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"
55 #include "buffer_head_io.h"
57 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
61 struct buffer_head **new_fe_bh,
62 struct buffer_head *parent_fe_bh,
64 struct ocfs2_alloc_context *inode_ac);
66 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
67 struct inode **ret_orphan_dir,
70 struct ocfs2_dir_lookup_result *lookup,
73 static int ocfs2_orphan_add(struct ocfs2_super *osb,
76 struct buffer_head *fe_bh,
78 struct ocfs2_dir_lookup_result *lookup,
79 struct inode *orphan_dir_inode,
82 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
87 static int ocfs2_double_lock(struct ocfs2_super *osb,
88 struct buffer_head **bh1,
90 struct buffer_head **bh2,
94 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2);
95 /* An orphan dir name is an 8 byte value, printed as a hex string */
96 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
98 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
103 struct inode *inode = NULL;
105 struct ocfs2_inode_info *oi;
107 trace_ocfs2_lookup(dir, dentry, dentry->d_name.len,
109 (unsigned long long)OCFS2_I(dir)->ip_blkno, 0);
111 if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
112 ret = ERR_PTR(-ENAMETOOLONG);
116 status = ocfs2_inode_lock_nested(dir, NULL, 0, OI_LS_PARENT);
118 if (status != -ENOENT)
120 ret = ERR_PTR(status);
124 status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
125 dentry->d_name.len, &blkno);
129 inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
131 ret = ERR_PTR(-EACCES);
136 /* Clear any orphaned state... If we were able to look up the
137 * inode from a directory, it certainly can't be orphaned. We
138 * might have the bad state from a node which intended to
139 * orphan this inode but crashed before it could commit the
141 spin_lock(&oi->ip_lock);
142 oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
143 spin_unlock(&oi->ip_lock);
146 ret = d_splice_alias(inode, dentry);
150 * If d_splice_alias() finds a DCACHE_DISCONNECTED
151 * dentry, it will d_move() it on top of ourse. The
152 * return value will indicate this however, so in
153 * those cases, we switch them around for the locking
156 * NOTE: This dentry already has ->d_op set from
157 * ocfs2_get_parent() and ocfs2_get_dentry()
159 if (!IS_ERR_OR_NULL(ret))
162 status = ocfs2_dentry_attach_lock(dentry, inode,
163 OCFS2_I(dir)->ip_blkno);
166 ret = ERR_PTR(status);
170 ocfs2_dentry_attach_gen(dentry);
173 /* Don't drop the cluster lock until *after* the d_add --
174 * unlink on another node will message us to remove that
175 * dentry under this lock so otherwise we can race this with
176 * the downconvert thread and have a stale dentry. */
177 ocfs2_inode_unlock(dir, 0);
181 trace_ocfs2_lookup_ret(ret);
186 static struct inode *ocfs2_get_init_inode(struct inode *dir, umode_t mode)
191 inode = new_inode(dir->i_sb);
193 mlog(ML_ERROR, "new_inode failed!\n");
194 return ERR_PTR(-ENOMEM);
197 /* populate as many fields early on as possible - many of
198 * these are used by the support functions here and in
202 inode_init_owner(&init_user_ns, inode, dir, mode);
203 status = dquot_initialize(inode);
205 return ERR_PTR(status);
210 static void ocfs2_cleanup_add_entry_failure(struct ocfs2_super *osb,
211 struct dentry *dentry, struct inode *inode)
213 struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
215 ocfs2_simple_drop_lockres(osb, &dl->dl_lockres);
216 ocfs2_lock_res_free(&dl->dl_lockres);
217 BUG_ON(dl->dl_count != 1);
218 spin_lock(&dentry_attach_lock);
219 dentry->d_fsdata = NULL;
220 spin_unlock(&dentry_attach_lock);
225 static int ocfs2_mknod(struct user_namespace *mnt_userns,
227 struct dentry *dentry,
232 struct buffer_head *parent_fe_bh = NULL;
233 handle_t *handle = NULL;
234 struct ocfs2_super *osb;
235 struct ocfs2_dinode *dirfe;
236 struct buffer_head *new_fe_bh = NULL;
237 struct inode *inode = NULL;
238 struct ocfs2_alloc_context *inode_ac = NULL;
239 struct ocfs2_alloc_context *data_ac = NULL;
240 struct ocfs2_alloc_context *meta_ac = NULL;
241 int want_clusters = 0;
243 int xattr_credits = 0;
244 struct ocfs2_security_xattr_info si = {
247 int did_quota_inode = 0;
248 struct ocfs2_dir_lookup_result lookup = { NULL, };
250 int did_block_signals = 0;
251 struct ocfs2_dentry_lock *dl = NULL;
253 trace_ocfs2_mknod(dir, dentry, dentry->d_name.len, dentry->d_name.name,
254 (unsigned long long)OCFS2_I(dir)->ip_blkno,
255 (unsigned long)dev, mode);
257 status = dquot_initialize(dir);
263 /* get our super block */
264 osb = OCFS2_SB(dir->i_sb);
266 status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
268 if (status != -ENOENT)
273 if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
278 dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
279 if (!ocfs2_read_links_count(dirfe)) {
280 /* can't make a file in a deleted directory. */
285 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
290 /* get a spot inside the dir. */
291 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
293 dentry->d_name.len, &lookup);
299 /* reserve an inode spot */
300 status = ocfs2_reserve_new_inode(osb, &inode_ac);
302 if (status != -ENOSPC)
307 inode = ocfs2_get_init_inode(dir, mode);
309 status = PTR_ERR(inode);
315 /* get security xattr */
316 status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
318 if (status == -EOPNOTSUPP)
326 /* calculate meta data/clusters for setting security and acl xattr */
327 status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
329 &xattr_credits, &want_meta);
335 /* Reserve a cluster if creating an extent based directory. */
336 if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
339 /* Dir indexing requires extra space as well */
340 if (ocfs2_supports_indexed_dirs(osb))
344 status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
346 if (status != -ENOSPC)
351 status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
353 if (status != -ENOSPC)
358 handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
361 if (IS_ERR(handle)) {
362 status = PTR_ERR(handle);
368 /* Starting to change things, restart is no longer possible. */
369 ocfs2_block_signals(&oldset);
370 did_block_signals = 1;
372 status = dquot_alloc_inode(inode);
377 /* do the real work now. */
378 status = ocfs2_mknod_locked(osb, dir, inode, dev,
379 &new_fe_bh, parent_fe_bh, handle,
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 ocfs2_commit_trans(osb, handle);
461 ocfs2_inode_unlock(dir, 1);
462 if (did_block_signals)
463 ocfs2_unblock_signals(&oldset);
466 brelse(parent_fe_bh);
469 ocfs2_free_dir_lookup_result(&lookup);
472 ocfs2_free_alloc_context(inode_ac);
475 ocfs2_free_alloc_context(data_ac);
478 ocfs2_free_alloc_context(meta_ac);
481 * We should call iput after the i_mutex of the bitmap been
482 * unlocked in ocfs2_free_alloc_context, or the
483 * ocfs2_delete_inode will mutex_lock again.
