perf/ring_buffer: Prevent concurent ring buffer access
[linux/fpc-iii.git] / fs / ocfs2 / namei.c
blobb7ca84bc3df7393b3a815081246caf063d36cdd8
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>
44 #include <linux/iversion.h>
46 #include <cluster/masklog.h>
48 #include "ocfs2.h"
50 #include "alloc.h"
51 #include "dcache.h"
52 #include "dir.h"
53 #include "dlmglue.h"
54 #include "extent_map.h"
55 #include "file.h"
56 #include "inode.h"
57 #include "journal.h"
58 #include "namei.h"
59 #include "suballoc.h"
60 #include "super.h"
61 #include "symlink.h"
62 #include "sysfile.h"
63 #include "uptodate.h"
64 #include "xattr.h"
65 #include "acl.h"
66 #include "ocfs2_trace.h"
68 #include "buffer_head_io.h"
70 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
71 struct inode *dir,
72 struct inode *inode,
73 dev_t dev,
74 struct buffer_head **new_fe_bh,
75 struct buffer_head *parent_fe_bh,
76 handle_t *handle,
77 struct ocfs2_alloc_context *inode_ac);
79 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
80 struct inode **ret_orphan_dir,
81 u64 blkno,
82 char *name,
83 struct ocfs2_dir_lookup_result *lookup,
84 bool dio);
86 static int ocfs2_orphan_add(struct ocfs2_super *osb,
87 handle_t *handle,
88 struct inode *inode,
89 struct buffer_head *fe_bh,
90 char *name,
91 struct ocfs2_dir_lookup_result *lookup,
92 struct inode *orphan_dir_inode,
93 bool dio);
95 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
96 handle_t *handle,
97 struct inode *inode,
98 const char *symname);
100 static int ocfs2_double_lock(struct ocfs2_super *osb,
101 struct buffer_head **bh1,
102 struct inode *inode1,
103 struct buffer_head **bh2,
104 struct inode *inode2,
105 int rename);
107 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2);
108 /* An orphan dir name is an 8 byte value, printed as a hex string */
109 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
111 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
112 unsigned int flags)
114 int status;
115 u64 blkno;
116 struct inode *inode = NULL;
117 struct dentry *ret;
118 struct ocfs2_inode_info *oi;
120 trace_ocfs2_lookup(dir, dentry, dentry->d_name.len,
121 dentry->d_name.name,
122 (unsigned long long)OCFS2_I(dir)->ip_blkno, 0);
124 if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
125 ret = ERR_PTR(-ENAMETOOLONG);
126 goto bail;
129 status = ocfs2_inode_lock_nested(dir, NULL, 0, OI_LS_PARENT);
130 if (status < 0) {
131 if (status != -ENOENT)
132 mlog_errno(status);
133 ret = ERR_PTR(status);
134 goto bail;
137 status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
138 dentry->d_name.len, &blkno);
139 if (status < 0)
140 goto bail_add;
142 inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
143 if (IS_ERR(inode)) {
144 ret = ERR_PTR(-EACCES);
145 goto bail_unlock;
148 oi = OCFS2_I(inode);
149 /* Clear any orphaned state... If we were able to look up the
150 * inode from a directory, it certainly can't be orphaned. We
151 * might have the bad state from a node which intended to
152 * orphan this inode but crashed before it could commit the
153 * unlink. */
154 spin_lock(&oi->ip_lock);
155 oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
156 spin_unlock(&oi->ip_lock);
158 bail_add:
159 ret = d_splice_alias(inode, dentry);
161 if (inode) {
163 * If d_splice_alias() finds a DCACHE_DISCONNECTED
164 * dentry, it will d_move() it on top of ourse. The
165 * return value will indicate this however, so in
166 * those cases, we switch them around for the locking
167 * code.
169 * NOTE: This dentry already has ->d_op set from
170 * ocfs2_get_parent() and ocfs2_get_dentry()
172 if (!IS_ERR_OR_NULL(ret))
173 dentry = ret;
175 status = ocfs2_dentry_attach_lock(dentry, inode,
176 OCFS2_I(dir)->ip_blkno);
177 if (status) {
178 mlog_errno(status);
179 ret = ERR_PTR(status);
180 goto bail_unlock;
182 } else
183 ocfs2_dentry_attach_gen(dentry);
185 bail_unlock:
186 /* Don't drop the cluster lock until *after* the d_add --
187 * unlink on another node will message us to remove that
188 * dentry under this lock so otherwise we can race this with
189 * the downconvert thread and have a stale dentry. */
190 ocfs2_inode_unlock(dir, 0);
192 bail:
194 trace_ocfs2_lookup_ret(ret);
196 return ret;
199 static struct inode *ocfs2_get_init_inode(struct inode *dir, umode_t mode)
201 struct inode *inode;
202 int status;
204 inode = new_inode(dir->i_sb);
205 if (!inode) {
206 mlog(ML_ERROR, "new_inode failed!\n");
207 return ERR_PTR(-ENOMEM);
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 status = dquot_initialize(inode);
217 if (status)
218 return ERR_PTR(status);
220 return inode;
223 static void ocfs2_cleanup_add_entry_failure(struct ocfs2_super *osb,
224 struct dentry *dentry, struct inode *inode)
226 struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
228 ocfs2_simple_drop_lockres(osb, &dl->dl_lockres);
229 ocfs2_lock_res_free(&dl->dl_lockres);
230 BUG_ON(dl->dl_count != 1);
231 spin_lock(&dentry_attach_lock);
232 dentry->d_fsdata = NULL;
233 spin_unlock(&dentry_attach_lock);
234 kfree(dl);
235 iput(inode);
238 static int ocfs2_mknod(struct inode *dir,
239 struct dentry *dentry,
240 umode_t mode,
241 dev_t dev)
243 int status = 0;
244 struct buffer_head *parent_fe_bh = NULL;
245 handle_t *handle = NULL;
246 struct ocfs2_super *osb;
247 struct ocfs2_dinode *dirfe;
248 struct buffer_head *new_fe_bh = NULL;
249 struct inode *inode = NULL;
250 struct ocfs2_alloc_context *inode_ac = NULL;
251 struct ocfs2_alloc_context *data_ac = NULL;
252 struct ocfs2_alloc_context *meta_ac = NULL;
253 int want_clusters = 0;
254 int want_meta = 0;
255 int xattr_credits = 0;
256 struct ocfs2_security_xattr_info si = {
257 .enable = 1,
259 int did_quota_inode = 0;
260 struct ocfs2_dir_lookup_result lookup = { NULL, };
261 sigset_t oldset;
262 int did_block_signals = 0;
263 struct ocfs2_dentry_lock *dl = NULL;
265 trace_ocfs2_mknod(dir, dentry, dentry->d_name.len, dentry->d_name.name,
266 (unsigned long long)OCFS2_I(dir)->ip_blkno,
267 (unsigned long)dev, mode);
269 status = dquot_initialize(dir);
270 if (status) {
271 mlog_errno(status);
272 return status;
275 /* get our super block */
276 osb = OCFS2_SB(dir->i_sb);
278 status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
279 if (status < 0) {
280 if (status != -ENOENT)
281 mlog_errno(status);
282 return status;
285 if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
286 status = -EMLINK;
287 goto leave;
290 dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
291 if (!ocfs2_read_links_count(dirfe)) {
292 /* can't make a file in a deleted directory. */
293 status = -ENOENT;
294 goto leave;
297 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
298 dentry->d_name.len);
299 if (status)
300 goto leave;
302 /* get a spot inside the dir. */
303 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
304 dentry->d_name.name,
305 dentry->d_name.len, &lookup);
