ceph: fix inode reference leak
[linux/fpc-iii.git] / fs / ocfs2 / namei.c
bloba8f1225e6d9b767f5551bca0699f8e5e8f045e4c
1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
4 * namei.c
6 * Create and rename file, directory, symlinks
8 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
10 * Portions of this code from linux/fs/ext3/dir.c
12 * Copyright (C) 1992, 1993, 1994, 1995
13 * Remy Card (card@masi.ibp.fr)
14 * Laboratoire MASI - Institut Blaise pascal
15 * Universite Pierre et Marie Curie (Paris VI)
17 * from
19 * linux/fs/minix/dir.c
21 * Copyright (C) 1991, 1992 Linux Torvalds
23 * This program is free software; you can redistribute it and/or
24 * modify it under the terms of the GNU General Public
25 * License as published by the Free Software Foundation; either
26 * version 2 of the License, or (at your option) any later version.
28 * This program is distributed in the hope that it will be useful,
29 * but WITHOUT ANY WARRANTY; without even the implied warranty of
30 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
31 * General Public License for more details.
33 * You should have received a copy of the GNU General Public
34 * License along with this program; if not, write to the
35 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
36 * Boston, MA 021110-1307, USA.
39 #include <linux/fs.h>
40 #include <linux/types.h>
41 #include <linux/slab.h>
42 #include <linux/highmem.h>
43 #include <linux/quotaops.h>
45 #include <cluster/masklog.h>
47 #include "ocfs2.h"
49 #include "alloc.h"
50 #include "dcache.h"
51 #include "dir.h"
52 #include "dlmglue.h"
53 #include "extent_map.h"
54 #include "file.h"
55 #include "inode.h"
56 #include "journal.h"
57 #include "namei.h"
58 #include "suballoc.h"
59 #include "super.h"
60 #include "symlink.h"
61 #include "sysfile.h"
62 #include "uptodate.h"
63 #include "xattr.h"
64 #include "acl.h"
65 #include "ocfs2_trace.h"
67 #include "buffer_head_io.h"
69 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
70 struct inode *dir,
71 struct inode *inode,
72 dev_t dev,
73 struct buffer_head **new_fe_bh,
74 struct buffer_head *parent_fe_bh,
75 handle_t *handle,
76 struct ocfs2_alloc_context *inode_ac);
78 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
79 struct inode **ret_orphan_dir,
80 u64 blkno,
81 char *name,
82 struct ocfs2_dir_lookup_result *lookup,
83 bool dio);
85 static int ocfs2_orphan_add(struct ocfs2_super *osb,
86 handle_t *handle,
87 struct inode *inode,
88 struct buffer_head *fe_bh,
89 char *name,
90 struct ocfs2_dir_lookup_result *lookup,
91 struct inode *orphan_dir_inode,
92 bool dio);
94 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
95 handle_t *handle,
96 struct inode *inode,
97 const char *symname);
99 static int ocfs2_double_lock(struct ocfs2_super *osb,
100 struct buffer_head **bh1,
101 struct inode *inode1,
102 struct buffer_head **bh2,
103 struct inode *inode2,
104 int rename);
106 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2);
107 /* An orphan dir name is an 8 byte value, printed as a hex string */
108 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
110 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
111 unsigned int flags)
113 int status;
114 u64 blkno;
115 struct inode *inode = NULL;
116 struct dentry *ret;
117 struct ocfs2_inode_info *oi;
119 trace_ocfs2_lookup(dir, dentry, dentry->d_name.len,
120 dentry->d_name.name,
121 (unsigned long long)OCFS2_I(dir)->ip_blkno, 0);
123 if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
124 ret = ERR_PTR(-ENAMETOOLONG);
125 goto bail;
128 status = ocfs2_inode_lock_nested(dir, NULL, 0, OI_LS_PARENT);
129 if (status < 0) {
130 if (status != -ENOENT)
131 mlog_errno(status);
132 ret = ERR_PTR(status);
133 goto bail;
136 status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
137 dentry->d_name.len, &blkno);
138 if (status < 0)
139 goto bail_add;
141 inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
142 if (IS_ERR(inode)) {
143 ret = ERR_PTR(-EACCES);
144 goto bail_unlock;
147 oi = OCFS2_I(inode);
148 /* Clear any orphaned state... If we were able to look up the
149 * inode from a directory, it certainly can't be orphaned. We
150 * might have the bad state from a node which intended to
151 * orphan this inode but crashed before it could commit the
152 * unlink. */
153 spin_lock(&oi->ip_lock);
154 oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
155 spin_unlock(&oi->ip_lock);
157 bail_add:
158 ret = d_splice_alias(inode, dentry);
160 if (inode) {
162 * If d_splice_alias() finds a DCACHE_DISCONNECTED
163 * dentry, it will d_move() it on top of ourse. The
164 * return value will indicate this however, so in
165 * those cases, we switch them around for the locking
166 * code.
168 * NOTE: This dentry already has ->d_op set from
169 * ocfs2_get_parent() and ocfs2_get_dentry()
171 if (!IS_ERR_OR_NULL(ret))
172 dentry = ret;
174 status = ocfs2_dentry_attach_lock(dentry, inode,
175 OCFS2_I(dir)->ip_blkno);
176 if (status) {
177 mlog_errno(status);
178 ret = ERR_PTR(status);
179 goto bail_unlock;
181 } else
182 ocfs2_dentry_attach_gen(dentry);
184 bail_unlock:
185 /* Don't drop the cluster lock until *after* the d_add --
186 * unlink on another node will message us to remove that
187 * dentry under this lock so otherwise we can race this with
188 * the downconvert thread and have a stale dentry. */
189 ocfs2_inode_unlock(dir, 0);
191 bail:
193 trace_ocfs2_lookup_ret(ret);
195 return ret;
198 static struct inode *ocfs2_get_init_inode(struct inode *dir, umode_t mode)
200 struct inode *inode;
201 int status;
203 inode = new_inode(dir->i_sb);
204 if (!inode) {
205 mlog(ML_ERROR, "new_inode failed!\n");
206 return ERR_PTR(-ENOMEM);
209 /* populate as many fields early on as possible - many of
210 * these are used by the support functions here and in
211 * callers. */
212 if (S_ISDIR(mode))
213 set_nlink(inode, 2);
214 inode_init_owner(inode, dir, mode);
215 status = dquot_initialize(inode);
216 if (status)
217 return ERR_PTR(status);
219 return inode;
222 static void ocfs2_cleanup_add_entry_failure(struct ocfs2_super *osb,
223 struct dentry *dentry, struct inode *inode)
225 struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
227 ocfs2_simple_drop_lockres(osb, &dl->dl_lockres);
228 ocfs2_lock_res_free(&dl->dl_lockres);
229 BUG_ON(dl->dl_count != 1);
230 spin_lock(&dentry_attach_lock);
231 dentry->d_fsdata = NULL;
232 spin_unlock(&dentry_attach_lock);
233 kfree(dl);
234 iput(inode);
237 static int ocfs2_mknod(struct inode *dir,
238 struct dentry *dentry,
239 umode_t mode,
240 dev_t dev)
242 int status = 0;
243 struct buffer_head *parent_fe_bh = NULL;
244 handle_t *handle = NULL;
245 struct ocfs2_super *osb;
246 struct ocfs2_dinode *dirfe;
247 struct buffer_head *new_fe_bh = NULL;
248 struct inode *inode = NULL;
249 struct ocfs2_alloc_context *inode_ac = NULL;
250 struct ocfs2_alloc_context *data_ac = NULL;
251 struct ocfs2_alloc_context *meta_ac = NULL;
252 int want_clusters = 0;
253 int want_meta = 0;
254 int xattr_credits = 0;
255 struct ocfs2_security_xattr_info si = {
256 .enable = 1,
258 int did_quota_inode = 0;
259 struct ocfs2_dir_lookup_result lookup = { NULL, };
260 sigset_t oldset;
261 int did_block_signals = 0;
262 struct ocfs2_dentry_lock *dl = NULL;
264 trace_ocfs2_mknod(dir, dentry, dentry->d_name.len, dentry->d_name.name,
265 (unsigned long long)OCFS2_I(dir)->ip_blkno,
266 (unsigned long)dev, mode);
268 status = dquot_initialize(dir);
269 if (status) {
270 mlog_errno(status);
271 return status;
274 /* get our super block */
275 osb = OCFS2_SB(dir->i_sb);
277 status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
278 if (status < 0) {
279 if (status != -ENOENT)
280 mlog_errno(status);
281 return status;
284 if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
285 status = -EMLINK;
286 goto leave;
289 dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
290 if (!ocfs2_read_links_count(dirfe)) {
291 /* can't make a file in a deleted directory. */
292 status = -ENOENT;
293 goto leave;
296 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
297 dentry->d_name.len);
298 if (status)
299 goto leave;
301 /* get a spot inside the dir. */
302 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
303 dentry->d_name.name,
304 dentry->d_name.len, &lookup);
305 if (status < 0) {
306 mlog_errno(status);
307 goto leave;
310 /* reserve an inode spot */
311 status = ocfs2_reserve_new_inode(osb, &inode_ac);
312 if (status < 0) {
313 if (status != -ENOSPC)
314 mlog_errno(status);
315 goto leave;
318 inode = ocfs2_get_init_inode(dir, mode);
319 if (IS_ERR(inode)) {
320 status = PTR_ERR(inode);
321 inode = NULL;
322 mlog_errno(status);
323 goto leave;
326 /* get security xattr */
327 status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
328 if (status) {
329 if (status == -EOPNOTSUPP)
330 si.enable = 0;
331 else {
332 mlog_errno(status);
333 goto leave;
337 /* calculate meta data/clusters for setting security and acl xattr */
338 status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
339 &si, &want_clusters,
340 &xattr_credits, &want_meta);
341 if (status < 0) {
342 mlog_errno(status);
343 goto leave;
346 /* Reserve a cluster if creating an extent based directory. */
347 if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
348 want_clusters += 1;
350 /* Dir indexing requires extra space as well */
351 if (ocfs2_supports_indexed_dirs(osb))
352 want_meta++;
355 status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
356 if (status < 0) {
357 if (status != -ENOSPC)
358 mlog_errno(status);
359 goto leave;
362 status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
363 if (status < 0) {
364 if (status != -ENOSPC)
365 mlog_errno(status);
366 goto leave;
369 handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
370 S_ISDIR(mode),
371 xattr_credits));
372 if (IS_ERR(handle)) {
373 status = PTR_ERR(handle);
374 handle = NULL;
375 mlog_errno(status);
376 goto leave;
379 /* Starting to change things, restart is no longer possible. */
380 ocfs2_block_signals(&oldset);
381 did_block_signals = 1;
383 status = dquot_alloc_inode(inode);
384 if (status)
385 goto leave;
386 did_quota_inode = 1;
388 /* do the real work now. */
389 status = ocfs2_mknod_locked(osb, dir, inode, dev,
390 &new_fe_bh, parent_fe_bh, handle,
391 inode_ac);
392 if (status < 0) {
393 mlog_errno(status);
394 goto leave;
397 if (S_ISDIR(mode)) {
398 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
399 new_fe_bh, data_ac, meta_ac);
400 if (status < 0) {
401 mlog_errno(status);
402 goto leave;
405 status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
406 parent_fe_bh,
407 OCFS2_JOURNAL_ACCESS_WRITE);
408 if (status < 0) {
409 mlog_errno(status);
410 goto leave;
412 ocfs2_add_links_count(dirfe, 1);
413 ocfs2_journal_dirty(handle, parent_fe_bh);
414 inc_nlink(dir);
417 status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
418 meta_ac, data_ac);
420 if (status < 0) {
421 mlog_errno(status);
422 goto leave;
425 if (si.enable) {
426 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
427 meta_ac, data_ac);
428 if (status < 0) {
429 mlog_errno(status);
430 goto leave;
435 * Do this before adding the entry to the directory. We add
436 * also set d_op after success so that ->d_iput() will cleanup
437 * the dentry lock even if ocfs2_add_entry() fails below.