485 if ((status < 0) && inode) {
487 ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
489 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
500 static int __ocfs2_mknod_locked(struct inode *dir,
503 struct buffer_head **new_fe_bh,
504 struct buffer_head *parent_fe_bh,
506 struct ocfs2_alloc_context *inode_ac,
507 u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
510 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
511 struct ocfs2_dinode *fe = NULL;
512 struct ocfs2_extent_list *fel;
514 struct ocfs2_inode_info *oi = OCFS2_I(inode);
515 struct timespec64 ts;
519 /* populate as many fields early on as possible - many of
520 * these are used by the support functions here and in
522 inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
523 oi->ip_blkno = fe_blkno;
524 spin_lock(&osb->osb_lock);
525 inode->i_generation = osb->s_next_generation++;
526 spin_unlock(&osb->osb_lock);
528 *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
534 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
536 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
538 OCFS2_JOURNAL_ACCESS_CREATE);
544 fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
545 memset(fe, 0, osb->sb->s_blocksize);
547 fe->i_generation = cpu_to_le32(inode->i_generation);
548 fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
549 fe->i_blkno = cpu_to_le64(fe_blkno);
550 fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
551 fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
552 fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
553 fe->i_uid = cpu_to_le32(i_uid_read(inode));
554 fe->i_gid = cpu_to_le32(i_gid_read(inode));
555 fe->i_mode = cpu_to_le16(inode->i_mode);
556 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
557 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
559 ocfs2_set_links_count(fe, inode->i_nlink);
561 fe->i_last_eb_blk = 0;
562 strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
563 fe->i_flags |= cpu_to_le32(OCFS2_VALID_FL);
564 ktime_get_real_ts64(&ts);
565 fe->i_atime = fe->i_ctime = fe->i_mtime =
566 cpu_to_le64(ts.tv_sec);
567 fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
568 cpu_to_le32(ts.tv_nsec);
572 * If supported, directories start with inline data. If inline
573 * isn't supported, but indexing is, we start them as indexed.
575 feat = le16_to_cpu(fe->i_dyn_features);
576 if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
577 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
579 fe->id2.i_data.id_count = cpu_to_le16(
580 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
582 fel = &fe->id2.i_list;
583 fel->l_tree_depth = 0;
584 fel->l_next_free_rec = 0;
585 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
588 ocfs2_journal_dirty(handle, *new_fe_bh);
590 ocfs2_populate_inode(inode, fe, 1);
591 ocfs2_ci_set_new(osb, INODE_CACHE(inode));
592 if (!ocfs2_mount_local(osb)) {
593 status = ocfs2_create_new_inode_locks(inode);
598 ocfs2_update_inode_fsync_trans(handle, inode, 1);
613 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
617 struct buffer_head **new_fe_bh,
618 struct buffer_head *parent_fe_bh,
620 struct ocfs2_alloc_context *inode_ac)
623 u64 suballoc_loc, fe_blkno = 0;
628 status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
629 inode_ac, &suballoc_loc,
630 &suballoc_bit, &fe_blkno);
636 status = __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
637 parent_fe_bh, handle, inode_ac,
638 fe_blkno, suballoc_loc, suballoc_bit);
640 u64 bg_blkno = ocfs2_which_suballoc_group(fe_blkno, suballoc_bit);
641 int tmp = ocfs2_free_suballoc_bits(handle, inode_ac->ac_inode,
642 inode_ac->ac_bh, suballoc_bit, bg_blkno, 1);
650 static int ocfs2_mkdir(struct user_namespace *mnt_userns,
652 struct dentry *dentry,
657 trace_ocfs2_mkdir(dir, dentry, dentry->d_name.len, dentry->d_name.name,
658 OCFS2_I(dir)->ip_blkno, mode);
659 ret = ocfs2_mknod(&init_user_ns, dir, dentry, mode | S_IFDIR, 0);
666 static int ocfs2_create(struct user_namespace *mnt_userns,
668 struct dentry *dentry,
674 trace_ocfs2_create(dir, dentry, dentry->d_name.len, dentry->d_name.name,
675 (unsigned long long)OCFS2_I(dir)->ip_blkno, mode);
676 ret = ocfs2_mknod(&init_user_ns, dir, dentry, mode | S_IFREG, 0);
683 static int ocfs2_link(struct dentry *old_dentry,
685 struct dentry *dentry)
688 struct inode *inode = d_inode(old_dentry);
689 struct inode *old_dir = d_inode(old_dentry->d_parent);
691 struct buffer_head *fe_bh = NULL;
692 struct buffer_head *old_dir_bh = NULL;
693 struct buffer_head *parent_fe_bh = NULL;
694 struct ocfs2_dinode *fe = NULL;
695 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
696 struct ocfs2_dir_lookup_result lookup = { NULL, };
700 trace_ocfs2_link((unsigned long long)OCFS2_I(inode)->ip_blkno,
701 old_dentry->d_name.len, old_dentry->d_name.name,
702 dentry->d_name.len, dentry->d_name.name);
704 if (S_ISDIR(inode->i_mode))
707 err = dquot_initialize(dir);
713 err = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
714 &parent_fe_bh, dir, 0);
721 /* make sure both dirs have bhs
722 * get an extra ref on old_dir_bh if old==new */
725 parent_fe_bh = old_dir_bh;
726 get_bh(parent_fe_bh);
728 mlog(ML_ERROR, "%s: no old_dir_bh!\n", osb->uuid_str);
739 err = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
740 old_dentry->d_name.len, &old_de_ino);
747 * Check whether another node removed the source inode while we
750 if (old_de_ino != OCFS2_I(inode)->ip_blkno) {
755 err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
760 err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
762 dentry->d_name.len, &lookup);
768 err = ocfs2_inode_lock(inode, &fe_bh, 1);
775 fe = (struct ocfs2_dinode *) fe_bh->b_data;
776 if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
778 goto out_unlock_inode;
781 handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
782 if (IS_ERR(handle)) {
783 err = PTR_ERR(handle);
786 goto out_unlock_inode;
789 /* Starting to change things, restart is no longer possible. */
790 ocfs2_block_signals(&oldset);
792 err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
793 OCFS2_JOURNAL_ACCESS_WRITE);
800 inode->i_ctime = current_time(inode);
801 ocfs2_set_links_count(fe, inode->i_nlink);
802 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
803 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
804 ocfs2_journal_dirty(handle, fe_bh);
806 err = ocfs2_add_entry(handle, dentry, inode,
807 OCFS2_I(inode)->ip_blkno,
808 parent_fe_bh, &lookup);
810 ocfs2_add_links_count(fe, -1);
816 err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
823 d_instantiate(dentry, inode);
826 ocfs2_commit_trans(osb, handle);
827 ocfs2_unblock_signals(&oldset);
829 ocfs2_inode_unlock(inode, 1);
832 ocfs2_double_unlock(old_dir, dir);
835 brelse(parent_fe_bh);
838 ocfs2_free_dir_lookup_result(&lookup);
847 * Takes and drops an exclusive lock on the given dentry. This will
848 * force other nodes to drop it.
