creating an extent based directory. */
335 if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
338 /* Dir indexing requires extra space as well */
339 if (ocfs2_supports_indexed_dirs(osb))
343 status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
345 if (status != -ENOSPC)
350 status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
352 if (status != -ENOSPC)
357 handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
360 if (IS_ERR(handle)) {
361 status = PTR_ERR(handle);
367 /* Starting to change things, restart is no longer possible. */
368 ocfs2_block_signals(&oldset);
369 did_block_signals = 1;
371 status = dquot_alloc_inode(inode);
376 /* do the real work now. */
377 status = ocfs2_mknod_locked(osb, dir, inode, dev,
378 &new_fe_bh, parent_fe_bh, handle,
385 fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
387 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
388 new_fe_bh, data_ac, meta_ac);
394 status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
396 OCFS2_JOURNAL_ACCESS_WRITE);
401 ocfs2_add_links_count(dirfe, 1);
402 ocfs2_journal_dirty(handle, parent_fe_bh);
406 status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
415 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
424 * Do this before adding the entry to the directory. We add
425 * also set d_op after success so that ->d_iput() will cleanup
426 * the dentry lock even if ocfs2_add_entry() fails below.
428 status = ocfs2_dentry_attach_lock(dentry, inode,
429 OCFS2_I(dir)->ip_blkno);
435 dl = dentry->d_fsdata;
437 status = ocfs2_add_entry(handle, dentry, inode,
438 OCFS2_I(inode)->ip_blkno, parent_fe_bh,
445 insert_inode_hash(inode);
446 d_instantiate(dentry, inode);
449 if (status < 0 && did_quota_inode)
450 dquot_free_inode(inode);
452 if (status < 0 && fe)
453 ocfs2_set_links_count(fe, 0);
454 ocfs2_commit_trans(osb, handle);
457 ocfs2_inode_unlock(dir, 1);
458 if (did_block_signals)
459 ocfs2_unblock_signals(&oldset);
462 brelse(parent_fe_bh);
465 ocfs2_free_dir_lookup_result(&lookup);
468 ocfs2_free_alloc_context(inode_ac);
471 ocfs2_free_alloc_context(data_ac);
474 ocfs2_free_alloc_context(meta_ac);
477 * We should call iput after the i_mutex of the bitmap been
478 * unlocked in ocfs2_free_alloc_context, or the
479 * ocfs2_delete_inode will mutex_lock again.
481 if ((status < 0) && inode) {
483 ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
485 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
496 static int __ocfs2_mknod_locked(struct inode *dir,
499 struct buffer_head **new_fe_bh,
500 struct buffer_head *parent_fe_bh,
502 struct ocfs2_alloc_context *inode_ac,
503 u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
506 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
507 struct ocfs2_dinode *fe = NULL;
508 struct ocfs2_extent_list *fel;
510 struct ocfs2_inode_info *oi = OCFS2_I(inode);
511 struct timespec64 ts;
515 /* populate as many fields early on as possible - many of
516 * these are used by the support functions here and in
518 inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
519 oi->ip_blkno = fe_blkno;
520 spin_lock(&osb->osb_lock);
521 inode->i_generation = osb->s_next_generation++;
522 spin_unlock(&osb->osb_lock);
524 *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
530 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
532 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
534 OCFS2_JOURNAL_ACCESS_CREATE);
540 fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
541 memset(fe, 0, osb->sb->s_blocksize);
543 fe->i_generation = cpu_to_le32(inode->i_generation);
544 fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
545 fe->i_blkno = cpu_to_le64(fe_blkno);
546 fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
547 fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
548 fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
549 fe->i_uid = cpu_to_le32(i_uid_read(inode));
550 fe->i_gid = cpu_to_le32(i_gid_read(inode));
551 fe->i_mode = cpu_to_le16(inode->i_mode);
552 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
553 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
555 ocfs2_set_links_count(fe, inode->i_nlink);
557 fe->i_last_eb_blk = 0;
558 strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
559 fe->i_flags |= cpu_to_le32(OCFS2_VALID_FL);
560 ktime_get_real_ts64(&ts);
561 fe->i_atime = fe->i_ctime = fe->i_mtime =
562 cpu_to_le64(ts.tv_sec);
563 fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
564 cpu_to_le32(ts.tv_nsec);
568 * If supported, directories start with inline data. If inline
569 * isn't supported, but indexing is, we start them as indexed.
571 feat = le16_to_cpu(fe->i_dyn_features);
572 if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
573 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
575 fe->id2.i_data.id_count = cpu_to_le16(
576 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
578 fel = &fe->id2.i_list;
579 fel->l_tree_depth = 0;
580 fel->l_next_free_rec = 0;
581 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
584 ocfs2_journal_dirty(handle, *new_fe_bh);
586 ocfs2_populate_inode(inode, fe, 1);
587 ocfs2_ci_set_new(osb, INODE_CACHE(inode));
588 if (!ocfs2_mount_local(osb)) {
589 status = ocfs2_create_new_inode_locks(inode);
594 oi->i_sync_tid = handle->h_transaction->t_tid;
595 oi->i_datasync_tid = handle->h_transaction->t_tid;
610 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
614 struct buffer_head **new_fe_bh,
615 struct buffer_head *parent_fe_bh,
617 struct ocfs2_alloc_context *inode_ac)
620 u64 suballoc_loc, fe_blkno = 0;
625 status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
626 inode_ac, &suballoc_loc,
627 &suballoc_bit, &fe_blkno);
633 return __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
634 parent_fe_bh, handle, inode_ac,
635 fe_blkno, suballoc_loc, suballoc_bit);
638 static int ocfs2_mkdir(struct inode *dir,
639 struct dentry *dentry,
644 trace_ocfs2_mkdir(dir, dentry, dentry->d_name.len, dentry->d_name.name,
645 OCFS2_I(dir)->ip_blkno, mode);
646 ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
653 static int ocfs2_create(struct inode *dir,
654 struct dentry *dentry,
660 trace_ocfs2_create(dir, dentry, dentry->d_name.len, dentry->d_name.name,
661 (unsigned long long)OCFS2_I(dir)->ip_blkno, mode);
662 ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
669 static int ocfs2_link(struct dentry *old_dentry,
671 struct dentry *dentry)
674 struct inode *inode = d_inode(old_dentry);
675 struct inode *old_dir = d_inode(old_dentry->d_parent);
677 struct buffer_head *fe_bh = NULL;
678 struct buffer_head *old_dir_bh = NULL;