439 status = ocfs2_dentry_attach_lock(dentry, inode,
440 OCFS2_I(dir)->ip_blkno);
441 if (status) {
442 mlog_errno(status);
443 goto leave;
446 dl = dentry->d_fsdata;
448 status = ocfs2_add_entry(handle, dentry, inode,
449 OCFS2_I(inode)->ip_blkno, parent_fe_bh,
450 &lookup);
451 if (status < 0) {
452 mlog_errno(status);
453 goto leave;
456 insert_inode_hash(inode);
457 d_instantiate(dentry, inode);
458 status = 0;
459 leave:
460 if (status < 0 && did_quota_inode)
461 dquot_free_inode(inode);
462 if (handle)
463 ocfs2_commit_trans(osb, handle);
465 ocfs2_inode_unlock(dir, 1);
466 if (did_block_signals)
467 ocfs2_unblock_signals(&oldset);
469 brelse(new_fe_bh);
470 brelse(parent_fe_bh);
471 kfree(si.value);
473 ocfs2_free_dir_lookup_result(&lookup);
475 if (inode_ac)
476 ocfs2_free_alloc_context(inode_ac);
478 if (data_ac)
479 ocfs2_free_alloc_context(data_ac);
481 if (meta_ac)
482 ocfs2_free_alloc_context(meta_ac);
485 * We should call iput after the i_mutex of the bitmap been
486 * unlocked in ocfs2_free_alloc_context, or the
487 * ocfs2_delete_inode will mutex_lock again.
489 if ((status < 0) && inode) {
490 if (dl)
491 ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
493 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
494 clear_nlink(inode);
495 iput(inode);
498 if (status)
499 mlog_errno(status);
501 return status;
504 static int __ocfs2_mknod_locked(struct inode *dir,
505 struct inode *inode,
506 dev_t dev,
507 struct buffer_head **new_fe_bh,
508 struct buffer_head *parent_fe_bh,
509 handle_t *handle,
510 struct ocfs2_alloc_context *inode_ac,
511 u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
513 int status = 0;
514 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
515 struct ocfs2_dinode *fe = NULL;
516 struct ocfs2_extent_list *fel;
517 u16 feat;
518 struct ocfs2_inode_info *oi = OCFS2_I(inode);
520 *new_fe_bh = NULL;
522 /* populate as many fields early on as possible - many of
523 * these are used by the support functions here and in
524 * callers. */
525 inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
526 OCFS2_I(inode)->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);
532 if (!*new_fe_bh) {
533 status = -ENOMEM;
534 mlog_errno(status);
535 goto leave;
537 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
539 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
540 *new_fe_bh,
541 OCFS2_JOURNAL_ACCESS_CREATE);
542 if (status < 0) {
543 mlog_errno(status);
544 goto leave;
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 fe->i_atime = fe->i_ctime = fe->i_mtime =
568 cpu_to_le64(CURRENT_TIME.tv_sec);
569 fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
570 cpu_to_le32(CURRENT_TIME.tv_nsec);
571 fe->i_dtime = 0;
574 * If supported, directories start with inline data. If inline
575 * isn't supported, but indexing is, we start them as indexed.
577 feat = le16_to_cpu(fe->i_dyn_features);
578 if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
579 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
581 fe->id2.i_data.id_count = cpu_to_le16(
582 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
583 } else {
584 fel = &fe->id2.i_list;
585 fel->l_tree_depth = 0;
586 fel->l_next_free_rec = 0;
587 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
590 ocfs2_journal_dirty(handle, *new_fe_bh);
592 ocfs2_populate_inode(inode, fe, 1);
593 ocfs2_ci_set_new(osb, INODE_CACHE(inode));
594 if (!ocfs2_mount_local(osb)) {
595 status = ocfs2_create_new_inode_locks(inode);
596 if (status < 0)
597 mlog_errno(status);
600 oi->i_sync_tid = handle->h_transaction->t_tid;
601 oi->i_datasync_tid = handle->h_transaction->t_tid;
603 leave:
604 if (status < 0) {
605 if (*new_fe_bh) {
606 brelse(*new_fe_bh);
607 *new_fe_bh = NULL;
611 if (status)
612 mlog_errno(status);
613 return status;
616 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
617 struct inode *dir,
618 struct inode *inode,
619 dev_t dev,
620 struct buffer_head **new_fe_bh,
621 struct buffer_head *parent_fe_bh,
622 handle_t *handle,
623 struct ocfs2_alloc_context *inode_ac)
625 int status = 0;
626 u64 suballoc_loc, fe_blkno = 0;
627 u16 suballoc_bit;
629 *new_fe_bh = NULL;
631 status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
632 inode_ac, &suballoc_loc,
633 &suballoc_bit, &fe_blkno);
634 if (status < 0) {
635 mlog_errno(status);
636 return status;
639 status = __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
640 parent_fe_bh, handle, inode_ac,
641 fe_blkno, suballoc_loc, suballoc_bit);
642 if (status < 0) {
643 u64 bg_blkno = ocfs2_which_suballoc_group(fe_blkno, suballoc_bit);
644 int tmp = ocfs2_free_suballoc_bits(handle, inode_ac->ac_inode,
645 inode_ac->ac_bh, suballoc_bit, bg_blkno, 1);
646 if (tmp)
647 mlog_errno(tmp);
650 return status;
653 static int ocfs2_mkdir(struct inode *dir,
654 struct dentry *dentry,
655 umode_t mode)
657 int ret;
659 trace_ocfs2_mkdir(dir, dentry, dentry->d_name.len, dentry->d_name.name,
660 OCFS2_I(dir)->ip_blkno, mode);
661 ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
662 if (ret)
663 mlog_errno(ret);
665 return ret;
668 static int ocfs2_create(struct inode *dir,
669 struct dentry *dentry,
670 umode_t mode,
671 bool excl)
673 int ret;
675 trace_ocfs2_create(dir, dentry, dentry->d_name.len, dentry->d_name.name,
676 (unsigned long long)OCFS2_I(dir)->ip_blkno, mode);
677 ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
678 if (ret)
679 mlog_errno(ret);
681 return ret;
684 static int ocfs2_link(struct dentry *old_dentry,
685 struct inode *dir,
686 struct dentry *dentry)
688 handle_t *handle;
689 struct inode *inode = d_inode(old_dentry);
690 struct inode *old_dir = d_inode(old_dentry->d_parent);
691 int err;
692 struct buffer_head *fe_bh = NULL;
693 struct buffer_head *old_dir_bh = NULL;
694 struct buffer_head *parent_fe_bh = NULL;
695 struct ocfs2_dinode *fe = NULL;
696 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
697 struct ocfs2_dir_lookup_result lookup = { NULL, };
698 sigset_t oldset;
699 u64 old_de_ino;
701 trace_ocfs2_link((unsigned long long)OCFS2_I(inode)->ip_blkno,
702 old_dentry->d_name.len, old_dentry->d_name.name,
703 dentry->d_name.len, dentry->d_name.name);
705 if (S_ISDIR(inode->i_mode))
706 return -EPERM;
708 err = dquot_initialize(dir);
709 if (err) {
710 mlog_errno(err);
711 return err;
714 err = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
715 &parent_fe_bh, dir, 0);
716 if (err < 0) {
717 if (err != -ENOENT)
718 mlog_errno(err);
719 return err;
722 /* make sure both dirs have bhs
723 * get an extra ref on old_dir_bh if old==new */
724 if (!parent_fe_bh) {
725 if (old_dir_bh) {
726 parent_fe_bh = old_dir_bh;
727 get_bh(parent_fe_bh);
728 } else {
729 mlog(ML_ERROR, "%s: no old_dir_bh!\n", osb->uuid_str);
730 err = -EIO;
731 goto out;
735 if (!dir->i_nlink) {
736 err = -ENOENT;
737 goto out;
740 err = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
741 old_dentry->d_name.len, &old_de_ino);
742 if (err) {
743 err = -ENOENT;
744 goto out;
748 * Check whether another node removed the source inode while we
749 * were in the vfs.