306 if (status < 0) {
307 mlog_errno(status);
308 goto leave;
311 /* reserve an inode spot */
312 status = ocfs2_reserve_new_inode(osb, &inode_ac);
313 if (status < 0) {
314 if (status != -ENOSPC)
315 mlog_errno(status);
316 goto leave;
319 inode = ocfs2_get_init_inode(dir, mode);
320 if (IS_ERR(inode)) {
321 status = PTR_ERR(inode);
322 inode = NULL;
323 mlog_errno(status);
324 goto leave;
327 /* get security xattr */
328 status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
329 if (status) {
330 if (status == -EOPNOTSUPP)
331 si.enable = 0;
332 else {
333 mlog_errno(status);
334 goto leave;
338 /* calculate meta data/clusters for setting security and acl xattr */
339 status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
340 &si, &want_clusters,
341 &xattr_credits, &want_meta);
342 if (status < 0) {
343 mlog_errno(status);
344 goto leave;
347 /* Reserve a cluster if creating an extent based directory. */
348 if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
349 want_clusters += 1;
351 /* Dir indexing requires extra space as well */
352 if (ocfs2_supports_indexed_dirs(osb))
353 want_meta++;
356 status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
357 if (status < 0) {
358 if (status != -ENOSPC)
359 mlog_errno(status);
360 goto leave;
363 status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
364 if (status < 0) {
365 if (status != -ENOSPC)
366 mlog_errno(status);
367 goto leave;
370 handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
371 S_ISDIR(mode),
372 xattr_credits));
373 if (IS_ERR(handle)) {
374 status = PTR_ERR(handle);
375 handle = NULL;
376 mlog_errno(status);
377 goto leave;
380 /* Starting to change things, restart is no longer possible. */
381 ocfs2_block_signals(&oldset);
382 did_block_signals = 1;
384 status = dquot_alloc_inode(inode);
385 if (status)
386 goto leave;
387 did_quota_inode = 1;
389 /* do the real work now. */
390 status = ocfs2_mknod_locked(osb, dir, inode, dev,
391 &new_fe_bh, parent_fe_bh, handle,
392 inode_ac);
393 if (status < 0) {
394 mlog_errno(status);
395 goto leave;
398 if (S_ISDIR(mode)) {
399 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
400 new_fe_bh, data_ac, meta_ac);
401 if (status < 0) {
402 mlog_errno(status);
403 goto leave;
406 status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
407 parent_fe_bh,
408 OCFS2_JOURNAL_ACCESS_WRITE);
409 if (status < 0) {
410 mlog_errno(status);
411 goto leave;
413 ocfs2_add_links_count(dirfe, 1);
414 ocfs2_journal_dirty(handle, parent_fe_bh);
415 inc_nlink(dir);
418 status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
419 meta_ac, data_ac);
421 if (status < 0) {
422 mlog_errno(status);
423 goto leave;
426 if (si.enable) {
427 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
428 meta_ac, data_ac);
429 if (status < 0) {
430 mlog_errno(status);
431 goto leave;
436 * Do this before adding the entry to the directory. We add
437 * also set d_op after success so that ->d_iput() will cleanup
438 * the dentry lock even if ocfs2_add_entry() fails below.
440 status = ocfs2_dentry_attach_lock(dentry, inode,
441 OCFS2_I(dir)->ip_blkno);
442 if (status) {
443 mlog_errno(status);
444 goto leave;
447 dl = dentry->d_fsdata;
449 status = ocfs2_add_entry(handle, dentry, inode,
450 OCFS2_I(inode)->ip_blkno, parent_fe_bh,
451 &lookup);
452 if (status < 0) {
453 mlog_errno(status);
454 goto leave;
457 insert_inode_hash(inode);
458 d_instantiate(dentry, inode);
459 status = 0;
460 leave:
461 if (status < 0 && did_quota_inode)
462 dquot_free_inode(inode);
463 if (handle)
464 ocfs2_commit_trans(osb, handle);
466 ocfs2_inode_unlock(dir, 1);
467 if (did_block_signals)
468 ocfs2_unblock_signals(&oldset);
470 brelse(new_fe_bh);
471 brelse(parent_fe_bh);
472 kfree(si.value);
474 ocfs2_free_dir_lookup_result(&lookup);
476 if (inode_ac)
477 ocfs2_free_alloc_context(inode_ac);
479 if (data_ac)
480 ocfs2_free_alloc_context(data_ac);
482 if (meta_ac)
483 ocfs2_free_alloc_context(meta_ac);
486 * We should call iput after the i_mutex of the bitmap been
487 * unlocked in ocfs2_free_alloc_context, or the
488 * ocfs2_delete_inode will mutex_lock again.
490 if ((status < 0) && inode) {
491 if (dl)
492 ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
494 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
495 clear_nlink(inode);
496 iput(inode);
499 if (status)
500 mlog_errno(status);
502 return status;
505 static int __ocfs2_mknod_locked(struct inode *dir,
506 struct inode *inode,
507 dev_t dev,
508 struct buffer_head **new_fe_bh,
509 struct buffer_head *parent_fe_bh,
510 handle_t *handle,
511 struct ocfs2_alloc_context *inode_ac,
512 u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
514 int status = 0;
515 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
516 struct ocfs2_dinode *fe = NULL;
517 struct ocfs2_extent_list *fel;
518 u16 feat;
519 struct ocfs2_inode_info *oi = OCFS2_I(inode);
520 struct timespec64 ts;
522 *new_fe_bh = NULL;
524 /* populate as many fields early on as possible - many of
525 * these are used by the support functions here and in
526 * callers. */
527 inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
528 oi->ip_blkno = fe_blkno;
529 spin_lock(&osb->osb_lock);
530 inode->i_generation = osb->s_next_generation++;
531 spin_unlock(&osb->osb_lock);
533 *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
534 if (!*new_fe_bh) {
535 status = -ENOMEM;
536 mlog_errno(status);
537 goto leave;
539 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
541 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
542 *new_fe_bh,
543 OCFS2_JOURNAL_ACCESS_CREATE);
544 if (status < 0) {
545 mlog_errno(status);
546 goto leave;
549 fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
550 memset(fe, 0, osb->sb->s_blocksize);
552 fe->i_generation = cpu_to_le32(inode->i_generation);
553 fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
554 fe->i_blkno = cpu_to_le64(fe_blkno);
555 fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
556 fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
557 fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
558 fe->i_uid = cpu_to_le32(i_uid_read(inode));
559 fe->i_gid = cpu_to_le32(i_gid_read(inode));
560 fe->i_mode = cpu_to_le16(inode->i_mode);
561 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
562 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
564 ocfs2_set_links_count(fe, inode->i_nlink);
566 fe->i_last_eb_blk = 0;
567 strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
568 fe->i_flags |= cpu_to_le32(OCFS2_VALID_FL);
569 ktime_get_real_ts64(&ts);
570 fe->i_atime = fe->i_ctime = fe->i_mtime =
571 cpu_to_le64(ts.tv_sec);
572 fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
573 cpu_to_le32(ts.tv_nsec);
574 fe->i_dtime = 0;
577 * If supported, directories start with inline data. If inline
578 * isn't supported, but indexing is, we start them as indexed.
