percpu-rwsem: kill CONFIG_PERCPU_RWSEM
[linux/fpc-iii.git] / fs / ocfs2 / namei.c
blob6e6abb93fda590468406951b54419b9d26bcbcc4
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)))
109 #define OCFS2_DIO_ORPHAN_PREFIX "dio-"
110 #define OCFS2_DIO_ORPHAN_PREFIX_LEN 4
112 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
113 unsigned int flags)
115 int status;
116 u64 blkno;
117 struct inode *inode = NULL;
118 struct dentry *ret;
119 struct ocfs2_inode_info *oi;
121 trace_ocfs2_lookup(dir, dentry, dentry->d_name.len,
122 dentry->d_name.name,
123 (unsigned long long)OCFS2_I(dir)->ip_blkno, 0);
125 if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
126 ret = ERR_PTR(-ENAMETOOLONG);
127 goto bail;
130 status = ocfs2_inode_lock_nested(dir, NULL, 0, OI_LS_PARENT);
131 if (status < 0) {
132 if (status != -ENOENT)
133 mlog_errno(status);
134 ret = ERR_PTR(status);
135 goto bail;
138 status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
139 dentry->d_name.len, &blkno);
140 if (status < 0)
141 goto bail_add;
143 inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
144 if (IS_ERR(inode)) {
145 ret = ERR_PTR(-EACCES);
146 goto bail_unlock;
149 oi = OCFS2_I(inode);
150 /* Clear any orphaned state... If we were able to look up the
151 * inode from a directory, it certainly can't be orphaned. We
152 * might have the bad state from a node which intended to
153 * orphan this inode but crashed before it could commit the
154 * unlink. */
155 spin_lock(&oi->ip_lock);
156 oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
157 spin_unlock(&oi->ip_lock);
159 bail_add:
160 ret = d_splice_alias(inode, dentry);
162 if (inode) {
164 * If d_splice_alias() finds a DCACHE_DISCONNECTED
165 * dentry, it will d_move() it on top of ourse. The
166 * return value will indicate this however, so in
167 * those cases, we switch them around for the locking
168 * code.
170 * NOTE: This dentry already has ->d_op set from
171 * ocfs2_get_parent() and ocfs2_get_dentry()
173 if (!IS_ERR_OR_NULL(ret))
174 dentry = ret;
176 status = ocfs2_dentry_attach_lock(dentry, inode,
177 OCFS2_I(dir)->ip_blkno);
178 if (status) {
179 mlog_errno(status);
180 ret = ERR_PTR(status);
181 goto bail_unlock;
183 } else
184 ocfs2_dentry_attach_gen(dentry);
186 bail_unlock:
187 /* Don't drop the cluster lock until *after* the d_add --
188 * unlink on another node will message us to remove that
189 * dentry under this lock so otherwise we can race this with
190 * the downconvert thread and have a stale dentry. */
191 ocfs2_inode_unlock(dir, 0);
193 bail:
195 trace_ocfs2_lookup_ret(ret);
197 return ret;
200 static struct inode *ocfs2_get_init_inode(struct inode *dir, umode_t mode)
202 struct inode *inode;
204 inode = new_inode(dir->i_sb);
205 if (!inode) {
206 mlog(ML_ERROR, "new_inode failed!\n");
207 return NULL;
210 /* populate as many fields early on as possible - many of
211 * these are used by the support functions here and in
212 * callers. */
213 if (S_ISDIR(mode))
214 set_nlink(inode, 2);
215 inode_init_owner(inode, dir, mode);
216 dquot_initialize(inode);
217 return inode;
220 static void ocfs2_cleanup_add_entry_failure(struct ocfs2_super *osb,
221 struct dentry *dentry, struct inode *inode)
223 struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
225 ocfs2_simple_drop_lockres(osb, &dl->dl_lockres);
226 ocfs2_lock_res_free(&dl->dl_lockres);
227 BUG_ON(dl->dl_count != 1);
228 spin_lock(&dentry_attach_lock);
229 dentry->d_fsdata = NULL;
230 spin_unlock(&dentry_attach_lock);
231 kfree(dl);
232 iput(inode);
235 static int ocfs2_mknod(struct inode *dir,
236 struct dentry *dentry,
237 umode_t mode,
238 dev_t dev)
240 int status = 0;
241 struct buffer_head *parent_fe_bh = NULL;
242 handle_t *handle = NULL;
243 struct ocfs2_super *osb;
244 struct ocfs2_dinode *dirfe;
245 struct buffer_head *new_fe_bh = NULL;
246 struct inode *inode = NULL;
247 struct ocfs2_alloc_context *inode_ac = NULL;
248 struct ocfs2_alloc_context *data_ac = NULL;
249 struct ocfs2_alloc_context *meta_ac = NULL;
250 int want_clusters = 0;
251 int want_meta = 0;
252 int xattr_credits = 0;
253 struct ocfs2_security_xattr_info si = {
254 .enable = 1,
256 int did_quota_inode = 0;
257 struct ocfs2_dir_lookup_result lookup = { NULL, };
258 sigset_t oldset;
259 int did_block_signals = 0;
260 struct posix_acl *default_acl = NULL, *acl = NULL;
261 struct ocfs2_dentry_lock *dl = NULL;
263 trace_ocfs2_mknod(dir, dentry, dentry->d_name.len, dentry->d_name.name,
264 (unsigned long long)OCFS2_I(dir)->ip_blkno,
265 (unsigned long)dev, mode);
267 dquot_initialize(dir);
269 /* get our super block */
270 osb = OCFS2_SB(dir->i_sb);
272 status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
273 if (status < 0) {
274 if (status != -ENOENT)
275 mlog_errno(status);
276 return status;
279 if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
280 status = -EMLINK;
281 goto leave;
284 dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
285 if (!ocfs2_read_links_count(dirfe)) {
286 /* can't make a file in a deleted directory. */
287 status = -ENOENT;
288 goto leave;
291 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
292 dentry->d_name.len);
293 if (status)
294 goto leave;
296 /* get a spot inside the dir. */
297 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
298 dentry->d_name.name,
299 dentry->d_name.len, &lookup);
300 if (status < 0) {
301 mlog_errno(status);
302 goto leave;
305 /* reserve an inode spot */
306 status = ocfs2_reserve_new_inode(osb, &inode_ac);
307 if (status < 0) {
308 if (status != -ENOSPC)
309 mlog_errno(status);
310 goto leave;
313 inode = ocfs2_get_init_inode(dir, mode);
314 if (!inode) {
315 status = -ENOMEM;
316 mlog_errno(status);
317 goto leave;
320 /* get security xattr */
321 status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
322 if (status) {
323 if (status == -EOPNOTSUPP)
324 si.enable = 0;
325 else {
326 mlog_errno(status);
327 goto leave;
331 /* calculate meta data/clusters for setting security and acl xattr */
332 status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
333 &si, &want_clusters,
334 &xattr_credits, &want_meta);
335 if (status < 0) {
336 mlog_errno(status);
337 goto leave;
340 /* Reserve a cluster if creating an extent based directory. */
341 if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
342 want_clusters += 1;
344 /* Dir indexing requires extra space as well */
345 if (ocfs2_supports_indexed_dirs(osb))
346 want_meta++;
349 status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
350 if (status < 0) {
351 if (status != -ENOSPC)
352 mlog_errno(status);
353 goto leave;
356 status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
357 if (status < 0) {
358 if (status != -ENOSPC)
359 mlog_errno(status);
360 goto leave;
363 status = posix_acl_create(dir, &mode, &default_acl, &acl);
364 if (status) {
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 if (default_acl) {
418 status = ocfs2_set_acl(handle, inode, new_fe_bh,
419 ACL_TYPE_DEFAULT, default_acl,
420 meta_ac, data_ac);
422 if (!status && acl) {
423 status = ocfs2_set_acl(handle, inode, new_fe_bh,
424 ACL_TYPE_ACCESS, acl,
425 meta_ac, data_ac);
428 if (status < 0) {
429 mlog_errno(status);
430 goto leave;
433 if (si.enable) {
434 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
435 meta_ac, data_ac);
436 if (status < 0) {
437 mlog_errno(status);
438 goto leave;
443 * Do this before adding the entry to the directory. We add
444 * also set d_op after success so that ->d_iput() will cleanup
445 * the dentry lock even if ocfs2_add_entry() fails below.
447 status = ocfs2_dentry_attach_lock(dentry, inode,
448 OCFS2_I(dir)->ip_blkno);
449 if (status) {
450 mlog_errno(status);
451 goto leave;
454 dl = dentry->d_fsdata;
456 status = ocfs2_add_entry(handle, dentry, inode,
457 OCFS2_I(inode)->ip_blkno, parent_fe_bh,
458 &lookup);
459 if (status < 0) {
460 mlog_errno(status);
461 goto leave;
464 insert_inode_hash(inode);
465 d_instantiate(dentry, inode);
466 status = 0;
467 leave:
468 if (default_acl)
469 posix_acl_release(default_acl);
470 if (acl)
471 posix_acl_release(acl);
472 if (status < 0 && did_quota_inode)
473 dquot_free_inode(inode);
474 if (handle)
475 ocfs2_commit_trans(osb, handle);
477 ocfs2_inode_unlock(dir, 1);
478 if (did_block_signals)
479 ocfs2_unblock_signals(&oldset);
481 brelse(new_fe_bh);
482 brelse(parent_fe_bh);
483 kfree(si.value);
485 ocfs2_free_dir_lookup_result(&lookup);
487 if (inode_ac)
488 ocfs2_free_alloc_context(inode_ac);
490 if (data_ac)
491 ocfs2_free_alloc_context(data_ac);
493 if (meta_ac)
494 ocfs2_free_alloc_context(meta_ac);
497 * We should call iput after the i_mutex of the bitmap been
498 * unlocked in ocfs2_free_alloc_context, or the
499 * ocfs2_delete_inode will mutex_lock again.