751 if (old_de_ino != OCFS2_I(inode)->ip_blkno) {
752 err = -ENOENT;
753 goto out;
756 err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
757 dentry->d_name.len);
758 if (err)
759 goto out;
761 err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
762 dentry->d_name.name,
763 dentry->d_name.len, &lookup);
764 if (err < 0) {
765 mlog_errno(err);
766 goto out;
769 err = ocfs2_inode_lock(inode, &fe_bh, 1);
770 if (err < 0) {
771 if (err != -ENOENT)
772 mlog_errno(err);
773 goto out;
776 fe = (struct ocfs2_dinode *) fe_bh->b_data;
777 if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
778 err = -EMLINK;
779 goto out_unlock_inode;
782 handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
783 if (IS_ERR(handle)) {
784 err = PTR_ERR(handle);
785 handle = NULL;
786 mlog_errno(err);
787 goto out_unlock_inode;
790 /* Starting to change things, restart is no longer possible. */
791 ocfs2_block_signals(&oldset);
793 err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
794 OCFS2_JOURNAL_ACCESS_WRITE);
795 if (err < 0) {
796 mlog_errno(err);
797 goto out_commit;
800 inc_nlink(inode);
801 inode->i_ctime = CURRENT_TIME;
802 ocfs2_set_links_count(fe, inode->i_nlink);
803 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
804 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
805 ocfs2_journal_dirty(handle, fe_bh);
807 err = ocfs2_add_entry(handle, dentry, inode,
808 OCFS2_I(inode)->ip_blkno,
809 parent_fe_bh, &lookup);
810 if (err) {
811 ocfs2_add_links_count(fe, -1);
812 drop_nlink(inode);
813 mlog_errno(err);
814 goto out_commit;
817 err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
818 if (err) {
819 mlog_errno(err);
820 goto out_commit;
823 ihold(inode);
824 d_instantiate(dentry, inode);
826 out_commit:
827 ocfs2_commit_trans(osb, handle);
828 ocfs2_unblock_signals(&oldset);
829 out_unlock_inode:
830 ocfs2_inode_unlock(inode, 1);
832 out:
833 ocfs2_double_unlock(old_dir, dir);
835 brelse(fe_bh);
836 brelse(parent_fe_bh);
837 brelse(old_dir_bh);
839 ocfs2_free_dir_lookup_result(&lookup);
841 if (err)
842 mlog_errno(err);
844 return err;
848 * Takes and drops an exclusive lock on the given dentry. This will
849 * force other nodes to drop it.
851 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
853 int ret;
855 ret = ocfs2_dentry_lock(dentry, 1);
856 if (ret)
857 mlog_errno(ret);
858 else
859 ocfs2_dentry_unlock(dentry, 1);
861 return ret;
864 static inline int ocfs2_inode_is_unlinkable(struct inode *inode)
866 if (S_ISDIR(inode->i_mode)) {
867 if (inode->i_nlink == 2)
868 return 1;
869 return 0;
872 if (inode->i_nlink == 1)
873 return 1;
874 return 0;
877 static int ocfs2_unlink(struct inode *dir,
878 struct dentry *dentry)
880 int status;
881 int child_locked = 0;
882 bool is_unlinkable = false;
883 struct inode *inode = d_inode(dentry);
884 struct inode *orphan_dir = NULL;
885 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
886 u64 blkno;
887 struct ocfs2_dinode *fe = NULL;
888 struct buffer_head *fe_bh = NULL;
889 struct buffer_head *parent_node_bh = NULL;
890 handle_t *handle = NULL;
891 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
892 struct ocfs2_dir_lookup_result lookup = { NULL, };
893 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
895 trace_ocfs2_unlink(dir, dentry, dentry->d_name.len,
896 dentry->d_name.name,
897 (unsigned long long)OCFS2_I(dir)->ip_blkno,
898 (unsigned long long)OCFS2_I(inode)->ip_blkno);
900 status = dquot_initialize(dir);
901 if (status) {
902 mlog_errno(status);
903 return status;
906 BUG_ON(d_inode(dentry->d_parent) != dir);
908 if (inode == osb->root_inode)
909 return -EPERM;
911 status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
912 OI_LS_PARENT);
913 if (status < 0) {
914 if (status != -ENOENT)
915 mlog_errno(status);
916 return status;
919 status = ocfs2_find_files_on_disk(dentry->d_name.name,
920 dentry->d_name.len, &blkno, dir,
921 &lookup);
922 if (status < 0) {
923 if (status != -ENOENT)
924 mlog_errno(status);
925 goto leave;
928 if (OCFS2_I(inode)->ip_blkno != blkno) {
929 status = -ENOENT;
931 trace_ocfs2_unlink_noent(
932 (unsigned long long)OCFS2_I(inode)->ip_blkno,
933 (unsigned long long)blkno,
934 OCFS2_I(inode)->ip_flags);
935 goto leave;
938 status = ocfs2_inode_lock(inode, &fe_bh, 1);
939 if (status < 0) {
940 if (status != -ENOENT)
941 mlog_errno(status);
942 goto leave;
944 child_locked = 1;
946 if (S_ISDIR(inode->i_mode)) {
947 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
948 status = -ENOTEMPTY;
949 goto leave;
953 status = ocfs2_remote_dentry_delete(dentry);
954 if (status < 0) {
955 /* This remote delete should succeed under all normal
956 * circumstances. */
957 mlog_errno(status);
958 goto leave;
961 if (ocfs2_inode_is_unlinkable(inode)) {
962 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
963 OCFS2_I(inode)->ip_blkno,
964 orphan_name, &orphan_insert,
965 false);
966 if (status < 0) {
967 mlog_errno(status);
968 goto leave;
970 is_unlinkable = true;
973 handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
974 if (IS_ERR(handle)) {
975 status = PTR_ERR(handle);
976 handle = NULL;
977 mlog_errno(status);
978 goto leave;
981 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
982 OCFS2_JOURNAL_ACCESS_WRITE);
983 if (status < 0) {
984 mlog_errno(status);
985 goto leave;
988 fe = (struct ocfs2_dinode *) fe_bh->b_data;
990 /* delete the name from the parent dir */
991 status = ocfs2_delete_entry(handle, dir, &lookup);
992 if (status < 0) {
993 mlog_errno(status);
994 goto leave;
997 if (S_ISDIR(inode->i_mode))
998 drop_nlink(inode);
999 drop_nlink(inode);
1000 ocfs2_set_links_count(fe, inode->i_nlink);
1001 ocfs2_journal_dirty(handle, fe_bh);
1003 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1004 if (S_ISDIR(inode->i_mode))
1005 drop_nlink(dir);
1007 status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
1008 if (status < 0) {
1009 mlog_errno(status);
1010 if (S_ISDIR(inode->i_mode))
1011 inc_nlink(dir);
1012 goto leave;
1015 if (is_unlinkable) {
1016 status = ocfs2_orphan_add(osb, handle, inode, fe_bh,
1017 orphan_name, &orphan_insert, orphan_dir, false);
1018 if (status < 0)
1019 mlog_errno(status);
1022 leave:
1023 if (handle)
1024 ocfs2_commit_trans(osb, handle);
1026 if (orphan_dir) {
1027 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1028 ocfs2_inode_unlock(orphan_dir, 1);
1029 inode_unlock(orphan_dir);
1030 iput(orphan_dir);
1033 if (child_locked)
1034 ocfs2_inode_unlock(inode, 1);
1036 ocfs2_inode_unlock(dir, 1);
1038 brelse(fe_bh);
1039 brelse(parent_node_bh);
1041 ocfs2_free_dir_lookup_result(&orphan_insert);
1042 ocfs2_free_dir_lookup_result(&lookup);
1044 if (status && (status != -ENOTEMPTY) && (status != -ENOENT))
1045 mlog_errno(status);
1047 return status;
1050 static int ocfs2_check_if_ancestor(struct ocfs2_super *osb,
1051 u64 src_inode_no, u64 dest_inode_no)
1053 int ret = 0, i = 0;
1054 u64 parent_inode_no = 0;
1055 u64 child_inode_no = src_inode_no;
1056 struct inode *child_inode;
1058 #define MAX_LOOKUP_TIMES 32
1059 while (1) {
1060 child_inode = ocfs2_iget(osb, child_inode_no, 0, 0);
1061 if (IS_ERR(child_inode)) {
1062 ret = PTR_ERR(child_inode);
1063 break;
1066 ret = ocfs2_inode_lock(child_inode, NULL, 0);
1067 if (ret < 0) {
1068 iput(child_inode);
1069 if (ret != -ENOENT)
1070 mlog_errno(ret);
1071 break;
1074 ret = ocfs2_lookup_ino_from_name(child_inode, "..", 2,
1075 &parent_inode_no);
1076 ocfs2_inode_unlock(child_inode, 0);
1077 iput(child_inode);
1078 if (ret < 0) {
1079 ret = -ENOENT;
1080 break;
1083 if (parent_inode_no == dest_inode_no) {
1084 ret = 1;
1085 break;
1088 if (parent_inode_no == osb->root_inode->i_ino) {
1089 ret = 0;
1090 break;
1093 child_inode_no = parent_inode_no;
1095 if (++i >= MAX_LOOKUP_TIMES) {
1096 mlog(ML_NOTICE, "max lookup times reached, filesystem "
1097 "may have nested directories, "
1098 "src inode: %llu, dest inode: %llu.\n",
1099 (unsigned long long)src_inode_no,
1100 (unsigned long long)dest_inode_no);
1101 ret = 0;
1102 break;
1106 return ret;
1110 * The only place this should be used is rename and link!
1111 * if they have the same id, then the 1st one is the only one locked.
1113 static int ocfs2_double_lock(struct ocfs2_super *osb,
1114 struct buffer_head **bh1,
1115 struct inode *inode1,
1116 struct buffer_head **bh2,
1117 struct inode *inode2,
1118 int rename)
1120 int status;
1121 int inode1_is_ancestor, inode2_is_ancestor;
1122 struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
1123 struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
1125 trace_ocfs2_double_lock((unsigned long long)oi1->ip_blkno,
1126 (unsigned long long)oi2->ip_blkno);
1128 if (*bh1)
1129 *bh1 = NULL;
1130 if (*bh2)
1131 *bh2 = NULL;
1133 /* we always want to lock the one with the lower lockid first.