850 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
854 ret = ocfs2_dentry_lock(dentry, 1);
858 ocfs2_dentry_unlock(dentry, 1);
863 static inline int ocfs2_inode_is_unlinkable(struct inode *inode)
865 if (S_ISDIR(inode->i_mode)) {
866 if (inode->i_nlink == 2)
871 if (inode->i_nlink == 1)
876 static int ocfs2_unlink(struct inode *dir,
877 struct dentry *dentry)
880 int child_locked = 0;
881 bool is_unlinkable = false;
882 struct inode *inode = d_inode(dentry);
883 struct inode *orphan_dir = NULL;
884 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
886 struct ocfs2_dinode *fe = NULL;
887 struct buffer_head *fe_bh = NULL;
888 struct buffer_head *parent_node_bh = NULL;
889 handle_t *handle = NULL;
890 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
891 struct ocfs2_dir_lookup_result lookup = { NULL, };
892 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
894 trace_ocfs2_unlink(dir, dentry, dentry->d_name.len,
896 (unsigned long long)OCFS2_I(dir)->ip_blkno,
897 (unsigned long long)OCFS2_I(inode)->ip_blkno);
899 status = dquot_initialize(dir);
905 BUG_ON(d_inode(dentry->d_parent) != dir);
907 if (inode == osb->root_inode)
910 status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
913 if (status != -ENOENT)
918 status = ocfs2_find_files_on_disk(dentry->d_name.name,
919 dentry->d_name.len, &blkno, dir,
922 if (status != -ENOENT)
927 if (OCFS2_I(inode)->ip_blkno != blkno) {
930 trace_ocfs2_unlink_noent(
931 (unsigned long long)OCFS2_I(inode)->ip_blkno,
932 (unsigned long long)blkno,
933 OCFS2_I(inode)->ip_flags);
937 status = ocfs2_inode_lock(inode, &fe_bh, 1);
939 if (status != -ENOENT)
945 if (S_ISDIR(inode->i_mode)) {
946 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
952 status = ocfs2_remote_dentry_delete(dentry);
954 /* This remote delete should succeed under all normal
960 if (ocfs2_inode_is_unlinkable(inode)) {
961 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
962 OCFS2_I(inode)->ip_blkno,
963 orphan_name, &orphan_insert,
969 is_unlinkable = true;
972 handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
973 if (IS_ERR(handle)) {
974 status = PTR_ERR(handle);
980 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
981 OCFS2_JOURNAL_ACCESS_WRITE);
987 fe = (struct ocfs2_dinode *) fe_bh->b_data;
989 /* delete the name from the parent dir */
990 status = ocfs2_delete_entry(handle, dir, &lookup);
996 if (S_ISDIR(inode->i_mode))
999 ocfs2_set_links_count(fe, inode->i_nlink);
1000 ocfs2_journal_dirty(handle, fe_bh);
1002 dir->i_ctime = dir->i_mtime = current_time(dir);
1003 if (S_ISDIR(inode->i_mode))
1006 status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
1009 if (S_ISDIR(inode->i_mode))
1014 if (is_unlinkable) {
1015 status = ocfs2_orphan_add(osb, handle, inode, fe_bh,
1016 orphan_name, &orphan_insert, orphan_dir, false);
1023 ocfs2_commit_trans(osb, handle);
1026 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1027 ocfs2_inode_unlock(orphan_dir, 1);
1028 inode_unlock(orphan_dir);
1033 ocfs2_inode_unlock(inode, 1);
1035 ocfs2_inode_unlock(dir, 1);
1038 brelse(parent_node_bh);
1040 ocfs2_free_dir_lookup_result(&orphan_insert);
1041 ocfs2_free_dir_lookup_result(&lookup);
1043 if (status && (status != -ENOTEMPTY) && (status != -ENOENT))
1049 static int ocfs2_check_if_ancestor(struct ocfs2_super *osb,
1050 u64 src_inode_no, u64 dest_inode_no)
1053 u64 parent_inode_no = 0;
1054 u64 child_inode_no = src_inode_no;
1055 struct inode *child_inode;
1057 #define MAX_LOOKUP_TIMES 32
1059 child_inode = ocfs2_iget(osb, child_inode_no, 0, 0);
1060 if (IS_ERR(child_inode)) {
1061 ret = PTR_ERR(child_inode);
1065 ret = ocfs2_inode_lock(child_inode, NULL, 0);
1073 ret = ocfs2_lookup_ino_from_name(child_inode, "..", 2,
1075 ocfs2_inode_unlock(child_inode, 0);
1082 if (parent_inode_no == dest_inode_no) {
1087 if (parent_inode_no == osb->root_inode->i_ino) {
1092 child_inode_no = parent_inode_no;
1094 if (++i >= MAX_LOOKUP_TIMES) {
1095 mlog_ratelimited(ML_NOTICE, "max lookup times reached, "
1096 "filesystem may have nested directories, "
1097 "src inode: %llu, dest inode: %llu.\n",
1098 (unsigned long long)src_inode_no,
1099 (unsigned long long)dest_inode_no);
1109 * The only place this should be used is rename and link!
1110 * if they have the same id, then the 1st one is the only one locked.
1112 static int ocfs2_double_lock(struct ocfs2_super *osb,
1113 struct buffer_head **bh1,
1114 struct inode *inode1,
1115 struct buffer_head **bh2,
1116 struct inode *inode2,
1120 int inode1_is_ancestor, inode2_is_ancestor;
1121 struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
1122 struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
1124 trace_ocfs2_double_lock((unsigned long long)oi1->ip_blkno,
1125 (unsigned long long)oi2->ip_blkno);
1132 /* we always want to lock the one with the lower lockid first.
1133 * and if they are nested, we lock ancestor first */
1134 if (oi1->ip_blkno != oi2->ip_blkno) {
1135 inode1_is_ancestor = ocfs2_check_if_ancestor(osb, oi2->ip_blkno,
1137 if (inode1_is_ancestor < 0) {
1138 status = inode1_is_ancestor;
1142 inode2_is_ancestor = ocfs2_check_if_ancestor(osb, oi1->ip_blkno,
1144 if (inode2_is_ancestor < 0) {
1145 status = inode2_is_ancestor;
1149 if ((inode1_is_ancestor == 1) ||
1150 (oi1->ip_blkno < oi2->ip_blkno &&
1151 inode2_is_ancestor == 0)) {
1152 /* switch id1 and id2 around */
1154 swap(inode2, inode1);
1157 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
1158 rename == 1 ? OI_LS_RENAME1 : OI_LS_PARENT);
1160 if (status != -ENOENT)
1167 status = ocfs2_inode_lock_nested(inode1, bh1, 1,
1168 rename == 1 ? OI_LS_RENAME2 : OI_LS_PARENT);
1171 * An error return must mean that no cluster locks
1172 * were held on function exit.