501 if ((status < 0) && inode) {
502 if (dl)
503 ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
505 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
506 clear_nlink(inode);
507 iput(inode);
510 if (status)
511 mlog_errno(status);
513 return status;
516 static int __ocfs2_mknod_locked(struct inode *dir,
517 struct inode *inode,
518 dev_t dev,
519 struct buffer_head **new_fe_bh,
520 struct buffer_head *parent_fe_bh,
521 handle_t *handle,
522 struct ocfs2_alloc_context *inode_ac,
523 u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
525 int status = 0;
526 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
527 struct ocfs2_dinode *fe = NULL;
528 struct ocfs2_extent_list *fel;
529 u16 feat;
530 struct ocfs2_inode_info *oi = OCFS2_I(inode);
532 *new_fe_bh = NULL;
534 /* populate as many fields early on as possible - many of
535 * these are used by the support functions here and in
536 * callers. */
537 inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
538 OCFS2_I(inode)->ip_blkno = fe_blkno;
539 spin_lock(&osb->osb_lock);
540 inode->i_generation = osb->s_next_generation++;
541 spin_unlock(&osb->osb_lock);
543 *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
544 if (!*new_fe_bh) {
545 status = -ENOMEM;
546 mlog_errno(status);
547 goto leave;
549 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
551 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
552 *new_fe_bh,
553 OCFS2_JOURNAL_ACCESS_CREATE);
554 if (status < 0) {
555 mlog_errno(status);
556 goto leave;
559 fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
560 memset(fe, 0, osb->sb->s_blocksize);
562 fe->i_generation = cpu_to_le32(inode->i_generation);
563 fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
564 fe->i_blkno = cpu_to_le64(fe_blkno);
565 fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
566 fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
567 fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
568 fe->i_uid = cpu_to_le32(i_uid_read(inode));
569 fe->i_gid = cpu_to_le32(i_gid_read(inode));
570 fe->i_mode = cpu_to_le16(inode->i_mode);
571 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
572 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
574 ocfs2_set_links_count(fe, inode->i_nlink);
576 fe->i_last_eb_blk = 0;
577 strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
578 fe->i_flags |= cpu_to_le32(OCFS2_VALID_FL);
579 fe->i_atime = fe->i_ctime = fe->i_mtime =
580 cpu_to_le64(CURRENT_TIME.tv_sec);
581 fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
582 cpu_to_le32(CURRENT_TIME.tv_nsec);
583 fe->i_dtime = 0;
586 * If supported, directories start with inline data. If inline
587 * isn't supported, but indexing is, we start them as indexed.
589 feat = le16_to_cpu(fe->i_dyn_features);
590 if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
591 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
593 fe->id2.i_data.id_count = cpu_to_le16(
594 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
595 } else {
596 fel = &fe->id2.i_list;
597 fel->l_tree_depth = 0;
598 fel->l_next_free_rec = 0;
599 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
602 ocfs2_journal_dirty(handle, *new_fe_bh);
604 ocfs2_populate_inode(inode, fe, 1);
605 ocfs2_ci_set_new(osb, INODE_CACHE(inode));
606 if (!ocfs2_mount_local(osb)) {
607 status = ocfs2_create_new_inode_locks(inode);
608 if (status < 0)
609 mlog_errno(status);
612 oi->i_sync_tid = handle->h_transaction->t_tid;
613 oi->i_datasync_tid = handle->h_transaction->t_tid;
615 leave:
616 if (status < 0) {
617 if (*new_fe_bh) {
618 brelse(*new_fe_bh);
619 *new_fe_bh = NULL;
623 if (status)
624 mlog_errno(status);
625 return status;
628 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
629 struct inode *dir,
630 struct inode *inode,
631 dev_t dev,
632 struct buffer_head **new_fe_bh,
633 struct buffer_head *parent_fe_bh,
634 handle_t *handle,
635 struct ocfs2_alloc_context *inode_ac)
637 int status = 0;
638 u64 suballoc_loc, fe_blkno = 0;
639 u16 suballoc_bit;
641 *new_fe_bh = NULL;
643 status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
644 inode_ac, &suballoc_loc,
645 &suballoc_bit, &fe_blkno);
646 if (status < 0) {
647 mlog_errno(status);
648 return status;
651 return __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
652 parent_fe_bh, handle, inode_ac,
653 fe_blkno, suballoc_loc, suballoc_bit);
656 static int ocfs2_mkdir(struct inode *dir,
657 struct dentry *dentry,
658 umode_t mode)
660 int ret;
662 trace_ocfs2_mkdir(dir, dentry, dentry->d_name.len, dentry->d_name.name,
663 OCFS2_I(dir)->ip_blkno, mode);
664 ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
665 if (ret)
666 mlog_errno(ret);
668 return ret;
671 static int ocfs2_create(struct inode *dir,
672 struct dentry *dentry,
673 umode_t mode,
674 bool excl)
676 int ret;
678 trace_ocfs2_create(dir, dentry, dentry->d_name.len, dentry->d_name.name,
679 (unsigned long long)OCFS2_I(dir)->ip_blkno, mode);
680 ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
681 if (ret)
682 mlog_errno(ret);
684 return ret;
687 static int ocfs2_link(struct dentry *old_dentry,
688 struct inode *dir,
689 struct dentry *dentry)
691 handle_t *handle;
692 struct inode *inode = d_inode(old_dentry);
693 struct inode *old_dir = d_inode(old_dentry->d_parent);
694 int err;
695 struct buffer_head *fe_bh = NULL;
696 struct buffer_head *old_dir_bh = NULL;
697 struct buffer_head *parent_fe_bh = NULL;
698 struct ocfs2_dinode *fe = NULL;
699 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
700 struct ocfs2_dir_lookup_result lookup = { NULL, };
701 sigset_t oldset;
702 u64 old_de_ino;
704 trace_ocfs2_link((unsigned long long)OCFS2_I(inode)->ip_blkno,
705 old_dentry->d_name.len, old_dentry->d_name.name,
706 dentry->d_name.len, dentry->d_name.name);
708 if (S_ISDIR(inode->i_mode))
709 return -EPERM;
711 dquot_initialize(dir);
713 err = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
714 &parent_fe_bh, dir, 0);
715 if (err < 0) {
716 if (err != -ENOENT)
717 mlog_errno(err);
718 return err;
721 /* make sure both dirs have bhs
722 * get an extra ref on old_dir_bh if old==new */
723 if (!parent_fe_bh) {
724 if (old_dir_bh) {
725 parent_fe_bh = old_dir_bh;
726 get_bh(parent_fe_bh);
727 } else {
728 mlog(ML_ERROR, "%s: no old_dir_bh!\n", osb->uuid_str);
729 err = -EIO;
730 goto out;
734 if (!dir->i_nlink) {
735 err = -ENOENT;
736 goto out;
739 err = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
740 old_dentry->d_name.len, &old_de_ino);
741 if (err) {
742 err = -ENOENT;
743 goto out;
747 * Check whether another node removed the source inode while we
748 * were in the vfs.
