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);
450 if (status < 0 && S_ISDIR(mode)) {
451 ocfs2_add_links_count(dirfe, -1);
456 if (status < 0 && did_quota_inode)
457 dquot_free_inode(inode);
459 if (status < 0 && fe)
460 ocfs2_set_links_count(fe, 0);
461 ocfs2_commit_trans(osb, handle);
464 ocfs2_inode_unlock(dir, 1);
465 if (did_block_signals)
466 ocfs2_unblock_signals(&oldset);
469 brelse(parent_fe_bh);
472 ocfs2_free_dir_lookup_result(&lookup);
475 ocfs2_free_alloc_context(inode_ac);
478 ocfs2_free_alloc_context(data_ac);
481 ocfs2_free_alloc_context(meta_ac);
484 * We should call iput after the i_mutex of the bitmap been
485 * unlocked in ocfs2_free_alloc_context, or the
486 * ocfs2_delete_inode will mutex_lock again.
488 if ((status < 0) && inode) {
490 ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
492 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
503 static int __ocfs2_mknod_locked(struct inode *dir,
506 struct buffer_head **new_fe_bh,
507 struct buffer_head *parent_fe_bh,
509 struct ocfs2_alloc_context *inode_ac,
510 u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
513 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
514 struct ocfs2_dinode *fe = NULL;
515 struct ocfs2_extent_list *fel;
517 struct ocfs2_inode_info *oi = OCFS2_I(inode);
518 struct timespec64 ts;
522 /* populate as many fields early on as possible - many of
523 * these are used by the support functions here and in
525 inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
526 oi->ip_blkno = fe_blkno;
527 spin_lock(&osb->osb_lock);
528 inode->i_generation = osb->s_next_generation++;
529 spin_unlock(&osb->osb_lock);
531 *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
537 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
539 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
541 OCFS2_JOURNAL_ACCESS_CREATE);
547 fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
548 memset(fe, 0, osb->sb->s_blocksize);
550 fe->i_generation = cpu_to_le32(inode->i_generation);
551 fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
552 fe->i_blkno = cpu_to_le64(fe_blkno);
553 fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
554 fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
555 fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
556 fe->i_uid = cpu_to_le32(i_uid_read(inode));
557 fe->i_gid = cpu_to_le32(i_gid_read(inode));
558 fe->i_mode = cpu_to_le16(inode->i_mode);
559 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
560 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
562 ocfs2_set_links_count(fe, inode->i_nlink);
564 fe->i_last_eb_blk = 0;
565 strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
566 fe->i_flags |= cpu_to_le32(OCFS2_VALID_FL);
567 ktime_get_real_ts64(&ts);
568 fe->i_atime = fe->i_ctime = fe->i_mtime =
569 cpu_to_le64(ts.tv_sec);
570 fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
571 cpu_to_le32(ts.tv_nsec);
575 * If supported, directories start with inline data. If inline
576 * isn't supported, but indexing is, we start them as indexed.
578 feat = le16_to_cpu(fe->i_dyn_features);
579 if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
580 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
582 fe->id2.i_data.id_count = cpu_to_le16(
583 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
585 fel = &fe->id2.i_list;
586 fel->l_tree_depth = 0;
587 fel->l_next_free_rec = 0;
588 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
591 ocfs2_journal_dirty(handle, *new_fe_bh);
593 ocfs2_populate_inode(inode, fe, 1);
594 ocfs2_ci_set_new(osb, INODE_CACHE(inode));
595 if (!ocfs2_mount_local(osb)) {
596 status = ocfs2_create_new_inode_locks(inode);
601 ocfs2_update_inode_fsync_trans(handle, inode, 1);
616 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
620 struct buffer_head **new_fe_bh,
621 struct buffer_head *parent_fe_bh,
623 struct ocfs2_alloc_context *inode_ac)
626 u64 suballoc_loc, fe_blkno = 0;
631 status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
632 inode_ac, &suballoc_loc,
633 &suballoc_bit, &fe_blkno);
639 return __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
640 parent_fe_bh, handle, inode_ac,
641 fe_blkno, suballoc_loc, suballoc_bit);
644 static int ocfs2_mkdir(struct inode *dir,
645 struct dentry *dentry,
650 trace_ocfs2_mkdir(dir, dentry, dentry->d_name.len, dentry->d_name.name,
651 OCFS2_I(dir)->ip_blkno, mode);
652 ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
659 static int ocfs2_create(struct inode *dir,
660 struct dentry *dentry,
666 trace_ocfs2_create(dir, dentry, dentry->d_name.len, dentry->d_name.name,
667 (unsigned long long)OCFS2_I(dir)->ip_blkno, mode);
668 ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
675 static int ocfs2_link(struct dentry *old_dentry,
677 struct dentry *dentry)
680 struct inode *inode = d_inode(old_dentry);
681 struct inode *old_dir = d_inode(old_dentry->d_parent);
683 struct buffer_head *fe_bh = NULL;
684 struct buffer_head *old_dir_bh = NULL;
685 struct buffer_head *parent_fe_bh = NULL;
686 struct ocfs2_dinode *fe = NULL;
687 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
688 struct ocfs2_dir_lookup_result lookup = { NULL, };
692 trace_ocfs2_link((unsigned long long)OCFS2_I(inode)->ip_blkno,
693 old_dentry->d_name.len, old_dentry->d_name.name,
694 dentry->d_name.len, dentry->d_name.name);
696 if (S_ISDIR(inode->i_mode))
699 err = dquot_initialize(dir);
705 err = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
706 &parent_fe_bh, dir, 0);
713 /* make sure both dirs have bhs
714 * get an extra ref on old_dir_bh if old==new */
717 parent_fe_bh = old_dir_bh;
718 get_bh(parent_fe_bh);
720 mlog(ML_ERROR, "%s: no old_dir_bh!\n", osb->uuid_str);
731 err = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
732 old_dentry->d_name.len, &old_de_ino);
739 * Check whether another node removed the source inode while we
742 if (old_de_ino != OCFS2_I(inode)->ip_blkno) {
747 err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
752 err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
754 dentry->d_name.len, &lookup);
760 err = ocfs2_inode_lock(inode, &fe_bh, 1);
767 fe = (struct ocfs2_dinode *) fe_bh->b_data;
768 if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
770 goto out_unlock_inode;
773 handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
774 if (IS_ERR(handle)) {
775 err = PTR_ERR(handle);
778 goto out_unlock_inode;
781 /* Starting to change things, restart is no longer possible. */
782 ocfs2_block_signals(&oldset);
784 err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
785 OCFS2_JOURNAL_ACCESS_WRITE);
792 inode->i_ctime = current_time(inode);
793 ocfs2_set_links_count(fe, inode->i_nlink);
794 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
795 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
796 ocfs2_journal_dirty(handle, fe_bh);
798 err = ocfs2_add_entry(handle, dentry, inode,
799 OCFS2_I(inode)->ip_blkno,
800 parent_fe_bh, &lookup);
802 ocfs2_add_links_count(fe, -1);
808 err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
815 d_instantiate(dentry, inode);
818 ocfs2_commit_trans(osb, handle);
819 ocfs2_unblock_signals(&oldset);
821 ocfs2_inode_unlock(inode, 1);
824 ocfs2_double_unlock(old_dir, dir);
827 brelse(parent_fe_bh);
830 ocfs2_free_dir_lookup_result(&lookup);
839 * Takes and drops an exclusive lock on the given dentry. This will
840 * force other nodes to drop it.