679 struct buffer_head *parent_fe_bh = NULL;
680 struct ocfs2_dinode *fe = NULL;
681 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
682 struct ocfs2_dir_lookup_result lookup = { NULL, };
686 trace_ocfs2_link((unsigned long long)OCFS2_I(inode)->ip_blkno,
687 old_dentry->d_name.len, old_dentry->d_name.name,
688 dentry->d_name.len, dentry->d_name.name);
690 if (S_ISDIR(inode->i_mode))
693 err = dquot_initialize(dir);
699 err = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
700 &parent_fe_bh, dir, 0);
707 /* make sure both dirs have bhs
708 * get an extra ref on old_dir_bh if old==new */
711 parent_fe_bh = old_dir_bh;
712 get_bh(parent_fe_bh);
714 mlog(ML_ERROR, "%s: no old_dir_bh!\n", osb->uuid_str);
725 err = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
726 old_dentry->d_name.len, &old_de_ino);
733 * Check whether another node removed the source inode while we
736 if (old_de_ino != OCFS2_I(inode)->ip_blkno) {
741 err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
746 err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
748 dentry->d_name.len, &lookup);
754 err = ocfs2_inode_lock(inode, &fe_bh, 1);
761 fe = (struct ocfs2_dinode *) fe_bh->b_data;
762 if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
764 goto out_unlock_inode;
767 handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
768 if (IS_ERR(handle)) {
769 err = PTR_ERR(handle);
772 goto out_unlock_inode;
775 /* Starting to change things, restart is no longer possible. */
776 ocfs2_block_signals(&oldset);
778 err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
779 OCFS2_JOURNAL_ACCESS_WRITE);
786 inode->i_ctime = current_time(inode);
787 ocfs2_set_links_count(fe, inode->i_nlink);
788 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
789 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
790 ocfs2_journal_dirty(handle, fe_bh);
792 err = ocfs2_add_entry(handle, dentry, inode,
793 OCFS2_I(inode)->ip_blkno,
794 parent_fe_bh, &lookup);
796 ocfs2_add_links_count(fe, -1);
802 err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
809 d_instantiate(dentry, inode);
812 ocfs2_commit_trans(osb, handle);
813 ocfs2_unblock_signals(&oldset);
815 ocfs2_inode_unlock(inode, 1);
818 ocfs2_double_unlock(old_dir, dir);
821 brelse(parent_fe_bh);
824 ocfs2_free_dir_lookup_result(&lookup);
833 * Takes and drops an exclusive lock on the given dentry. This will
834 * force other nodes to drop it.
836 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
840 ret = ocfs2_dentry_lock(dentry, 1);
844 ocfs2_dentry_unlock(dentry, 1);
849 static inline int ocfs2_inode_is_unlinkable(struct inode *inode)
851 if (S_ISDIR(inode->i_mode)) {
852 if (inode->i_nlink == 2)
857 if (inode->i_nlink == 1)
862 static int ocfs2_unlink(struct inode *dir,
863 struct dentry *dentry)
866 int child_locked = 0;
867 bool is_unlinkable = false;
868 struct inode *inode = d_inode(dentry);
869 struct inode *orphan_dir = NULL;
870 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
872 struct ocfs2_dinode *fe = NULL;
873 struct buffer_head *fe_bh = NULL;
874 struct buffer_head *parent_node_bh = NULL;
875 handle_t *handle = NULL;
876 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
877 struct ocfs2_dir_lookup_result lookup = { NULL, };
878 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
880 trace_ocfs2_unlink(dir, dentry, dentry->d_name.len,
882 (unsigned long long)OCFS2_I(dir)->ip_blkno,
883 (unsigned long long)OCFS2_I(inode)->ip_blkno);
885 status = dquot_initialize(dir);
891 BUG_ON(d_inode(dentry->d_parent) != dir);
893 if (inode == osb->root_inode)
896 status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
899 if (status != -ENOENT)
904 status = ocfs2_find_files_on_disk(dentry->d_name.name,
905 dentry->d_name.len, &blkno, dir,
908 if (status != -ENOENT)
913 if (OCFS2_I(inode)->ip_blkno != blkno) {
916 trace_ocfs2_unlink_noent(
917 (unsigned long long)OCFS2_I(inode)->ip_blkno,
918 (unsigned long long)blkno,
919 OCFS2_I(inode)->ip_flags);
923 status = ocfs2_inode_lock(inode, &fe_bh, 1);
925 if (status != -ENOENT)
931 if (S_ISDIR(inode->i_mode)) {
932 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
938 status = ocfs2_remote_dentry_delete(dentry);
940 /* This remote delete should succeed under all normal
946 if (ocfs2_inode_is_unlinkable(inode)) {
947 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
948 OCFS2_I(inode)->ip_blkno,
949 orphan_name, &orphan_insert,
955 is_unlinkable = true;
958 handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
959 if (IS_ERR(handle)) {
960 status = PTR_ERR(handle);
966 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
967 OCFS2_JOURNAL_ACCESS_WRITE);
973 fe = (struct ocfs2_dinode *) fe_bh->b_data;
975 /* delete the name from the parent dir */
976 status = ocfs2_delete_entry(handle, dir, &lookup);
982 if (S_ISDIR(inode->i_mode))
985 ocfs2_set_links_count(fe, inode->i_nlink);
986 ocfs2_journal_dirty(handle, fe_bh);
988 dir->i_ctime = dir->i_mtime = current_time(dir);
989 if (S_ISDIR(inode->i_mode))
992 status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
995 if (S_ISDIR(inode->i_mode))
1000 if (is_unlinkable) {
1001 status = ocfs2_orphan_add(osb, handle, inode, fe_bh,
1002 orphan_name, &orphan_insert, orphan_dir, false);
1009 ocfs2_commit_trans(osb, handle);
1012 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1013 ocfs2_inode_unlock(orphan_dir, 1);
1014 inode_unlock(orphan_dir);
1019 ocfs2_inode_unlock(inode, 1);
1021 ocfs2_inode_unlock(dir, 1);
1024 brelse(parent_node_bh);
1026 ocfs2_free_dir_lookup_result(&orphan_insert);
1027 ocfs2_free_dir_lookup_result(&lookup);
1029 if (status && (status != -ENOTEMPTY) && (status != -ENOENT))
1035 static int ocfs2_check_if_ancestor(struct ocfs2_super *osb,
1036 u64 src_inode_no, u64 dest_inode_no)
1039 u64 parent_inode_no = 0;
1040 u64 child_inode_no = src_inode_no;
1041 struct inode *child_inode;
1043 #define MAX_LOOKUP_TIMES 32
1045 child_inode = ocfs2_iget(osb, child_inode_no, 0, 0);
1046 if (IS_ERR(child_inode)) {
1047 ret = PTR_ERR(child_inode);
1051 ret = ocfs2_inode_lock(child_inode, NULL, 0);
1059 ret = ocfs2_lookup_ino_from_name(child_inode, "..", 2,
1061 ocfs2_inode_unlock(child_inode, 0);
1068 if (parent_inode_no == dest_inode_no) {
1073 if (parent_inode_no == osb->root_inode->i_ino) {
1078 child_inode_no = parent_inode_no;
1080 if (++i >= MAX_LOOKUP_TIMES) {
1081 mlog(ML_NOTICE, "max lookup times reached, filesystem "
1082 "may have nested directories, "
1083 "src inode: %llu, dest inode: %llu.\n",
1084 (unsigned long long)src_inode_no,
1085 (unsigned long long)dest_inode_no);
1095 * The only place this should be used is rename and link!