1134 * and if they are nested, we lock ancestor first */
1135 if (oi1->ip_blkno != oi2->ip_blkno) {
1136 inode1_is_ancestor = ocfs2_check_if_ancestor(osb, oi2->ip_blkno,
1137 oi1->ip_blkno);
1138 if (inode1_is_ancestor < 0) {
1139 status = inode1_is_ancestor;
1140 goto bail;
1143 inode2_is_ancestor = ocfs2_check_if_ancestor(osb, oi1->ip_blkno,
1144 oi2->ip_blkno);
1145 if (inode2_is_ancestor < 0) {
1146 status = inode2_is_ancestor;
1147 goto bail;
1150 if ((inode1_is_ancestor == 1) ||
1151 (oi1->ip_blkno < oi2->ip_blkno &&
1152 inode2_is_ancestor == 0)) {
1153 /* switch id1 and id2 around */
1154 swap(bh2, bh1);
1155 swap(inode2, inode1);
1157 /* lock id2 */
1158 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
1159 rename == 1 ? OI_LS_RENAME1 : OI_LS_PARENT);
1160 if (status < 0) {
1161 if (status != -ENOENT)
1162 mlog_errno(status);
1163 goto bail;
1167 /* lock id1 */
1168 status = ocfs2_inode_lock_nested(inode1, bh1, 1,
1169 rename == 1 ? OI_LS_RENAME2 : OI_LS_PARENT);
1170 if (status < 0) {
1172 * An error return must mean that no cluster locks
1173 * were held on function exit.
1175 if (oi1->ip_blkno != oi2->ip_blkno) {
1176 ocfs2_inode_unlock(inode2, 1);
1177 brelse(*bh2);
1178 *bh2 = NULL;
1181 if (status != -ENOENT)
1182 mlog_errno(status);
1185 trace_ocfs2_double_lock_end(
1186 (unsigned long long)OCFS2_I(inode1)->ip_blkno,
1187 (unsigned long long)OCFS2_I(inode2)->ip_blkno);
1189 bail:
1190 if (status)
1191 mlog_errno(status);
1192 return status;
1195 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1197 ocfs2_inode_unlock(inode1, 1);
1199 if (inode1 != inode2)
1200 ocfs2_inode_unlock(inode2, 1);
1203 static int ocfs2_rename(struct inode *old_dir,
1204 struct dentry *old_dentry,
1205 struct inode *new_dir,
1206 struct dentry *new_dentry)
1208 int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1209 int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1210 struct inode *old_inode = d_inode(old_dentry);
1211 struct inode *new_inode = d_inode(new_dentry);
1212 struct inode *orphan_dir = NULL;
1213 struct ocfs2_dinode *newfe = NULL;
1214 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1215 struct buffer_head *newfe_bh = NULL;
1216 struct buffer_head *old_inode_bh = NULL;
1217 struct ocfs2_super *osb = NULL;
1218 u64 newfe_blkno, old_de_ino;
1219 handle_t *handle = NULL;
1220 struct buffer_head *old_dir_bh = NULL;
1221 struct buffer_head *new_dir_bh = NULL;
1222 u32 old_dir_nlink = old_dir->i_nlink;
1223 struct ocfs2_dinode *old_di;
1224 struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1225 struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1226 struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1227 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1228 struct ocfs2_dir_lookup_result target_insert = { NULL, };
1229 bool should_add_orphan = false;
1231 /* At some point it might be nice to break this function up a
1232 * bit. */
1234 trace_ocfs2_rename(old_dir, old_dentry, new_dir, new_dentry,
1235 old_dentry->d_name.len, old_dentry->d_name.name,
1236 new_dentry->d_name.len, new_dentry->d_name.name);
1238 status = dquot_initialize(old_dir);
1239 if (status) {
1240 mlog_errno(status);
1241 goto bail;
1243 status = dquot_initialize(new_dir);
1244 if (status) {
1245 mlog_errno(status);
1246 goto bail;
1249 osb = OCFS2_SB(old_dir->i_sb);
1251 if (new_inode) {
1252 if (!igrab(new_inode))
1253 BUG();
1256 /* Assume a directory hierarchy thusly:
1257 * a/b/c
1258 * a/d
1259 * a,b,c, and d are all directories.
1261 * from cwd of 'a' on both nodes:
1262 * node1: mv b/c d
1263 * node2: mv d b/c
1265 * And that's why, just like the VFS, we need a file system
1266 * rename lock. */
1267 if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1268 status = ocfs2_rename_lock(osb);
1269 if (status < 0) {
1270 mlog_errno(status);
1271 goto bail;
1273 rename_lock = 1;
1275 /* here we cannot guarantee the inodes haven't just been
1276 * changed, so check if they are nested again */
1277 status = ocfs2_check_if_ancestor(osb, new_dir->i_ino,
1278 old_inode->i_ino);
1279 if (status < 0) {
1280 mlog_errno(status);
1281 goto bail;
1282 } else if (status == 1) {
1283 status = -EPERM;
1284 trace_ocfs2_rename_not_permitted(
1285 (unsigned long long)old_inode->i_ino,
1286 (unsigned long long)new_dir->i_ino);
1287 goto bail;
1291 /* if old and new are the same, this'll just do one lock. */
1292 status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1293 &new_dir_bh, new_dir, 1);
1294 if (status < 0) {
1295 mlog_errno(status);
1296 goto bail;
1298 parents_locked = 1;
1300 if (!new_dir->i_nlink) {
1301 status = -EACCES;
1302 goto bail;
1305 /* make sure both dirs have bhs
1306 * get an extra ref on old_dir_bh if old==new */
1307 if (!new_dir_bh) {
1308 if (old_dir_bh) {
1309 new_dir_bh = old_dir_bh;
1310 get_bh(new_dir_bh);
1311 } else {
1312 mlog(ML_ERROR, "no old_dir_bh!\n");
1313 status = -EIO;
1314 goto bail;
1319 * Aside from allowing a meta data update, the locking here
1320 * also ensures that the downconvert thread on other nodes
1321 * won't have to concurrently downconvert the inode and the
1322 * dentry locks.
1324 status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1325 OI_LS_PARENT);
1326 if (status < 0) {
1327 if (status != -ENOENT)
1328 mlog_errno(status);
1329 goto bail;
1331 old_child_locked = 1;
1333 status = ocfs2_remote_dentry_delete(old_dentry);
1334 if (status < 0) {
1335 mlog_errno(status);
1336 goto bail;
1339 if (S_ISDIR(old_inode->i_mode)) {
1340 u64 old_inode_parent;
1342 update_dot_dot = 1;
1343 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1344 old_inode,
1345 &old_inode_dot_dot_res);
1346 if (status) {
1347 status = -EIO;
1348 goto bail;
1351 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1352 status = -EIO;
1353 goto bail;
1356 if (!new_inode && new_dir != old_dir &&
1357 new_dir->i_nlink >= ocfs2_link_max(osb)) {
1358 status = -EMLINK;
1359 goto bail;
1363 status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1364 old_dentry->d_name.len,
1365 &old_de_ino);
1366 if (status) {
1367 status = -ENOENT;
1368 goto bail;
1372 * Check for inode number is _not_ due to possible IO errors.
1373 * We might rmdir the source, keep it as pwd of some process
1374 * and merrily kill the link to whatever was created under the
1375 * same name. Goodbye sticky bit ;-<
1377 if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1378 status = -ENOENT;
1379 goto bail;
1382 /* check if the target already exists (in which case we need
1383 * to delete it */
1384 status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1385 new_dentry->d_name.len,
1386 &newfe_blkno, new_dir,
1387 &target_lookup_res);
1388 /* The only error we allow here is -ENOENT because the new
1389 * file not existing is perfectly valid. */
1390 if ((status < 0) && (status != -ENOENT)) {
1391 /* If we cannot find the file specified we should just */
1392 /* return the error... */
1393 mlog_errno(status);
1394 goto bail;
1396 if (status == 0)
1397 target_exists = 1;
1399 if (!target_exists && new_inode) {
1401 * Target was unlinked by another node while we were
1402 * waiting to get to ocfs2_rename(). There isn't
1403 * anything we can do here to help the situation, so
1404 * bubble up the appropriate error.
1406 status = -ENOENT;
1407 goto bail;
1410 /* In case we need to overwrite an existing file, we blow it
1411 * away first */
1412 if (target_exists) {
1413 /* VFS didn't think there existed an inode here, but
1414 * someone else in the cluster must have raced our
1415 * rename to create one. Today we error cleanly, in
1416 * the future we should consider calling iget to build
1417 * a new struct inode for this entry. */
1418 if (!new_inode) {
1419 status = -EACCES;
1421 trace_ocfs2_rename_target_exists(new_dentry->d_name.len,
1422 new_dentry->d_name.name);
1423 goto bail;
1426 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1427 status = -EACCES;
1429 trace_ocfs2_rename_disagree(
1430 (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1431 (unsigned long long)newfe_blkno,
1432 OCFS2_I(new_inode)->ip_flags);
1433 goto bail;
1436 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1437 if (status < 0) {
1438 if (status != -ENOENT)
1439 mlog_errno(status);
1440 goto bail;
1442 new_child_locked = 1;
1444 status = ocfs2_remote_dentry_delete(new_dentry);
1445 if (status < 0) {
1446 mlog_errno(status);
1447 goto bail;
1450 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1452 trace_ocfs2_rename_over_existing(
1453 (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1454 (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1456 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1457 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1458 OCFS2_I(new_inode)->ip_blkno,
1459 orphan_name, &orphan_insert,
1460 false);
1461 if (status < 0) {
1462 mlog_errno(status);
1463 goto bail;
1465 should_add_orphan = true;
1467 } else {
1468 BUG_ON(d_inode(new_dentry->d_parent) != new_dir);
1470 status = ocfs2_check_dir_for_entry(new_dir,
1471 new_dentry->d_name.name,
1472 new_dentry->d_name.len);
1473 if (status)
1474 goto bail;
1476 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1477 new_dentry->d_name.name,
1478 new_dentry->d_name.len,
1479 &target_insert);
1480 if (status < 0) {
1481 mlog_errno(status);
1482 goto bail;
1486 handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1487 if (IS_ERR(handle)) {
1488 status = PTR_ERR(handle);
1489 handle = NULL;
1490 mlog_errno(status);
1491 goto bail;
1494 if (target_exists) {
1495 if (S_ISDIR(new_inode->i_mode)) {
1496 if (new_inode->i_nlink != 2 ||
1497 !ocfs2_empty_dir(new_inode)) {
1498 status = -ENOTEMPTY;
1499 goto bail;
1502 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1503 newfe_bh,
1504 OCFS2_JOURNAL_ACCESS_WRITE);
1505 if (status < 0) {
1506 mlog_errno(status);
1507 goto bail;
1510 /* change the dirent to point to the correct inode */
1511 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1512 old_inode);
1513 if (status < 0) {
1514 mlog_errno(status);
1515 goto bail;
1517 new_dir->i_version++;
1519 if (S_ISDIR(new_inode->i_mode))
1520 ocfs2_set_links_count(newfe, 0);
1521 else
1522 ocfs2_add_links_count(newfe, -1);
1523 ocfs2_journal_dirty(handle, newfe_bh);
1524 if (should_add_orphan) {
1525 status = ocfs2_orphan_add(osb, handle, new_inode,
1526 newfe_bh, orphan_name,
1527 &orphan_insert, orphan_dir, false);
1528 if (status < 0) {
1529 mlog_errno(status);
1530 goto bail;
1533 } else {
1534 /* if the name was not found in new_dir, add it now */
1535 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1536 OCFS2_I(old_inode)->ip_blkno,
1537 new_dir_bh, &target_insert);
1540 old_inode->i_ctime = CURRENT_TIME;
1541 mark_inode_dirty(old_inode);
1543 status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1544 old_inode_bh,
1545 OCFS2_JOURNAL_ACCESS_WRITE);
1546 if (status >= 0) {
1547 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1549 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1550 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1551 ocfs2_journal_dirty(handle, old_inode_bh);
1552 } else
1553 mlog_errno(status);
1556 * Now that the name has been added to new_dir, remove the old name.