842 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
846 ret = ocfs2_dentry_lock(dentry, 1);
850 ocfs2_dentry_unlock(dentry, 1);
855 static inline int ocfs2_inode_is_unlinkable(struct inode *inode)
857 if (S_ISDIR(inode->i_mode)) {
858 if (inode->i_nlink == 2)
863 if (inode->i_nlink == 1)
868 static int ocfs2_unlink(struct inode *dir,
869 struct dentry *dentry)
872 int child_locked = 0;
873 bool is_unlinkable = false;
874 struct inode *inode = d_inode(dentry);
875 struct inode *orphan_dir = NULL;
876 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
878 struct ocfs2_dinode *fe = NULL;
879 struct buffer_head *fe_bh = NULL;
880 struct buffer_head *parent_node_bh = NULL;
881 handle_t *handle = NULL;
882 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
883 struct ocfs2_dir_lookup_result lookup = { NULL, };
884 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
886 trace_ocfs2_unlink(dir, dentry, dentry->d_name.len,
888 (unsigned long long)OCFS2_I(dir)->ip_blkno,
889 (unsigned long long)OCFS2_I(inode)->ip_blkno);
891 status = dquot_initialize(dir);
897 BUG_ON(d_inode(dentry->d_parent) != dir);
899 if (inode == osb->root_inode)
902 status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
905 if (status != -ENOENT)
910 status = ocfs2_find_files_on_disk(dentry->d_name.name,
911 dentry->d_name.len, &blkno, dir,
914 if (status != -ENOENT)
919 if (OCFS2_I(inode)->ip_blkno != blkno) {
922 trace_ocfs2_unlink_noent(
923 (unsigned long long)OCFS2_I(inode)->ip_blkno,
924 (unsigned long long)blkno,
925 OCFS2_I(inode)->ip_flags);
929 status = ocfs2_inode_lock(inode, &fe_bh, 1);
931 if (status != -ENOENT)
937 if (S_ISDIR(inode->i_mode)) {
938 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
944 status = ocfs2_remote_dentry_delete(dentry);
946 /* This remote delete should succeed under all normal
952 if (ocfs2_inode_is_unlinkable(inode)) {
953 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
954 OCFS2_I(inode)->ip_blkno,
955 orphan_name, &orphan_insert,
961 is_unlinkable = true;
964 handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
965 if (IS_ERR(handle)) {
966 status = PTR_ERR(handle);
972 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
973 OCFS2_JOURNAL_ACCESS_WRITE);
979 fe = (struct ocfs2_dinode *) fe_bh->b_data;
981 /* delete the name from the parent dir */
982 status = ocfs2_delete_entry(handle, dir, &lookup);
988 if (S_ISDIR(inode->i_mode))
991 ocfs2_set_links_count(fe, inode->i_nlink);
992 ocfs2_journal_dirty(handle, fe_bh);
994 dir->i_ctime = dir->i_mtime = current_time(dir);
995 if (S_ISDIR(inode->i_mode))
998 status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
1001 if (S_ISDIR(inode->i_mode))
1006 if (is_unlinkable) {
1007 status = ocfs2_orphan_add(osb, handle, inode, fe_bh,
1008 orphan_name, &orphan_insert, orphan_dir, false);
1015 ocfs2_commit_trans(osb, handle);
1018 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1019 ocfs2_inode_unlock(orphan_dir, 1);
1020 inode_unlock(orphan_dir);
1025 ocfs2_inode_unlock(inode, 1);
1027 ocfs2_inode_unlock(dir, 1);
1030 brelse(parent_node_bh);
1032 ocfs2_free_dir_lookup_result(&orphan_insert);
1033 ocfs2_free_dir_lookup_result(&lookup);
1035 if (status && (status != -ENOTEMPTY) && (status != -ENOENT))
1041 static int ocfs2_check_if_ancestor(struct ocfs2_super *osb,
1042 u64 src_inode_no, u64 dest_inode_no)
1045 u64 parent_inode_no = 0;
1046 u64 child_inode_no = src_inode_no;
1047 struct inode *child_inode;
1049 #define MAX_LOOKUP_TIMES 32
1051 child_inode = ocfs2_iget(osb, child_inode_no, 0, 0);
1052 if (IS_ERR(child_inode)) {
1053 ret = PTR_ERR(child_inode);
1057 ret = ocfs2_inode_lock(child_inode, NULL, 0);
1065 ret = ocfs2_lookup_ino_from_name(child_inode, "..", 2,
1067 ocfs2_inode_unlock(child_inode, 0);
1074 if (parent_inode_no == dest_inode_no) {
1079 if (parent_inode_no == osb->root_inode->i_ino) {
1084 child_inode_no = parent_inode_no;
1086 if (++i >= MAX_LOOKUP_TIMES) {
1087 mlog(ML_NOTICE, "max lookup times reached, filesystem "
1088 "may have nested directories, "
1089 "src inode: %llu, dest inode: %llu.\n",
1090 (unsigned long long)src_inode_no,
1091 (unsigned long long)dest_inode_no);
1101 * The only place this should be used is rename and link!