1174 if (oi1->ip_blkno != oi2->ip_blkno) {
1175 ocfs2_inode_unlock(inode2, 1);
1180 if (status != -ENOENT)
1184 trace_ocfs2_double_lock_end(
1185 (unsigned long long)oi1->ip_blkno,
1186 (unsigned long long)oi2->ip_blkno);
1194 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1196 ocfs2_inode_unlock(inode1, 1);
1198 if (inode1 != inode2)
1199 ocfs2_inode_unlock(inode2, 1);
1202 static int ocfs2_rename(struct user_namespace *mnt_userns,
1203 struct inode *old_dir,
1204 struct dentry *old_dentry,
1205 struct inode *new_dir,
1206 struct dentry *new_dentry,
1209 int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1210 int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1211 struct inode *old_inode = d_inode(old_dentry);
1212 struct inode *new_inode = d_inode(new_dentry);
1213 struct inode *orphan_dir = NULL;
1214 struct ocfs2_dinode *newfe = NULL;
1215 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1216 struct buffer_head *newfe_bh = NULL;
1217 struct buffer_head *old_inode_bh = NULL;
1218 struct ocfs2_super *osb = NULL;
1219 u64 newfe_blkno, old_de_ino;
1220 handle_t *handle = NULL;
1221 struct buffer_head *old_dir_bh = NULL;
1222 struct buffer_head *new_dir_bh = NULL;
1223 u32 old_dir_nlink = old_dir->i_nlink;
1224 struct ocfs2_dinode *old_di;
1225 struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1226 struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1227 struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1228 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1229 struct ocfs2_dir_lookup_result target_insert = { NULL, };
1230 bool should_add_orphan = false;
1235 /* At some point it might be nice to break this function up a
1238 trace_ocfs2_rename(old_dir, old_dentry, new_dir, new_dentry,
1239 old_dentry->d_name.len, old_dentry->d_name.name,
1240 new_dentry->d_name.len, new_dentry->d_name.name);
1242 status = dquot_initialize(old_dir);
1247 status = dquot_initialize(new_dir);
1253 osb = OCFS2_SB(old_dir->i_sb);
1256 if (!igrab(new_inode))
1260 /* Assume a directory hierarchy thusly:
1263 * a,b,c, and d are all directories.
1265 * from cwd of 'a' on both nodes:
1269 * And that's why, just like the VFS, we need a file system
1271 if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1272 status = ocfs2_rename_lock(osb);
1279 /* here we cannot guarantee the inodes haven't just been
1280 * changed, so check if they are nested again */
1281 status = ocfs2_check_if_ancestor(osb, new_dir->i_ino,
1286 } else if (status == 1) {
1288 trace_ocfs2_rename_not_permitted(
1289 (unsigned long long)old_inode->i_ino,
1290 (unsigned long long)new_dir->i_ino);
1295 /* if old and new are the same, this'll just do one lock. */
1296 status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1297 &new_dir_bh, new_dir, 1);
1304 if (!new_dir->i_nlink) {
1309 /* make sure both dirs have bhs
1310 * get an extra ref on old_dir_bh if old==new */
1313 new_dir_bh = old_dir_bh;
1316 mlog(ML_ERROR, "no old_dir_bh!\n");
1323 * Aside from allowing a meta data update, the locking here
1324 * also ensures that the downconvert thread on other nodes
1325 * won't have to concurrently downconvert the inode and the
1328 status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1331 if (status != -ENOENT)
1335 old_child_locked = 1;
1337 status = ocfs2_remote_dentry_delete(old_dentry);
1343 if (S_ISDIR(old_inode->i_mode)) {
1344 u64 old_inode_parent;
1347 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1349 &old_inode_dot_dot_res);
1355 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1360 if (!new_inode && new_dir != old_dir &&
1361 new_dir->i_nlink >= ocfs2_link_max(osb)) {
1367 status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1368 old_dentry->d_name.len,
1376 * Check for inode number is _not_ due to possible IO errors.
1377 * We might rmdir the source, keep it as pwd of some process
1378 * and merrily kill the link to whatever was created under the
1379 * same name. Goodbye sticky bit ;-<
1381 if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1386 /* check if the target already exists (in which case we need
1388 status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1389 new_dentry->d_name.len,
1390 &newfe_blkno, new_dir,
1391 &target_lookup_res);
1392 /* The only error we allow here is -ENOENT because the new
1393 * file not existing is perfectly valid. */
1394 if ((status < 0) && (status != -ENOENT)) {
1395 /* If we cannot find the file specified we should just */
1396 /* return the error... */
1403 if (!target_exists && new_inode) {
1405 * Target was unlinked by another node while we were
1406 * waiting to get to ocfs2_rename(). There isn't
1407 * anything we can do here to help the situation, so
1408 * bubble up the appropriate error.
1414 /* In case we need to overwrite an existing file, we blow it
1416 if (target_exists) {
1417 /* VFS didn't think there existed an inode here, but
1418 * someone else in the cluster must have raced our
1419 * rename to create one. Today we error cleanly, in
1420 * the future we should consider calling iget to build
1421 * a new struct inode for this entry. */
1425 trace_ocfs2_rename_target_exists(new_dentry->d_name.len,
1426 new_dentry->d_name.name);
1430 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1433 trace_ocfs2_rename_disagree(
1434 (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1435 (unsigned long long)newfe_blkno,
1436 OCFS2_I(new_inode)->ip_flags);
1440 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1442 if (status != -ENOENT)
1446 new_child_locked = 1;
1448 status = ocfs2_remote_dentry_delete(new_dentry);
1454 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1456 trace_ocfs2_rename_over_existing(
1457 (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1458 (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1460 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1461 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1462 OCFS2_I(new_inode)->ip_blkno,
1463 orphan_name, &orphan_insert,
1469 should_add_orphan = true;
1472 BUG_ON(d_inode(new_dentry->d_parent) != new_dir);
1474 status = ocfs2_check_dir_for_entry(new_dir,
1475 new_dentry->d_name.name,
1476 new_dentry->d_name.len);
1480 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1481 new_dentry->d_name.name,
1482 new_dentry->d_name.len,
1490 handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1491 if (IS_ERR(handle)) {
1492 status = PTR_ERR(handle);
1498 if (target_exists) {
1499 if (S_ISDIR(new_inode->i_mode)) {
1500 if (new_inode->i_nlink != 2 ||
1501 !ocfs2_empty_dir(new_inode)) {
1502 status = -ENOTEMPTY;
1506 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1508 OCFS2_JOURNAL_ACCESS_WRITE);
1514 /* change the dirent to point to the correct inode */
1515 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1521 inode_inc_iversion(new_dir);
1523 if (S_ISDIR(new_inode->i_mode))
1524 ocfs2_set_links_count(newfe, 0);
1526 ocfs2_add_links_count(newfe, -1);
1527 ocfs2_journal_dirty(handle, newfe_bh);
1528 if (should_add_orphan) {
1529 status = ocfs2_orphan_add(osb, handle, new_inode,
1530 newfe_bh, orphan_name,
1531 &orphan_insert, orphan_dir, false);
1538 /* if the name was not found in new_dir, add it now */
1539 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1540 OCFS2_I(old_inode)->ip_blkno,
1541 new_dir_bh, &target_insert);
1544 old_inode->i_ctime = current_time(old_inode);
1545 mark_inode_dirty(old_inode);
1547 status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1549 OCFS2_JOURNAL_ACCESS_WRITE);
1551 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1553 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1554 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1555 ocfs2_journal_dirty(handle, old_inode_bh);
1560 * Now that the name has been added to new_dir, remove the old name.