580 feat = le16_to_cpu(fe->i_dyn_features);
581 if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
582 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
584 fe->id2.i_data.id_count = cpu_to_le16(
585 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
586 } else {
587 fel = &fe->id2.i_list;
588 fel->l_tree_depth = 0;
589 fel->l_next_free_rec = 0;
590 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
593 ocfs2_journal_dirty(handle, *new_fe_bh);
595 ocfs2_populate_inode(inode, fe, 1);
596 ocfs2_ci_set_new(osb, INODE_CACHE(inode));
597 if (!ocfs2_mount_local(osb)) {
598 status = ocfs2_create_new_inode_locks(inode);
599 if (status < 0)
600 mlog_errno(status);
603 oi->i_sync_tid = handle->h_transaction->t_tid;
604 oi->i_datasync_tid = handle->h_transaction->t_tid;
606 leave:
607 if (status < 0) {
608 if (*new_fe_bh) {
609 brelse(*new_fe_bh);
610 *new_fe_bh = NULL;
614 if (status)
615 mlog_errno(status);
616 return status;
619 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
620 struct inode *dir,
621 struct inode *inode,
622 dev_t dev,
623 struct buffer_head **new_fe_bh,
624 struct buffer_head *parent_fe_bh,
625 handle_t *handle,
626 struct ocfs2_alloc_context *inode_ac)
628 int status = 0;
629 u64 suballoc_loc, fe_blkno = 0;
630 u16 suballoc_bit;
632 *new_fe_bh = NULL;
634 status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
635 inode_ac, &suballoc_loc,
636 &suballoc_bit, &fe_blkno);
637 if (status < 0) {
638 mlog_errno(status);
639 return status;
642 status = __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
643 parent_fe_bh, handle, inode_ac,
644 fe_blkno, suballoc_loc, suballoc_bit);
645 if (status < 0) {
646 u64 bg_blkno = ocfs2_which_suballoc_group(fe_blkno, suballoc_bit);
647 int tmp = ocfs2_free_suballoc_bits(handle, inode_ac->ac_inode,
648 inode_ac->ac_bh, suballoc_bit, bg_blkno, 1);
649 if (tmp)
650 mlog_errno(tmp);
653 return status;
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 err = dquot_initialize(dir);
712 if (err) {
713 mlog_errno(err);
714 return err;
717 err = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
718 &parent_fe_bh, dir, 0);
719 if (err < 0) {
720 if (err != -ENOENT)
721 mlog_errno(err);
722 return err;
725 /* make sure both dirs have bhs
726 * get an extra ref on old_dir_bh if old==new */
727 if (!parent_fe_bh) {
728 if (old_dir_bh) {
729 parent_fe_bh = old_dir_bh;
730 get_bh(parent_fe_bh);
731 } else {
732 mlog(ML_ERROR, "%s: no old_dir_bh!\n", osb->uuid_str);
733 err = -EIO;
734 goto out;
738 if (!dir->i_nlink) {
739 err = -ENOENT;
740 goto out;
743 err = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
744 old_dentry->d_name.len, &old_de_ino);
745 if (err) {
746 err = -ENOENT;
747 goto out;
751 * Check whether another node removed the source inode while we
752 * were in the vfs.
754 if (old_de_ino != OCFS2_I(inode)->ip_blkno) {
755 err = -ENOENT;
756 goto out;
759 err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
760 dentry->d_name.len);
761 if (err)
762 goto out;
764 err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
765 dentry->d_name.name,
766 dentry->d_name.len, &lookup);
767 if (err < 0) {
768 mlog_errno(err);
769 goto out;
772 err = ocfs2_inode_lock(inode, &fe_bh, 1);
773 if (err < 0) {
774 if (err != -ENOENT)
775 mlog_errno(err);
776 goto out;
779 fe = (struct ocfs2_dinode *) fe_bh->b_data;
780 if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
781 err = -EMLINK;
782 goto out_unlock_inode;
785 handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
786 if (IS_ERR(handle)) {
787 err = PTR_ERR(handle);
788 handle = NULL;
789 mlog_errno(err);
790 goto out_unlock_inode;
793 /* Starting to change things, restart is no longer possible. */
794 ocfs2_block_signals(&oldset);
796 err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
797 OCFS2_JOURNAL_ACCESS_WRITE);
798 if (err < 0) {
799 mlog_errno(err);
800 goto out_commit;
803 inc_nlink(inode);
804 inode->i_ctime = current_time(inode);
805 ocfs2_set_links_count(fe, inode->i_nlink);
806 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
807 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
808 ocfs2_journal_dirty(handle, fe_bh);
810 err = ocfs2_add_entry(handle, dentry, inode,
811 OCFS2_I(inode)->ip_blkno,
812 parent_fe_bh, &lookup);
813 if (err) {
814 ocfs2_add_links_count(fe, -1);
815 drop_nlink(inode);
816 mlog_errno(err);
817 goto out_commit;
820 err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
821 if (err) {
822 mlog_errno(err);
823 goto out_commit;
826 ihold(inode);
827 d_instantiate(dentry, inode);
829 out_commit:
830 ocfs2_commit_trans(osb, handle);
831 ocfs2_unblock_signals(&oldset);
832 out_unlock_inode:
833 ocfs2_inode_unlock(inode, 1);
835 out:
836 ocfs2_double_unlock(old_dir, dir);
838 brelse(fe_bh);
839 brelse(parent_fe_bh);
840 brelse(old_dir_bh);
842 ocfs2_free_dir_lookup_result(&lookup);
844 if (err)
845 mlog_errno(err);
847 return err;
851 * Takes and drops an exclusive lock on the given dentry. This will
852 * force other nodes to drop it.
854 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
856 int ret;
858 ret = ocfs2_dentry_lock(dentry, 1);
859 if (ret)
860 mlog_errno(ret);
861 else
862 ocfs2_dentry_unlock(dentry, 1);
864 return ret;
867 static inline int ocfs2_inode_is_unlinkable(struct inode *inode)
869 if (S_ISDIR(inode->i_mode)) {
870 if (inode->i_nlink == 2)
871 return 1;
872 return 0;
875 if (inode->i_nlink == 1)
876 return 1;
877 return 0;
880 static int ocfs2_unlink(struct inode *dir,
881 struct dentry *dentry)
883 int status;
884 int child_locked = 0;
885 bool is_unlinkable = false;
886 struct inode *inode = d_inode(dentry);
887 struct inode *orphan_dir = NULL;
888 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
889 u64 blkno;
890 struct ocfs2_dinode *fe = NULL;
891 struct buffer_head *fe_bh = NULL;
892 struct buffer_head *parent_node_bh = NULL;
893 handle_t *handle = NULL;
894 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
895 struct ocfs2_dir_lookup_result lookup = { NULL, };
896 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
898 trace_ocfs2_unlink(dir, dentry, dentry->d_name.len,
899 dentry->d_name.name,
900 (unsigned long long)OCFS2_I(dir)->ip_blkno,
901 (unsigned long long)OCFS2_I(inode)->ip_blkno);
903 status = dquot_initialize(dir);
904 if (status) {
905 mlog_errno(status);
906 return status;