1102 * if they have the same id, then the 1st one is the only one locked.
1104 static int ocfs2_double_lock(struct ocfs2_super *osb,
1105 struct buffer_head **bh1,
1106 struct inode *inode1,
1107 struct buffer_head **bh2,
1108 struct inode *inode2,
1112 int inode1_is_ancestor, inode2_is_ancestor;
1113 struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
1114 struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
1116 trace_ocfs2_double_lock((unsigned long long)oi1->ip_blkno,
1117 (unsigned long long)oi2->ip_blkno);
1124 /* we always want to lock the one with the lower lockid first.
1125 * and if they are nested, we lock ancestor first */
1126 if (oi1->ip_blkno != oi2->ip_blkno) {
1127 inode1_is_ancestor = ocfs2_check_if_ancestor(osb, oi2->ip_blkno,
1129 if (inode1_is_ancestor < 0) {
1130 status = inode1_is_ancestor;
1134 inode2_is_ancestor = ocfs2_check_if_ancestor(osb, oi1->ip_blkno,
1136 if (inode2_is_ancestor < 0) {
1137 status = inode2_is_ancestor;
1141 if ((inode1_is_ancestor == 1) ||
1142 (oi1->ip_blkno < oi2->ip_blkno &&
1143 inode2_is_ancestor == 0)) {
1144 /* switch id1 and id2 around */
1146 swap(inode2, inode1);
1149 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
1150 rename == 1 ? OI_LS_RENAME1 : OI_LS_PARENT);
1152 if (status != -ENOENT)
1159 status = ocfs2_inode_lock_nested(inode1, bh1, 1,
1160 rename == 1 ? OI_LS_RENAME2 : OI_LS_PARENT);
1163 * An error return must mean that no cluster locks
1164 * were held on function exit.
1166 if (oi1->ip_blkno != oi2->ip_blkno) {
1167 ocfs2_inode_unlock(inode2, 1);
1172 if (status != -ENOENT)
1176 trace_ocfs2_double_lock_end(
1177 (unsigned long long)oi1->ip_blkno,
1178 (unsigned long long)oi2->ip_blkno);
1186 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1188 ocfs2_inode_unlock(inode1, 1);
1190 if (inode1 != inode2)
1191 ocfs2_inode_unlock(inode2, 1);
1194 static int ocfs2_rename(struct inode *old_dir,
1195 struct dentry *old_dentry,
1196 struct inode *new_dir,
1197 struct dentry *new_dentry,
1200 int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1201 int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1202 struct inode *old_inode = d_inode(old_dentry);
1203 struct inode *new_inode = d_inode(new_dentry);
1204 struct inode *orphan_dir = NULL;
1205 struct ocfs2_dinode *newfe = NULL;
1206 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1207 struct buffer_head *newfe_bh = NULL;
1208 struct buffer_head *old_inode_bh = NULL;
1209 struct ocfs2_super *osb = NULL;
1210 u64 newfe_blkno, old_de_ino;
1211 handle_t *handle = NULL;
1212 struct buffer_head *old_dir_bh = NULL;
1213 struct buffer_head *new_dir_bh = NULL;
1214 u32 old_dir_nlink = old_dir->i_nlink;
1215 struct ocfs2_dinode *old_di;
1216 struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1217 struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1218 struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1219 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1220 struct ocfs2_dir_lookup_result target_insert = { NULL, };
1221 bool should_add_orphan = false;
1226 /* At some point it might be nice to break this function up a
1229 trace_ocfs2_rename(old_dir, old_dentry, new_dir, new_dentry,
1230 old_dentry->d_name.len, old_dentry->d_name.name,
1231 new_dentry->d_name.len, new_dentry->d_name.name);
1233 status = dquot_initialize(old_dir);
1238 status = dquot_initialize(new_dir);
1244 osb = OCFS2_SB(old_dir->i_sb);
1247 if (!igrab(new_inode))
1251 /* Assume a directory hierarchy thusly:
1254 * a,b,c, and d are all directories.
1256 * from cwd of 'a' on both nodes:
1260 * And that's why, just like the VFS, we need a file system
1262 if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1263 status = ocfs2_rename_lock(osb);
1270 /* here we cannot guarantee the inodes haven't just been
1271 * changed, so check if they are nested again */
1272 status = ocfs2_check_if_ancestor(osb, new_dir->i_ino,
1277 } else if (status == 1) {
1279 trace_ocfs2_rename_not_permitted(
1280 (unsigned long long)old_inode->i_ino,
1281 (unsigned long long)new_dir->i_ino);
1286 /* if old and new are the same, this'll just do one lock. */
1287 status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1288 &new_dir_bh, new_dir, 1);
1295 if (!new_dir->i_nlink) {
1300 /* make sure both dirs have bhs
1301 * get an extra ref on old_dir_bh if old==new */
1304 new_dir_bh = old_dir_bh;
1307 mlog(ML_ERROR, "no old_dir_bh!\n");
1314 * Aside from allowing a meta data update, the locking here
1315 * also ensures that the downconvert thread on other nodes
1316 * won't have to concurrently downconvert the inode and the
1319 status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1322 if (status != -ENOENT)
1326 old_child_locked = 1;
1328 status = ocfs2_remote_dentry_delete(old_dentry);
1334 if (S_ISDIR(old_inode->i_mode)) {
1335 u64 old_inode_parent;
1338 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1340 &old_inode_dot_dot_res);
1346 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1351 if (!new_inode && new_dir != old_dir &&
1352 new_dir->i_nlink >= ocfs2_link_max(osb)) {
1358 status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1359 old_dentry->d_name.len,
1367 * Check for inode number is _not_ due to possible IO errors.