1562 * We don't keep any directory entry context around until now
1563 * because the insert might have changed the type of directory
1564 * we're dealing with.
1566 status = ocfs2_find_entry(old_dentry->d_name.name,
1567 old_dentry->d_name.len, old_dir,
1570 if (!is_journal_aborted(osb->journal->j_journal)) {
1571 ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1573 new_dentry->d_name.len, new_dentry->d_name.name,
1574 old_dentry->d_name.len, old_dentry->d_name.name);
1579 status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1582 if (!is_journal_aborted(osb->journal->j_journal)) {
1583 ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1585 new_dentry->d_name.len, new_dentry->d_name.name,
1586 old_dentry->d_name.len, old_dentry->d_name.name);
1592 drop_nlink(new_inode);
1593 new_inode->i_ctime = current_time(new_inode);
1595 old_dir->i_ctime = old_dir->i_mtime = current_time(old_dir);
1597 if (update_dot_dot) {
1598 status = ocfs2_update_entry(old_inode, handle,
1599 &old_inode_dot_dot_res, new_dir);
1600 drop_nlink(old_dir);
1602 drop_nlink(new_inode);
1605 mark_inode_dirty(new_dir);
1608 mark_inode_dirty(old_dir);
1609 ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1611 mark_inode_dirty(new_inode);
1612 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1615 if (old_dir != new_dir) {
1616 /* Keep the same times on both directories.*/
1617 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1620 * This will also pick up the i_nlink change from the
1623 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1626 if (old_dir_nlink != old_dir->i_nlink) {
1628 mlog(ML_ERROR, "need to change nlink for old dir "
1629 "%llu from %d to %d but bh is NULL!\n",
1630 (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1631 (int)old_dir_nlink, old_dir->i_nlink);
1633 struct ocfs2_dinode *fe;
1634 status = ocfs2_journal_access_di(handle,
1635 INODE_CACHE(old_dir),
1637 OCFS2_JOURNAL_ACCESS_WRITE);
1638 fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1639 ocfs2_set_links_count(fe, old_dir->i_nlink);
1640 ocfs2_journal_dirty(handle, old_dir_bh);
1643 ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1647 ocfs2_commit_trans(osb, handle);
1650 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1651 ocfs2_inode_unlock(orphan_dir, 1);
1652 inode_unlock(orphan_dir);
1656 if (new_child_locked)
1657 ocfs2_inode_unlock(new_inode, 1);
1659 if (old_child_locked)
1660 ocfs2_inode_unlock(old_inode, 1);
1663 ocfs2_double_unlock(old_dir, new_dir);
1666 ocfs2_rename_unlock(osb);
1669 sync_mapping_buffers(old_inode->i_mapping);
1673 ocfs2_free_dir_lookup_result(&target_lookup_res);
1674 ocfs2_free_dir_lookup_result(&old_entry_lookup);
1675 ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1676 ocfs2_free_dir_lookup_result(&orphan_insert);
1677 ocfs2_free_dir_lookup_result(&target_insert);
1680 brelse(old_inode_bh);
1691 * we expect i_size = strlen(symname). Copy symname into the file
1692 * data, including the null terminator.
1694 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1696 struct inode *inode,
1697 const char *symname)
1699 struct buffer_head **bhs = NULL;
1701 struct super_block *sb = osb->sb;
1702 u64 p_blkno, p_blocks;
1703 int virtual, blocks, status, i, bytes_left;
1705 bytes_left = i_size_read(inode) + 1;
1706 /* we can't trust i_blocks because we're actually going to
1707 * write i_size + 1 bytes. */
1708 blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1710 trace_ocfs2_create_symlink_data((unsigned long long)inode->i_blocks,
1711 i_size_read(inode), blocks);
1713 /* Sanity check -- make sure we're going to fit. */
1715 ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1721 bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1728 status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1735 /* links can never be larger than one cluster so we know this
1736 * is all going to be contiguous, but do a sanity check
1738 if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1745 while(bytes_left > 0) {
1746 c = &symname[virtual * sb->s_blocksize];
1748 bhs[virtual] = sb_getblk(sb, p_blkno);
1749 if (!bhs[virtual]) {
1754 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1757 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1759 OCFS2_JOURNAL_ACCESS_CREATE);
1765 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1767 memcpy(bhs[virtual]->b_data, c,
1768 (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1771 ocfs2_journal_dirty(handle, bhs[virtual]);
1775 bytes_left -= sb->s_blocksize;
1782 for(i = 0; i < blocks; i++)
1792 static int ocfs2_symlink(struct user_namespace *mnt_userns,
1794 struct dentry *dentry,
1795 const char *symname)
1797 int status, l, credits;
1799 struct ocfs2_super *osb = NULL;
1800 struct inode *inode = NULL;
1801 struct super_block *sb;
1802 struct buffer_head *new_fe_bh = NULL;
1803 struct buffer_head *parent_fe_bh = NULL;
1804 struct ocfs2_dinode *fe = NULL;
1805 struct ocfs2_dinode *dirfe;
1806 handle_t *handle = NULL;
1807 struct ocfs2_alloc_context *inode_ac = NULL;
1808 struct ocfs2_alloc_context *data_ac = NULL;
1809 struct ocfs2_alloc_context *xattr_ac = NULL;
1810 int want_clusters = 0;
1811 int xattr_credits = 0;
1812 struct ocfs2_security_xattr_info si = {
1815 int did_quota = 0, did_quota_inode = 0;
1816 struct ocfs2_dir_lookup_result lookup = { NULL, };
1818 int did_block_signals = 0;
1819 struct ocfs2_dentry_lock *dl = NULL;
1821 trace_ocfs2_symlink_begin(dir, dentry, symname,
1822 dentry->d_name.len, dentry->d_name.name);
1824 status = dquot_initialize(dir);
1833 l = strlen(symname) + 1;
1835 credits = ocfs2_calc_symlink_credits(sb);
1837 /* lock the parent directory */
1838 status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1840 if (status != -ENOENT)
1845 dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1846 if (!ocfs2_read_links_count(dirfe)) {
1847 /* can't make a file in a deleted directory. */
1852 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1853 dentry->d_name.len);
1857 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1858 dentry->d_name.name,
1859 dentry->d_name.len, &lookup);
1865 status = ocfs2_reserve_new_inode(osb, &inode_ac);
1867 if (status != -ENOSPC)
1872 inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1873 if (IS_ERR(inode)) {
1874 status = PTR_ERR(inode);
1880 /* get security xattr */
1881 status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
1883 if (status == -EOPNOTSUPP)