1096 * if they have the same id, then the 1st one is the only one locked.
1098 static int ocfs2_double_lock(struct ocfs2_super *osb,
1099 struct buffer_head **bh1,
1100 struct inode *inode1,
1101 struct buffer_head **bh2,
1102 struct inode *inode2,
1106 int inode1_is_ancestor, inode2_is_ancestor;
1107 struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
1108 struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
1110 trace_ocfs2_double_lock((unsigned long long)oi1->ip_blkno,
1111 (unsigned long long)oi2->ip_blkno);
1118 /* we always want to lock the one with the lower lockid first.
1119 * and if they are nested, we lock ancestor first */
1120 if (oi1->ip_blkno != oi2->ip_blkno) {
1121 inode1_is_ancestor = ocfs2_check_if_ancestor(osb, oi2->ip_blkno,
1123 if (inode1_is_ancestor < 0) {
1124 status = inode1_is_ancestor;
1128 inode2_is_ancestor = ocfs2_check_if_ancestor(osb, oi1->ip_blkno,
1130 if (inode2_is_ancestor < 0) {
1131 status = inode2_is_ancestor;
1135 if ((inode1_is_ancestor == 1) ||
1136 (oi1->ip_blkno < oi2->ip_blkno &&
1137 inode2_is_ancestor == 0)) {
1138 /* switch id1 and id2 around */
1140 swap(inode2, inode1);
1143 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
1144 rename == 1 ? OI_LS_RENAME1 : OI_LS_PARENT);
1146 if (status != -ENOENT)
1153 status = ocfs2_inode_lock_nested(inode1, bh1, 1,
1154 rename == 1 ? OI_LS_RENAME2 : OI_LS_PARENT);
1157 * An error return must mean that no cluster locks
1158 * were held on function exit.
1160 if (oi1->ip_blkno != oi2->ip_blkno) {
1161 ocfs2_inode_unlock(inode2, 1);
1166 if (status != -ENOENT)
1170 trace_ocfs2_double_lock_end(
1171 (unsigned long long)oi1->ip_blkno,
1172 (unsigned long long)oi2->ip_blkno);
1180 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1182 ocfs2_inode_unlock(inode1, 1);
1184 if (inode1 != inode2)
1185 ocfs2_inode_unlock(inode2, 1);
1188 static int ocfs2_rename(struct inode *old_dir,
1189 struct dentry *old_dentry,
1190 struct inode *new_dir,
1191 struct dentry *new_dentry,
1194 int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1195 int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1196 struct inode *old_inode = d_inode(old_dentry);
1197 struct inode *new_inode = d_inode(new_dentry);
1198 struct inode *orphan_dir = NULL;
1199 struct ocfs2_dinode *newfe = NULL;
1200 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1201 struct buffer_head *newfe_bh = NULL;
1202 struct buffer_head *old_inode_bh = NULL;
1203 struct ocfs2_super *osb = NULL;
1204 u64 newfe_blkno, old_de_ino;
1205 handle_t *handle = NULL;
1206 struct buffer_head *old_dir_bh = NULL;
1207 struct buffer_head *new_dir_bh = NULL;
1208 u32 old_dir_nlink = old_dir->i_nlink;
1209 struct ocfs2_dinode *old_di;
1210 struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1211 struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1212 struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1213 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1214 struct ocfs2_dir_lookup_result target_insert = { NULL, };
1215 bool should_add_orphan = false;
1220 /* At some point it might be nice to break this function up a
1223 trace_ocfs2_rename(old_dir, old_dentry, new_dir, new_dentry,
1224 old_dentry->d_name.len, old_dentry->d_name.name,
1225 new_dentry->d_name.len, new_dentry->d_name.name);
1227 status = dquot_initialize(old_dir);
1232 status = dquot_initialize(new_dir);
1238 osb = OCFS2_SB(old_dir->i_sb);
1241 if (!igrab(new_inode))
1245 /* Assume a directory hierarchy thusly:
1248 * a,b,c, and d are all directories.
1250 * from cwd of 'a' on both nodes:
1254 * And that's why, just like the VFS, we need a file system
1256 if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1257 status = ocfs2_rename_lock(osb);
1264 /* here we cannot guarantee the inodes haven't just been
1265 * changed, so check if they are nested again */
1266 status = ocfs2_check_if_ancestor(osb, new_dir->i_ino,
1271 } else if (status == 1) {
1273 trace_ocfs2_rename_not_permitted(
1274 (unsigned long long)old_inode->i_ino,
1275 (unsigned long long)new_dir->i_ino);
1280 /* if old and new are the same, this'll just do one lock. */
1281 status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1282 &new_dir_bh, new_dir, 1);
1289 if (!new_dir->i_nlink) {
1294 /* make sure both dirs have bhs
1295 * get an extra ref on old_dir_bh if old==new */
1298 new_dir_bh = old_dir_bh;
1301 mlog(ML_ERROR, "no old_dir_bh!\n");
1308 * Aside from allowing a meta data update, the locking here
1309 * also ensures that the downconvert thread on other nodes
1310 * won't have to concurrently downconvert the inode and the
1313 status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1316 if (status != -ENOENT)
1320 old_child_locked = 1;
1322 status = ocfs2_remote_dentry_delete(old_dentry);
1328 if (S_ISDIR(old_inode->i_mode)) {
1329 u64 old_inode_parent;
1332 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1334 &old_inode_dot_dot_res);
1340 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1345 if (!new_inode && new_dir != old_dir &&
1346 new_dir->i_nlink >= ocfs2_link_max(osb)) {
1352 status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1353 old_dentry->d_name.len,
1361 * Check for inode number is _not_ due to possible IO errors.