1368 * We might rmdir the source, keep it as pwd of some process
1369 * and merrily kill the link to whatever was created under the
1370 * same name. Goodbye sticky bit ;-<
1372 if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1377 /* check if the target already exists (in which case we need
1379 status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1380 new_dentry->d_name.len,
1381 &newfe_blkno, new_dir,
1382 &target_lookup_res);
1383 /* The only error we allow here is -ENOENT because the new
1384 * file not existing is perfectly valid. */
1385 if ((status < 0) && (status != -ENOENT)) {
1386 /* If we cannot find the file specified we should just */
1387 /* return the error... */
1394 if (!target_exists && new_inode) {
1396 * Target was unlinked by another node while we were
1397 * waiting to get to ocfs2_rename(). There isn't
1398 * anything we can do here to help the situation, so
1399 * bubble up the appropriate error.
1405 /* In case we need to overwrite an existing file, we blow it
1407 if (target_exists) {
1408 /* VFS didn't think there existed an inode here, but
1409 * someone else in the cluster must have raced our
1410 * rename to create one. Today we error cleanly, in
1411 * the future we should consider calling iget to build
1412 * a new struct inode for this entry. */
1416 trace_ocfs2_rename_target_exists(new_dentry->d_name.len,
1417 new_dentry->d_name.name);
1421 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1424 trace_ocfs2_rename_disagree(
1425 (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1426 (unsigned long long)newfe_blkno,
1427 OCFS2_I(new_inode)->ip_flags);
1431 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1433 if (status != -ENOENT)
1437 new_child_locked = 1;
1439 status = ocfs2_remote_dentry_delete(new_dentry);
1445 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1447 trace_ocfs2_rename_over_existing(
1448 (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1449 (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1451 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1452 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1453 OCFS2_I(new_inode)->ip_blkno,
1454 orphan_name, &orphan_insert,
1460 should_add_orphan = true;
1463 BUG_ON(d_inode(new_dentry->d_parent) != new_dir);
1465 status = ocfs2_check_dir_for_entry(new_dir,
1466 new_dentry->d_name.name,
1467 new_dentry->d_name.len);
1471 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1472 new_dentry->d_name.name,
1473 new_dentry->d_name.len,
1481 handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1482 if (IS_ERR(handle)) {
1483 status = PTR_ERR(handle);
1489 if (target_exists) {
1490 if (S_ISDIR(new_inode->i_mode)) {
1491 if (new_inode->i_nlink != 2 ||
1492 !ocfs2_empty_dir(new_inode)) {
1493 status = -ENOTEMPTY;
1497 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1499 OCFS2_JOURNAL_ACCESS_WRITE);
1505 /* change the dirent to point to the correct inode */
1506 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1512 inode_inc_iversion(new_dir);
1514 if (S_ISDIR(new_inode->i_mode))
1515 ocfs2_set_links_count(newfe, 0);
1517 ocfs2_add_links_count(newfe, -1);
1518 ocfs2_journal_dirty(handle, newfe_bh);
1519 if (should_add_orphan) {
1520 status = ocfs2_orphan_add(osb, handle, new_inode,
1521 newfe_bh, orphan_name,
1522 &orphan_insert, orphan_dir, false);
1529 /* if the name was not found in new_dir, add it now */
1530 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1531 OCFS2_I(old_inode)->ip_blkno,
1532 new_dir_bh, &target_insert);
1535 old_inode->i_ctime = current_time(old_inode);
1536 mark_inode_dirty(old_inode);
1538 status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1540 OCFS2_JOURNAL_ACCESS_WRITE);
1542 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1544 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1545 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1546 ocfs2_journal_dirty(handle, old_inode_bh);
1551 * Now that the name has been added to new_dir, remove the old name.
1553 * We don't keep any directory entry context around until now
1554 * because the insert might have changed the type of directory
1555 * we're dealing with.
1557 status = ocfs2_find_entry(old_dentry->d_name.name,
1558 old_dentry->d_name.len, old_dir,
1561 if (!is_journal_aborted(osb->journal->j_journal)) {
1562 ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1564 new_dentry->d_name.len, new_dentry->d_name.name,
1565 old_dentry->d_name.len, old_dentry->d_name.name);
1570 status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1573 if (!is_journal_aborted(osb->journal->j_journal)) {
1574 ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1576 new_dentry->d_name.len, new_dentry->d_name.name,
1577 old_dentry->d_name.len, old_dentry->d_name.name);
1583 drop_nlink(new_inode);
1584 new_inode->i_ctime = current_time(new_inode);
1586 old_dir->i_ctime = old_dir->i_mtime = current_time(old_dir);
1588 if (update_dot_dot) {
1589 status = ocfs2_update_entry(old_inode, handle,
1590 &old_inode_dot_dot_res, new_dir);
1591 drop_nlink(old_dir);
1593 drop_nlink(new_inode);
1596 mark_inode_dirty(new_dir);
1599 mark_inode_dirty(old_dir);
1600 ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1602 mark_inode_dirty(new_inode);
1603 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1606 if (old_dir != new_dir) {
1607 /* Keep the same times on both directories.*/
1608 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1611 * This will also pick up the i_nlink change from the
1614 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1617 if (old_dir_nlink != old_dir->i_nlink) {
1619 mlog(ML_ERROR, "need to change nlink for old dir "
1620 "%llu from %d to %d but bh is NULL!\n",
1621 (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1622 (int)old_dir_nlink, old_dir->i_nlink);
1624 struct ocfs2_dinode *fe;
1625 status = ocfs2_journal_access_di(handle,
1626 INODE_CACHE(old_dir),
1628 OCFS2_JOURNAL_ACCESS_WRITE);
1629 fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1630 ocfs2_set_links_count(fe, old_dir->i_nlink);
1631 ocfs2_journal_dirty(handle, old_dir_bh);
1634 ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1638 ocfs2_commit_trans(osb, handle);
1641 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1642 ocfs2_inode_unlock(orphan_dir, 1);
1643 inode_unlock(orphan_dir);
1647 if (new_child_locked)
1648 ocfs2_inode_unlock(new_inode, 1);
1650 if (old_child_locked)
1651 ocfs2_inode_unlock(old_inode, 1);
1654 ocfs2_double_unlock(old_dir, new_dir);
1657 ocfs2_rename_unlock(osb);
1660 sync_mapping_buffers(old_inode->i_mapping);
1664 ocfs2_free_dir_lookup_result(&target_lookup_res);
1665 ocfs2_free_dir_lookup_result(&old_entry_lookup);
1666 ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1667 ocfs2_free_dir_lookup_result(&orphan_insert);
1668 ocfs2_free_dir_lookup_result(&target_insert);
1671 brelse(old_inode_bh);
1682 * we expect i_size = strlen(symname). Copy symname into the file
1683 * data, including the null terminator.