909 BUG_ON(d_inode(dentry->d_parent) != dir);
911 if (inode == osb->root_inode)
912 return -EPERM;
914 status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
915 OI_LS_PARENT);
916 if (status < 0) {
917 if (status != -ENOENT)
918 mlog_errno(status);
919 return status;
922 status = ocfs2_find_files_on_disk(dentry->d_name.name,
923 dentry->d_name.len, &blkno, dir,
924 &lookup);
925 if (status < 0) {
926 if (status != -ENOENT)
927 mlog_errno(status);
928 goto leave;
931 if (OCFS2_I(inode)->ip_blkno != blkno) {
932 status = -ENOENT;
934 trace_ocfs2_unlink_noent(
935 (unsigned long long)OCFS2_I(inode)->ip_blkno,
936 (unsigned long long)blkno,
937 OCFS2_I(inode)->ip_flags);
938 goto leave;
941 status = ocfs2_inode_lock(inode, &fe_bh, 1);
942 if (status < 0) {
943 if (status != -ENOENT)
944 mlog_errno(status);
945 goto leave;
947 child_locked = 1;
949 if (S_ISDIR(inode->i_mode)) {
950 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
951 status = -ENOTEMPTY;
952 goto leave;
956 status = ocfs2_remote_dentry_delete(dentry);
957 if (status < 0) {
958 /* This remote delete should succeed under all normal
959 * circumstances. */
960 mlog_errno(status);
961 goto leave;
964 if (ocfs2_inode_is_unlinkable(inode)) {
965 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
966 OCFS2_I(inode)->ip_blkno,
967 orphan_name, &orphan_insert,
968 false);
969 if (status < 0) {
970 mlog_errno(status);
971 goto leave;
973 is_unlinkable = true;
976 handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
977 if (IS_ERR(handle)) {
978 status = PTR_ERR(handle);
979 handle = NULL;
980 mlog_errno(status);
981 goto leave;
984 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
985 OCFS2_JOURNAL_ACCESS_WRITE);
986 if (status < 0) {
987 mlog_errno(status);
988 goto leave;
991 fe = (struct ocfs2_dinode *) fe_bh->b_data;
993 /* delete the name from the parent dir */
994 status = ocfs2_delete_entry(handle, dir, &lookup);
995 if (status < 0) {
996 mlog_errno(status);
997 goto leave;
1000 if (S_ISDIR(inode->i_mode))
1001 drop_nlink(inode);
1002 drop_nlink(inode);
1003 ocfs2_set_links_count(fe, inode->i_nlink);
1004 ocfs2_journal_dirty(handle, fe_bh);
1006 dir->i_ctime = dir->i_mtime = current_time(dir);
1007 if (S_ISDIR(inode->i_mode))
1008 drop_nlink(dir);
1010 status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
1011 if (status < 0) {
1012 mlog_errno(status);
1013 if (S_ISDIR(inode->i_mode))
1014 inc_nlink(dir);
1015 goto leave;
1018 if (is_unlinkable) {
1019 status = ocfs2_orphan_add(osb, handle, inode, fe_bh,
1020 orphan_name, &orphan_insert, orphan_dir, false);
1021 if (status < 0)
1022 mlog_errno(status);
1025 leave:
1026 if (handle)
1027 ocfs2_commit_trans(osb, handle);
1029 if (orphan_dir) {
1030 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1031 ocfs2_inode_unlock(orphan_dir, 1);
1032 inode_unlock(orphan_dir);
1033 iput(orphan_dir);
1036 if (child_locked)
1037 ocfs2_inode_unlock(inode, 1);
1039 ocfs2_inode_unlock(dir, 1);
1041 brelse(fe_bh);
1042 brelse(parent_node_bh);
1044 ocfs2_free_dir_lookup_result(&orphan_insert);
1045 ocfs2_free_dir_lookup_result(&lookup);
1047 if (status && (status != -ENOTEMPTY) && (status != -ENOENT))
1048 mlog_errno(status);
1050 return status;
1053 static int ocfs2_check_if_ancestor(struct ocfs2_super *osb,
1054 u64 src_inode_no, u64 dest_inode_no)
1056 int ret = 0, i = 0;
1057 u64 parent_inode_no = 0;
1058 u64 child_inode_no = src_inode_no;
1059 struct inode *child_inode;
1061 #define MAX_LOOKUP_TIMES 32
1062 while (1) {
1063 child_inode = ocfs2_iget(osb, child_inode_no, 0, 0);
1064 if (IS_ERR(child_inode)) {
1065 ret = PTR_ERR(child_inode);
1066 break;
1069 ret = ocfs2_inode_lock(child_inode, NULL, 0);
1070 if (ret < 0) {
1071 iput(child_inode);
1072 if (ret != -ENOENT)
1073 mlog_errno(ret);
1074 break;
1077 ret = ocfs2_lookup_ino_from_name(child_inode, "..", 2,
1078 &parent_inode_no);
1079 ocfs2_inode_unlock(child_inode, 0);
1080 iput(child_inode);
1081 if (ret < 0) {
1082 ret = -ENOENT;
1083 break;
1086 if (parent_inode_no == dest_inode_no) {
1087 ret = 1;
1088 break;
1091 if (parent_inode_no == osb->root_inode->i_ino) {
1092 ret = 0;
1093 break;
1096 child_inode_no = parent_inode_no;
1098 if (++i >= MAX_LOOKUP_TIMES) {
1099 mlog(ML_NOTICE, "max lookup times reached, filesystem "
1100 "may have nested directories, "
1101 "src inode: %llu, dest inode: %llu.\n",
1102 (unsigned long long)src_inode_no,
1103 (unsigned long long)dest_inode_no);
1104 ret = 0;
1105 break;
1109 return ret;
1113 * The only place this should be used is rename and link!
1114 * if they have the same id, then the 1st one is the only one locked.
1116 static int ocfs2_double_lock(struct ocfs2_super *osb,
1117 struct buffer_head **bh1,
1118 struct inode *inode1,
1119 struct buffer_head **bh2,
1120 struct inode *inode2,
1121 int rename)
1123 int status;
1124 int inode1_is_ancestor, inode2_is_ancestor;
1125 struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
1126 struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
1128 trace_ocfs2_double_lock((unsigned long long)oi1->ip_blkno,
1129 (unsigned long long)oi2->ip_blkno);
1131 if (*bh1)
1132 *bh1 = NULL;
1133 if (*bh2)
1134 *bh2 = NULL;
1136 /* we always want to lock the one with the lower lockid first.
1137 * and if they are nested, we lock ancestor first */
1138 if (oi1->ip_blkno != oi2->ip_blkno) {
1139 inode1_is_ancestor = ocfs2_check_if_ancestor(osb, oi2->ip_blkno,
1140 oi1->ip_blkno);
1141 if (inode1_is_ancestor < 0) {
1142 status = inode1_is_ancestor;
1143 goto bail;
1146 inode2_is_ancestor = ocfs2_check_if_ancestor(osb, oi1->ip_blkno,
1147 oi2->ip_blkno);
1148 if (inode2_is_ancestor < 0) {
1149 status = inode2_is_ancestor;
1150 goto bail;
1153 if ((inode1_is_ancestor == 1) ||
1154 (oi1->ip_blkno < oi2->ip_blkno &&
1155 inode2_is_ancestor == 0)) {
1156 /* switch id1 and id2 around */
1157 swap(bh2, bh1);
1158 swap(inode2, inode1);
1160 /* lock id2 */
1161 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
1162 rename == 1 ? OI_LS_RENAME1 : OI_LS_PARENT);
1163 if (status < 0) {
1164 if (status != -ENOENT)
1165 mlog_errno(status);
1166 goto bail;
1170 /* lock id1 */
1171 status = ocfs2_inode_lock_nested(inode1, bh1, 1,
1172 rename == 1 ? OI_LS_RENAME2 : OI_LS_PARENT);
1173 if (status < 0) {
1175 * An error return must mean that no cluster locks
1176 * were held on function exit.