1362 * We might rmdir the source, keep it as pwd of some process
1363 * and merrily kill the link to whatever was created under the
1364 * same name. Goodbye sticky bit ;-<
1366 if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1371 /* check if the target already exists (in which case we need
1373 status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1374 new_dentry->d_name.len,
1375 &newfe_blkno, new_dir,
1376 &target_lookup_res);
1377 /* The only error we allow here is -ENOENT because the new
1378 * file not existing is perfectly valid. */
1379 if ((status < 0) && (status != -ENOENT)) {
1380 /* If we cannot find the file specified we should just */
1381 /* return the error... */
1388 if (!target_exists && new_inode) {
1390 * Target was unlinked by another node while we were
1391 * waiting to get to ocfs2_rename(). There isn't
1392 * anything we can do here to help the situation, so
1393 * bubble up the appropriate error.
1399 /* In case we need to overwrite an existing file, we blow it
1401 if (target_exists) {
1402 /* VFS didn't think there existed an inode here, but
1403 * someone else in the cluster must have raced our
1404 * rename to create one. Today we error cleanly, in
1405 * the future we should consider calling iget to build
1406 * a new struct inode for this entry. */
1410 trace_ocfs2_rename_target_exists(new_dentry->d_name.len,
1411 new_dentry->d_name.name);
1415 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1418 trace_ocfs2_rename_disagree(
1419 (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1420 (unsigned long long)newfe_blkno,
1421 OCFS2_I(new_inode)->ip_flags);
1425 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1427 if (status != -ENOENT)
1431 new_child_locked = 1;
1433 status = ocfs2_remote_dentry_delete(new_dentry);
1439 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1441 trace_ocfs2_rename_over_existing(
1442 (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1443 (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1445 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1446 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1447 OCFS2_I(new_inode)->ip_blkno,
1448 orphan_name, &orphan_insert,
1454 should_add_orphan = true;
1457 BUG_ON(d_inode(new_dentry->d_parent) != new_dir);
1459 status = ocfs2_check_dir_for_entry(new_dir,
1460 new_dentry->d_name.name,
1461 new_dentry->d_name.len);
1465 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1466 new_dentry->d_name.name,
1467 new_dentry->d_name.len,
1475 handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1476 if (IS_ERR(handle)) {
1477 status = PTR_ERR(handle);
1483 if (target_exists) {
1484 if (S_ISDIR(new_inode->i_mode)) {
1485 if (new_inode->i_nlink != 2 ||
1486 !ocfs2_empty_dir(new_inode)) {
1487 status = -ENOTEMPTY;
1491 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1493 OCFS2_JOURNAL_ACCESS_WRITE);
1499 /* change the dirent to point to the correct inode */
1500 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1506 inode_inc_iversion(new_dir);
1508 if (S_ISDIR(new_inode->i_mode))
1509 ocfs2_set_links_count(newfe, 0);
1511 ocfs2_add_links_count(newfe, -1);
1512 ocfs2_journal_dirty(handle, newfe_bh);
1513 if (should_add_orphan) {
1514 status = ocfs2_orphan_add(osb, handle, new_inode,
1515 newfe_bh, orphan_name,
1516 &orphan_insert, orphan_dir, false);
1523 /* if the name was not found in new_dir, add it now */
1524 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1525 OCFS2_I(old_inode)->ip_blkno,
1526 new_dir_bh, &target_insert);
1533 old_inode->i_ctime = current_time(old_inode);
1534 mark_inode_dirty(old_inode);
1536 status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1538 OCFS2_JOURNAL_ACCESS_WRITE);
1540 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1542 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1543 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1544 ocfs2_journal_dirty(handle, old_inode_bh);
1549 * Now that the name has been added to new_dir, remove the old name.
1551 * We don't keep any directory entry context around until now
1552 * because the insert might have changed the type of directory
1553 * we're dealing with.
1555 status = ocfs2_find_entry(old_dentry->d_name.name,
1556 old_dentry->d_name.len, old_dir,
1559 if (!is_journal_aborted(osb->journal->j_journal)) {
1560 ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1562 new_dentry->d_name.len, new_dentry->d_name.name,
1563 old_dentry->d_name.len, old_dentry->d_name.name);
1568 status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1571 if (!is_journal_aborted(osb->journal->j_journal)) {
1572 ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1574 new_dentry->d_name.len, new_dentry->d_name.name,
1575 old_dentry->d_name.len, old_dentry->d_name.name);
1581 drop_nlink(new_inode);
1582 new_inode->i_ctime = current_time(new_inode);
1584 old_dir->i_ctime = old_dir->i_mtime = current_time(old_dir);
1586 if (update_dot_dot) {
1587 status = ocfs2_update_entry(old_inode, handle,
1588 &old_inode_dot_dot_res, new_dir);
1589 drop_nlink(old_dir);
1591 drop_nlink(new_inode);
1594 mark_inode_dirty(new_dir);
1597 mark_inode_dirty(old_dir);
1598 ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1600 mark_inode_dirty(new_inode);
1601 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1604 if (old_dir != new_dir) {
1605 /* Keep the same times on both directories.*/
1606 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1609 * This will also pick up the i_nlink change from the
1612 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1615 if (old_dir_nlink != old_dir->i_nlink) {
1617 mlog(ML_ERROR, "need to change nlink for old dir "
1618 "%llu from %d to %d but bh is NULL!\n",
1619 (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1620 (int)old_dir_nlink, old_dir->i_nlink);
1622 struct ocfs2_dinode *fe;
1623 status = ocfs2_journal_access_di(handle,
1624 INODE_CACHE(old_dir),
1626 OCFS2_JOURNAL_ACCESS_WRITE);
1627 fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1628 ocfs2_set_links_count(fe, old_dir->i_nlink);
1629 ocfs2_journal_dirty(handle, old_dir_bh);
1632 ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1636 ocfs2_commit_trans(osb, handle);
1639 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1640 ocfs2_inode_unlock(orphan_dir, 1);
1641 inode_unlock(orphan_dir);
1645 if (new_child_locked)
1646 ocfs2_inode_unlock(new_inode, 1);
1648 if (old_child_locked)
1649 ocfs2_inode_unlock(old_inode, 1);
1652 ocfs2_double_unlock(old_dir, new_dir);
1655 ocfs2_rename_unlock(osb);
1658 sync_mapping_buffers(old_inode->i_mapping);
1662 ocfs2_free_dir_lookup_result(&target_lookup_res);
1663 ocfs2_free_dir_lookup_result(&old_entry_lookup);
1664 ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1665 ocfs2_free_dir_lookup_result(&orphan_insert);
1666 ocfs2_free_dir_lookup_result(&target_insert);
1669 brelse(old_inode_bh);
1680 * we expect i_size = strlen(symname). Copy symname into the file
1681 * data, including the null terminator.