1685 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1687 struct inode *inode,
1688 const char *symname)
1690 struct buffer_head **bhs = NULL;
1692 struct super_block *sb = osb->sb;
1693 u64 p_blkno, p_blocks;
1694 int virtual, blocks, status, i, bytes_left;
1696 bytes_left = i_size_read(inode) + 1;
1697 /* we can't trust i_blocks because we're actually going to
1698 * write i_size + 1 bytes. */
1699 blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1701 trace_ocfs2_create_symlink_data((unsigned long long)inode->i_blocks,
1702 i_size_read(inode), blocks);
1704 /* Sanity check -- make sure we're going to fit. */
1706 ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1712 bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1719 status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1726 /* links can never be larger than one cluster so we know this
1727 * is all going to be contiguous, but do a sanity check
1729 if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1736 while(bytes_left > 0) {
1737 c = &symname[virtual * sb->s_blocksize];
1739 bhs[virtual] = sb_getblk(sb, p_blkno);
1740 if (!bhs[virtual]) {
1745 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1748 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1750 OCFS2_JOURNAL_ACCESS_CREATE);
1756 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1758 memcpy(bhs[virtual]->b_data, c,
1759 (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1762 ocfs2_journal_dirty(handle, bhs[virtual]);
1766 bytes_left -= sb->s_blocksize;
1773 for(i = 0; i < blocks; i++)
1783 static int ocfs2_symlink(struct inode *dir,
1784 struct dentry *dentry,
1785 const char *symname)
1787 int status, l, credits;
1789 struct ocfs2_super *osb = NULL;
1790 struct inode *inode = NULL;
1791 struct super_block *sb;
1792 struct buffer_head *new_fe_bh = NULL;
1793 struct buffer_head *parent_fe_bh = NULL;
1794 struct ocfs2_dinode *fe = NULL;
1795 struct ocfs2_dinode *dirfe;
1796 handle_t *handle = NULL;
1797 struct ocfs2_alloc_context *inode_ac = NULL;
1798 struct ocfs2_alloc_context *data_ac = NULL;
1799 struct ocfs2_alloc_context *xattr_ac = NULL;
1800 int want_clusters = 0;
1801 int xattr_credits = 0;
1802 struct ocfs2_security_xattr_info si = {
1805 int did_quota = 0, did_quota_inode = 0;
1806 struct ocfs2_dir_lookup_result lookup = { NULL, };
1808 int did_block_signals = 0;
1809 struct ocfs2_dentry_lock *dl = NULL;
1811 trace_ocfs2_symlink_begin(dir, dentry, symname,
1812 dentry->d_name.len, dentry->d_name.name);
1814 status = dquot_initialize(dir);
1823 l = strlen(symname) + 1;
1825 credits = ocfs2_calc_symlink_credits(sb);
1827 /* lock the parent directory */
1828 status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1830 if (status != -ENOENT)
1835 dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1836 if (!ocfs2_read_links_count(dirfe)) {
1837 /* can't make a file in a deleted directory. */
1842 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1843 dentry->d_name.len);
1847 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1848 dentry->d_name.name,
1849 dentry->d_name.len, &lookup);
1855 status = ocfs2_reserve_new_inode(osb, &inode_ac);
1857 if (status != -ENOSPC)
1862 inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1863 if (IS_ERR(inode)) {
1864 status = PTR_ERR(inode);
1870 /* get security xattr */
1871 status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
1873 if (status == -EOPNOTSUPP)
1881 /* calculate meta data/clusters for setting security xattr */
1883 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1884 &xattr_credits, &xattr_ac);
1891 /* don't reserve bitmap space for fast symlinks. */
1892 if (l > ocfs2_fast_symlink_chars(sb))
1895 status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1897 if (status != -ENOSPC)
1902 handle = ocfs2_start_trans(osb, credits + xattr_credits);
1903 if (IS_ERR(handle)) {
1904 status = PTR_ERR(handle);
1910 /* Starting to change things, restart is no longer possible. */
1911 ocfs2_block_signals(&oldset);
1912 did_block_signals = 1;
1914 status = dquot_alloc_inode(inode);
1917 did_quota_inode = 1;
1919 trace_ocfs2_symlink_create(dir, dentry, dentry->d_name.len,
1920 dentry->d_name.name,
1921 (unsigned long long)OCFS2_I(dir)->ip_blkno,
1924 status = ocfs2_mknod_locked(osb, dir, inode,
1925 0, &new_fe_bh, parent_fe_bh, handle,
1932 fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1935 inode->i_op = &ocfs2_symlink_inode_operations;
1936 inode_nohighmem(inode);
1937 if (l > ocfs2_fast_symlink_chars(sb)) {
1940 status = dquot_alloc_space_nodirty(inode,
1941 ocfs2_clusters_to_bytes(osb->sb, 1));
1945 inode->i_mapping->a_ops = &ocfs2_aops;
1946 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1948 handle, data_ac, NULL,
1951 if (status != -ENOSPC && status != -EINTR) {
1953 "Failed to extend file to %llu\n",
1954 (unsigned long long)newsize);
1960 i_size_write(inode, newsize);
1961 inode->i_blocks = ocfs2_inode_sector_count(inode);
1963 inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
1964 memcpy((char *) fe->id2.i_symlink, symname, l);
1965 i_size_write(inode, newsize);
1966 inode->i_blocks = 0;
1969 status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1975 if (!ocfs2_inode_is_fast_symlink(inode)) {
1976 status = ocfs2_create_symlink_data(osb, handle, inode,
1985 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1994 * Do this before adding the entry to the directory. We add
1995 * also set d_op after success so that ->d_iput() will cleanup
1996 * the dentry lock even if ocfs2_add_entry() fails below.