1178 if (oi1->ip_blkno != oi2->ip_blkno) {
1179 ocfs2_inode_unlock(inode2, 1);
1180 brelse(*bh2);
1181 *bh2 = NULL;
1184 if (status != -ENOENT)
1185 mlog_errno(status);
1188 trace_ocfs2_double_lock_end(
1189 (unsigned long long)oi1->ip_blkno,
1190 (unsigned long long)oi2->ip_blkno);
1192 bail:
1193 if (status)
1194 mlog_errno(status);
1195 return status;
1198 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1200 ocfs2_inode_unlock(inode1, 1);
1202 if (inode1 != inode2)
1203 ocfs2_inode_unlock(inode2, 1);
1206 static int ocfs2_rename(struct inode *old_dir,
1207 struct dentry *old_dentry,
1208 struct inode *new_dir,
1209 struct dentry *new_dentry,
1210 unsigned int flags)
1212 int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1213 int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1214 struct inode *old_inode = d_inode(old_dentry);
1215 struct inode *new_inode = d_inode(new_dentry);
1216 struct inode *orphan_dir = NULL;
1217 struct ocfs2_dinode *newfe = NULL;
1218 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1219 struct buffer_head *newfe_bh = NULL;
1220 struct buffer_head *old_inode_bh = NULL;
1221 struct ocfs2_super *osb = NULL;
1222 u64 newfe_blkno, old_de_ino;
1223 handle_t *handle = NULL;
1224 struct buffer_head *old_dir_bh = NULL;
1225 struct buffer_head *new_dir_bh = NULL;
1226 u32 old_dir_nlink = old_dir->i_nlink;
1227 struct ocfs2_dinode *old_di;
1228 struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1229 struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1230 struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1231 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1232 struct ocfs2_dir_lookup_result target_insert = { NULL, };
1233 bool should_add_orphan = false;
1235 if (flags)
1236 return -EINVAL;
1238 /* At some point it might be nice to break this function up a
1239 * bit. */
1241 trace_ocfs2_rename(old_dir, old_dentry, new_dir, new_dentry,
1242 old_dentry->d_name.len, old_dentry->d_name.name,
1243 new_dentry->d_name.len, new_dentry->d_name.name);
1245 status = dquot_initialize(old_dir);
1246 if (status) {
1247 mlog_errno(status);
1248 goto bail;
1250 status = dquot_initialize(new_dir);
1251 if (status) {
1252 mlog_errno(status);
1253 goto bail;
1256 osb = OCFS2_SB(old_dir->i_sb);
1258 if (new_inode) {
1259 if (!igrab(new_inode))
1260 BUG();
1263 /* Assume a directory hierarchy thusly:
1264 * a/b/c
1265 * a/d
1266 * a,b,c, and d are all directories.
1268 * from cwd of 'a' on both nodes:
1269 * node1: mv b/c d
1270 * node2: mv d b/c
1272 * And that's why, just like the VFS, we need a file system
1273 * rename lock. */
1274 if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1275 status = ocfs2_rename_lock(osb);
1276 if (status < 0) {
1277 mlog_errno(status);
1278 goto bail;
1280 rename_lock = 1;
1282 /* here we cannot guarantee the inodes haven't just been
1283 * changed, so check if they are nested again */
1284 status = ocfs2_check_if_ancestor(osb, new_dir->i_ino,
1285 old_inode->i_ino);
1286 if (status < 0) {
1287 mlog_errno(status);
1288 goto bail;
1289 } else if (status == 1) {
1290 status = -EPERM;
1291 trace_ocfs2_rename_not_permitted(
1292 (unsigned long long)old_inode->i_ino,
1293 (unsigned long long)new_dir->i_ino);
1294 goto bail;
1298 /* if old and new are the same, this'll just do one lock. */
1299 status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1300 &new_dir_bh, new_dir, 1);
1301 if (status < 0) {
1302 mlog_errno(status);
1303 goto bail;
1305 parents_locked = 1;
1307 if (!new_dir->i_nlink) {
1308 status = -EACCES;
1309 goto bail;
1312 /* make sure both dirs have bhs
1313 * get an extra ref on old_dir_bh if old==new */
1314 if (!new_dir_bh) {
1315 if (old_dir_bh) {
1316 new_dir_bh = old_dir_bh;
1317 get_bh(new_dir_bh);
1318 } else {
1319 mlog(ML_ERROR, "no old_dir_bh!\n");
1320 status = -EIO;
1321 goto bail;
1326 * Aside from allowing a meta data update, the locking here
1327 * also ensures that the downconvert thread on other nodes
1328 * won't have to concurrently downconvert the inode and the
1329 * dentry locks.
1331 status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1332 OI_LS_PARENT);
1333 if (status < 0) {
1334 if (status != -ENOENT)
1335 mlog_errno(status);
1336 goto bail;
1338 old_child_locked = 1;
1340 status = ocfs2_remote_dentry_delete(old_dentry);
1341 if (status < 0) {
1342 mlog_errno(status);
1343 goto bail;
1346 if (S_ISDIR(old_inode->i_mode)) {
1347 u64 old_inode_parent;
1349 update_dot_dot = 1;
1350 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1351 old_inode,
1352 &old_inode_dot_dot_res);
1353 if (status) {
1354 status = -EIO;
1355 goto bail;
1358 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1359 status = -EIO;
1360 goto bail;
1363 if (!new_inode && new_dir != old_dir &&
1364 new_dir->i_nlink >= ocfs2_link_max(osb)) {
1365 status = -EMLINK;
1366 goto bail;
1370 status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1371 old_dentry->d_name.len,
1372 &old_de_ino);
1373 if (status) {
1374 status = -ENOENT;
1375 goto bail;
1379 * Check for inode number is _not_ due to possible IO errors.
1380 * We might rmdir the source, keep it as pwd of some process
1381 * and merrily kill the link to whatever was created under the
1382 * same name. Goodbye sticky bit ;-<
1384 if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1385 status = -ENOENT;
1386 goto bail;
1389 /* check if the target already exists (in which case we need
1390 * to delete it */
1391 status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1392 new_dentry->d_name.len,
1393 &newfe_blkno, new_dir,
1394 &target_lookup_res);
1395 /* The only error we allow here is -ENOENT because the new
1396 * file not existing is perfectly valid. */
1397 if ((status < 0) && (status != -ENOENT)) {
1398 /* If we cannot find the file specified we should just */
1399 /* return the error... */
1400 mlog_errno(status);
1401 goto bail;
1403 if (status == 0)
1404 target_exists = 1;
1406 if (!target_exists && new_inode) {
1408 * Target was unlinked by another node while we were
1409 * waiting to get to ocfs2_rename(). There isn't
1410 * anything we can do here to help the situation, so
1411 * bubble up the appropriate error.
1413 status = -ENOENT;
1414 goto bail;
1417 /* In case we need to overwrite an existing file, we blow it
1418 * away first */
1419 if (target_exists) {
1420 /* VFS didn't think there existed an inode here, but
1421 * someone else in the cluster must have raced our
1422 * rename to create one. Today we error cleanly, in
1423 * the future we should consider calling iget to build
1424 * a new struct inode for this entry. */
1425 if (!new_inode) {
1426 status = -EACCES;
1428 trace_ocfs2_rename_target_exists(new_dentry->d_name.len,
1429 new_dentry->d_name.name);
1430 goto bail;
1433 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1434 status = -EACCES;
1436 trace_ocfs2_rename_disagree(
1437 (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1438 (unsigned long long)newfe_blkno,
1439 OCFS2_I(new_inode)->ip_flags);
1440 goto bail;
1443 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1444 if (status < 0) {
1445 if (status != -ENOENT)
1446 mlog_errno(status);
1447 goto bail;
1449 new_child_locked = 1;
1451 status = ocfs2_remote_dentry_delete(new_dentry);
1452 if (status < 0) {
1453 mlog_errno(status);
1454 goto bail;
1457 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1459 trace_ocfs2_rename_over_existing(
1460 (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1461 (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1463 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1464 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1465 OCFS2_I(new_inode)->ip_blkno,
1466 orphan_name, &orphan_insert,
1467 false);
1468 if (status < 0) {
1469 mlog_errno(status);
1470 goto bail;
1472 should_add_orphan = true;
1474 } else {
1475 BUG_ON(d_inode(new_dentry->d_parent) != new_dir);
1477 status = ocfs2_check_dir_for_entry(new_dir,
1478 new_dentry->d_name.name,
1479 new_dentry->d_name.len);
1480 if (status)
1481 goto bail;
1483 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1484 new_dentry->d_name.name,
1485 new_dentry->d_name.len,
1486 &target_insert);
1487 if (status < 0) {
1488 mlog_errno(status);
1489 goto bail;
1493 handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1494 if (IS_ERR(handle)) {
1495 status = PTR_ERR(handle);
1496 handle = NULL;
1497 mlog_errno(status);
1498 goto bail;
1501 if (target_exists) {
1502 if (S_ISDIR(new_inode->i_mode)) {
1503 if (new_inode->i_nlink != 2 ||
1504 !ocfs2_empty_dir(new_inode)) {
1505 status = -ENOTEMPTY;
1506 goto bail;
1509 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1510 newfe_bh,
1511 OCFS2_JOURNAL_ACCESS_WRITE);
1512 if (status < 0) {
1513 mlog_errno(status);
1514 goto bail;
1517 /* change the dirent to point to the correct inode */
1518 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1519 old_inode);
1520 if (status < 0) {
1521 mlog_errno(status);
1522 goto bail;
1524 inode_inc_iversion(new_dir);
1526 if (S_ISDIR(new_inode->i_mode))
1527 ocfs2_set_links_count(newfe, 0);
1528 else
1529 ocfs2_add_links_count(newfe, -1);
1530 ocfs2_journal_dirty(handle, newfe_bh);
1531 if (should_add_orphan) {
1532 status = ocfs2_orphan_add(osb, handle, new_inode,
1533 newfe_bh, orphan_name,
1534 &orphan_insert, orphan_dir, false);
1535 if (status < 0) {
1536 mlog_errno(status);
1537 goto bail;
1540 } else {
1541 /* if the name was not found in new_dir, add it now */
1542 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1543 OCFS2_I(old_inode)->ip_blkno,
1544 new_dir_bh, &target_insert);
1547 old_inode->i_ctime = current_time(old_inode);
1548 mark_inode_dirty(old_inode);
1550 status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1551 old_inode_bh,
1552 OCFS2_JOURNAL_ACCESS_WRITE);
1553 if (status >= 0) {
1554 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1556 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1557 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1558 ocfs2_journal_dirty(handle, old_inode_bh);
1559 } else
1560 mlog_errno(status);
1563 * Now that the name has been added to new_dir, remove the old name.