750 if (old_de_ino != OCFS2_I(inode)->ip_blkno) {
751 err = -ENOENT;
752 goto out;
755 err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
756 dentry->d_name.len);
757 if (err)
758 goto out;
760 err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
761 dentry->d_name.name,
762 dentry->d_name.len, &lookup);
763 if (err < 0) {
764 mlog_errno(err);
765 goto out;
768 err = ocfs2_inode_lock(inode, &fe_bh, 1);
769 if (err < 0) {
770 if (err != -ENOENT)
771 mlog_errno(err);
772 goto out;
775 fe = (struct ocfs2_dinode *) fe_bh->b_data;
776 if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
777 err = -EMLINK;
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);
784 handle = NULL;
785 mlog_errno(err);
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);
794 if (err < 0) {
795 mlog_errno(err);
796 goto out_commit;
799 inc_nlink(inode);
800 inode->i_ctime = CURRENT_TIME;
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);
809 if (err) {
810 ocfs2_add_links_count(fe, -1);
811 drop_nlink(inode);
812 mlog_errno(err);
813 goto out_commit;
816 err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
817 if (err) {
818 mlog_errno(err);
819 goto out_commit;
822 ihold(inode);
823 d_instantiate(dentry, inode);
825 out_commit:
826 ocfs2_commit_trans(osb, handle);
827 ocfs2_unblock_signals(&oldset);
828 out_unlock_inode:
829 ocfs2_inode_unlock(inode, 1);
831 out:
832 ocfs2_double_unlock(old_dir, dir);
834 brelse(fe_bh);
835 brelse(parent_fe_bh);
836 brelse(old_dir_bh);
838 ocfs2_free_dir_lookup_result(&lookup);
840 if (err)
841 mlog_errno(err);
843 return err;
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)
852 int ret;
854 ret = ocfs2_dentry_lock(dentry, 1);
855 if (ret)
856 mlog_errno(ret);
857 else
858 ocfs2_dentry_unlock(dentry, 1);
860 return ret;
863 static inline int ocfs2_inode_is_unlinkable(struct inode *inode)
865 if (S_ISDIR(inode->i_mode)) {
866 if (inode->i_nlink == 2)
867 return 1;
868 return 0;
871 if (inode->i_nlink == 1)
872 return 1;
873 return 0;
876 static int ocfs2_unlink(struct inode *dir,
877 struct dentry *dentry)
879 int status;
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);
885 u64 blkno;
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,
895 dentry->d_name.name,
896 (unsigned long long)OCFS2_I(dir)->ip_blkno,
897 (unsigned long long)OCFS2_I(inode)->ip_blkno);
899 dquot_initialize(dir);
901 BUG_ON(d_inode(dentry->d_parent) != dir);
903 if (inode == osb->root_inode)
904 return -EPERM;
906 status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
907 OI_LS_PARENT);
908 if (status < 0) {
909 if (status != -ENOENT)
910 mlog_errno(status);
911 return status;
914 status = ocfs2_find_files_on_disk(dentry->d_name.name,
915 dentry->d_name.len, &blkno, dir,
916 &lookup);
917 if (status < 0) {
918 if (status != -ENOENT)
919 mlog_errno(status);
920 goto leave;
923 if (OCFS2_I(inode)->ip_blkno != blkno) {
924 status = -ENOENT;
926 trace_ocfs2_unlink_noent(
927 (unsigned long long)OCFS2_I(inode)->ip_blkno,
928 (unsigned long long)blkno,
929 OCFS2_I(inode)->ip_flags);
930 goto leave;
933 status = ocfs2_inode_lock(inode, &fe_bh, 1);
934 if (status < 0) {
935 if (status != -ENOENT)
936 mlog_errno(status);
937 goto leave;
939 child_locked = 1;
941 if (S_ISDIR(inode->i_mode)) {
942 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
943 status = -ENOTEMPTY;
944 goto leave;
948 status = ocfs2_remote_dentry_delete(dentry);
949 if (status < 0) {
950 /* This remote delete should succeed under all normal
951 * circumstances. */
952 mlog_errno(status);
953 goto leave;
956 if (ocfs2_inode_is_unlinkable(inode)) {
957 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
958 OCFS2_I(inode)->ip_blkno,
959 orphan_name, &orphan_insert,
960 false);
961 if (status < 0) {
962 mlog_errno(status);
963 goto leave;
965 is_unlinkable = true;
968 handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
969 if (IS_ERR(handle)) {
970 status = PTR_ERR(handle);
971 handle = NULL;
972 mlog_errno(status);
973 goto leave;
976 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
977 OCFS2_JOURNAL_ACCESS_WRITE);
978 if (status < 0) {
979 mlog_errno(status);
980 goto leave;
983 fe = (struct ocfs2_dinode *) fe_bh->b_data;
985 /* delete the name from the parent dir */
986 status = ocfs2_delete_entry(handle, dir, &lookup);
987 if (status < 0) {
988 mlog_errno(status);
989 goto leave;
992 if (S_ISDIR(inode->i_mode))
993 drop_nlink(inode);
994 drop_nlink(inode);
995 ocfs2_set_links_count(fe, inode->i_nlink);
996 ocfs2_journal_dirty(handle, fe_bh);
998 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
999 if (S_ISDIR(inode->i_mode))
1000 drop_nlink(dir);
1002 status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
1003 if (status < 0) {
1004 mlog_errno(status);
1005 if (S_ISDIR(inode->i_mode))
1006 inc_nlink(dir);
1007 goto leave;
1010 if (is_unlinkable) {
1011 status = ocfs2_orphan_add(osb, handle, inode, fe_bh,
1012 orphan_name, &orphan_insert, orphan_dir, false);
1013 if (status < 0)
1014 mlog_errno(status);
1017 leave:
1018 if (handle)
1019 ocfs2_commit_trans(osb, handle);
1021 if (child_locked)
1022 ocfs2_inode_unlock(inode, 1);
1024 ocfs2_inode_unlock(dir, 1);
1026 if (orphan_dir) {
1027 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1028 ocfs2_inode_unlock(orphan_dir, 1);
1029 mutex_unlock(&orphan_dir->i_mutex);
1030 iput(orphan_dir);
1033 brelse(fe_bh);
1034 brelse(parent_node_bh);
1036 ocfs2_free_dir_lookup_result(&orphan_insert);
1037 ocfs2_free_dir_lookup_result(&lookup);
1039 if (status && (status != -ENOTEMPTY) && (status != -ENOENT))
1040 mlog_errno(status);
1042 return status;
1045 static int ocfs2_check_if_ancestor(struct ocfs2_super *osb,
1046 u64 src_inode_no, u64 dest_inode_no)
1048 int ret = 0, i = 0;
1049 u64 parent_inode_no = 0;
1050 u64 child_inode_no = src_inode_no;
1051 struct inode *child_inode;
1053 #define MAX_LOOKUP_TIMES 32
1054 while (1) {
1055 child_inode = ocfs2_iget(osb, child_inode_no, 0, 0);
1056 if (IS_ERR(child_inode)) {
1057 ret = PTR_ERR(child_inode);
1058 break;
1061 ret = ocfs2_inode_lock(child_inode, NULL, 0);
1062 if (ret < 0) {
1063 iput(child_inode);
1064 if (ret != -ENOENT)
1065 mlog_errno(ret);
1066 break;
1069 ret = ocfs2_lookup_ino_from_name(child_inode, "..", 2,
1070 &parent_inode_no);
1071 ocfs2_inode_unlock(child_inode, 0);
1072 iput(child_inode);
1073 if (ret < 0) {
1074 ret = -ENOENT;
1075 break;
1078 if (parent_inode_no == dest_inode_no) {
1079 ret = 1;
1080 break;
1083 if (parent_inode_no == osb->root_inode->i_ino) {
1084 ret = 0;
1085 break;
1088 child_inode_no = parent_inode_no;
1090 if (++i >= MAX_LOOKUP_TIMES) {
1091 mlog(ML_NOTICE, "max lookup times reached, filesystem "
1092 "may have nested directories, "
1093 "src inode: %llu, dest inode: %llu.\n",
1094 (unsigned long long)src_inode_no,
1095 (unsigned long long)dest_inode_no);
1096 ret = 0;
1097 break;
1101 return ret;
1105 * The only place this should be used is rename and link!
1106 * if they have the same id, then the 1st one is the only one locked.
1108 static int ocfs2_double_lock(struct ocfs2_super *osb,
1109 struct buffer_head **bh1,
1110 struct inode *inode1,
1111 struct buffer_head **bh2,
1112 struct inode *inode2,
1113 int rename)
1115 int status;
1116 int inode1_is_ancestor, inode2_is_ancestor;
1117 struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
1118 struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
1120 trace_ocfs2_double_lock((unsigned long long)oi1->ip_blkno,
1121 (unsigned long long)oi2->ip_blkno);
1123 if (*bh1)
1124 *bh1 = NULL;
1125 if (*bh2)
1126 *bh2 = NULL;
1128 /* we always want to lock the one with the lower lockid first.
1129 * and if they are nested, we lock ancestor first */
1130 if (oi1->ip_blkno != oi2->ip_blkno) {
1131 inode1_is_ancestor = ocfs2_check_if_ancestor(osb, oi2->ip_blkno,
1132 oi1->ip_blkno);
1133 if (inode1_is_ancestor < 0) {
1134 status = inode1_is_ancestor;
1135 goto bail;
1138 inode2_is_ancestor = ocfs2_check_if_ancestor(osb, oi1->ip_blkno,
1139 oi2->ip_blkno);
1140 if (inode2_is_ancestor < 0) {
1141 status = inode2_is_ancestor;
1142 goto bail;
1145 if ((inode1_is_ancestor == 1) ||
1146 (oi1->ip_blkno < oi2->ip_blkno &&
1147 inode2_is_ancestor == 0)) {
1148 /* switch id1 and id2 around */
1149 swap(bh2, bh1);
1150 swap(inode2, inode1);
1152 /* lock id2 */
1153 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
1154 rename == 1 ? OI_LS_RENAME1 : OI_LS_PARENT);
1155 if (status < 0) {
1156 if (status != -ENOENT)
1157 mlog_errno(status);
1158 goto bail;
1162 /* lock id1 */
1163 status = ocfs2_inode_lock_nested(inode1, bh1, 1,
1164 rename == 1 ? OI_LS_RENAME2 : OI_LS_PARENT);
1165 if (status < 0) {
1167 * An error return must mean that no cluster locks
1168 * were held on function exit.
1170 if (oi1->ip_blkno != oi2->ip_blkno) {
1171 ocfs2_inode_unlock(inode2, 1);
1172 brelse(*bh2);
1173 *bh2 = NULL;
1176 if (status != -ENOENT)
1177 mlog_errno(status);
1180 trace_ocfs2_double_lock_end(
1181 (unsigned long long)OCFS2_I(inode1)->ip_blkno,
1182 (unsigned long long)OCFS2_I(inode2)->ip_blkno);
1184 bail:
1185 if (status)
1186 mlog_errno(status);
1187 return status;
1190 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1192 ocfs2_inode_unlock(inode1, 1);
1194 if (inode1 != inode2)
1195 ocfs2_inode_unlock(inode2, 1);
1198 static int ocfs2_rename(struct inode *old_dir,
1199 struct dentry *old_dentry,
1200 struct inode *new_dir,
1201 struct dentry *new_dentry)
1203 int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1204 int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1205 struct inode *old_inode = d_inode(old_dentry);
1206 struct inode *new_inode = d_inode(new_dentry);
1207 struct inode *orphan_dir = NULL;
1208 struct ocfs2_dinode *newfe = NULL;
1209 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1210 struct buffer_head *newfe_bh = NULL;
1211 struct buffer_head *old_inode_bh = NULL;
1212 struct ocfs2_super *osb = NULL;
1213 u64 newfe_blkno, old_de_ino;
1214 handle_t *handle = NULL;
1215 struct buffer_head *old_dir_bh = NULL;
1216 struct buffer_head *new_dir_bh = NULL;
1217 u32 old_dir_nlink = old_dir->i_nlink;
1218 struct ocfs2_dinode *old_di;
1219 struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1220 struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1221 struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1222 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1223 struct ocfs2_dir_lookup_result target_insert = { NULL, };
1224 bool should_add_orphan = false;
1226 /* At some point it might be nice to break this function up a
1227 * bit. */
1229 trace_ocfs2_rename(old_dir, old_dentry, new_dir, new_dentry,
1230 old_dentry->d_name.len, old_dentry->d_name.name,
1231 new_dentry->d_name.len, new_dentry->d_name.name);
1233 dquot_initialize(old_dir);
1234 dquot_initialize(new_dir);
1236 osb = OCFS2_SB(old_dir->i_sb);
1238 if (new_inode) {
1239 if (!igrab(new_inode))
1240 BUG();
1243 /* Assume a directory hierarchy thusly:
1244 * a/b/c
1245 * a/d
1246 * a,b,c, and d are all directories.