1683 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1685 struct inode *inode,
1686 const char *symname)
1688 struct buffer_head **bhs = NULL;
1690 struct super_block *sb = osb->sb;
1691 u64 p_blkno, p_blocks;
1692 int virtual, blocks, status, i, bytes_left;
1694 bytes_left = i_size_read(inode) + 1;
1695 /* we can't trust i_blocks because we're actually going to
1696 * write i_size + 1 bytes. */
1697 blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1699 trace_ocfs2_create_symlink_data((unsigned long long)inode->i_blocks,
1700 i_size_read(inode), blocks);
1702 /* Sanity check -- make sure we're going to fit. */
1704 ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1710 bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1717 status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1724 /* links can never be larger than one cluster so we know this
1725 * is all going to be contiguous, but do a sanity check
1727 if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1734 while(bytes_left > 0) {
1735 c = &symname[virtual * sb->s_blocksize];
1737 bhs[virtual] = sb_getblk(sb, p_blkno);
1738 if (!bhs[virtual]) {
1743 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1746 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1748 OCFS2_JOURNAL_ACCESS_CREATE);
1754 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1756 memcpy(bhs[virtual]->b_data, c,
1757 (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1760 ocfs2_journal_dirty(handle, bhs[virtual]);
1764 bytes_left -= sb->s_blocksize;
1771 for(i = 0; i < blocks; i++)
1781 static int ocfs2_symlink(struct inode *dir,
1782 struct dentry *dentry,
1783 const char *symname)
1785 int status, l, credits;
1787 struct ocfs2_super *osb = NULL;
1788 struct inode *inode = NULL;
1789 struct super_block *sb;
1790 struct buffer_head *new_fe_bh = NULL;
1791 struct buffer_head *parent_fe_bh = NULL;
1792 struct ocfs2_dinode *fe = NULL;
1793 struct ocfs2_dinode *dirfe;
1794 handle_t *handle = NULL;
1795 struct ocfs2_alloc_context *inode_ac = NULL;
1796 struct ocfs2_alloc_context *data_ac = NULL;
1797 struct ocfs2_alloc_context *xattr_ac = NULL;
1798 int want_clusters = 0;
1799 int xattr_credits = 0;
1800 struct ocfs2_security_xattr_info si = {
1803 int did_quota = 0, did_quota_inode = 0;
1804 struct ocfs2_dir_lookup_result lookup = { NULL, };
1806 int did_block_signals = 0;
1807 struct ocfs2_dentry_lock *dl = NULL;
1809 trace_ocfs2_symlink_begin(dir, dentry, symname,
1810 dentry->d_name.len, dentry->d_name.name);
1812 status = dquot_initialize(dir);
1821 l = strlen(symname) + 1;
1823 credits = ocfs2_calc_symlink_credits(sb);
1825 /* lock the parent directory */
1826 status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1828 if (status != -ENOENT)
1833 dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1834 if (!ocfs2_read_links_count(dirfe)) {
1835 /* can't make a file in a deleted directory. */
1840 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1841 dentry->d_name.len);
1845 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1846 dentry->d_name.name,
1847 dentry->d_name.len, &lookup);
1853 status = ocfs2_reserve_new_inode(osb, &inode_ac);
1855 if (status != -ENOSPC)
1860 inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1861 if (IS_ERR(inode)) {
1862 status = PTR_ERR(inode);
1868 /* get security xattr */
1869 status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
1871 if (status == -EOPNOTSUPP)
1879 /* calculate meta data/clusters for setting security xattr */
1881 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1882 &xattr_credits, &xattr_ac);
1889 /* don't reserve bitmap space for fast symlinks. */
1890 if (l > ocfs2_fast_symlink_chars(sb))
1893 status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1895 if (status != -ENOSPC)
1900 handle = ocfs2_start_trans(osb, credits + xattr_credits);
1901 if (IS_ERR(handle)) {
1902 status = PTR_ERR(handle);
1908 /* Starting to change things, restart is no longer possible. */
1909 ocfs2_block_signals(&oldset);
1910 did_block_signals = 1;
1912 status = dquot_alloc_inode(inode);
1915 did_quota_inode = 1;
1917 trace_ocfs2_symlink_create(dir, dentry, dentry->d_name.len,
1918 dentry->d_name.name,
1919 (unsigned long long)OCFS2_I(dir)->ip_blkno,
1922 status = ocfs2_mknod_locked(osb, dir, inode,
1923 0, &new_fe_bh, parent_fe_bh, handle,
1930 fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1933 inode->i_op = &ocfs2_symlink_inode_operations;
1934 inode_nohighmem(inode);
1935 if (l > ocfs2_fast_symlink_chars(sb)) {
1938 status = dquot_alloc_space_nodirty(inode,
1939 ocfs2_clusters_to_bytes(osb->sb, 1));
1943 inode->i_mapping->a_ops = &ocfs2_aops;
1944 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1946 handle, data_ac, NULL,
1949 if (status != -ENOSPC && status != -EINTR) {
1951 "Failed to extend file to %llu\n",
1952 (unsigned long long)newsize);
1958 i_size_write(inode, newsize);
1959 inode->i_blocks = ocfs2_inode_sector_count(inode);
1961 inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
1962 memcpy((char *) fe->id2.i_symlink, symname, l);
1963 i_size_write(inode, newsize);
1964 inode->i_blocks = 0;
1967 status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1973 if (!ocfs2_inode_is_fast_symlink(inode)) {
1974 status = ocfs2_create_symlink_data(osb, handle, inode,
1983 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1992 * Do this before adding the entry to the directory. We add
1993 * also set d_op after success so that ->d_iput() will cleanup
1994 * the dentry lock even if ocfs2_add_entry() fails below.