1565 * We don't keep any directory entry context around until now
1566 * because the insert might have changed the type of directory
1567 * we're dealing with.
1569 status = ocfs2_find_entry(old_dentry->d_name.name,
1570 old_dentry->d_name.len, old_dir,
1571 &old_entry_lookup);
1572 if (status) {
1573 if (!is_journal_aborted(osb->journal->j_journal)) {
1574 ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1575 "is not deleted.",
1576 new_dentry->d_name.len, new_dentry->d_name.name,
1577 old_dentry->d_name.len, old_dentry->d_name.name);
1579 goto bail;
1582 status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1583 if (status < 0) {
1584 mlog_errno(status);
1585 if (!is_journal_aborted(osb->journal->j_journal)) {
1586 ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1587 "is not deleted.",
1588 new_dentry->d_name.len, new_dentry->d_name.name,
1589 old_dentry->d_name.len, old_dentry->d_name.name);
1591 goto bail;
1594 if (new_inode) {
1595 drop_nlink(new_inode);
1596 new_inode->i_ctime = current_time(new_inode);
1598 old_dir->i_ctime = old_dir->i_mtime = current_time(old_dir);
1600 if (update_dot_dot) {
1601 status = ocfs2_update_entry(old_inode, handle,
1602 &old_inode_dot_dot_res, new_dir);
1603 drop_nlink(old_dir);
1604 if (new_inode) {
1605 drop_nlink(new_inode);
1606 } else {
1607 inc_nlink(new_dir);
1608 mark_inode_dirty(new_dir);
1611 mark_inode_dirty(old_dir);
1612 ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1613 if (new_inode) {
1614 mark_inode_dirty(new_inode);
1615 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1618 if (old_dir != new_dir) {
1619 /* Keep the same times on both directories.*/
1620 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1623 * This will also pick up the i_nlink change from the
1624 * block above.
1626 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1629 if (old_dir_nlink != old_dir->i_nlink) {
1630 if (!old_dir_bh) {
1631 mlog(ML_ERROR, "need to change nlink for old dir "
1632 "%llu from %d to %d but bh is NULL!\n",
1633 (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1634 (int)old_dir_nlink, old_dir->i_nlink);
1635 } else {
1636 struct ocfs2_dinode *fe;
1637 status = ocfs2_journal_access_di(handle,
1638 INODE_CACHE(old_dir),
1639 old_dir_bh,
1640 OCFS2_JOURNAL_ACCESS_WRITE);
1641 fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1642 ocfs2_set_links_count(fe, old_dir->i_nlink);
1643 ocfs2_journal_dirty(handle, old_dir_bh);
1646 ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1647 status = 0;
1648 bail:
1649 if (handle)
1650 ocfs2_commit_trans(osb, handle);
1652 if (orphan_dir) {
1653 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1654 ocfs2_inode_unlock(orphan_dir, 1);
1655 inode_unlock(orphan_dir);
1656 iput(orphan_dir);
1659 if (new_child_locked)
1660 ocfs2_inode_unlock(new_inode, 1);
1662 if (old_child_locked)
1663 ocfs2_inode_unlock(old_inode, 1);
1665 if (parents_locked)
1666 ocfs2_double_unlock(old_dir, new_dir);
1668 if (rename_lock)
1669 ocfs2_rename_unlock(osb);
1671 if (new_inode)
1672 sync_mapping_buffers(old_inode->i_mapping);
1674 iput(new_inode);
1676 ocfs2_free_dir_lookup_result(&target_lookup_res);
1677 ocfs2_free_dir_lookup_result(&old_entry_lookup);
1678 ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1679 ocfs2_free_dir_lookup_result(&orphan_insert);
1680 ocfs2_free_dir_lookup_result(&target_insert);
1682 brelse(newfe_bh);
1683 brelse(old_inode_bh);
1684 brelse(old_dir_bh);
1685 brelse(new_dir_bh);
1687 if (status)
1688 mlog_errno(status);
1690 return status;
1694 * we expect i_size = strlen(symname). Copy symname into the file
1695 * data, including the null terminator.