1248 * from cwd of 'a' on both nodes:
1249 * node1: mv b/c d
1250 * node2: mv d b/c
1252 * And that's why, just like the VFS, we need a file system
1253 * rename lock. */
1254 if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1255 status = ocfs2_rename_lock(osb);
1256 if (status < 0) {
1257 mlog_errno(status);
1258 goto bail;
1260 rename_lock = 1;
1262 /* here we cannot guarantee the inodes haven't just been
1263 * changed, so check if they are nested again */
1264 status = ocfs2_check_if_ancestor(osb, new_dir->i_ino,
1265 old_inode->i_ino);
1266 if (status < 0) {
1267 mlog_errno(status);
1268 goto bail;
1269 } else if (status == 1) {
1270 status = -EPERM;
1271 trace_ocfs2_rename_not_permitted(
1272 (unsigned long long)old_inode->i_ino,
1273 (unsigned long long)new_dir->i_ino);
1274 goto bail;
1278 /* if old and new are the same, this'll just do one lock. */
1279 status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1280 &new_dir_bh, new_dir, 1);
1281 if (status < 0) {
1282 mlog_errno(status);
1283 goto bail;
1285 parents_locked = 1;
1287 /* make sure both dirs have bhs
1288 * get an extra ref on old_dir_bh if old==new */
1289 if (!new_dir_bh) {
1290 if (old_dir_bh) {
1291 new_dir_bh = old_dir_bh;
1292 get_bh(new_dir_bh);
1293 } else {
1294 mlog(ML_ERROR, "no old_dir_bh!\n");
1295 status = -EIO;
1296 goto bail;
1301 * Aside from allowing a meta data update, the locking here
1302 * also ensures that the downconvert thread on other nodes
1303 * won't have to concurrently downconvert the inode and the
1304 * dentry locks.
1306 status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1307 OI_LS_PARENT);
1308 if (status < 0) {
1309 if (status != -ENOENT)
1310 mlog_errno(status);
1311 goto bail;
1313 old_child_locked = 1;
1315 status = ocfs2_remote_dentry_delete(old_dentry);
1316 if (status < 0) {
1317 mlog_errno(status);
1318 goto bail;
1321 if (S_ISDIR(old_inode->i_mode)) {
1322 u64 old_inode_parent;
1324 update_dot_dot = 1;
1325 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1326 old_inode,
1327 &old_inode_dot_dot_res);
1328 if (status) {
1329 status = -EIO;
1330 goto bail;
1333 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1334 status = -EIO;
1335 goto bail;
1338 if (!new_inode && new_dir != old_dir &&
1339 new_dir->i_nlink >= ocfs2_link_max(osb)) {
1340 status = -EMLINK;
1341 goto bail;
1345 status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1346 old_dentry->d_name.len,
1347 &old_de_ino);
1348 if (status) {
1349 status = -ENOENT;
1350 goto bail;
1354 * Check for inode number is _not_ due to possible IO errors.
1355 * We might rmdir the source, keep it as pwd of some process
1356 * and merrily kill the link to whatever was created under the
1357 * same name. Goodbye sticky bit ;-<
1359 if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1360 status = -ENOENT;
1361 goto bail;
1364 /* check if the target already exists (in which case we need
1365 * to delete it */
1366 status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1367 new_dentry->d_name.len,
1368 &newfe_blkno, new_dir,
1369 &target_lookup_res);
1370 /* The only error we allow here is -ENOENT because the new
1371 * file not existing is perfectly valid. */
1372 if ((status < 0) && (status != -ENOENT)) {
1373 /* If we cannot find the file specified we should just */
1374 /* return the error... */
1375 mlog_errno(status);
1376 goto bail;
1378 if (status == 0)
1379 target_exists = 1;
1381 if (!target_exists && new_inode) {
1383 * Target was unlinked by another node while we were
1384 * waiting to get to ocfs2_rename(). There isn't
1385 * anything we can do here to help the situation, so
1386 * bubble up the appropriate error.
1388 status = -ENOENT;
1389 goto bail;
1392 /* In case we need to overwrite an existing file, we blow it
1393 * away first */
1394 if (target_exists) {
1395 /* VFS didn't think there existed an inode here, but
1396 * someone else in the cluster must have raced our
1397 * rename to create one. Today we error cleanly, in
1398 * the future we should consider calling iget to build
1399 * a new struct inode for this entry. */
1400 if (!new_inode) {
1401 status = -EACCES;
1403 trace_ocfs2_rename_target_exists(new_dentry->d_name.len,
1404 new_dentry->d_name.name);
1405 goto bail;
1408 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1409 status = -EACCES;
1411 trace_ocfs2_rename_disagree(
1412 (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1413 (unsigned long long)newfe_blkno,
1414 OCFS2_I(new_inode)->ip_flags);
1415 goto bail;
1418 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1419 if (status < 0) {
1420 if (status != -ENOENT)
1421 mlog_errno(status);
1422 goto bail;
1424 new_child_locked = 1;
1426 status = ocfs2_remote_dentry_delete(new_dentry);
1427 if (status < 0) {
1428 mlog_errno(status);
1429 goto bail;
1432 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1434 trace_ocfs2_rename_over_existing(
1435 (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1436 (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1438 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1439 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1440 OCFS2_I(new_inode)->ip_blkno,
1441 orphan_name, &orphan_insert,
1442 false);
1443 if (status < 0) {
1444 mlog_errno(status);
1445 goto bail;
1447 should_add_orphan = true;
1449 } else {
1450 BUG_ON(d_inode(new_dentry->d_parent) != new_dir);
1452 status = ocfs2_check_dir_for_entry(new_dir,
1453 new_dentry->d_name.name,
1454 new_dentry->d_name.len);
1455 if (status)
1456 goto bail;
1458 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1459 new_dentry->d_name.name,
1460 new_dentry->d_name.len,
1461 &target_insert);
1462 if (status < 0) {
1463 mlog_errno(status);
1464 goto bail;
1468 handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1469 if (IS_ERR(handle)) {
1470 status = PTR_ERR(handle);
1471 handle = NULL;
1472 mlog_errno(status);
1473 goto bail;
1476 if (target_exists) {
1477 if (S_ISDIR(new_inode->i_mode)) {
1478 if (new_inode->i_nlink != 2 ||
1479 !ocfs2_empty_dir(new_inode)) {
1480 status = -ENOTEMPTY;
1481 goto bail;
1484 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1485 newfe_bh,
1486 OCFS2_JOURNAL_ACCESS_WRITE);
1487 if (status < 0) {
1488 mlog_errno(status);
1489 goto bail;
1492 /* change the dirent to point to the correct inode */
1493 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1494 old_inode);
1495 if (status < 0) {
1496 mlog_errno(status);
1497 goto bail;
1499 new_dir->i_version++;
1501 if (S_ISDIR(new_inode->i_mode))
1502 ocfs2_set_links_count(newfe, 0);
1503 else
1504 ocfs2_add_links_count(newfe, -1);
1505 ocfs2_journal_dirty(handle, newfe_bh);
1506 if (should_add_orphan) {
1507 status = ocfs2_orphan_add(osb, handle, new_inode,
1508 newfe_bh, orphan_name,
1509 &orphan_insert, orphan_dir, false);
1510 if (status < 0) {
1511 mlog_errno(status);
1512 goto bail;
1515 } else {
1516 /* if the name was not found in new_dir, add it now */
1517 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1518 OCFS2_I(old_inode)->ip_blkno,
1519 new_dir_bh, &target_insert);
1522 old_inode->i_ctime = CURRENT_TIME;
1523 mark_inode_dirty(old_inode);
1525 status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1526 old_inode_bh,
1527 OCFS2_JOURNAL_ACCESS_WRITE);
1528 if (status >= 0) {
1529 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1531 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1532 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1533 ocfs2_journal_dirty(handle, old_inode_bh);
1534 } else
1535 mlog_errno(status);
1538 * Now that the name has been added to new_dir, remove the old name.
1540 * We don't keep any directory entry context around until now
1541 * because the insert might have changed the type of directory
1542 * we're dealing with.