1998 status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
2004 dl = dentry->d_fsdata;
2006 status = ocfs2_add_entry(handle, dentry, inode,
2007 le64_to_cpu(fe->i_blkno), parent_fe_bh,
2014 insert_inode_hash(inode);
2015 d_instantiate(dentry, inode);
2017 if (status < 0 && did_quota)
2018 dquot_free_space_nodirty(inode,
2019 ocfs2_clusters_to_bytes(osb->sb, 1));
2020 if (status < 0 && did_quota_inode)
2021 dquot_free_inode(inode);
2023 if (status < 0 && fe)
2024 ocfs2_set_links_count(fe, 0);
2025 ocfs2_commit_trans(osb, handle);
2028 ocfs2_inode_unlock(dir, 1);
2029 if (did_block_signals)
2030 ocfs2_unblock_signals(&oldset);
2033 brelse(parent_fe_bh);
2035 ocfs2_free_dir_lookup_result(&lookup);
2037 ocfs2_free_alloc_context(inode_ac);
2039 ocfs2_free_alloc_context(data_ac);
2041 ocfs2_free_alloc_context(xattr_ac);
2042 if ((status < 0) && inode) {
2044 ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
2046 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
2057 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2059 int status, namelen;
2061 namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2071 if (namelen != OCFS2_ORPHAN_NAMELEN) {
2077 trace_ocfs2_blkno_stringify(blkno, name, namelen);
2086 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
2087 struct inode **ret_orphan_dir,
2088 struct buffer_head **ret_orphan_dir_bh)
2090 struct inode *orphan_dir_inode;
2091 struct buffer_head *orphan_dir_bh = NULL;
2094 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2095 ORPHAN_DIR_SYSTEM_INODE,
2097 if (!orphan_dir_inode) {
2103 inode_lock(orphan_dir_inode);
2105 ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2107 inode_unlock(orphan_dir_inode);
2108 iput(orphan_dir_inode);
2114 *ret_orphan_dir = orphan_dir_inode;
2115 *ret_orphan_dir_bh = orphan_dir_bh;
2120 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
2121 struct buffer_head *orphan_dir_bh,
2124 struct ocfs2_dir_lookup_result *lookup,
2128 struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
2130 (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2131 OCFS2_ORPHAN_NAMELEN;
2134 ret = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2135 OCFS2_DIO_ORPHAN_PREFIX);
2136 if (ret != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2142 ret = ocfs2_blkno_stringify(blkno,
2143 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2145 ret = ocfs2_blkno_stringify(blkno, name);
2151 ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2152 orphan_dir_bh, name,
2163 * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
2164 * insertion of an orphan.
2165 * @osb: ocfs2 file system
2166 * @ret_orphan_dir: Orphan dir inode - returned locked!
2167 * @blkno: Actual block number of the inode to be inserted into orphan dir.
2168 * @lookup: dir lookup result, to be passed back into functions like
2171 * Returns zero on success and the ret_orphan_dir, name and lookup
2172 * fields will be populated.
2174 * Returns non-zero on failure.
2176 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2177 struct inode **ret_orphan_dir,
2180 struct ocfs2_dir_lookup_result *lookup,
2183 struct inode *orphan_dir_inode = NULL;
2184 struct buffer_head *orphan_dir_bh = NULL;
2187 ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
2194 ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
2195 blkno, name, lookup, dio);
2201 *ret_orphan_dir = orphan_dir_inode;
2204 brelse(orphan_dir_bh);
2207 ocfs2_inode_unlock(orphan_dir_inode, 1);
2208 inode_unlock(orphan_dir_inode);
2209 iput(orphan_dir_inode);
2217 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2219 struct inode *inode,
2220 struct buffer_head *fe_bh,
2222 struct ocfs2_dir_lookup_result *lookup,
2223 struct inode *orphan_dir_inode,
2226 struct buffer_head *orphan_dir_bh = NULL;
2228 struct ocfs2_dinode *orphan_fe;
2229 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
2231 (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2232 OCFS2_ORPHAN_NAMELEN;
2234 trace_ocfs2_orphan_add_begin(
2235 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2237 status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2243 status = ocfs2_journal_access_di(handle,
2244 INODE_CACHE(orphan_dir_inode),
2246 OCFS2_JOURNAL_ACCESS_WRITE);
2253 * We're going to journal the change of i_flags and i_orphaned_slot.
2254 * It's safe anyway, though some callers may duplicate the journaling.
2255 * Journaling within the func just make the logic look more
2258 status = ocfs2_journal_access_di(handle,
2261 OCFS2_JOURNAL_ACCESS_WRITE);
2267 /* we're a cluster, and nlink can change on disk from
2268 * underneath us... */
2269 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2270 if (S_ISDIR(inode->i_mode))
2271 ocfs2_add_links_count(orphan_fe, 1);
2272 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2273 ocfs2_journal_dirty(handle, orphan_dir_bh);
2275 status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2277 OCFS2_I(inode)->ip_blkno,
2278 orphan_dir_bh, lookup);
2285 /* Update flag OCFS2_DIO_ORPHANED_FL and record the orphan
2288 fe->i_flags |= cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2289 fe->i_dio_orphaned_slot = cpu_to_le16(osb->slot_num);
2291 fe->i_flags |= cpu_to_le32(OCFS2_ORPHANED_FL);
2292 OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2294 /* Record which orphan dir our inode now resides
2295 * in. delete_inode will use this to determine which orphan
2297 fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2300 ocfs2_journal_dirty(handle, fe_bh);
2302 trace_ocfs2_orphan_add_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
2307 if (S_ISDIR(inode->i_mode))
2308 ocfs2_add_links_count(orphan_fe, -1);
2309 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2313 brelse(orphan_dir_bh);
2318 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2319 int ocfs2_orphan_del(struct ocfs2_super *osb,
2321 struct inode *orphan_dir_inode,
2322 struct inode *inode,
2323 struct buffer_head *orphan_dir_bh,
2326 char name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2327 struct ocfs2_dinode *orphan_fe;
2329 struct ocfs2_dir_lookup_result lookup = { NULL, };
2332 status = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2333 OCFS2_DIO_ORPHAN_PREFIX);
2334 if (status != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2340 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno,
2341 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2343 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2349 trace_ocfs2_orphan_del(
2350 (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2351 name, strlen(name));
2353 status = ocfs2_journal_access_di(handle,
2354 INODE_CACHE(orphan_dir_inode),
2356 OCFS2_JOURNAL_ACCESS_WRITE);
2362 /* find it's spot in the orphan directory */
2363 status = ocfs2_find_entry(name, strlen(name), orphan_dir_inode,
2370 /* remove it from the orphan directory */
2371 status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2377 /* do the i_nlink dance! :) */
2378 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2379 if (S_ISDIR(inode->i_mode))
2380 ocfs2_add_links_count(orphan_fe, -1);
2381 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2382 ocfs2_journal_dirty(handle, orphan_dir_bh);
2385 ocfs2_free_dir_lookup_result(&lookup);
2393 * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to receive a newly
2394 * allocated file. This is different from the typical 'add to orphan dir'
2395 * operation in that the inode does not yet exist. This is a problem because
2396 * the orphan dir stringifies the inode block number to come up with it's
2397 * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2398 * problem. This function works around it by calling deeper into the orphan
2399 * and suballoc code than other callers. Use this only by necessity.