1996 status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
2002 dl = dentry->d_fsdata;
2004 status = ocfs2_add_entry(handle, dentry, inode,
2005 le64_to_cpu(fe->i_blkno), parent_fe_bh,
2012 insert_inode_hash(inode);
2013 d_instantiate(dentry, inode);
2015 if (status < 0 && did_quota)
2016 dquot_free_space_nodirty(inode,
2017 ocfs2_clusters_to_bytes(osb->sb, 1));
2018 if (status < 0 && did_quota_inode)
2019 dquot_free_inode(inode);
2021 if (status < 0 && fe)
2022 ocfs2_set_links_count(fe, 0);
2023 ocfs2_commit_trans(osb, handle);
2026 ocfs2_inode_unlock(dir, 1);
2027 if (did_block_signals)
2028 ocfs2_unblock_signals(&oldset);
2031 brelse(parent_fe_bh);
2033 ocfs2_free_dir_lookup_result(&lookup);
2035 ocfs2_free_alloc_context(inode_ac);
2037 ocfs2_free_alloc_context(data_ac);
2039 ocfs2_free_alloc_context(xattr_ac);
2040 if ((status < 0) && inode) {
2042 ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
2044 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
2055 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2057 int status, namelen;
2059 namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2069 if (namelen != OCFS2_ORPHAN_NAMELEN) {
2075 trace_ocfs2_blkno_stringify(blkno, name, namelen);
2084 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
2085 struct inode **ret_orphan_dir,
2086 struct buffer_head **ret_orphan_dir_bh)
2088 struct inode *orphan_dir_inode;
2089 struct buffer_head *orphan_dir_bh = NULL;
2092 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2093 ORPHAN_DIR_SYSTEM_INODE,
2095 if (!orphan_dir_inode) {
2101 inode_lock(orphan_dir_inode);
2103 ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2105 inode_unlock(orphan_dir_inode);
2106 iput(orphan_dir_inode);
2112 *ret_orphan_dir = orphan_dir_inode;
2113 *ret_orphan_dir_bh = orphan_dir_bh;
2118 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
2119 struct buffer_head *orphan_dir_bh,
2122 struct ocfs2_dir_lookup_result *lookup,
2126 struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
2128 (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2129 OCFS2_ORPHAN_NAMELEN;
2132 ret = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2133 OCFS2_DIO_ORPHAN_PREFIX);
2134 if (ret != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2140 ret = ocfs2_blkno_stringify(blkno,
2141 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2143 ret = ocfs2_blkno_stringify(blkno, name);
2149 ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2150 orphan_dir_bh, name,
2161 * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
2162 * insertion of an orphan.
2163 * @osb: ocfs2 file system
2164 * @ret_orphan_dir: Orphan dir inode - returned locked!
2165 * @blkno: Actual block number of the inode to be inserted into orphan dir.
2166 * @lookup: dir lookup result, to be passed back into functions like
2169 * Returns zero on success and the ret_orphan_dir, name and lookup
2170 * fields will be populated.
2172 * Returns non-zero on failure.
2174 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2175 struct inode **ret_orphan_dir,
2178 struct ocfs2_dir_lookup_result *lookup,
2181 struct inode *orphan_dir_inode = NULL;
2182 struct buffer_head *orphan_dir_bh = NULL;
2185 ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
2192 ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
2193 blkno, name, lookup, dio);
2199 *ret_orphan_dir = orphan_dir_inode;
2202 brelse(orphan_dir_bh);
2205 ocfs2_inode_unlock(orphan_dir_inode, 1);
2206 inode_unlock(orphan_dir_inode);
2207 iput(orphan_dir_inode);
2215 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2217 struct inode *inode,
2218 struct buffer_head *fe_bh,
2220 struct ocfs2_dir_lookup_result *lookup,
2221 struct inode *orphan_dir_inode,
2224 struct buffer_head *orphan_dir_bh = NULL;
2226 struct ocfs2_dinode *orphan_fe;
2227 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
2229 (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2230 OCFS2_ORPHAN_NAMELEN;
2232 trace_ocfs2_orphan_add_begin(
2233 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2235 status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2241 status = ocfs2_journal_access_di(handle,
2242 INODE_CACHE(orphan_dir_inode),
2244 OCFS2_JOURNAL_ACCESS_WRITE);
2251 * We're going to journal the change of i_flags and i_orphaned_slot.
2252 * It's safe anyway, though some callers may duplicate the journaling.