1544 status = ocfs2_find_entry(old_dentry->d_name.name,
1545 old_dentry->d_name.len, old_dir,
1546 &old_entry_lookup);
1547 if (status)
1548 goto bail;
1550 status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1551 if (status < 0) {
1552 mlog_errno(status);
1553 goto bail;
1556 if (new_inode) {
1557 drop_nlink(new_inode);
1558 new_inode->i_ctime = CURRENT_TIME;
1560 old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1562 if (update_dot_dot) {
1563 status = ocfs2_update_entry(old_inode, handle,
1564 &old_inode_dot_dot_res, new_dir);
1565 drop_nlink(old_dir);
1566 if (new_inode) {
1567 drop_nlink(new_inode);
1568 } else {
1569 inc_nlink(new_dir);
1570 mark_inode_dirty(new_dir);
1573 mark_inode_dirty(old_dir);
1574 ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1575 if (new_inode) {
1576 mark_inode_dirty(new_inode);
1577 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1580 if (old_dir != new_dir) {
1581 /* Keep the same times on both directories.*/
1582 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1585 * This will also pick up the i_nlink change from the
1586 * block above.
1588 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1591 if (old_dir_nlink != old_dir->i_nlink) {
1592 if (!old_dir_bh) {
1593 mlog(ML_ERROR, "need to change nlink for old dir "
1594 "%llu from %d to %d but bh is NULL!\n",
1595 (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1596 (int)old_dir_nlink, old_dir->i_nlink);
1597 } else {
1598 struct ocfs2_dinode *fe;
1599 status = ocfs2_journal_access_di(handle,
1600 INODE_CACHE(old_dir),
1601 old_dir_bh,
1602 OCFS2_JOURNAL_ACCESS_WRITE);
1603 fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1604 ocfs2_set_links_count(fe, old_dir->i_nlink);
1605 ocfs2_journal_dirty(handle, old_dir_bh);
1608 ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1609 status = 0;
1610 bail:
1611 if (rename_lock)
1612 ocfs2_rename_unlock(osb);
1614 if (handle)
1615 ocfs2_commit_trans(osb, handle);
1617 if (parents_locked)
1618 ocfs2_double_unlock(old_dir, new_dir);
1620 if (old_child_locked)
1621 ocfs2_inode_unlock(old_inode, 1);
1623 if (new_child_locked)
1624 ocfs2_inode_unlock(new_inode, 1);
1626 if (orphan_dir) {
1627 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1628 ocfs2_inode_unlock(orphan_dir, 1);
1629 mutex_unlock(&orphan_dir->i_mutex);
1630 iput(orphan_dir);
1633 if (new_inode)
1634 sync_mapping_buffers(old_inode->i_mapping);
1636 if (new_inode)
1637 iput(new_inode);
1639 ocfs2_free_dir_lookup_result(&target_lookup_res);
1640 ocfs2_free_dir_lookup_result(&old_entry_lookup);
1641 ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1642 ocfs2_free_dir_lookup_result(&orphan_insert);
1643 ocfs2_free_dir_lookup_result(&target_insert);
1645 brelse(newfe_bh);
1646 brelse(old_inode_bh);
1647 brelse(old_dir_bh);
1648 brelse(new_dir_bh);
1650 if (status)
1651 mlog_errno(status);
1653 return status;
1657 * we expect i_size = strlen(symname). Copy symname into the file
1658 * data, including the null terminator.
1660 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1661 handle_t *handle,
1662 struct inode *inode,
1663 const char *symname)
1665 struct buffer_head **bhs = NULL;
1666 const char *c;
1667 struct super_block *sb = osb->sb;
1668 u64 p_blkno, p_blocks;
1669 int virtual, blocks, status, i, bytes_left;
1671 bytes_left = i_size_read(inode) + 1;
1672 /* we can't trust i_blocks because we're actually going to
1673 * write i_size + 1 bytes. */
1674 blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1676 trace_ocfs2_create_symlink_data((unsigned long long)inode->i_blocks,
1677 i_size_read(inode), blocks);
1679 /* Sanity check -- make sure we're going to fit. */
1680 if (bytes_left >
1681 ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1682 status = -EIO;
1683 mlog_errno(status);
1684 goto bail;
1687 bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1688 if (!bhs) {
1689 status = -ENOMEM;
1690 mlog_errno(status);
1691 goto bail;
1694 status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1695 NULL);
1696 if (status < 0) {
1697 mlog_errno(status);
1698 goto bail;
1701 /* links can never be larger than one cluster so we know this
1702 * is all going to be contiguous, but do a sanity check
1703 * anyway. */
1704 if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1705 status = -EIO;
1706 mlog_errno(status);
1707 goto bail;
1710 virtual = 0;
1711 while(bytes_left > 0) {
1712 c = &symname[virtual * sb->s_blocksize];
1714 bhs[virtual] = sb_getblk(sb, p_blkno);
1715 if (!bhs[virtual]) {
1716 status = -ENOMEM;
1717 mlog_errno(status);
1718 goto bail;
1720 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1721 bhs[virtual]);
1723 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1724 bhs[virtual],
1725 OCFS2_JOURNAL_ACCESS_CREATE);
1726 if (status < 0) {
1727 mlog_errno(status);
1728 goto bail;
1731 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1733 memcpy(bhs[virtual]->b_data, c,
1734 (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1735 bytes_left);
1737 ocfs2_journal_dirty(handle, bhs[virtual]);
1739 virtual++;
1740 p_blkno++;
1741 bytes_left -= sb->s_blocksize;
1744 status = 0;
1745 bail:
1747 if (bhs) {
1748 for(i = 0; i < blocks; i++)
1749 brelse(bhs[i]);
1750 kfree(bhs);
1753 if (status)
1754 mlog_errno(status);
1755 return status;
1758 static int ocfs2_symlink(struct inode *dir,
1759 struct dentry *dentry,
1760 const char *symname)
1762 int status, l, credits;
1763 u64 newsize;
1764 struct ocfs2_super *osb = NULL;
1765 struct inode *inode = NULL;
1766 struct super_block *sb;
1767 struct buffer_head *new_fe_bh = NULL;
1768 struct buffer_head *parent_fe_bh = NULL;
1769 struct ocfs2_dinode *fe = NULL;
1770 struct ocfs2_dinode *dirfe;
1771 handle_t *handle = NULL;
1772 struct ocfs2_alloc_context *inode_ac = NULL;
1773 struct ocfs2_alloc_context *data_ac = NULL;
1774 struct ocfs2_alloc_context *xattr_ac = NULL;
1775 int want_clusters = 0;
1776 int xattr_credits = 0;
1777 struct ocfs2_security_xattr_info si = {
1778 .enable = 1,
1780 int did_quota = 0, did_quota_inode = 0;
1781 struct ocfs2_dir_lookup_result lookup = { NULL, };
1782 sigset_t oldset;
1783 int did_block_signals = 0;
1784 struct ocfs2_dentry_lock *dl = NULL;
1786 trace_ocfs2_symlink_begin(dir, dentry, symname,
1787 dentry->d_name.len, dentry->d_name.name);
1789 dquot_initialize(dir);
1791 sb = dir->i_sb;
1792 osb = OCFS2_SB(sb);
1794 l = strlen(symname) + 1;
1796 credits = ocfs2_calc_symlink_credits(sb);
1798 /* lock the parent directory */
1799 status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1800 if (status < 0) {
1801 if (status != -ENOENT)
1802 mlog_errno(status);
1803 return status;
1806 dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1807 if (!ocfs2_read_links_count(dirfe)) {
1808 /* can't make a file in a deleted directory. */
1809 status = -ENOENT;
1810 goto bail;
1813 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1814 dentry->d_name.len);
1815 if (status)
1816 goto bail;
1818 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1819 dentry->d_name.name,
1820 dentry->d_name.len, &lookup);
1821 if (status < 0) {
1822 mlog_errno(status);
1823 goto bail;
1826 status = ocfs2_reserve_new_inode(osb, &inode_ac);
1827 if (status < 0) {
1828 if (status != -ENOSPC)
1829 mlog_errno(status);
1830 goto bail;
1833 inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1834 if (!inode) {
1835 status = -ENOMEM;
1836 mlog_errno(status);
1837 goto bail;
1840 /* get security xattr */
1841 status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
1842 if (status) {
1843 if (status == -EOPNOTSUPP)
1844 si.enable = 0;
1845 else {
1846 mlog_errno(status);
1847 goto bail;
1851 /* calculate meta data/clusters for setting security xattr */
1852 if (si.enable) {
1853 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1854 &xattr_credits, &xattr_ac);
1855 if (status < 0) {
1856 mlog_errno(status);
1857 goto bail;
1861 /* don't reserve bitmap space for fast symlinks. */
1862 if (l > ocfs2_fast_symlink_chars(sb))
1863 want_clusters += 1;
1865 status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1866 if (status < 0) {
1867 if (status != -ENOSPC)
1868 mlog_errno(status);
1869 goto bail;
1872 handle = ocfs2_start_trans(osb, credits + xattr_credits);
1873 if (IS_ERR(handle)) {
1874 status = PTR_ERR(handle);
1875 handle = NULL;
1876 mlog_errno(status);
1877 goto bail;
1880 /* Starting to change things, restart is no longer possible. */
1881 ocfs2_block_signals(&oldset);
1882 did_block_signals = 1;
1884 status = dquot_alloc_inode(inode);
1885 if (status)
1886 goto bail;
1887 did_quota_inode = 1;
1889 trace_ocfs2_symlink_create(dir, dentry, dentry->d_name.len,
1890 dentry->d_name.name,
1891 (unsigned long long)OCFS2_I(dir)->ip_blkno,
1892 inode->i_mode);
1894 status = ocfs2_mknod_locked(osb, dir, inode,
1895 0, &new_fe_bh, parent_fe_bh, handle,
1896 inode_ac);
1897 if (status < 0) {
1898 mlog_errno(status);
1899 goto bail;
1902 fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1903 inode->i_rdev = 0;
1904 newsize = l - 1;
1905 inode->i_op = &ocfs2_symlink_inode_operations;
1906 if (l > ocfs2_fast_symlink_chars(sb)) {
1907 u32 offset = 0;
1909 status = dquot_alloc_space_nodirty(inode,
1910 ocfs2_clusters_to_bytes(osb->sb, 1));
1911 if (status)
1912 goto bail;
1913 did_quota = 1;
1914 inode->i_mapping->a_ops = &ocfs2_aops;
1915 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1916 new_fe_bh,
1917 handle, data_ac, NULL,
1918 NULL);
1919 if (status < 0) {
1920 if (status != -ENOSPC && status != -EINTR) {
1921 mlog(ML_ERROR,
1922 "Failed to extend file to %llu\n",
1923 (unsigned long long)newsize);
1924 mlog_errno(status);
1925 status = -ENOSPC;
1927 goto bail;
1929 i_size_write(inode, newsize);
1930 inode->i_blocks = ocfs2_inode_sector_count(inode);
1931 } else {
1932 inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
1933 memcpy((char *) fe->id2.i_symlink, symname, l);
1934 i_size_write(inode, newsize);
1935 inode->i_blocks = 0;
1938 status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1939 if (status < 0) {
1940 mlog_errno(status);
1941 goto bail;
1944 if (!ocfs2_inode_is_fast_symlink(inode)) {
1945 status = ocfs2_create_symlink_data(osb, handle, inode,
1946 symname);
1947 if (status < 0) {
1948 mlog_errno(status);
1949 goto bail;
1953 if (si.enable) {
1954 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1955 xattr_ac, data_ac);
1956 if (status < 0) {
1957 mlog_errno(status);
1958 goto bail;
1963 * Do this before adding the entry to the directory. We add
1964 * also set d_op after success so that ->d_iput() will cleanup
1965 * the dentry lock even if ocfs2_add_entry() fails below.