1891 /* calculate meta data/clusters for setting security xattr */
1893 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1894 &xattr_credits, &xattr_ac);
1901 /* don't reserve bitmap space for fast symlinks. */
1902 if (l > ocfs2_fast_symlink_chars(sb))
1905 status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1907 if (status != -ENOSPC)
1912 handle = ocfs2_start_trans(osb, credits + xattr_credits);
1913 if (IS_ERR(handle)) {
1914 status = PTR_ERR(handle);
1920 /* Starting to change things, restart is no longer possible. */
1921 ocfs2_block_signals(&oldset);
1922 did_block_signals = 1;
1924 status = dquot_alloc_inode(inode);
1927 did_quota_inode = 1;
1929 trace_ocfs2_symlink_create(dir, dentry, dentry->d_name.len,
1930 dentry->d_name.name,
1931 (unsigned long long)OCFS2_I(dir)->ip_blkno,
1934 status = ocfs2_mknod_locked(osb, dir, inode,
1935 0, &new_fe_bh, parent_fe_bh, handle,
1942 fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1945 inode->i_op = &ocfs2_symlink_inode_operations;
1946 inode_nohighmem(inode);
1947 if (l > ocfs2_fast_symlink_chars(sb)) {
1950 status = dquot_alloc_space_nodirty(inode,
1951 ocfs2_clusters_to_bytes(osb->sb, 1));
1955 inode->i_mapping->a_ops = &ocfs2_aops;
1956 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1958 handle, data_ac, NULL,
1961 if (status != -ENOSPC && status != -EINTR) {
1963 "Failed to extend file to %llu\n",
1964 (unsigned long long)newsize);
1970 i_size_write(inode, newsize);
1971 inode->i_blocks = ocfs2_inode_sector_count(inode);
1973 inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
1974 memcpy((char *) fe->id2.i_symlink, symname, l);
1975 i_size_write(inode, newsize);
1976 inode->i_blocks = 0;
1979 status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1985 if (!ocfs2_inode_is_fast_symlink(inode)) {
1986 status = ocfs2_create_symlink_data(osb, handle, inode,
1995 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
2004 * Do this before adding the entry to the directory. We add
2005 * also set d_op after success so that ->d_iput() will cleanup
2006 * the dentry lock even if ocfs2_add_entry() fails below.
2008 status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
2014 dl = dentry->d_fsdata;
2016 status = ocfs2_add_entry(handle, dentry, inode,
2017 le64_to_cpu(fe->i_blkno), parent_fe_bh,
2024 insert_inode_hash(inode);
2025 d_instantiate(dentry, inode);
2027 if (status < 0 && did_quota)
2028 dquot_free_space_nodirty(inode,
2029 ocfs2_clusters_to_bytes(osb->sb, 1));
2030 if (status < 0 && did_quota_inode)
2031 dquot_free_inode(inode);
2033 ocfs2_commit_trans(osb, handle);
2035 ocfs2_inode_unlock(dir, 1);
2036 if (did_block_signals)
2037 ocfs2_unblock_signals(&oldset);
2040 brelse(parent_fe_bh);
2042 ocfs2_free_dir_lookup_result(&lookup);
2044 ocfs2_free_alloc_context(inode_ac);
2046 ocfs2_free_alloc_context(data_ac);
2048 ocfs2_free_alloc_context(xattr_ac);
2049 if ((status < 0) && inode) {
2051 ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
2053 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
2064 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2066 int status, namelen;
2068 namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2078 if (namelen != OCFS2_ORPHAN_NAMELEN) {
2084 trace_ocfs2_blkno_stringify(blkno, name, namelen);
2093 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
2094 struct inode **ret_orphan_dir,
2095 struct buffer_head **ret_orphan_dir_bh)
2097 struct inode *orphan_dir_inode;
2098 struct buffer_head *orphan_dir_bh = NULL;
2101 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2102 ORPHAN_DIR_SYSTEM_INODE,
2104 if (!orphan_dir_inode) {
2110 inode_lock(orphan_dir_inode);
2112 ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2114 inode_unlock(orphan_dir_inode);
2115 iput(orphan_dir_inode);
2121 *ret_orphan_dir = orphan_dir_inode;
2122 *ret_orphan_dir_bh = orphan_dir_bh;
2127 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
2128 struct buffer_head *orphan_dir_bh,
2131 struct ocfs2_dir_lookup_result *lookup,
2135 struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
2137 (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2138 OCFS2_ORPHAN_NAMELEN;
2141 ret = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2142 OCFS2_DIO_ORPHAN_PREFIX);
2143 if (ret != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2149 ret = ocfs2_blkno_stringify(blkno,
2150 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2152 ret = ocfs2_blkno_stringify(blkno, name);
2158 ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2159 orphan_dir_bh, name,
2170 * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
2171 * insertion of an orphan.
2172 * @osb: ocfs2 file system
2173 * @ret_orphan_dir: Orphan dir inode - returned locked!
2174 * @blkno: Actual block number of the inode to be inserted into orphan dir.
2175 * @lookup: dir lookup result, to be passed back into functions like
2178 * Returns zero on success and the ret_orphan_dir, name and lookup
2179 * fields will be populated.
2181 * Returns non-zero on failure.
2183 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2184 struct inode **ret_orphan_dir,
2187 struct ocfs2_dir_lookup_result *lookup,
2190 struct inode *orphan_dir_inode = NULL;
2191 struct buffer_head *orphan_dir_bh = NULL;
2194 ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
2201 ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
2202 blkno, name, lookup, dio);
2208 *ret_orphan_dir = orphan_dir_inode;
2211 brelse(orphan_dir_bh);
2214 ocfs2_inode_unlock(orphan_dir_inode, 1);
2215 inode_unlock(orphan_dir_inode);
2216 iput(orphan_dir_inode);
2224 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2226 struct inode *inode,
2227 struct buffer_head *fe_bh,
2229 struct ocfs2_dir_lookup_result *lookup,
2230 struct inode *orphan_dir_inode,
2233 struct buffer_head *orphan_dir_bh = NULL;
2235 struct ocfs2_dinode *orphan_fe;
2236 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
2238 (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2239 OCFS2_ORPHAN_NAMELEN;
2241 trace_ocfs2_orphan_add_begin(
2242 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2244 status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2250 status = ocfs2_journal_access_di(handle,
2251 INODE_CACHE(orphan_dir_inode),
2253 OCFS2_JOURNAL_ACCESS_WRITE);
2260 * We're going to journal the change of i_flags and i_orphaned_slot.
2261 * It's safe anyway, though some callers may duplicate the journaling.