1697 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1698 handle_t *handle,
1699 struct inode *inode,
1700 const char *symname)
1702 struct buffer_head **bhs = NULL;
1703 const char *c;
1704 struct super_block *sb = osb->sb;
1705 u64 p_blkno, p_blocks;
1706 int virtual, blocks, status, i, bytes_left;
1708 bytes_left = i_size_read(inode) + 1;
1709 /* we can't trust i_blocks because we're actually going to
1710 * write i_size + 1 bytes. */
1711 blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1713 trace_ocfs2_create_symlink_data((unsigned long long)inode->i_blocks,
1714 i_size_read(inode), blocks);
1716 /* Sanity check -- make sure we're going to fit. */
1717 if (bytes_left >
1718 ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1719 status = -EIO;
1720 mlog_errno(status);
1721 goto bail;
1724 bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1725 if (!bhs) {
1726 status = -ENOMEM;
1727 mlog_errno(status);
1728 goto bail;
1731 status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1732 NULL);
1733 if (status < 0) {
1734 mlog_errno(status);
1735 goto bail;
1738 /* links can never be larger than one cluster so we know this
1739 * is all going to be contiguous, but do a sanity check
1740 * anyway. */
1741 if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1742 status = -EIO;
1743 mlog_errno(status);
1744 goto bail;
1747 virtual = 0;
1748 while(bytes_left > 0) {
1749 c = &symname[virtual * sb->s_blocksize];
1751 bhs[virtual] = sb_getblk(sb, p_blkno);
1752 if (!bhs[virtual]) {
1753 status = -ENOMEM;
1754 mlog_errno(status);
1755 goto bail;
1757 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1758 bhs[virtual]);
1760 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1761 bhs[virtual],
1762 OCFS2_JOURNAL_ACCESS_CREATE);
1763 if (status < 0) {
1764 mlog_errno(status);
1765 goto bail;
1768 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1770 memcpy(bhs[virtual]->b_data, c,
1771 (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1772 bytes_left);
1774 ocfs2_journal_dirty(handle, bhs[virtual]);
1776 virtual++;
1777 p_blkno++;
1778 bytes_left -= sb->s_blocksize;
1781 status = 0;
1782 bail:
1784 if (bhs) {
1785 for(i = 0; i < blocks; i++)
1786 brelse(bhs[i]);
1787 kfree(bhs);
1790 if (status)
1791 mlog_errno(status);
1792 return status;
1795 static int ocfs2_symlink(struct inode *dir,
1796 struct dentry *dentry,
1797 const char *symname)
1799 int status, l, credits;
1800 u64 newsize;
1801 struct ocfs2_super *osb = NULL;
1802 struct inode *inode = NULL;
1803 struct super_block *sb;
1804 struct buffer_head *new_fe_bh = NULL;
1805 struct buffer_head *parent_fe_bh = NULL;
1806 struct ocfs2_dinode *fe = NULL;
1807 struct ocfs2_dinode *dirfe;
1808 handle_t *handle = NULL;
1809 struct ocfs2_alloc_context *inode_ac = NULL;
1810 struct ocfs2_alloc_context *data_ac = NULL;
1811 struct ocfs2_alloc_context *xattr_ac = NULL;
1812 int want_clusters = 0;
1813 int xattr_credits = 0;
1814 struct ocfs2_security_xattr_info si = {
1815 .enable = 1,
1817 int did_quota = 0, did_quota_inode = 0;
1818 struct ocfs2_dir_lookup_result lookup = { NULL, };
1819 sigset_t oldset;
1820 int did_block_signals = 0;
1821 struct ocfs2_dentry_lock *dl = NULL;
1823 trace_ocfs2_symlink_begin(dir, dentry, symname,
1824 dentry->d_name.len, dentry->d_name.name);
1826 status = dquot_initialize(dir);
1827 if (status) {
1828 mlog_errno(status);
1829 goto bail;
1832 sb = dir->i_sb;
1833 osb = OCFS2_SB(sb);
1835 l = strlen(symname) + 1;
1837 credits = ocfs2_calc_symlink_credits(sb);
1839 /* lock the parent directory */
1840 status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1841 if (status < 0) {
1842 if (status != -ENOENT)
1843 mlog_errno(status);
1844 return status;
1847 dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1848 if (!ocfs2_read_links_count(dirfe)) {
1849 /* can't make a file in a deleted directory. */
1850 status = -ENOENT;
1851 goto bail;
1854 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1855 dentry->d_name.len);
1856 if (status)
1857 goto bail;
1859 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1860 dentry->d_name.name,
1861 dentry->d_name.len, &lookup);
1862 if (status < 0) {
1863 mlog_errno(status);
1864 goto bail;
1867 status = ocfs2_reserve_new_inode(osb, &inode_ac);
1868 if (status < 0) {
1869 if (status != -ENOSPC)
1870 mlog_errno(status);
1871 goto bail;
1874 inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1875 if (IS_ERR(inode)) {
1876 status = PTR_ERR(inode);
1877 inode = NULL;
1878 mlog_errno(status);
1879 goto bail;
1882 /* get security xattr */
1883 status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
1884 if (status) {
1885 if (status == -EOPNOTSUPP)
1886 si.enable = 0;
1887 else {
1888 mlog_errno(status);
1889 goto bail;
1893 /* calculate meta data/clusters for setting security xattr */
1894 if (si.enable) {
1895 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1896 &xattr_credits, &xattr_ac);
1897 if (status < 0) {
1898 mlog_errno(status);
1899 goto bail;
1903 /* don't reserve bitmap space for fast symlinks. */
1904 if (l > ocfs2_fast_symlink_chars(sb))
1905 want_clusters += 1;
1907 status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1908 if (status < 0) {
1909 if (status != -ENOSPC)
1910 mlog_errno(status);
1911 goto bail;
1914 handle = ocfs2_start_trans(osb, credits + xattr_credits);
1915 if (IS_ERR(handle)) {
1916 status = PTR_ERR(handle);
1917 handle = NULL;
1918 mlog_errno(status);
1919 goto bail;
1922 /* Starting to change things, restart is no longer possible. */
1923 ocfs2_block_signals(&oldset);
1924 did_block_signals = 1;
1926 status = dquot_alloc_inode(inode);
1927 if (status)
1928 goto bail;
1929 did_quota_inode = 1;
1931 trace_ocfs2_symlink_create(dir, dentry, dentry->d_name.len,
1932 dentry->d_name.name,
1933 (unsigned long long)OCFS2_I(dir)->ip_blkno,
1934 inode->i_mode);
1936 status = ocfs2_mknod_locked(osb, dir, inode,
1937 0, &new_fe_bh, parent_fe_bh, handle,
1938 inode_ac);
1939 if (status < 0) {
1940 mlog_errno(status);
1941 goto bail;
1944 fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1945 inode->i_rdev = 0;
1946 newsize = l - 1;
1947 inode->i_op = &ocfs2_symlink_inode_operations;
1948 inode_nohighmem(inode);
1949 if (l > ocfs2_fast_symlink_chars(sb)) {
1950 u32 offset = 0;
1952 status = dquot_alloc_space_nodirty(inode,
1953 ocfs2_clusters_to_bytes(osb->sb, 1));
1954 if (status)
1955 goto bail;
1956 did_quota = 1;
1957 inode->i_mapping->a_ops = &ocfs2_aops;
1958 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1959 new_fe_bh,
1960 handle, data_ac, NULL,
1961 NULL);
1962 if (status < 0) {
1963 if (status != -ENOSPC && status != -EINTR) {
1964 mlog(ML_ERROR,
1965 "Failed to extend file to %llu\n",
1966 (unsigned long long)newsize);
1967 mlog_errno(status);
1968 status = -ENOSPC;
1970 goto bail;
1972 i_size_write(inode, newsize);
1973 inode->i_blocks = ocfs2_inode_sector_count(inode);
1974 } else {
1975 inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
1976 memcpy((char *) fe->id2.i_symlink, symname, l);
1977 i_size_write(inode, newsize);
1978 inode->i_blocks = 0;
1981 status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1982 if (status < 0) {
1983 mlog_errno(status);
1984 goto bail;
1987 if (!ocfs2_inode_is_fast_symlink(inode)) {
1988 status = ocfs2_create_symlink_data(osb, handle, inode,
1989 symname);
1990 if (status < 0) {
1991 mlog_errno(status);
1992 goto bail;
1996 if (si.enable) {
1997 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1998 xattr_ac, data_ac);
1999 if (status < 0) {
2000 mlog_errno(status);
2001 goto bail;
2006 * Do this before adding the entry to the directory. We add
2007 * also set d_op after success so that ->d_iput() will cleanup
2008 * the dentry lock even if ocfs2_add_entry() fails below.