1558 * We don't keep any directory entry context around until now
1559 * because the insert might have changed the type of directory
1560 * we're dealing with.
1562 status = ocfs2_find_entry(old_dentry->d_name.name,
1563 old_dentry->d_name.len, old_dir,
1564 &old_entry_lookup);
1565 if (status) {
1566 if (!is_journal_aborted(osb->journal->j_journal)) {
1567 ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1568 "is not deleted.",
1569 new_dentry->d_name.len, new_dentry->d_name.name,
1570 old_dentry->d_name.len, old_dentry->d_name.name);
1572 goto bail;
1575 status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1576 if (status < 0) {
1577 mlog_errno(status);
1578 if (!is_journal_aborted(osb->journal->j_journal)) {
1579 ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1580 "is not deleted.",
1581 new_dentry->d_name.len, new_dentry->d_name.name,
1582 old_dentry->d_name.len, old_dentry->d_name.name);
1584 goto bail;
1587 if (new_inode) {
1588 drop_nlink(new_inode);
1589 new_inode->i_ctime = CURRENT_TIME;
1591 old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1593 if (update_dot_dot) {
1594 status = ocfs2_update_entry(old_inode, handle,
1595 &old_inode_dot_dot_res, new_dir);
1596 drop_nlink(old_dir);
1597 if (new_inode) {
1598 drop_nlink(new_inode);
1599 } else {
1600 inc_nlink(new_dir);
1601 mark_inode_dirty(new_dir);
1604 mark_inode_dirty(old_dir);
1605 ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1606 if (new_inode) {
1607 mark_inode_dirty(new_inode);
1608 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1611 if (old_dir != new_dir) {
1612 /* Keep the same times on both directories.*/
1613 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1616 * This will also pick up the i_nlink change from the
1617 * block above.
1619 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1622 if (old_dir_nlink != old_dir->i_nlink) {
1623 if (!old_dir_bh) {
1624 mlog(ML_ERROR, "need to change nlink for old dir "
1625 "%llu from %d to %d but bh is NULL!\n",
1626 (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1627 (int)old_dir_nlink, old_dir->i_nlink);
1628 } else {
1629 struct ocfs2_dinode *fe;
1630 status = ocfs2_journal_access_di(handle,
1631 INODE_CACHE(old_dir),
1632 old_dir_bh,
1633 OCFS2_JOURNAL_ACCESS_WRITE);
1634 fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1635 ocfs2_set_links_count(fe, old_dir->i_nlink);
1636 ocfs2_journal_dirty(handle, old_dir_bh);
1639 ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1640 status = 0;
1641 bail:
1642 if (handle)
1643 ocfs2_commit_trans(osb, handle);
1645 if (orphan_dir) {
1646 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1647 ocfs2_inode_unlock(orphan_dir, 1);
1648 inode_unlock(orphan_dir);
1649 iput(orphan_dir);
1652 if (new_child_locked)
1653 ocfs2_inode_unlock(new_inode, 1);
1655 if (old_child_locked)
1656 ocfs2_inode_unlock(old_inode, 1);
1658 if (parents_locked)
1659 ocfs2_double_unlock(old_dir, new_dir);
1661 if (rename_lock)
1662 ocfs2_rename_unlock(osb);
1664 if (new_inode)
1665 sync_mapping_buffers(old_inode->i_mapping);
1667 iput(new_inode);
1669 ocfs2_free_dir_lookup_result(&target_lookup_res);
1670 ocfs2_free_dir_lookup_result(&old_entry_lookup);
1671 ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1672 ocfs2_free_dir_lookup_result(&orphan_insert);
1673 ocfs2_free_dir_lookup_result(&target_insert);
1675 brelse(newfe_bh);
1676 brelse(old_inode_bh);
1677 brelse(old_dir_bh);
1678 brelse(new_dir_bh);
1680 if (status)
1681 mlog_errno(status);
1683 return status;
1687 * we expect i_size = strlen(symname). Copy symname into the file
1688 * data, including the null terminator.
1690 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1691 handle_t *handle,
1692 struct inode *inode,
1693 const char *symname)
1695 struct buffer_head **bhs = NULL;
1696 const char *c;
1697 struct super_block *sb = osb->sb;
1698 u64 p_blkno, p_blocks;
1699 int virtual, blocks, status, i, bytes_left;
1701 bytes_left = i_size_read(inode) + 1;
1702 /* we can't trust i_blocks because we're actually going to
1703 * write i_size + 1 bytes. */
1704 blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1706 trace_ocfs2_create_symlink_data((unsigned long long)inode->i_blocks,
1707 i_size_read(inode), blocks);
1709 /* Sanity check -- make sure we're going to fit. */
1710 if (bytes_left >
1711 ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1712 status = -EIO;
1713 mlog_errno(status);
1714 goto bail;
1717 bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1718 if (!bhs) {
1719 status = -ENOMEM;
1720 mlog_errno(status);
1721 goto bail;
1724 status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1725 NULL);
1726 if (status < 0) {
1727 mlog_errno(status);
1728 goto bail;
1731 /* links can never be larger than one cluster so we know this
1732 * is all going to be contiguous, but do a sanity check
1733 * anyway. */
1734 if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1735 status = -EIO;
1736 mlog_errno(status);
1737 goto bail;
1740 virtual = 0;
1741 while(bytes_left > 0) {
1742 c = &symname[virtual * sb->s_blocksize];
1744 bhs[virtual] = sb_getblk(sb, p_blkno);
1745 if (!bhs[virtual]) {
1746 status = -ENOMEM;
1747 mlog_errno(status);
1748 goto bail;
1750 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1751 bhs[virtual]);
1753 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1754 bhs[virtual],
1755 OCFS2_JOURNAL_ACCESS_CREATE);
1756 if (status < 0) {
1757 mlog_errno(status);
1758 goto bail;
1761 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1763 memcpy(bhs[virtual]->b_data, c,
1764 (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1765 bytes_left);
1767 ocfs2_journal_dirty(handle, bhs[virtual]);
1769 virtual++;
1770 p_blkno++;
1771 bytes_left -= sb->s_blocksize;
1774 status = 0;
1775 bail:
1777 if (bhs) {
1778 for(i = 0; i < blocks; i++)
1779 brelse(bhs[i]);
1780 kfree(bhs);
1783 if (status)
1784 mlog_errno(status);
1785 return status;
1788 static int ocfs2_symlink(struct inode *dir,
1789 struct dentry *dentry,
1790 const char *symname)
1792 int status, l, credits;
1793 u64 newsize;
1794 struct ocfs2_super *osb = NULL;
1795 struct inode *inode = NULL;
1796 struct super_block *sb;
1797 struct buffer_head *new_fe_bh = NULL;
1798 struct buffer_head *parent_fe_bh = NULL;
1799 struct ocfs2_dinode *fe = NULL;
1800 struct ocfs2_dinode *dirfe;
1801 handle_t *handle = NULL;
1802 struct ocfs2_alloc_context *inode_ac = NULL;
1803 struct ocfs2_alloc_context *data_ac = NULL;
1804 struct ocfs2_alloc_context *xattr_ac = NULL;
1805 int want_clusters = 0;
1806 int xattr_credits = 0;
1807 struct ocfs2_security_xattr_info si = {
1808 .enable = 1,
1810 int did_quota = 0, did_quota_inode = 0;
1811 struct ocfs2_dir_lookup_result lookup = { NULL, };
1812 sigset_t oldset;
1813 int did_block_signals = 0;
1814 struct ocfs2_dentry_lock *dl = NULL;
1816 trace_ocfs2_symlink_begin(dir, dentry, symname,
1817 dentry->d_name.len, dentry->d_name.name);
1819 status = dquot_initialize(dir);
1820 if (status) {
1821 mlog_errno(status);
1822 goto bail;
1825 sb = dir->i_sb;
1826 osb = OCFS2_SB(sb);
1828 l = strlen(symname) + 1;
1830 credits = ocfs2_calc_symlink_credits(sb);
1832 /* lock the parent directory */
1833 status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1834 if (status < 0) {
1835 if (status != -ENOENT)
1836 mlog_errno(status);
1837 return status;
1840 dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1841 if (!ocfs2_read_links_count(dirfe)) {
1842 /* can't make a file in a deleted directory. */
1843 status = -ENOENT;
1844 goto bail;
1847 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1848 dentry->d_name.len);
1849 if (status)
1850 goto bail;
1852 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1853 dentry->d_name.name,
1854 dentry->d_name.len, &lookup);
1855 if (status < 0) {
1856 mlog_errno(status);
1857 goto bail;
1860 status = ocfs2_reserve_new_inode(osb, &inode_ac);
1861 if (status < 0) {
1862 if (status != -ENOSPC)
1863 mlog_errno(status);
1864 goto bail;
1867 inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1868 if (IS_ERR(inode)) {
1869 status = PTR_ERR(inode);
1870 inode = NULL;
1871 mlog_errno(status);
1872 goto bail;
1875 /* get security xattr */
1876 status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
1877 if (status) {
1878 if (status == -EOPNOTSUPP)
1879 si.enable = 0;
1880 else {
1881 mlog_errno(status);
1882 goto bail;