2253 * Journaling within the func just make the logic look more
2256 status = ocfs2_journal_access_di(handle,
2259 OCFS2_JOURNAL_ACCESS_WRITE);
2265 /* we're a cluster, and nlink can change on disk from
2266 * underneath us... */
2267 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2268 if (S_ISDIR(inode->i_mode))
2269 ocfs2_add_links_count(orphan_fe, 1);
2270 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2271 ocfs2_journal_dirty(handle, orphan_dir_bh);
2273 status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2275 OCFS2_I(inode)->ip_blkno,
2276 orphan_dir_bh, lookup);
2283 /* Update flag OCFS2_DIO_ORPHANED_FL and record the orphan
2286 fe->i_flags |= cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2287 fe->i_dio_orphaned_slot = cpu_to_le16(osb->slot_num);
2289 fe->i_flags |= cpu_to_le32(OCFS2_ORPHANED_FL);
2290 OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2292 /* Record which orphan dir our inode now resides
2293 * in. delete_inode will use this to determine which orphan
2295 fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2298 ocfs2_journal_dirty(handle, fe_bh);
2300 trace_ocfs2_orphan_add_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
2305 if (S_ISDIR(inode->i_mode))
2306 ocfs2_add_links_count(orphan_fe, -1);
2307 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2311 brelse(orphan_dir_bh);
2316 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2317 int ocfs2_orphan_del(struct ocfs2_super *osb,
2319 struct inode *orphan_dir_inode,
2320 struct inode *inode,
2321 struct buffer_head *orphan_dir_bh,
2324 char name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2325 struct ocfs2_dinode *orphan_fe;
2327 struct ocfs2_dir_lookup_result lookup = { NULL, };
2330 status = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2331 OCFS2_DIO_ORPHAN_PREFIX);
2332 if (status != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2338 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno,
2339 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2341 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2347 trace_ocfs2_orphan_del(
2348 (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2349 name, strlen(name));
2351 status = ocfs2_journal_access_di(handle,
2352 INODE_CACHE(orphan_dir_inode),
2354 OCFS2_JOURNAL_ACCESS_WRITE);
2360 /* find it's spot in the orphan directory */
2361 status = ocfs2_find_entry(name, strlen(name), orphan_dir_inode,
2368 /* remove it from the orphan directory */
2369 status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2375 /* do the i_nlink dance! :) */
2376 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2377 if (S_ISDIR(inode->i_mode))
2378 ocfs2_add_links_count(orphan_fe, -1);
2379 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2380 ocfs2_journal_dirty(handle, orphan_dir_bh);
2383 ocfs2_free_dir_lookup_result(&lookup);
2391 * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to receive a newly
2392 * allocated file. This is different from the typical 'add to orphan dir'
2393 * operation in that the inode does not yet exist. This is a problem because
2394 * the orphan dir stringifies the inode block number to come up with it's
2395 * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2396 * problem. This function works around it by calling deeper into the orphan
2397 * and suballoc code than other callers. Use this only by necessity.
2398 * @dir: The directory which this inode will ultimately wind up under - not the
2400 * @dir_bh: buffer_head the @dir inode block
2401 * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2402 * with the string to be used for orphan dirent. Pass back to the orphan dir
2404 * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2406 * @ret_di_blkno: block number where the new inode will be allocated.
2407 * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2408 * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2410 * Returns zero on success and the ret_orphan_dir, name and lookup
2411 * fields will be populated.
2413 * Returns non-zero on failure.
2415 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2416 struct buffer_head *dir_bh,
2418 struct inode **ret_orphan_dir,
2420 struct ocfs2_dir_lookup_result *orphan_insert,
2421 struct ocfs2_alloc_context **ret_inode_ac)
2425 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2426 struct inode *orphan_dir = NULL;
2427 struct buffer_head *orphan_dir_bh = NULL;
2428 struct ocfs2_alloc_context *inode_ac = NULL;
2430 ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2436 /* reserve an inode spot */
2437 ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2444 ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2451 ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2452 di_blkno, orphan_name, orphan_insert,
2461 *ret_orphan_dir = orphan_dir;
2462 *ret_di_blkno = di_blkno;
2463 *ret_inode_ac = inode_ac;
2465 * orphan_name and orphan_insert are already up to
2466 * date via prepare_orphan_dir
2469 /* Unroll reserve_new_inode* */
2471 ocfs2_free_alloc_context(inode_ac);
2473 /* Unroll orphan dir locking */
2474 inode_unlock(orphan_dir);
2475 ocfs2_inode_unlock(orphan_dir, 1);
2479 brelse(orphan_dir_bh);
2484 int ocfs2_create_inode_in_orphan(struct inode *dir,
2486 struct inode **new_inode)
2488 int status, did_quota_inode = 0;
2489 struct inode *inode = NULL;
2490 struct inode *orphan_dir = NULL;
2491 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2492 handle_t *handle = NULL;
2493 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2494 struct buffer_head *parent_di_bh = NULL;
2495 struct buffer_head *new_di_bh = NULL;
2496 struct ocfs2_alloc_context *inode_ac = NULL;
2497 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2498 u64 di_blkno, suballoc_loc;
2501 status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2503 if (status != -ENOENT)
2508 status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2509 orphan_name, &orphan_dir,
2510 &di_blkno, &orphan_insert, &inode_ac);
2512 if (status != -ENOSPC)
2517 inode = ocfs2_get_init_inode(dir, mode);
2518 if (IS_ERR(inode)) {
2519 status = PTR_ERR(inode);
2525 handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2526 if (IS_ERR(handle)) {
2527 status = PTR_ERR(handle);
2533 status = dquot_alloc_inode(inode);
2536 did_quota_inode = 1;
2538 status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2540 &suballoc_bit, di_blkno);
2547 /* do the real work now. */
2548 status = __ocfs2_mknod_locked(dir, inode,
2549 0, &new_di_bh, parent_di_bh, handle,
2550 inode_ac, di_blkno, suballoc_loc,
2557 status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2558 &orphan_insert, orphan_dir, false);
2564 /* get open lock so that only nodes can't remove it from orphan dir. */
2565 status = ocfs2_open_lock(inode);
2569 insert_inode_hash(inode);
2571 if (status < 0 && did_quota_inode)
2572 dquot_free_inode(inode);
2574 ocfs2_commit_trans(osb, handle);
2577 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2578 ocfs2_inode_unlock(orphan_dir, 1);
2579 inode_unlock(orphan_dir);
2583 if ((status < 0) && inode) {
2589 ocfs2_free_alloc_context(inode_ac);
2596 ocfs2_free_dir_lookup_result(&orphan_insert);
2598 ocfs2_inode_unlock(dir, 1);
2599 brelse(parent_di_bh);
2603 int ocfs2_add_inode_to_orphan(struct ocfs2_super *osb,
2604 struct inode *inode)
2606 char orphan_name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2607 struct inode *orphan_dir_inode = NULL;
2608 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2609 struct buffer_head *di_bh = NULL;
2611 handle_t *handle = NULL;
2612 struct ocfs2_dinode *di = NULL;
2614 status = ocfs2_inode_lock(inode, &di_bh, 1);
2620 di = (struct ocfs2_dinode *) di_bh->b_data;
2622 * Another append dio crashed?
2623 * If so, manually recover it first.