1967 status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1968 if (status) {
1969 mlog_errno(status);
1970 goto bail;
1973 dl = dentry->d_fsdata;
1975 status = ocfs2_add_entry(handle, dentry, inode,
1976 le64_to_cpu(fe->i_blkno), parent_fe_bh,
1977 &lookup);
1978 if (status < 0) {
1979 mlog_errno(status);
1980 goto bail;
1983 insert_inode_hash(inode);
1984 d_instantiate(dentry, inode);
1985 bail:
1986 if (status < 0 && did_quota)
1987 dquot_free_space_nodirty(inode,
1988 ocfs2_clusters_to_bytes(osb->sb, 1));
1989 if (status < 0 && did_quota_inode)
1990 dquot_free_inode(inode);
1991 if (handle)
1992 ocfs2_commit_trans(osb, handle);
1994 ocfs2_inode_unlock(dir, 1);
1995 if (did_block_signals)
1996 ocfs2_unblock_signals(&oldset);
1998 brelse(new_fe_bh);
1999 brelse(parent_fe_bh);
2000 kfree(si.value);
2001 ocfs2_free_dir_lookup_result(&lookup);
2002 if (inode_ac)
2003 ocfs2_free_alloc_context(inode_ac);
2004 if (data_ac)
2005 ocfs2_free_alloc_context(data_ac);
2006 if (xattr_ac)
2007 ocfs2_free_alloc_context(xattr_ac);
2008 if ((status < 0) && inode) {
2009 if (dl)
2010 ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
2012 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
2013 clear_nlink(inode);
2014 iput(inode);
2017 if (status)
2018 mlog_errno(status);
2020 return status;
2023 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2025 int status, namelen;
2027 namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2028 (long long)blkno);
2029 if (namelen <= 0) {
2030 if (namelen)
2031 status = namelen;
2032 else
2033 status = -EINVAL;
2034 mlog_errno(status);
2035 goto bail;
2037 if (namelen != OCFS2_ORPHAN_NAMELEN) {
2038 status = -EINVAL;
2039 mlog_errno(status);
2040 goto bail;
2043 trace_ocfs2_blkno_stringify(blkno, name, namelen);
2045 status = 0;
2046 bail:
2047 if (status < 0)
2048 mlog_errno(status);
2049 return status;
2052 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
2053 struct inode **ret_orphan_dir,
2054 struct buffer_head **ret_orphan_dir_bh)
2056 struct inode *orphan_dir_inode;
2057 struct buffer_head *orphan_dir_bh = NULL;
2058 int ret = 0;
2060 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2061 ORPHAN_DIR_SYSTEM_INODE,
2062 osb->slot_num);
2063 if (!orphan_dir_inode) {
2064 ret = -ENOENT;
2065 mlog_errno(ret);
2066 return ret;
2069 mutex_lock(&orphan_dir_inode->i_mutex);
2071 ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2072 if (ret < 0) {
2073 mutex_unlock(&orphan_dir_inode->i_mutex);
2074 iput(orphan_dir_inode);
2076 mlog_errno(ret);
2077 return ret;
2080 *ret_orphan_dir = orphan_dir_inode;
2081 *ret_orphan_dir_bh = orphan_dir_bh;
2083 return 0;
2086 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
2087 struct buffer_head *orphan_dir_bh,
2088 u64 blkno,
2089 char *name,
2090 struct ocfs2_dir_lookup_result *lookup,
2091 bool dio)
2093 int ret;
2094 struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
2095 int namelen = dio ?
2096 (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2097 OCFS2_ORPHAN_NAMELEN;
2099 if (dio) {
2100 ret = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2101 OCFS2_DIO_ORPHAN_PREFIX);
2102 if (ret != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2103 ret = -EINVAL;
2104 mlog_errno(ret);
2105 return ret;
2108 ret = ocfs2_blkno_stringify(blkno,
2109 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2110 } else
2111 ret = ocfs2_blkno_stringify(blkno, name);
2112 if (ret < 0) {
2113 mlog_errno(ret);
2114 return ret;
2117 ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2118 orphan_dir_bh, name,
2119 namelen, lookup);
2120 if (ret < 0) {
2121 mlog_errno(ret);
2122 return ret;
2125 return 0;
2129 * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
2130 * insertion of an orphan.
2131 * @osb: ocfs2 file system
2132 * @ret_orphan_dir: Orphan dir inode - returned locked!
2133 * @blkno: Actual block number of the inode to be inserted into orphan dir.
2134 * @lookup: dir lookup result, to be passed back into functions like
2135 * ocfs2_orphan_add
2137 * Returns zero on success and the ret_orphan_dir, name and lookup
2138 * fields will be populated.
2140 * Returns non-zero on failure.
2142 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2143 struct inode **ret_orphan_dir,
2144 u64 blkno,
2145 char *name,
2146 struct ocfs2_dir_lookup_result *lookup,
2147 bool dio)
2149 struct inode *orphan_dir_inode = NULL;
2150 struct buffer_head *orphan_dir_bh = NULL;
2151 int ret = 0;
2153 ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
2154 &orphan_dir_bh);
2155 if (ret < 0) {
2156 mlog_errno(ret);
2157 return ret;
2160 ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
2161 blkno, name, lookup, dio);
2162 if (ret < 0) {
2163 mlog_errno(ret);
2164 goto out;
2167 *ret_orphan_dir = orphan_dir_inode;
2169 out:
2170 brelse(orphan_dir_bh);
2172 if (ret) {
2173 ocfs2_inode_unlock(orphan_dir_inode, 1);
2174 mutex_unlock(&orphan_dir_inode->i_mutex);
2175 iput(orphan_dir_inode);
2178 if (ret)
2179 mlog_errno(ret);
2180 return ret;
2183 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2184 handle_t *handle,
2185 struct inode *inode,
2186 struct buffer_head *fe_bh,
2187 char *name,
2188 struct ocfs2_dir_lookup_result *lookup,
2189 struct inode *orphan_dir_inode,
2190 bool dio)
2192 struct buffer_head *orphan_dir_bh = NULL;
2193 int status = 0;
2194 struct ocfs2_dinode *orphan_fe;
2195 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
2196 int namelen = dio ?
2197 (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2198 OCFS2_ORPHAN_NAMELEN;
2200 trace_ocfs2_orphan_add_begin(
2201 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2203 status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2204 if (status < 0) {
2205 mlog_errno(status);
2206 goto leave;
2209 status = ocfs2_journal_access_di(handle,
2210 INODE_CACHE(orphan_dir_inode),
2211 orphan_dir_bh,
2212 OCFS2_JOURNAL_ACCESS_WRITE);
2213 if (status < 0) {
2214 mlog_errno(status);
2215 goto leave;
2219 * We're going to journal the change of i_flags and i_orphaned_slot.
2220 * It's safe anyway, though some callers may duplicate the journaling.
2221 * Journaling within the func just make the logic look more
2222 * straightforward.
2224 status = ocfs2_journal_access_di(handle,
2225 INODE_CACHE(inode),
2226 fe_bh,
2227 OCFS2_JOURNAL_ACCESS_WRITE);
2228 if (status < 0) {
2229 mlog_errno(status);
2230 goto leave;
2233 /* we're a cluster, and nlink can change on disk from
2234 * underneath us... */
2235 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2236 if (S_ISDIR(inode->i_mode))
2237 ocfs2_add_links_count(orphan_fe, 1);
2238 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2239 ocfs2_journal_dirty(handle, orphan_dir_bh);
2241 status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2242 namelen, inode,
2243 OCFS2_I(inode)->ip_blkno,
2244 orphan_dir_bh, lookup);
2245 if (status < 0) {
2246 mlog_errno(status);
2247 goto rollback;
2250 if (dio) {
2251 /* Update flag OCFS2_DIO_ORPHANED_FL and record the orphan
2252 * slot.