2400 * @dir: The directory which this inode will ultimately wind up under - not the
2402 * @dir_bh: buffer_head the @dir inode block
2403 * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2404 * with the string to be used for orphan dirent. Pass back to the orphan dir
2406 * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2408 * @ret_di_blkno: block number where the new inode will be allocated.
2409 * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2410 * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2412 * Returns zero on success and the ret_orphan_dir, name and lookup
2413 * fields will be populated.
2415 * Returns non-zero on failure.
2417 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2418 struct buffer_head *dir_bh,
2420 struct inode **ret_orphan_dir,
2422 struct ocfs2_dir_lookup_result *orphan_insert,
2423 struct ocfs2_alloc_context **ret_inode_ac)
2427 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2428 struct inode *orphan_dir = NULL;
2429 struct buffer_head *orphan_dir_bh = NULL;
2430 struct ocfs2_alloc_context *inode_ac = NULL;
2432 ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2438 /* reserve an inode spot */
2439 ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2446 ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2453 ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2454 di_blkno, orphan_name, orphan_insert,
2463 *ret_orphan_dir = orphan_dir;
2464 *ret_di_blkno = di_blkno;
2465 *ret_inode_ac = inode_ac;
2467 * orphan_name and orphan_insert are already up to
2468 * date via prepare_orphan_dir
2471 /* Unroll reserve_new_inode* */
2473 ocfs2_free_alloc_context(inode_ac);
2475 /* Unroll orphan dir locking */
2476 inode_unlock(orphan_dir);
2477 ocfs2_inode_unlock(orphan_dir, 1);
2481 brelse(orphan_dir_bh);
2486 int ocfs2_create_inode_in_orphan(struct inode *dir,
2488 struct inode **new_inode)
2490 int status, did_quota_inode = 0;
2491 struct inode *inode = NULL;
2492 struct inode *orphan_dir = NULL;
2493 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2494 handle_t *handle = NULL;
2495 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2496 struct buffer_head *parent_di_bh = NULL;
2497 struct buffer_head *new_di_bh = NULL;
2498 struct ocfs2_alloc_context *inode_ac = NULL;
2499 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2500 u64 di_blkno, suballoc_loc;
2503 status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2505 if (status != -ENOENT)
2510 status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2511 orphan_name, &orphan_dir,
2512 &di_blkno, &orphan_insert, &inode_ac);
2514 if (status != -ENOSPC)
2519 inode = ocfs2_get_init_inode(dir, mode);
2520 if (IS_ERR(inode)) {
2521 status = PTR_ERR(inode);
2527 handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2528 if (IS_ERR(handle)) {
2529 status = PTR_ERR(handle);
2535 status = dquot_alloc_inode(inode);
2538 did_quota_inode = 1;
2540 status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2542 &suballoc_bit, di_blkno);
2549 /* do the real work now. */
2550 status = __ocfs2_mknod_locked(dir, inode,
2551 0, &new_di_bh, parent_di_bh, handle,
2552 inode_ac, di_blkno, suballoc_loc,
2559 status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2560 &orphan_insert, orphan_dir, false);
2566 /* get open lock so that only nodes can't remove it from orphan dir. */
2567 status = ocfs2_open_lock(inode);
2571 insert_inode_hash(inode);
2573 if (status < 0 && did_quota_inode)
2574 dquot_free_inode(inode);
2576 ocfs2_commit_trans(osb, handle);
2579 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2580 ocfs2_inode_unlock(orphan_dir, 1);
2581 inode_unlock(orphan_dir);
2585 if ((status < 0) && inode) {
2591 ocfs2_free_alloc_context(inode_ac);
2598 ocfs2_free_dir_lookup_result(&orphan_insert);
2600 ocfs2_inode_unlock(dir, 1);
2601 brelse(parent_di_bh);
2605 int ocfs2_add_inode_to_orphan(struct ocfs2_super *osb,
2606 struct inode *inode)
2608 char orphan_name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2609 struct inode *orphan_dir_inode = NULL;
2610 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2611 struct buffer_head *di_bh = NULL;
2613 handle_t *handle = NULL;
2614 struct ocfs2_dinode *di = NULL;
2616 status = ocfs2_inode_lock(inode, &di_bh, 1);
2622 di = (struct ocfs2_dinode *) di_bh->b_data;
2624 * Another append dio crashed?
2625 * If so, manually recover it first.