2625 if (unlikely(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL))) {
2626 status = ocfs2_truncate_file(inode, di_bh, i_size_read(inode));
2628 if (status != -ENOSPC)
2630 goto bail_unlock_inode;
2633 status = ocfs2_del_inode_from_orphan(osb, inode, di_bh, 0, 0);
2636 goto bail_unlock_inode;
2640 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir_inode,
2641 OCFS2_I(inode)->ip_blkno,
2647 goto bail_unlock_inode;
2650 handle = ocfs2_start_trans(osb,
2651 OCFS2_INODE_ADD_TO_ORPHAN_CREDITS);
2652 if (IS_ERR(handle)) {
2653 status = PTR_ERR(handle);
2654 goto bail_unlock_orphan;
2657 status = ocfs2_orphan_add(osb, handle, inode, di_bh, orphan_name,
2658 &orphan_insert, orphan_dir_inode, true);
2662 ocfs2_commit_trans(osb, handle);
2665 ocfs2_inode_unlock(orphan_dir_inode, 1);
2666 inode_unlock(orphan_dir_inode);
2667 iput(orphan_dir_inode);
2669 ocfs2_free_dir_lookup_result(&orphan_insert);
2672 ocfs2_inode_unlock(inode, 1);
2679 int ocfs2_del_inode_from_orphan(struct ocfs2_super *osb,
2680 struct inode *inode, struct buffer_head *di_bh,
2681 int update_isize, loff_t end)
2683 struct inode *orphan_dir_inode = NULL;
2684 struct buffer_head *orphan_dir_bh = NULL;
2685 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2686 handle_t *handle = NULL;
2689 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2690 ORPHAN_DIR_SYSTEM_INODE,
2691 le16_to_cpu(di->i_dio_orphaned_slot));
2692 if (!orphan_dir_inode) {
2698 inode_lock(orphan_dir_inode);
2699 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2701 inode_unlock(orphan_dir_inode);
2702 iput(orphan_dir_inode);
2707 handle = ocfs2_start_trans(osb,
2708 OCFS2_INODE_DEL_FROM_ORPHAN_CREDITS);
2709 if (IS_ERR(handle)) {
2710 status = PTR_ERR(handle);
2711 goto bail_unlock_orphan;
2714 BUG_ON(!(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)));
2716 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode,
2717 inode, orphan_dir_bh, true);
2723 status = ocfs2_journal_access_di(handle,
2726 OCFS2_JOURNAL_ACCESS_WRITE);
2732 di->i_flags &= ~cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2733 di->i_dio_orphaned_slot = 0;
2736 status = ocfs2_set_inode_size(handle, inode, di_bh, end);
2740 ocfs2_journal_dirty(handle, di_bh);
2743 ocfs2_commit_trans(osb, handle);
2746 ocfs2_inode_unlock(orphan_dir_inode, 1);
2747 inode_unlock(orphan_dir_inode);
2748 brelse(orphan_dir_bh);
2749 iput(orphan_dir_inode);
2755 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2756 struct inode *inode,
2757 struct dentry *dentry)
2760 struct buffer_head *parent_di_bh = NULL;
2761 handle_t *handle = NULL;
2762 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2763 struct ocfs2_dinode *dir_di, *di;
2764 struct inode *orphan_dir_inode = NULL;
2765 struct buffer_head *orphan_dir_bh = NULL;
2766 struct buffer_head *di_bh = NULL;
2767 struct ocfs2_dir_lookup_result lookup = { NULL, };
2769 trace_ocfs2_mv_orphaned_inode_to_new(dir, dentry,
2770 dentry->d_name.len, dentry->d_name.name,
2771 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2772 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2774 status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2776 if (status != -ENOENT)
2781 dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2782 if (!dir_di->i_links_count) {
2783 /* can't make a file in a deleted directory. */
2788 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2789 dentry->d_name.len);
2793 /* get a spot inside the dir. */
2794 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2795 dentry->d_name.name,
2796 dentry->d_name.len, &lookup);
2802 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2803 ORPHAN_DIR_SYSTEM_INODE,
2805 if (!orphan_dir_inode) {
2811 inode_lock(orphan_dir_inode);
2813 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2816 inode_unlock(orphan_dir_inode);
2817 iput(orphan_dir_inode);
2821 status = ocfs2_read_inode_block(inode, &di_bh);
2827 handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2828 if (IS_ERR(handle)) {
2829 status = PTR_ERR(handle);
2835 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2836 di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2842 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2843 orphan_dir_bh, false);
2849 di = (struct ocfs2_dinode *)di_bh->b_data;
2850 di->i_flags &= ~cpu_to_le32(OCFS2_ORPHANED_FL);
2851 di->i_orphaned_slot = 0;
2852 set_nlink(inode, 1);
2853 ocfs2_set_links_count(di, inode->i_nlink);
2854 ocfs2_update_inode_fsync_trans(handle, inode, 1);
2855 ocfs2_journal_dirty(handle, di_bh);
2857 status = ocfs2_add_entry(handle, dentry, inode,
2858 OCFS2_I(inode)->ip_blkno, parent_di_bh,
2865 status = ocfs2_dentry_attach_lock(dentry, inode,
2866 OCFS2_I(dir)->ip_blkno);
2872 d_instantiate(dentry, inode);
2875 ocfs2_commit_trans(osb, handle);
2877 ocfs2_inode_unlock(orphan_dir_inode, 1);
2878 inode_unlock(orphan_dir_inode);
2879 iput(orphan_dir_inode);
2882 ocfs2_inode_unlock(dir, 1);
2885 brelse(parent_di_bh);
2886 brelse(orphan_dir_bh);
2888 ocfs2_free_dir_lookup_result(&lookup);
2896 const struct inode_operations ocfs2_dir_iops = {
2897 .create = ocfs2_create,
2898 .lookup = ocfs2_lookup,
2900 .unlink = ocfs2_unlink,
2901 .rmdir = ocfs2_unlink,
2902 .symlink = ocfs2_symlink,
2903 .mkdir = ocfs2_mkdir,
2904 .mknod = ocfs2_mknod,
2905 .rename = ocfs2_rename,
2906 .setattr = ocfs2_setattr,
2907 .getattr = ocfs2_getattr,
2908 .permission = ocfs2_permission,
2909 .listxattr = ocfs2_listxattr,
2910 .fiemap = ocfs2_fiemap,
2911 .get_acl = ocfs2_iop_get_acl,
2912 .set_acl = ocfs2_iop_set_acl,