2254 fe->i_flags |= cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2255 fe->i_dio_orphaned_slot = cpu_to_le16(osb->slot_num);
2256 } else {
2257 fe->i_flags |= cpu_to_le32(OCFS2_ORPHANED_FL);
2258 OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2260 /* Record which orphan dir our inode now resides
2261 * in. delete_inode will use this to determine which orphan
2262 * dir to lock. */
2263 fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2266 ocfs2_journal_dirty(handle, fe_bh);
2268 trace_ocfs2_orphan_add_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
2269 osb->slot_num);
2271 rollback:
2272 if (status < 0) {
2273 if (S_ISDIR(inode->i_mode))
2274 ocfs2_add_links_count(orphan_fe, -1);
2275 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2278 leave:
2279 brelse(orphan_dir_bh);
2281 return status;
2284 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2285 int ocfs2_orphan_del(struct ocfs2_super *osb,
2286 handle_t *handle,
2287 struct inode *orphan_dir_inode,
2288 struct inode *inode,
2289 struct buffer_head *orphan_dir_bh,
2290 bool dio)
2292 const int namelen = OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN;
2293 char name[namelen + 1];
2294 struct ocfs2_dinode *orphan_fe;
2295 int status = 0;
2296 struct ocfs2_dir_lookup_result lookup = { NULL, };
2298 if (dio) {
2299 status = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2300 OCFS2_DIO_ORPHAN_PREFIX);
2301 if (status != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2302 status = -EINVAL;
2303 mlog_errno(status);
2304 return status;
2307 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno,
2308 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2309 } else
2310 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2311 if (status < 0) {
2312 mlog_errno(status);
2313 goto leave;
2316 trace_ocfs2_orphan_del(
2317 (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2318 name, strlen(name));
2320 /* find it's spot in the orphan directory */
2321 status = ocfs2_find_entry(name, strlen(name), orphan_dir_inode,
2322 &lookup);
2323 if (status) {
2324 mlog_errno(status);
2325 goto leave;
2328 /* remove it from the orphan directory */
2329 status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2330 if (status < 0) {
2331 mlog_errno(status);
2332 goto leave;
2335 status = ocfs2_journal_access_di(handle,
2336 INODE_CACHE(orphan_dir_inode),
2337 orphan_dir_bh,
2338 OCFS2_JOURNAL_ACCESS_WRITE);
2339 if (status < 0) {
2340 mlog_errno(status);
2341 goto leave;
2344 /* do the i_nlink dance! :) */
2345 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2346 if (S_ISDIR(inode->i_mode))
2347 ocfs2_add_links_count(orphan_fe, -1);
2348 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2349 ocfs2_journal_dirty(handle, orphan_dir_bh);
2351 leave:
2352 ocfs2_free_dir_lookup_result(&lookup);
2354 if (status)
2355 mlog_errno(status);
2356 return status;
2360 * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to receive a newly
2361 * allocated file. This is different from the typical 'add to orphan dir'
2362 * operation in that the inode does not yet exist. This is a problem because
2363 * the orphan dir stringifies the inode block number to come up with it's
2364 * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2365 * problem. This function works around it by calling deeper into the orphan
2366 * and suballoc code than other callers. Use this only by necessity.
2367 * @dir: The directory which this inode will ultimately wind up under - not the
2368 * orphan dir!
2369 * @dir_bh: buffer_head the @dir inode block
2370 * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2371 * with the string to be used for orphan dirent. Pass back to the orphan dir
2372 * code.
2373 * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2374 * dir code.
2375 * @ret_di_blkno: block number where the new inode will be allocated.
2376 * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2377 * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2379 * Returns zero on success and the ret_orphan_dir, name and lookup
2380 * fields will be populated.
2382 * Returns non-zero on failure.
2384 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2385 struct buffer_head *dir_bh,
2386 char *orphan_name,
2387 struct inode **ret_orphan_dir,
2388 u64 *ret_di_blkno,
2389 struct ocfs2_dir_lookup_result *orphan_insert,
2390 struct ocfs2_alloc_context **ret_inode_ac)
2392 int ret;
2393 u64 di_blkno;
2394 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2395 struct inode *orphan_dir = NULL;
2396 struct buffer_head *orphan_dir_bh = NULL;
2397 struct ocfs2_alloc_context *inode_ac = NULL;
2399 ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2400 if (ret < 0) {
2401 mlog_errno(ret);
2402 return ret;
2405 /* reserve an inode spot */
2406 ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2407 if (ret < 0) {
2408 if (ret != -ENOSPC)
2409 mlog_errno(ret);
2410 goto out;
2413 ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2414 &di_blkno);
2415 if (ret) {
2416 mlog_errno(ret);
2417 goto out;
2420 ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2421 di_blkno, orphan_name, orphan_insert,
2422 false);
2423 if (ret < 0) {
2424 mlog_errno(ret);
2425 goto out;
2428 out:
2429 if (ret == 0) {
2430 *ret_orphan_dir = orphan_dir;
2431 *ret_di_blkno = di_blkno;
2432 *ret_inode_ac = inode_ac;
2434 * orphan_name and orphan_insert are already up to
2435 * date via prepare_orphan_dir
2437 } else {
2438 /* Unroll reserve_new_inode* */
2439 if (inode_ac)
2440 ocfs2_free_alloc_context(inode_ac);
2442 /* Unroll orphan dir locking */
2443 mutex_unlock(&orphan_dir->i_mutex);
2444 ocfs2_inode_unlock(orphan_dir, 1);
2445 iput(orphan_dir);
2448 brelse(orphan_dir_bh);
2450 return ret;
2453 int ocfs2_create_inode_in_orphan(struct inode *dir,
2454 int mode,
2455 struct inode **new_inode)
2457 int status, did_quota_inode = 0;
2458 struct inode *inode = NULL;
2459 struct inode *orphan_dir = NULL;
2460 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2461 struct ocfs2_dinode *di = NULL;
2462 handle_t *handle = NULL;
2463 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2464 struct buffer_head *parent_di_bh = NULL;
2465 struct buffer_head *new_di_bh = NULL;
2466 struct ocfs2_alloc_context *inode_ac = NULL;
2467 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2468 u64 uninitialized_var(di_blkno), suballoc_loc;
2469 u16 suballoc_bit;
2471 status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2472 if (status < 0) {
2473 if (status != -ENOENT)
2474 mlog_errno(status);
2475 return status;
2478 status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2479 orphan_name, &orphan_dir,
2480 &di_blkno, &orphan_insert, &inode_ac);
2481 if (status < 0) {
2482 if (status != -ENOSPC)
2483 mlog_errno(status);
2484 goto leave;
2487 inode = ocfs2_get_init_inode(dir, mode);
2488 if (!inode) {
2489 status = -ENOMEM;
2490 mlog_errno(status);
2491 goto leave;
2494 handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2495 if (IS_ERR(handle)) {
2496 status = PTR_ERR(handle);
2497 handle = NULL;
2498 mlog_errno(status);
2499 goto leave;
2502 status = dquot_alloc_inode(inode);
2503 if (status)
2504 goto leave;
2505 did_quota_inode = 1;
2507 status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2508 &suballoc_loc,
2509 &suballoc_bit, di_blkno);
2510 if (status < 0) {
2511 mlog_errno(status);
2512 goto leave;
2515 clear_nlink(inode);
2516 /* do the real work now. */
2517 status = __ocfs2_mknod_locked(dir, inode,
2518 0, &new_di_bh, parent_di_bh, handle,
2519 inode_ac, di_blkno, suballoc_loc,
2520 suballoc_bit);
2521 if (status < 0) {
2522 mlog_errno(status);
2523 goto leave;
2526 di = (struct ocfs2_dinode *)new_di_bh->b_data;
2527 status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2528 &orphan_insert, orphan_dir, false);
2529 if (status < 0) {
2530 mlog_errno(status);
2531 goto leave;
2534 /* get open lock so that only nodes can't remove it from orphan dir. */
2535 status = ocfs2_open_lock(inode);
2536 if (status < 0)
2537 mlog_errno(status);
2539 insert_inode_hash(inode);
2540 leave:
2541 if (status < 0 && did_quota_inode)
2542 dquot_free_inode(inode);
2543 if (handle)
2544 ocfs2_commit_trans(osb, handle);
2546 if (orphan_dir) {
2547 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2548 ocfs2_inode_unlock(orphan_dir, 1);
2549 mutex_unlock(&orphan_dir->i_mutex);
2550 iput(orphan_dir);
2553 if ((status < 0) && inode) {
2554 clear_nlink(inode);
2555 iput(inode);
2558 if (inode_ac)
2559 ocfs2_free_alloc_context(inode_ac);
2561 brelse(new_di_bh);
2563 if (!status)
2564 *new_inode = inode;
2566 ocfs2_free_dir_lookup_result(&orphan_insert);
2568 ocfs2_inode_unlock(dir, 1);
2569 brelse(parent_di_bh);
2570 return status;
2573 static int ocfs2_dio_orphan_recovered(struct inode *inode)
2575 int ret;
2576 struct buffer_head *di_bh = NULL;
2577 struct ocfs2_dinode *di = NULL;
2579 ret = ocfs2_inode_lock(inode, &di_bh, 1);
2580 if (ret < 0) {
2581 mlog_errno(ret);
2582 return 0;
2585 di = (struct ocfs2_dinode *) di_bh->b_data;
2586 ret = !(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL));
2587 ocfs2_inode_unlock(inode, 1);
2588 brelse(di_bh);
2590 return ret;
2593 #define OCFS2_DIO_ORPHANED_FL_CHECK_INTERVAL 10000
2594 int ocfs2_add_inode_to_orphan(struct ocfs2_super *osb,
2595 struct inode *inode)
2597 char orphan_name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2598 struct inode *orphan_dir_inode = NULL;
2599 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2600 struct buffer_head *di_bh = NULL;
2601 int status = 0;
2602 handle_t *handle = NULL;
2603 struct ocfs2_dinode *di = NULL;
2605 restart:
2606 status = ocfs2_inode_lock(inode, &di_bh, 1);
2607 if (status < 0) {
2608 mlog_errno(status);
2609 goto bail;
2612 di = (struct ocfs2_dinode *) di_bh->b_data;
2614 * Another append dio crashed?