2262 * Journaling within the func just make the logic look more
2265 status = ocfs2_journal_access_di(handle,
2268 OCFS2_JOURNAL_ACCESS_WRITE);
2274 /* we're a cluster, and nlink can change on disk from
2275 * underneath us... */
2276 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2277 if (S_ISDIR(inode->i_mode))
2278 ocfs2_add_links_count(orphan_fe, 1);
2279 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2280 ocfs2_journal_dirty(handle, orphan_dir_bh);
2282 status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2284 OCFS2_I(inode)->ip_blkno,
2285 orphan_dir_bh, lookup);
2292 /* Update flag OCFS2_DIO_ORPHANED_FL and record the orphan
2295 fe->i_flags |= cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2296 fe->i_dio_orphaned_slot = cpu_to_le16(osb->slot_num);
2298 fe->i_flags |= cpu_to_le32(OCFS2_ORPHANED_FL);
2299 OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2301 /* Record which orphan dir our inode now resides
2302 * in. delete_inode will use this to determine which orphan
2304 fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2307 ocfs2_journal_dirty(handle, fe_bh);
2309 trace_ocfs2_orphan_add_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
2314 if (S_ISDIR(inode->i_mode))
2315 ocfs2_add_links_count(orphan_fe, -1);
2316 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2320 brelse(orphan_dir_bh);
2325 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2326 int ocfs2_orphan_del(struct ocfs2_super *osb,
2328 struct inode *orphan_dir_inode,
2329 struct inode *inode,
2330 struct buffer_head *orphan_dir_bh,
2333 char name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2334 struct ocfs2_dinode *orphan_fe;
2336 struct ocfs2_dir_lookup_result lookup = { NULL, };
2339 status = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2340 OCFS2_DIO_ORPHAN_PREFIX);
2341 if (status != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2347 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno,
2348 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2350 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2356 trace_ocfs2_orphan_del(
2357 (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2358 name, strlen(name));
2360 status = ocfs2_journal_access_di(handle,
2361 INODE_CACHE(orphan_dir_inode),
2363 OCFS2_JOURNAL_ACCESS_WRITE);
2369 /* find it's spot in the orphan directory */
2370 status = ocfs2_find_entry(name, strlen(name), orphan_dir_inode,
2377 /* remove it from the orphan directory */
2378 status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2384 /* do the i_nlink dance! :) */
2385 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2386 if (S_ISDIR(inode->i_mode))
2387 ocfs2_add_links_count(orphan_fe, -1);
2388 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2389 ocfs2_journal_dirty(handle, orphan_dir_bh);
2392 ocfs2_free_dir_lookup_result(&lookup);
2400 * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to receive a newly
2401 * allocated file. This is different from the typical 'add to orphan dir'
2402 * operation in that the inode does not yet exist. This is a problem because
2403 * the orphan dir stringifies the inode block number to come up with it's
2404 * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2405 * problem. This function works around it by calling deeper into the orphan
2406 * and suballoc code than other callers. Use this only by necessity.
2407 * @dir: The directory which this inode will ultimately wind up under - not the
2409 * @dir_bh: buffer_head the @dir inode block
2410 * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2411 * with the string to be used for orphan dirent. Pass back to the orphan dir
2413 * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2415 * @ret_di_blkno: block number where the new inode will be allocated.
2416 * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2417 * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2419 * Returns zero on success and the ret_orphan_dir, name and lookup
2420 * fields will be populated.
2422 * Returns non-zero on failure.
2424 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2425 struct buffer_head *dir_bh,
2427 struct inode **ret_orphan_dir,
2429 struct ocfs2_dir_lookup_result *orphan_insert,
2430 struct ocfs2_alloc_context **ret_inode_ac)
2434 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2435 struct inode *orphan_dir = NULL;
2436 struct buffer_head *orphan_dir_bh = NULL;
2437 struct ocfs2_alloc_context *inode_ac = NULL;
2439 ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2445 /* reserve an inode spot */
2446 ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2453 ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2460 ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2461 di_blkno, orphan_name, orphan_insert,
2470 *ret_orphan_dir = orphan_dir;
2471 *ret_di_blkno = di_blkno;
2472 *ret_inode_ac = inode_ac;
2474 * orphan_name and orphan_insert are already up to
2475 * date via prepare_orphan_dir
2478 /* Unroll reserve_new_inode* */
2480 ocfs2_free_alloc_context(inode_ac);
2482 /* Unroll orphan dir locking */
2483 inode_unlock(orphan_dir);
2484 ocfs2_inode_unlock(orphan_dir, 1);
2488 brelse(orphan_dir_bh);
2493 int ocfs2_create_inode_in_orphan(struct inode *dir,
2495 struct inode **new_inode)
2497 int status, did_quota_inode = 0;
2498 struct inode *inode = NULL;
2499 struct inode *orphan_dir = NULL;
2500 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2501 handle_t *handle = NULL;
2502 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2503 struct buffer_head *parent_di_bh = NULL;
2504 struct buffer_head *new_di_bh = NULL;
2505 struct ocfs2_alloc_context *inode_ac = NULL;
2506 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2507 u64 di_blkno, suballoc_loc;
2510 status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2512 if (status != -ENOENT)
2517 status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2518 orphan_name, &orphan_dir,
2519 &di_blkno, &orphan_insert, &inode_ac);
2521 if (status != -ENOSPC)
2526 inode = ocfs2_get_init_inode(dir, mode);
2527 if (IS_ERR(inode)) {
2528 status = PTR_ERR(inode);
2534 handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2535 if (IS_ERR(handle)) {
2536 status = PTR_ERR(handle);
2542 status = dquot_alloc_inode(inode);
2545 did_quota_inode = 1;
2547 status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2549 &suballoc_bit, di_blkno);
2556 /* do the real work now. */
2557 status = __ocfs2_mknod_locked(dir, inode,
2558 0, &new_di_bh, parent_di_bh, handle,
2559 inode_ac, di_blkno, suballoc_loc,
2566 status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2567 &orphan_insert, orphan_dir, false);
2573 /* get open lock so that only nodes can't remove it from orphan dir. */
2574 status = ocfs2_open_lock(inode);
2578 insert_inode_hash(inode);
2580 if (status < 0 && did_quota_inode)
2581 dquot_free_inode(inode);
2583 ocfs2_commit_trans(osb, handle);
2586 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2587 ocfs2_inode_unlock(orphan_dir, 1);
2588 inode_unlock(orphan_dir);
2592 if ((status < 0) && inode) {
2598 ocfs2_free_alloc_context(inode_ac);
2605 ocfs2_free_dir_lookup_result(&orphan_insert);
2607 ocfs2_inode_unlock(dir, 1);
2608 brelse(parent_di_bh);
2612 int ocfs2_add_inode_to_orphan(struct ocfs2_super *osb,
2613 struct inode *inode)
2615 char orphan_name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2616 struct inode *orphan_dir_inode = NULL;
2617 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2618 struct buffer_head *di_bh = NULL;
2620 handle_t *handle = NULL;
2621 struct ocfs2_dinode *di = NULL;
2623 status = ocfs2_inode_lock(inode, &di_bh, 1);
2629 di = (struct ocfs2_dinode *) di_bh->b_data;
2631 * Another append dio crashed?