1886 /* calculate meta data/clusters for setting security xattr */
1887 if (si.enable) {
1888 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1889 &xattr_credits, &xattr_ac);
1890 if (status < 0) {
1891 mlog_errno(status);
1892 goto bail;
1896 /* don't reserve bitmap space for fast symlinks. */
1897 if (l > ocfs2_fast_symlink_chars(sb))
1898 want_clusters += 1;
1900 status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1901 if (status < 0) {
1902 if (status != -ENOSPC)
1903 mlog_errno(status);
1904 goto bail;
1907 handle = ocfs2_start_trans(osb, credits + xattr_credits);
1908 if (IS_ERR(handle)) {
1909 status = PTR_ERR(handle);
1910 handle = NULL;
1911 mlog_errno(status);
1912 goto bail;
1915 /* Starting to change things, restart is no longer possible. */
1916 ocfs2_block_signals(&oldset);
1917 did_block_signals = 1;
1919 status = dquot_alloc_inode(inode);
1920 if (status)
1921 goto bail;
1922 did_quota_inode = 1;
1924 trace_ocfs2_symlink_create(dir, dentry, dentry->d_name.len,
1925 dentry->d_name.name,
1926 (unsigned long long)OCFS2_I(dir)->ip_blkno,
1927 inode->i_mode);
1929 status = ocfs2_mknod_locked(osb, dir, inode,
1930 0, &new_fe_bh, parent_fe_bh, handle,
1931 inode_ac);
1932 if (status < 0) {
1933 mlog_errno(status);
1934 goto bail;
1937 fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1938 inode->i_rdev = 0;
1939 newsize = l - 1;
1940 inode->i_op = &ocfs2_symlink_inode_operations;
1941 inode_nohighmem(inode);
1942 if (l > ocfs2_fast_symlink_chars(sb)) {
1943 u32 offset = 0;
1945 status = dquot_alloc_space_nodirty(inode,
1946 ocfs2_clusters_to_bytes(osb->sb, 1));
1947 if (status)
1948 goto bail;
1949 did_quota = 1;
1950 inode->i_mapping->a_ops = &ocfs2_aops;
1951 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1952 new_fe_bh,
1953 handle, data_ac, NULL,
1954 NULL);
1955 if (status < 0) {
1956 if (status != -ENOSPC && status != -EINTR) {
1957 mlog(ML_ERROR,
1958 "Failed to extend file to %llu\n",
1959 (unsigned long long)newsize);
1960 mlog_errno(status);
1961 status = -ENOSPC;
1963 goto bail;
1965 i_size_write(inode, newsize);
1966 inode->i_blocks = ocfs2_inode_sector_count(inode);
1967 } else {
1968 inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
1969 memcpy((char *) fe->id2.i_symlink, symname, l);
1970 i_size_write(inode, newsize);
1971 inode->i_blocks = 0;
1974 status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1975 if (status < 0) {
1976 mlog_errno(status);
1977 goto bail;
1980 if (!ocfs2_inode_is_fast_symlink(inode)) {
1981 status = ocfs2_create_symlink_data(osb, handle, inode,
1982 symname);
1983 if (status < 0) {
1984 mlog_errno(status);
1985 goto bail;
1989 if (si.enable) {
1990 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1991 xattr_ac, data_ac);
1992 if (status < 0) {
1993 mlog_errno(status);
1994 goto bail;
1999 * Do this before adding the entry to the directory. We add
2000 * also set d_op after success so that ->d_iput() will cleanup
2001 * the dentry lock even if ocfs2_add_entry() fails below.
2003 status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
2004 if (status) {
2005 mlog_errno(status);
2006 goto bail;
2009 dl = dentry->d_fsdata;
2011 status = ocfs2_add_entry(handle, dentry, inode,
2012 le64_to_cpu(fe->i_blkno), parent_fe_bh,
2013 &lookup);
2014 if (status < 0) {
2015 mlog_errno(status);
2016 goto bail;
2019 insert_inode_hash(inode);
2020 d_instantiate(dentry, inode);
2021 bail:
2022 if (status < 0 && did_quota)
2023 dquot_free_space_nodirty(inode,
2024 ocfs2_clusters_to_bytes(osb->sb, 1));
2025 if (status < 0 && did_quota_inode)
2026 dquot_free_inode(inode);
2027 if (handle)
2028 ocfs2_commit_trans(osb, handle);
2030 ocfs2_inode_unlock(dir, 1);
2031 if (did_block_signals)
2032 ocfs2_unblock_signals(&oldset);
2034 brelse(new_fe_bh);
2035 brelse(parent_fe_bh);
2036 kfree(si.value);
2037 ocfs2_free_dir_lookup_result(&lookup);
2038 if (inode_ac)
2039 ocfs2_free_alloc_context(inode_ac);
2040 if (data_ac)
2041 ocfs2_free_alloc_context(data_ac);
2042 if (xattr_ac)
2043 ocfs2_free_alloc_context(xattr_ac);
2044 if ((status < 0) && inode) {
2045 if (dl)
2046 ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
2048 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
2049 clear_nlink(inode);
2050 iput(inode);
2053 if (status)
2054 mlog_errno(status);
2056 return status;
2059 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2061 int status, namelen;
2063 namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2064 (long long)blkno);
2065 if (namelen <= 0) {
2066 if (namelen)
2067 status = namelen;
2068 else
2069 status = -EINVAL;
2070 mlog_errno(status);
2071 goto bail;
2073 if (namelen != OCFS2_ORPHAN_NAMELEN) {
2074 status = -EINVAL;
2075 mlog_errno(status);
2076 goto bail;
2079 trace_ocfs2_blkno_stringify(blkno, name, namelen);
2081 status = 0;
2082 bail:
2083 if (status < 0)
2084 mlog_errno(status);
2085 return status;
2088 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
2089 struct inode **ret_orphan_dir,
2090 struct buffer_head **ret_orphan_dir_bh)
2092 struct inode *orphan_dir_inode;
2093 struct buffer_head *orphan_dir_bh = NULL;
2094 int ret = 0;
2096 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2097 ORPHAN_DIR_SYSTEM_INODE,
2098 osb->slot_num);
2099 if (!orphan_dir_inode) {
2100 ret = -ENOENT;
2101 mlog_errno(ret);
2102 return ret;
2105 inode_lock(orphan_dir_inode);
2107 ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2108 if (ret < 0) {
2109 inode_unlock(orphan_dir_inode);
2110 iput(orphan_dir_inode);
2112 mlog_errno(ret);
2113 return ret;
2116 *ret_orphan_dir = orphan_dir_inode;
2117 *ret_orphan_dir_bh = orphan_dir_bh;
2119 return 0;
2122 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
2123 struct buffer_head *orphan_dir_bh,
2124 u64 blkno,
2125 char *name,
2126 struct ocfs2_dir_lookup_result *lookup,
2127 bool dio)
2129 int ret;
2130 struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
2131 int namelen = dio ?
2132 (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2133 OCFS2_ORPHAN_NAMELEN;
2135 if (dio) {
2136 ret = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2137 OCFS2_DIO_ORPHAN_PREFIX);
2138 if (ret != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2139 ret = -EINVAL;
2140 mlog_errno(ret);
2141 return ret;
2144 ret = ocfs2_blkno_stringify(blkno,
2145 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2146 } else
2147 ret = ocfs2_blkno_stringify(blkno, name);
2148 if (ret < 0) {
2149 mlog_errno(ret);
2150 return ret;
2153 ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2154 orphan_dir_bh, name,
2155 namelen, lookup);
2156 if (ret < 0) {
2157 mlog_errno(ret);
2158 return ret;
2161 return 0;
2165 * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
2166 * insertion of an orphan.
2167 * @osb: ocfs2 file system
2168 * @ret_orphan_dir: Orphan dir inode - returned locked!
2169 * @blkno: Actual block number of the inode to be inserted into orphan dir.
2170 * @lookup: dir lookup result, to be passed back into functions like
2171 * ocfs2_orphan_add
2173 * Returns zero on success and the ret_orphan_dir, name and lookup
2174 * fields will be populated.
2176 * Returns non-zero on failure.
2178 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2179 struct inode **ret_orphan_dir,
2180 u64 blkno,
2181 char *name,
2182 struct ocfs2_dir_lookup_result *lookup,
2183 bool dio)
2185 struct inode *orphan_dir_inode = NULL;
2186 struct buffer_head *orphan_dir_bh = NULL;
2187 int ret = 0;
2189 ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
2190 &orphan_dir_bh);
2191 if (ret < 0) {
2192 mlog_errno(ret);
2193 return ret;
2196 ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
2197 blkno, name, lookup, dio);
2198 if (ret < 0) {
2199 mlog_errno(ret);
2200 goto out;
2203 *ret_orphan_dir = orphan_dir_inode;
2205 out:
2206 brelse(orphan_dir_bh);
2208 if (ret) {
2209 ocfs2_inode_unlock(orphan_dir_inode, 1);
2210 inode_unlock(orphan_dir_inode);
2211 iput(orphan_dir_inode);
2214 if (ret)
2215 mlog_errno(ret);
2216 return ret;
2219 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2220 handle_t *handle,
2221 struct inode *inode,
2222 struct buffer_head *fe_bh,
2223 char *name,
2224 struct ocfs2_dir_lookup_result *lookup,
2225 struct inode *orphan_dir_inode,
2226 bool dio)
2228 struct buffer_head *orphan_dir_bh = NULL;
2229 int status = 0;
2230 struct ocfs2_dinode *orphan_fe;
2231 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
2232 int namelen = dio ?