2627 if (unlikely(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL))) {
2628 status = ocfs2_truncate_file(inode, di_bh, i_size_read(inode));
2630 if (status != -ENOSPC)
2632 goto bail_unlock_inode;
2635 status = ocfs2_del_inode_from_orphan(osb, inode, di_bh, 0, 0);
2638 goto bail_unlock_inode;
2642 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir_inode,
2643 OCFS2_I(inode)->ip_blkno,
2649 goto bail_unlock_inode;
2652 handle = ocfs2_start_trans(osb,
2653 OCFS2_INODE_ADD_TO_ORPHAN_CREDITS);
2654 if (IS_ERR(handle)) {
2655 status = PTR_ERR(handle);
2656 goto bail_unlock_orphan;
2659 status = ocfs2_orphan_add(osb, handle, inode, di_bh, orphan_name,
2660 &orphan_insert, orphan_dir_inode, true);
2664 ocfs2_commit_trans(osb, handle);
2667 ocfs2_inode_unlock(orphan_dir_inode, 1);
2668 inode_unlock(orphan_dir_inode);
2669 iput(orphan_dir_inode);
2671 ocfs2_free_dir_lookup_result(&orphan_insert);
2674 ocfs2_inode_unlock(inode, 1);
2681 int ocfs2_del_inode_from_orphan(struct ocfs2_super *osb,
2682 struct inode *inode, struct buffer_head *di_bh,
2683 int update_isize, loff_t end)
2685 struct inode *orphan_dir_inode = NULL;
2686 struct buffer_head *orphan_dir_bh = NULL;
2687 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2688 handle_t *handle = NULL;
2691 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2692 ORPHAN_DIR_SYSTEM_INODE,
2693 le16_to_cpu(di->i_dio_orphaned_slot));
2694 if (!orphan_dir_inode) {
2700 inode_lock(orphan_dir_inode);
2701 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2703 inode_unlock(orphan_dir_inode);
2704 iput(orphan_dir_inode);
2709 handle = ocfs2_start_trans(osb,
2710 OCFS2_INODE_DEL_FROM_ORPHAN_CREDITS);
2711 if (IS_ERR(handle)) {
2712 status = PTR_ERR(handle);
2713 goto bail_unlock_orphan;
2716 BUG_ON(!(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)));
2718 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode,
2719 inode, orphan_dir_bh, true);
2725 status = ocfs2_journal_access_di(handle,
2728 OCFS2_JOURNAL_ACCESS_WRITE);
2734 di->i_flags &= ~cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2735 di->i_dio_orphaned_slot = 0;
2738 status = ocfs2_set_inode_size(handle, inode, di_bh, end);
2742 ocfs2_journal_dirty(handle, di_bh);
2745 ocfs2_commit_trans(osb, handle);
2748 ocfs2_inode_unlock(orphan_dir_inode, 1);
2749 inode_unlock(orphan_dir_inode);
2750 brelse(orphan_dir_bh);
2751 iput(orphan_dir_inode);
2757 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2758 struct inode *inode,
2759 struct dentry *dentry)
2762 struct buffer_head *parent_di_bh = NULL;
2763 handle_t *handle = NULL;
2764 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2765 struct ocfs2_dinode *dir_di, *di;
2766 struct inode *orphan_dir_inode = NULL;
2767 struct buffer_head *orphan_dir_bh = NULL;
2768 struct buffer_head *di_bh = NULL;
2769 struct ocfs2_dir_lookup_result lookup = { NULL, };
2771 trace_ocfs2_mv_orphaned_inode_to_new(dir, dentry,
2772 dentry->d_name.len, dentry->d_name.name,
2773 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2774 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2776 status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2778 if (status != -ENOENT)
2783 dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2784 if (!dir_di->i_links_count) {
2785 /* can't make a file in a deleted directory. */
2790 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2791 dentry->d_name.len);
2795 /* get a spot inside the dir. */
2796 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2797 dentry->d_name.name,
2798 dentry->d_name.len, &lookup);
2804 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2805 ORPHAN_DIR_SYSTEM_INODE,
2807 if (!orphan_dir_inode) {
2813 inode_lock(orphan_dir_inode);
2815 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2818 inode_unlock(orphan_dir_inode);
2819 iput(orphan_dir_inode);
2823 status = ocfs2_read_inode_block(inode, &di_bh);
2829 handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2830 if (IS_ERR(handle)) {
2831 status = PTR_ERR(handle);
2837 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2838 di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2844 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2845 orphan_dir_bh, false);
2851 di = (struct ocfs2_dinode *)di_bh->b_data;
2852 di->i_flags &= ~cpu_to_le32(OCFS2_ORPHANED_FL);
2853 di->i_orphaned_slot = 0;
2854 set_nlink(inode, 1);
2855 ocfs2_set_links_count(di, inode->i_nlink);
2856 ocfs2_update_inode_fsync_trans(handle, inode, 1);
2857 ocfs2_journal_dirty(handle, di_bh);
2859 status = ocfs2_add_entry(handle, dentry, inode,
2860 OCFS2_I(inode)->ip_blkno, parent_di_bh,
2867 status = ocfs2_dentry_attach_lock(dentry, inode,
2868 OCFS2_I(dir)->ip_blkno);
2874 d_instantiate(dentry, inode);
2877 ocfs2_commit_trans(osb, handle);
2879 ocfs2_inode_unlock(orphan_dir_inode, 1);
2880 inode_unlock(orphan_dir_inode);
2881 iput(orphan_dir_inode);
2884 ocfs2_inode_unlock(dir, 1);
2887 brelse(parent_di_bh);
2888 brelse(orphan_dir_bh);
2890 ocfs2_free_dir_lookup_result(&lookup);
2898 const struct inode_operations ocfs2_dir_iops = {
2899 .create = ocfs2_create,
2900 .lookup = ocfs2_lookup,
2902 .unlink = ocfs2_unlink,
2903 .rmdir = ocfs2_unlink,
2904 .symlink = ocfs2_symlink,
2905 .mkdir = ocfs2_mkdir,
2906 .mknod = ocfs2_mknod,
2907 .rename = ocfs2_rename,
2908 .setattr = ocfs2_setattr,
2909 .getattr = ocfs2_getattr,
2910 .permission = ocfs2_permission,
2911 .listxattr = ocfs2_listxattr,
2912 .fiemap = ocfs2_fiemap,
2913 .get_acl = ocfs2_iop_get_acl,
2914 .set_acl = ocfs2_iop_set_acl,