2632 * If so, manually recover it first.
2634 if (unlikely(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL))) {
2635 status = ocfs2_truncate_file(inode, di_bh, i_size_read(inode));
2637 if (status != -ENOSPC)
2639 goto bail_unlock_inode;
2642 status = ocfs2_del_inode_from_orphan(osb, inode, di_bh, 0, 0);
2645 goto bail_unlock_inode;
2649 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir_inode,
2650 OCFS2_I(inode)->ip_blkno,
2656 goto bail_unlock_inode;
2659 handle = ocfs2_start_trans(osb,
2660 OCFS2_INODE_ADD_TO_ORPHAN_CREDITS);
2661 if (IS_ERR(handle)) {
2662 status = PTR_ERR(handle);
2663 goto bail_unlock_orphan;
2666 status = ocfs2_orphan_add(osb, handle, inode, di_bh, orphan_name,
2667 &orphan_insert, orphan_dir_inode, true);
2671 ocfs2_commit_trans(osb, handle);
2674 ocfs2_inode_unlock(orphan_dir_inode, 1);
2675 inode_unlock(orphan_dir_inode);
2676 iput(orphan_dir_inode);
2678 ocfs2_free_dir_lookup_result(&orphan_insert);
2681 ocfs2_inode_unlock(inode, 1);
2688 int ocfs2_del_inode_from_orphan(struct ocfs2_super *osb,
2689 struct inode *inode, struct buffer_head *di_bh,
2690 int update_isize, loff_t end)
2692 struct inode *orphan_dir_inode = NULL;
2693 struct buffer_head *orphan_dir_bh = NULL;
2694 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2695 handle_t *handle = NULL;
2698 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2699 ORPHAN_DIR_SYSTEM_INODE,
2700 le16_to_cpu(di->i_dio_orphaned_slot));
2701 if (!orphan_dir_inode) {
2707 inode_lock(orphan_dir_inode);
2708 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2710 inode_unlock(orphan_dir_inode);
2711 iput(orphan_dir_inode);
2716 handle = ocfs2_start_trans(osb,
2717 OCFS2_INODE_DEL_FROM_ORPHAN_CREDITS);
2718 if (IS_ERR(handle)) {
2719 status = PTR_ERR(handle);
2720 goto bail_unlock_orphan;
2723 BUG_ON(!(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)));
2725 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode,
2726 inode, orphan_dir_bh, true);
2732 status = ocfs2_journal_access_di(handle,
2735 OCFS2_JOURNAL_ACCESS_WRITE);
2741 di->i_flags &= ~cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2742 di->i_dio_orphaned_slot = 0;
2745 status = ocfs2_set_inode_size(handle, inode, di_bh, end);
2749 ocfs2_journal_dirty(handle, di_bh);
2752 ocfs2_commit_trans(osb, handle);
2755 ocfs2_inode_unlock(orphan_dir_inode, 1);
2756 inode_unlock(orphan_dir_inode);
2757 brelse(orphan_dir_bh);
2758 iput(orphan_dir_inode);
2764 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2765 struct inode *inode,
2766 struct dentry *dentry)
2769 struct buffer_head *parent_di_bh = NULL;
2770 handle_t *handle = NULL;
2771 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2772 struct ocfs2_dinode *dir_di, *di;
2773 struct inode *orphan_dir_inode = NULL;
2774 struct buffer_head *orphan_dir_bh = NULL;
2775 struct buffer_head *di_bh = NULL;
2776 struct ocfs2_dir_lookup_result lookup = { NULL, };
2778 trace_ocfs2_mv_orphaned_inode_to_new(dir, dentry,
2779 dentry->d_name.len, dentry->d_name.name,
2780 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2781 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2783 status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2785 if (status != -ENOENT)
2790 dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2791 if (!dir_di->i_links_count) {
2792 /* can't make a file in a deleted directory. */
2797 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2798 dentry->d_name.len);
2802 /* get a spot inside the dir. */
2803 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2804 dentry->d_name.name,
2805 dentry->d_name.len, &lookup);
2811 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2812 ORPHAN_DIR_SYSTEM_INODE,
2814 if (!orphan_dir_inode) {
2820 inode_lock(orphan_dir_inode);
2822 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2825 inode_unlock(orphan_dir_inode);
2826 iput(orphan_dir_inode);
2830 status = ocfs2_read_inode_block(inode, &di_bh);
2836 handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2837 if (IS_ERR(handle)) {
2838 status = PTR_ERR(handle);
2844 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2845 di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2851 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2852 orphan_dir_bh, false);
2858 di = (struct ocfs2_dinode *)di_bh->b_data;
2859 di->i_flags &= ~cpu_to_le32(OCFS2_ORPHANED_FL);
2860 di->i_orphaned_slot = 0;
2861 set_nlink(inode, 1);
2862 ocfs2_set_links_count(di, inode->i_nlink);
2863 ocfs2_update_inode_fsync_trans(handle, inode, 1);
2864 ocfs2_journal_dirty(handle, di_bh);
2866 status = ocfs2_add_entry(handle, dentry, inode,
2867 OCFS2_I(inode)->ip_blkno, parent_di_bh,
2874 status = ocfs2_dentry_attach_lock(dentry, inode,
2875 OCFS2_I(dir)->ip_blkno);
2881 d_instantiate(dentry, inode);
2884 ocfs2_commit_trans(osb, handle);
2886 ocfs2_inode_unlock(orphan_dir_inode, 1);
2887 inode_unlock(orphan_dir_inode);
2888 iput(orphan_dir_inode);
2891 ocfs2_inode_unlock(dir, 1);
2894 brelse(parent_di_bh);
2895 brelse(orphan_dir_bh);
2897 ocfs2_free_dir_lookup_result(&lookup);
2905 const struct inode_operations ocfs2_dir_iops = {
2906 .create = ocfs2_create,
2907 .lookup = ocfs2_lookup,
2909 .unlink = ocfs2_unlink,
2910 .rmdir = ocfs2_unlink,
2911 .symlink = ocfs2_symlink,
2912 .mkdir = ocfs2_mkdir,
2913 .mknod = ocfs2_mknod,
2914 .rename = ocfs2_rename,
2915 .setattr = ocfs2_setattr,
2916 .getattr = ocfs2_getattr,
2917 .permission = ocfs2_permission,
2918 .listxattr = ocfs2_listxattr,
2919 .fiemap = ocfs2_fiemap,
2920 .get_acl = ocfs2_iop_get_acl,
2921 .set_acl = ocfs2_iop_set_acl,
2922 .fileattr_get = ocfs2_fileattr_get,
2923 .fileattr_set = ocfs2_fileattr_set,