2233 (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2234 OCFS2_ORPHAN_NAMELEN;
2236 trace_ocfs2_orphan_add_begin(
2237 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2239 status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2240 if (status < 0) {
2241 mlog_errno(status);
2242 goto leave;
2245 status = ocfs2_journal_access_di(handle,
2246 INODE_CACHE(orphan_dir_inode),
2247 orphan_dir_bh,
2248 OCFS2_JOURNAL_ACCESS_WRITE);
2249 if (status < 0) {
2250 mlog_errno(status);
2251 goto leave;
2255 * We're going to journal the change of i_flags and i_orphaned_slot.
2256 * It's safe anyway, though some callers may duplicate the journaling.
2257 * Journaling within the func just make the logic look more
2258 * straightforward.
2260 status = ocfs2_journal_access_di(handle,
2261 INODE_CACHE(inode),
2262 fe_bh,
2263 OCFS2_JOURNAL_ACCESS_WRITE);
2264 if (status < 0) {
2265 mlog_errno(status);
2266 goto leave;
2269 /* we're a cluster, and nlink can change on disk from
2270 * underneath us... */
2271 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2272 if (S_ISDIR(inode->i_mode))
2273 ocfs2_add_links_count(orphan_fe, 1);
2274 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2275 ocfs2_journal_dirty(handle, orphan_dir_bh);
2277 status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2278 namelen, inode,
2279 OCFS2_I(inode)->ip_blkno,
2280 orphan_dir_bh, lookup);
2281 if (status < 0) {
2282 mlog_errno(status);
2283 goto rollback;
2286 if (dio) {
2287 /* Update flag OCFS2_DIO_ORPHANED_FL and record the orphan
2288 * slot.
2290 fe->i_flags |= cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2291 fe->i_dio_orphaned_slot = cpu_to_le16(osb->slot_num);
2292 } else {
2293 fe->i_flags |= cpu_to_le32(OCFS2_ORPHANED_FL);
2294 OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2296 /* Record which orphan dir our inode now resides
2297 * in. delete_inode will use this to determine which orphan
2298 * dir to lock. */
2299 fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2302 ocfs2_journal_dirty(handle, fe_bh);
2304 trace_ocfs2_orphan_add_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
2305 osb->slot_num);
2307 rollback:
2308 if (status < 0) {
2309 if (S_ISDIR(inode->i_mode))
2310 ocfs2_add_links_count(orphan_fe, -1);
2311 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2314 leave:
2315 brelse(orphan_dir_bh);
2317 return status;
2320 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2321 int ocfs2_orphan_del(struct ocfs2_super *osb,
2322 handle_t *handle,
2323 struct inode *orphan_dir_inode,
2324 struct inode *inode,
2325 struct buffer_head *orphan_dir_bh,
2326 bool dio)
2328 const int namelen = OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN;
2329 char name[namelen + 1];
2330 struct ocfs2_dinode *orphan_fe;
2331 int status = 0;
2332 struct ocfs2_dir_lookup_result lookup = { NULL, };
2334 if (dio) {
2335 status = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2336 OCFS2_DIO_ORPHAN_PREFIX);
2337 if (status != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2338 status = -EINVAL;
2339 mlog_errno(status);
2340 return status;
2343 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno,
2344 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2345 } else
2346 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2347 if (status < 0) {
2348 mlog_errno(status);
2349 goto leave;
2352 trace_ocfs2_orphan_del(
2353 (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2354 name, strlen(name));
2356 status = ocfs2_journal_access_di(handle,
2357 INODE_CACHE(orphan_dir_inode),
2358 orphan_dir_bh,
2359 OCFS2_JOURNAL_ACCESS_WRITE);
2360 if (status < 0) {
2361 mlog_errno(status);
2362 goto leave;
2365 /* find it's spot in the orphan directory */
2366 status = ocfs2_find_entry(name, strlen(name), orphan_dir_inode,
2367 &lookup);
2368 if (status) {
2369 mlog_errno(status);
2370 goto leave;
2373 /* remove it from the orphan directory */
2374 status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2375 if (status < 0) {
2376 mlog_errno(status);
2377 goto leave;
2380 /* do the i_nlink dance! :) */
2381 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2382 if (S_ISDIR(inode->i_mode))
2383 ocfs2_add_links_count(orphan_fe, -1);
2384 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2385 ocfs2_journal_dirty(handle, orphan_dir_bh);
2387 leave:
2388 ocfs2_free_dir_lookup_result(&lookup);
2390 if (status)
2391 mlog_errno(status);
2392 return status;
2396 * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to receive a newly
2397 * allocated file. This is different from the typical 'add to orphan dir'
2398 * operation in that the inode does not yet exist. This is a problem because
2399 * the orphan dir stringifies the inode block number to come up with it's
2400 * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2401 * problem. This function works around it by calling deeper into the orphan
2402 * and suballoc code than other callers. Use this only by necessity.
2403 * @dir: The directory which this inode will ultimately wind up under - not the
2404 * orphan dir!
2405 * @dir_bh: buffer_head the @dir inode block
2406 * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2407 * with the string to be used for orphan dirent. Pass back to the orphan dir
2408 * code.
2409 * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2410 * dir code.
2411 * @ret_di_blkno: block number where the new inode will be allocated.
2412 * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2413 * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2415 * Returns zero on success and the ret_orphan_dir, name and lookup
2416 * fields will be populated.
2418 * Returns non-zero on failure.
2420 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2421 struct buffer_head *dir_bh,
2422 char *orphan_name,
2423 struct inode **ret_orphan_dir,
2424 u64 *ret_di_blkno,
2425 struct ocfs2_dir_lookup_result *orphan_insert,
2426 struct ocfs2_alloc_context **ret_inode_ac)
2428 int ret;
2429 u64 di_blkno;
2430 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2431 struct inode *orphan_dir = NULL;
2432 struct buffer_head *orphan_dir_bh = NULL;
2433 struct ocfs2_alloc_context *inode_ac = NULL;
2435 ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2436 if (ret < 0) {
2437 mlog_errno(ret);
2438 return ret;
2441 /* reserve an inode spot */
2442 ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2443 if (ret < 0) {
2444 if (ret != -ENOSPC)
2445 mlog_errno(ret);
2446 goto out;
2449 ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2450 &di_blkno);
2451 if (ret) {
2452 mlog_errno(ret);
2453 goto out;
2456 ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2457 di_blkno, orphan_name, orphan_insert,
2458 false);
2459 if (ret < 0) {
2460 mlog_errno(ret);
2461 goto out;
2464 out:
2465 if (ret == 0) {
2466 *ret_orphan_dir = orphan_dir;
2467 *ret_di_blkno = di_blkno;
2468 *ret_inode_ac = inode_ac;
2470 * orphan_name and orphan_insert are already up to
2471 * date via prepare_orphan_dir
2473 } else {
2474 /* Unroll reserve_new_inode* */
2475 if (inode_ac)
2476 ocfs2_free_alloc_context(inode_ac);
2478 /* Unroll orphan dir locking */
2479 inode_unlock(orphan_dir);
2480 ocfs2_inode_unlock(orphan_dir, 1);
2481 iput(orphan_dir);
2484 brelse(orphan_dir_bh);
2486 return ret;
2489 int ocfs2_create_inode_in_orphan(struct inode *dir,
2490 int mode,
2491 struct inode **new_inode)
2493 int status, did_quota_inode = 0;
2494 struct inode *inode = NULL;
2495 struct inode *orphan_dir = NULL;
2496 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2497 struct ocfs2_dinode *di = NULL;
2498 handle_t *handle = NULL;
2499 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2500 struct buffer_head *parent_di_bh = NULL;
2501 struct buffer_head *new_di_bh = NULL;
2502 struct ocfs2_alloc_context *inode_ac = NULL;
2503 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2504 u64 uninitialized_var(di_blkno), suballoc_loc;
2505 u16 suballoc_bit;
2507 status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2508 if (status < 0) {
2509 if (status != -ENOENT)
2510 mlog_errno(status);
2511 return status;
2514 status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2515 orphan_name, &orphan_dir,
2516 &di_blkno, &orphan_insert, &inode_ac);
2517 if (status < 0) {
2518 if (status != -ENOSPC)
2519 mlog_errno(status);
2520 goto leave;
2523 inode = ocfs2_get_init_inode(dir, mode);
2524 if (IS_ERR(inode)) {
2525 status = PTR_ERR(inode);
2526 inode = NULL;
2527 mlog_errno(status);
2528 goto leave;
2531 handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2532 if (IS_ERR(handle)) {
2533 status = PTR_ERR(handle);
2534 handle = NULL;
2535 mlog_errno(status);
2536 goto leave;
2539 status = dquot_alloc_inode(inode);
2540 if (status)
2541 goto leave;
2542 did_quota_inode = 1;
2544 status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2545 &suballoc_loc,
2546 &suballoc_bit, di_blkno);
2547 if (status < 0) {
2548 mlog_errno(status);
2549 goto leave;
2552 clear_nlink(inode);
2553 /* do the real work now. */
2554 status = __ocfs2_mknod_locked(dir, inode,
2555 0, &new_di_bh, parent_di_bh, handle,
2556 inode_ac, di_blkno, suballoc_loc,
2557 suballoc_bit);
2558 if (status < 0) {
2559 mlog_errno(status);
2560 goto leave;
2563 di = (struct ocfs2_dinode *)new_di_bh->b_data;
2564 status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2565 &orphan_insert, orphan_dir, false);
2566 if (status < 0) {
2567 mlog_errno(status);
2568 goto leave;
2571 /* get open lock so that only nodes can't remove it from orphan dir. */
2572 status = ocfs2_open_lock(inode);
2573 if (status < 0)
2574 mlog_errno(status);
2576 insert_inode_hash(inode);
2577 leave:
2578 if (status < 0 && did_quota_inode)
2579 dquot_free_inode(inode);
2580 if (handle)
2581 ocfs2_commit_trans(osb, handle);
2583 if (orphan_dir) {
2584 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2585 ocfs2_inode_unlock(orphan_dir, 1);
2586 inode_unlock(orphan_dir);
2587 iput(orphan_dir);
2590 if ((status < 0) && inode) {
2591 clear_nlink(inode);
2592 iput(inode);
2595 if (inode_ac)
2596 ocfs2_free_alloc_context(inode_ac);
2598 brelse(new_di_bh);
2600 if (!status)
2601 *new_inode = inode;
2603 ocfs2_free_dir_lookup_result(&orphan_insert);
2605 ocfs2_inode_unlock(dir, 1);
2606 brelse(parent_di_bh);
2607 return status;
2610 int ocfs2_add_inode_to_orphan(struct ocfs2_super *osb,
2611 struct inode *inode)
2613 char orphan_name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2614 struct inode *orphan_dir_inode = NULL;
2615 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2616 struct buffer_head *di_bh = NULL;
2617 int status = 0;
2618 handle_t *handle = NULL;
2619 struct ocfs2_dinode *di = NULL;
2621 status = ocfs2_inode_lock(inode, &di_bh, 1);
2622 if (status < 0) {
2623 mlog_errno(status);
2624 goto bail;
2627 di = (struct ocfs2_dinode *) di_bh->b_data;
2629 * Another append dio crashed?