2615 * If so, wait for recovery first.
2617 if (unlikely(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL))) {
2618 ocfs2_inode_unlock(inode, 1);
2619 brelse(di_bh);
2620 wait_event_interruptible_timeout(OCFS2_I(inode)->append_dio_wq,
2621 ocfs2_dio_orphan_recovered(inode),
2622 msecs_to_jiffies(OCFS2_DIO_ORPHANED_FL_CHECK_INTERVAL));
2623 goto restart;
2626 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir_inode,
2627 OCFS2_I(inode)->ip_blkno,
2628 orphan_name,
2629 &orphan_insert,
2630 true);
2631 if (status < 0) {
2632 mlog_errno(status);
2633 goto bail_unlock_inode;
2636 handle = ocfs2_start_trans(osb,
2637 OCFS2_INODE_ADD_TO_ORPHAN_CREDITS);
2638 if (IS_ERR(handle)) {
2639 status = PTR_ERR(handle);
2640 goto bail_unlock_orphan;
2643 status = ocfs2_orphan_add(osb, handle, inode, di_bh, orphan_name,
2644 &orphan_insert, orphan_dir_inode, true);
2645 if (status)
2646 mlog_errno(status);
2648 ocfs2_commit_trans(osb, handle);
2650 bail_unlock_orphan:
2651 ocfs2_inode_unlock(orphan_dir_inode, 1);
2652 mutex_unlock(&orphan_dir_inode->i_mutex);
2653 iput(orphan_dir_inode);
2655 ocfs2_free_dir_lookup_result(&orphan_insert);
2657 bail_unlock_inode:
2658 ocfs2_inode_unlock(inode, 1);
2659 brelse(di_bh);
2661 bail:
2662 return status;
2665 int ocfs2_del_inode_from_orphan(struct ocfs2_super *osb,
2666 struct inode *inode, struct buffer_head *di_bh,
2667 int update_isize, loff_t end)
2669 struct inode *orphan_dir_inode = NULL;
2670 struct buffer_head *orphan_dir_bh = NULL;
2671 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2672 handle_t *handle = NULL;
2673 int status = 0;
2675 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2676 ORPHAN_DIR_SYSTEM_INODE,
2677 le16_to_cpu(di->i_dio_orphaned_slot));
2678 if (!orphan_dir_inode) {
2679 status = -ENOENT;
2680 mlog_errno(status);
2681 goto bail;
2684 mutex_lock(&orphan_dir_inode->i_mutex);
2685 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2686 if (status < 0) {
2687 mutex_unlock(&orphan_dir_inode->i_mutex);
2688 iput(orphan_dir_inode);
2689 mlog_errno(status);
2690 goto bail;
2693 handle = ocfs2_start_trans(osb,
2694 OCFS2_INODE_DEL_FROM_ORPHAN_CREDITS);
2695 if (IS_ERR(handle)) {
2696 status = PTR_ERR(handle);
2697 goto bail_unlock_orphan;
2700 BUG_ON(!(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)));
2702 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode,
2703 inode, orphan_dir_bh, true);
2704 if (status < 0) {
2705 mlog_errno(status);
2706 goto bail_commit;
2709 status = ocfs2_journal_access_di(handle,
2710 INODE_CACHE(inode),
2711 di_bh,
2712 OCFS2_JOURNAL_ACCESS_WRITE);
2713 if (status < 0) {
2714 mlog_errno(status);
2715 goto bail_commit;
2718 di->i_flags &= ~cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2719 di->i_dio_orphaned_slot = 0;
2721 if (update_isize) {
2722 status = ocfs2_set_inode_size(handle, inode, di_bh, end);
2723 if (status)
2724 mlog_errno(status);
2725 } else
2726 ocfs2_journal_dirty(handle, di_bh);
2728 bail_commit:
2729 ocfs2_commit_trans(osb, handle);
2731 bail_unlock_orphan:
2732 ocfs2_inode_unlock(orphan_dir_inode, 1);
2733 mutex_unlock(&orphan_dir_inode->i_mutex);
2734 brelse(orphan_dir_bh);
2735 iput(orphan_dir_inode);
2737 bail:
2738 return status;
2741 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2742 struct inode *inode,
2743 struct dentry *dentry)
2745 int status = 0;
2746 struct buffer_head *parent_di_bh = NULL;
2747 handle_t *handle = NULL;
2748 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2749 struct ocfs2_dinode *dir_di, *di;
2750 struct inode *orphan_dir_inode = NULL;
2751 struct buffer_head *orphan_dir_bh = NULL;
2752 struct buffer_head *di_bh = NULL;
2753 struct ocfs2_dir_lookup_result lookup = { NULL, };
2755 trace_ocfs2_mv_orphaned_inode_to_new(dir, dentry,
2756 dentry->d_name.len, dentry->d_name.name,
2757 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2758 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2760 status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2761 if (status < 0) {
2762 if (status != -ENOENT)
2763 mlog_errno(status);
2764 return status;
2767 dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2768 if (!dir_di->i_links_count) {
2769 /* can't make a file in a deleted directory. */
2770 status = -ENOENT;
2771 goto leave;
2774 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2775 dentry->d_name.len);
2776 if (status)
2777 goto leave;
2779 /* get a spot inside the dir. */
2780 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2781 dentry->d_name.name,
2782 dentry->d_name.len, &lookup);
2783 if (status < 0) {
2784 mlog_errno(status);
2785 goto leave;
2788 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2789 ORPHAN_DIR_SYSTEM_INODE,
2790 osb->slot_num);
2791 if (!orphan_dir_inode) {
2792 status = -ENOENT;
2793 mlog_errno(status);
2794 goto leave;
2797 mutex_lock(&orphan_dir_inode->i_mutex);
2799 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2800 if (status < 0) {
2801 mlog_errno(status);
2802 mutex_unlock(&orphan_dir_inode->i_mutex);
2803 iput(orphan_dir_inode);
2804 goto leave;
2807 status = ocfs2_read_inode_block(inode, &di_bh);
2808 if (status < 0) {
2809 mlog_errno(status);
2810 goto orphan_unlock;
2813 handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2814 if (IS_ERR(handle)) {
2815 status = PTR_ERR(handle);
2816 handle = NULL;
2817 mlog_errno(status);
2818 goto orphan_unlock;
2821 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2822 di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2823 if (status < 0) {
2824 mlog_errno(status);
2825 goto out_commit;
2828 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2829 orphan_dir_bh, false);
2830 if (status < 0) {
2831 mlog_errno(status);
2832 goto out_commit;
2835 di = (struct ocfs2_dinode *)di_bh->b_data;
2836 di->i_flags &= ~cpu_to_le32(OCFS2_ORPHANED_FL);
2837 di->i_orphaned_slot = 0;
2838 set_nlink(inode, 1);
2839 ocfs2_set_links_count(di, inode->i_nlink);
2840 ocfs2_update_inode_fsync_trans(handle, inode, 1);
2841 ocfs2_journal_dirty(handle, di_bh);
2843 status = ocfs2_add_entry(handle, dentry, inode,
2844 OCFS2_I(inode)->ip_blkno, parent_di_bh,
2845 &lookup);
2846 if (status < 0) {
2847 mlog_errno(status);
2848 goto out_commit;
2851 status = ocfs2_dentry_attach_lock(dentry, inode,
2852 OCFS2_I(dir)->ip_blkno);
2853 if (status) {
2854 mlog_errno(status);
2855 goto out_commit;
2858 d_instantiate(dentry, inode);
2859 status = 0;
2860 out_commit:
2861 ocfs2_commit_trans(osb, handle);
2862 orphan_unlock:
2863 ocfs2_inode_unlock(orphan_dir_inode, 1);
2864 mutex_unlock(&orphan_dir_inode->i_mutex);
2865 iput(orphan_dir_inode);
2866 leave:
2868 ocfs2_inode_unlock(dir, 1);
2870 brelse(di_bh);
2871 brelse(parent_di_bh);
2872 brelse(orphan_dir_bh);
2874 ocfs2_free_dir_lookup_result(&lookup);
2876 if (status)
2877 mlog_errno(status);
2879 return status;
2882 const struct inode_operations ocfs2_dir_iops = {
2883 .create = ocfs2_create,
2884 .lookup = ocfs2_lookup,
2885 .link = ocfs2_link,
2886 .unlink = ocfs2_unlink,
2887 .rmdir = ocfs2_unlink,
2888 .symlink = ocfs2_symlink,
2889 .mkdir = ocfs2_mkdir,
2890 .mknod = ocfs2_mknod,
2891 .rename = ocfs2_rename,
2892 .setattr = ocfs2_setattr,
2893 .getattr = ocfs2_getattr,
2894 .permission = ocfs2_permission,
2895 .setxattr = generic_setxattr,
2896 .getxattr = generic_getxattr,
2897 .listxattr = ocfs2_listxattr,
2898 .removexattr = generic_removexattr,
2899 .fiemap = ocfs2_fiemap,
2900 .get_acl = ocfs2_iop_get_acl,
2901 .set_acl = ocfs2_iop_set_acl,