2630 * If so, manually recover it first.
2632 if (unlikely(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL))) {
2633 status = ocfs2_truncate_file(inode, di_bh, i_size_read(inode));
2634 if (status < 0) {
2635 if (status != -ENOSPC)
2636 mlog_errno(status);
2637 goto bail_unlock_inode;
2640 status = ocfs2_del_inode_from_orphan(osb, inode, di_bh, 0, 0);
2641 if (status < 0) {
2642 mlog_errno(status);
2643 goto bail_unlock_inode;
2647 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir_inode,
2648 OCFS2_I(inode)->ip_blkno,
2649 orphan_name,
2650 &orphan_insert,
2651 true);
2652 if (status < 0) {
2653 mlog_errno(status);
2654 goto bail_unlock_inode;
2657 handle = ocfs2_start_trans(osb,
2658 OCFS2_INODE_ADD_TO_ORPHAN_CREDITS);
2659 if (IS_ERR(handle)) {
2660 status = PTR_ERR(handle);
2661 goto bail_unlock_orphan;
2664 status = ocfs2_orphan_add(osb, handle, inode, di_bh, orphan_name,
2665 &orphan_insert, orphan_dir_inode, true);
2666 if (status)
2667 mlog_errno(status);
2669 ocfs2_commit_trans(osb, handle);
2671 bail_unlock_orphan:
2672 ocfs2_inode_unlock(orphan_dir_inode, 1);
2673 inode_unlock(orphan_dir_inode);
2674 iput(orphan_dir_inode);
2676 ocfs2_free_dir_lookup_result(&orphan_insert);
2678 bail_unlock_inode:
2679 ocfs2_inode_unlock(inode, 1);
2680 brelse(di_bh);
2682 bail:
2683 return status;
2686 int ocfs2_del_inode_from_orphan(struct ocfs2_super *osb,
2687 struct inode *inode, struct buffer_head *di_bh,
2688 int update_isize, loff_t end)
2690 struct inode *orphan_dir_inode = NULL;
2691 struct buffer_head *orphan_dir_bh = NULL;
2692 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2693 handle_t *handle = NULL;
2694 int status = 0;
2696 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2697 ORPHAN_DIR_SYSTEM_INODE,
2698 le16_to_cpu(di->i_dio_orphaned_slot));
2699 if (!orphan_dir_inode) {
2700 status = -ENOENT;
2701 mlog_errno(status);
2702 goto bail;
2705 inode_lock(orphan_dir_inode);
2706 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2707 if (status < 0) {
2708 inode_unlock(orphan_dir_inode);
2709 iput(orphan_dir_inode);
2710 mlog_errno(status);
2711 goto bail;
2714 handle = ocfs2_start_trans(osb,
2715 OCFS2_INODE_DEL_FROM_ORPHAN_CREDITS);
2716 if (IS_ERR(handle)) {
2717 status = PTR_ERR(handle);
2718 goto bail_unlock_orphan;
2721 BUG_ON(!(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)));
2723 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode,
2724 inode, orphan_dir_bh, true);
2725 if (status < 0) {
2726 mlog_errno(status);
2727 goto bail_commit;
2730 status = ocfs2_journal_access_di(handle,
2731 INODE_CACHE(inode),
2732 di_bh,
2733 OCFS2_JOURNAL_ACCESS_WRITE);
2734 if (status < 0) {
2735 mlog_errno(status);
2736 goto bail_commit;
2739 di->i_flags &= ~cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2740 di->i_dio_orphaned_slot = 0;
2742 if (update_isize) {
2743 status = ocfs2_set_inode_size(handle, inode, di_bh, end);
2744 if (status)
2745 mlog_errno(status);
2746 } else
2747 ocfs2_journal_dirty(handle, di_bh);
2749 bail_commit:
2750 ocfs2_commit_trans(osb, handle);
2752 bail_unlock_orphan:
2753 ocfs2_inode_unlock(orphan_dir_inode, 1);
2754 inode_unlock(orphan_dir_inode);
2755 brelse(orphan_dir_bh);
2756 iput(orphan_dir_inode);
2758 bail:
2759 return status;
2762 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2763 struct inode *inode,
2764 struct dentry *dentry)
2766 int status = 0;
2767 struct buffer_head *parent_di_bh = NULL;
2768 handle_t *handle = NULL;
2769 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2770 struct ocfs2_dinode *dir_di, *di;
2771 struct inode *orphan_dir_inode = NULL;
2772 struct buffer_head *orphan_dir_bh = NULL;
2773 struct buffer_head *di_bh = NULL;
2774 struct ocfs2_dir_lookup_result lookup = { NULL, };
2776 trace_ocfs2_mv_orphaned_inode_to_new(dir, dentry,
2777 dentry->d_name.len, dentry->d_name.name,
2778 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2779 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2781 status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2782 if (status < 0) {
2783 if (status != -ENOENT)
2784 mlog_errno(status);
2785 return status;
2788 dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2789 if (!dir_di->i_links_count) {
2790 /* can't make a file in a deleted directory. */
2791 status = -ENOENT;
2792 goto leave;
2795 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2796 dentry->d_name.len);
2797 if (status)
2798 goto leave;
2800 /* get a spot inside the dir. */
2801 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2802 dentry->d_name.name,
2803 dentry->d_name.len, &lookup);
2804 if (status < 0) {
2805 mlog_errno(status);
2806 goto leave;
2809 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2810 ORPHAN_DIR_SYSTEM_INODE,
2811 osb->slot_num);
2812 if (!orphan_dir_inode) {
2813 status = -ENOENT;
2814 mlog_errno(status);
2815 goto leave;
2818 inode_lock(orphan_dir_inode);
2820 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2821 if (status < 0) {
2822 mlog_errno(status);
2823 inode_unlock(orphan_dir_inode);
2824 iput(orphan_dir_inode);
2825 goto leave;
2828 status = ocfs2_read_inode_block(inode, &di_bh);
2829 if (status < 0) {
2830 mlog_errno(status);
2831 goto orphan_unlock;
2834 handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2835 if (IS_ERR(handle)) {
2836 status = PTR_ERR(handle);
2837 handle = NULL;
2838 mlog_errno(status);
2839 goto orphan_unlock;
2842 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2843 di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2844 if (status < 0) {
2845 mlog_errno(status);
2846 goto out_commit;
2849 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2850 orphan_dir_bh, false);
2851 if (status < 0) {
2852 mlog_errno(status);
2853 goto out_commit;
2856 di = (struct ocfs2_dinode *)di_bh->b_data;
2857 di->i_flags &= ~cpu_to_le32(OCFS2_ORPHANED_FL);
2858 di->i_orphaned_slot = 0;
2859 set_nlink(inode, 1);
2860 ocfs2_set_links_count(di, inode->i_nlink);
2861 ocfs2_update_inode_fsync_trans(handle, inode, 1);
2862 ocfs2_journal_dirty(handle, di_bh);
2864 status = ocfs2_add_entry(handle, dentry, inode,
2865 OCFS2_I(inode)->ip_blkno, parent_di_bh,
2866 &lookup);
2867 if (status < 0) {
2868 mlog_errno(status);
2869 goto out_commit;
2872 status = ocfs2_dentry_attach_lock(dentry, inode,
2873 OCFS2_I(dir)->ip_blkno);
2874 if (status) {
2875 mlog_errno(status);
2876 goto out_commit;
2879 d_instantiate(dentry, inode);
2880 status = 0;
2881 out_commit:
2882 ocfs2_commit_trans(osb, handle);
2883 orphan_unlock:
2884 ocfs2_inode_unlock(orphan_dir_inode, 1);
2885 inode_unlock(orphan_dir_inode);
2886 iput(orphan_dir_inode);
2887 leave:
2889 ocfs2_inode_unlock(dir, 1);
2891 brelse(di_bh);
2892 brelse(parent_di_bh);
2893 brelse(orphan_dir_bh);
2895 ocfs2_free_dir_lookup_result(&lookup);
2897 if (status)
2898 mlog_errno(status);
2900 return status;
2903 const struct inode_operations ocfs2_dir_iops = {
2904 .create = ocfs2_create,
2905 .lookup = ocfs2_lookup,
2906 .link = ocfs2_link,
2907 .unlink = ocfs2_unlink,
2908 .rmdir = ocfs2_unlink,
2909 .symlink = ocfs2_symlink,
2910 .mkdir = ocfs2_mkdir,
2911 .mknod = ocfs2_mknod,
2912 .rename = ocfs2_rename,
2913 .setattr = ocfs2_setattr,
2914 .getattr = ocfs2_getattr,
2915 .permission = ocfs2_permission,
2916 .setxattr = generic_setxattr,
2917 .getxattr = generic_getxattr,
2918 .listxattr = ocfs2_listxattr,
2919 .removexattr = generic_removexattr,
2920 .fiemap = ocfs2_fiemap,
2921 .get_acl = ocfs2_iop_get_acl,
2922 .set_acl = ocfs2_iop_set_acl,