2010 status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
2011 if (status) {
2012 mlog_errno(status);
2013 goto bail;
2016 dl = dentry->d_fsdata;
2018 status = ocfs2_add_entry(handle, dentry, inode,
2019 le64_to_cpu(fe->i_blkno), parent_fe_bh,
2020 &lookup);
2021 if (status < 0) {
2022 mlog_errno(status);
2023 goto bail;
2026 insert_inode_hash(inode);
2027 d_instantiate(dentry, inode);
2028 bail:
2029 if (status < 0 && did_quota)
2030 dquot_free_space_nodirty(inode,
2031 ocfs2_clusters_to_bytes(osb->sb, 1));
2032 if (status < 0 && did_quota_inode)
2033 dquot_free_inode(inode);
2034 if (handle)
2035 ocfs2_commit_trans(osb, handle);
2037 ocfs2_inode_unlock(dir, 1);
2038 if (did_block_signals)
2039 ocfs2_unblock_signals(&oldset);
2041 brelse(new_fe_bh);
2042 brelse(parent_fe_bh);
2043 kfree(si.value);
2044 ocfs2_free_dir_lookup_result(&lookup);
2045 if (inode_ac)
2046 ocfs2_free_alloc_context(inode_ac);
2047 if (data_ac)
2048 ocfs2_free_alloc_context(data_ac);
2049 if (xattr_ac)
2050 ocfs2_free_alloc_context(xattr_ac);
2051 if ((status < 0) && inode) {
2052 if (dl)
2053 ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
2055 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
2056 clear_nlink(inode);
2057 iput(inode);
2060 if (status)
2061 mlog_errno(status);
2063 return status;
2066 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2068 int status, namelen;
2070 namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2071 (long long)blkno);
2072 if (namelen <= 0) {
2073 if (namelen)
2074 status = namelen;
2075 else
2076 status = -EINVAL;
2077 mlog_errno(status);
2078 goto bail;
2080 if (namelen != OCFS2_ORPHAN_NAMELEN) {
2081 status = -EINVAL;
2082 mlog_errno(status);
2083 goto bail;
2086 trace_ocfs2_blkno_stringify(blkno, name, namelen);
2088 status = 0;
2089 bail:
2090 if (status < 0)
2091 mlog_errno(status);
2092 return status;
2095 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
2096 struct inode **ret_orphan_dir,
2097 struct buffer_head **ret_orphan_dir_bh)
2099 struct inode *orphan_dir_inode;
2100 struct buffer_head *orphan_dir_bh = NULL;
2101 int ret = 0;
2103 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2104 ORPHAN_DIR_SYSTEM_INODE,
2105 osb->slot_num);
2106 if (!orphan_dir_inode) {
2107 ret = -ENOENT;
2108 mlog_errno(ret);
2109 return ret;
2112 inode_lock(orphan_dir_inode);
2114 ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2115 if (ret < 0) {
2116 inode_unlock(orphan_dir_inode);
2117 iput(orphan_dir_inode);
2119 mlog_errno(ret);
2120 return ret;
2123 *ret_orphan_dir = orphan_dir_inode;
2124 *ret_orphan_dir_bh = orphan_dir_bh;
2126 return 0;
2129 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
2130 struct buffer_head *orphan_dir_bh,
2131 u64 blkno,
2132 char *name,
2133 struct ocfs2_dir_lookup_result *lookup,
2134 bool dio)
2136 int ret;
2137 struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
2138 int namelen = dio ?
2139 (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2140 OCFS2_ORPHAN_NAMELEN;
2142 if (dio) {
2143 ret = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2144 OCFS2_DIO_ORPHAN_PREFIX);
2145 if (ret != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2146 ret = -EINVAL;
2147 mlog_errno(ret);
2148 return ret;
2151 ret = ocfs2_blkno_stringify(blkno,
2152 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2153 } else
2154 ret = ocfs2_blkno_stringify(blkno, name);
2155 if (ret < 0) {
2156 mlog_errno(ret);
2157 return ret;
2160 ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2161 orphan_dir_bh, name,
2162 namelen, lookup);
2163 if (ret < 0) {
2164 mlog_errno(ret);
2165 return ret;
2168 return 0;
2172 * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
2173 * insertion of an orphan.
2174 * @osb: ocfs2 file system
2175 * @ret_orphan_dir: Orphan dir inode - returned locked!
2176 * @blkno: Actual block number of the inode to be inserted into orphan dir.
2177 * @lookup: dir lookup result, to be passed back into functions like
2178 * ocfs2_orphan_add
2180 * Returns zero on success and the ret_orphan_dir, name and lookup
2181 * fields will be populated.
2183 * Returns non-zero on failure.
2185 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2186 struct inode **ret_orphan_dir,
2187 u64 blkno,
2188 char *name,
2189 struct ocfs2_dir_lookup_result *lookup,
2190 bool dio)
2192 struct inode *orphan_dir_inode = NULL;
2193 struct buffer_head *orphan_dir_bh = NULL;
2194 int ret = 0;
2196 ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
2197 &orphan_dir_bh);
2198 if (ret < 0) {
2199 mlog_errno(ret);
2200 return ret;
2203 ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
2204 blkno, name, lookup, dio);
2205 if (ret < 0) {
2206 mlog_errno(ret);
2207 goto out;
2210 *ret_orphan_dir = orphan_dir_inode;
2212 out:
2213 brelse(orphan_dir_bh);
2215 if (ret) {
2216 ocfs2_inode_unlock(orphan_dir_inode, 1);
2217 inode_unlock(orphan_dir_inode);
2218 iput(orphan_dir_inode);
2221 if (ret)
2222 mlog_errno(ret);
2223 return ret;
2226 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2227 handle_t *handle,
2228 struct inode *inode,
2229 struct buffer_head *fe_bh,
2230 char *name,
2231 struct ocfs2_dir_lookup_result *lookup,
2232 struct inode *orphan_dir_inode,
2233 bool dio)
2235 struct buffer_head *orphan_dir_bh = NULL;
2236 int status = 0;
2237 struct ocfs2_dinode *orphan_fe;
2238 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
2239 int namelen = dio ?
2240 (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2241 OCFS2_ORPHAN_NAMELEN;
2243 trace_ocfs2_orphan_add_begin(
2244 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2246 status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2247 if (status < 0) {
2248 mlog_errno(status);
2249 goto leave;
2252 status = ocfs2_journal_access_di(handle,
2253 INODE_CACHE(orphan_dir_inode),
2254 orphan_dir_bh,
2255 OCFS2_JOURNAL_ACCESS_WRITE);
2256 if (status < 0) {
2257 mlog_errno(status);
2258 goto leave;
2262 * We're going to journal the change of i_flags and i_orphaned_slot.
2263 * It's safe anyway, though some callers may duplicate the journaling.
2264 * Journaling within the func just make the logic look more
2265 * straightforward.
2267 status = ocfs2_journal_access_di(handle,
2268 INODE_CACHE(inode),
2269 fe_bh,
2270 OCFS2_JOURNAL_ACCESS_WRITE);
2271 if (status < 0) {
2272 mlog_errno(status);
2273 goto leave;
2276 /* we're a cluster, and nlink can change on disk from
2277 * underneath us... */
2278 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2279 if (S_ISDIR(inode->i_mode))
2280 ocfs2_add_links_count(orphan_fe, 1);
2281 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2282 ocfs2_journal_dirty(handle, orphan_dir_bh);
2284 status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2285 namelen, inode,
2286 OCFS2_I(inode)->ip_blkno,
2287 orphan_dir_bh, lookup);
2288 if (status < 0) {
2289 mlog_errno(status);
2290 goto rollback;
2293 if (dio) {
2294 /* Update flag OCFS2_DIO_ORPHANED_FL and record the orphan
2295 * slot.
2297 fe->i_flags |= cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2298 fe->i_dio_orphaned_slot = cpu_to_le16(osb->slot_num);
2299 } else {
2300 fe->i_flags |= cpu_to_le32(OCFS2_ORPHANED_FL);
2301 OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2303 /* Record which orphan dir our inode now resides
2304 * in. delete_inode will use this to determine which orphan
2305 * dir to lock. */
2306 fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2309 ocfs2_journal_dirty(handle, fe_bh);
2311 trace_ocfs2_orphan_add_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
2312 osb->slot_num);
2314 rollback:
2315 if (status < 0) {
2316 if (S_ISDIR(inode->i_mode))
2317 ocfs2_add_links_count(orphan_fe, -1);
2318 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2321 leave:
2322 brelse(orphan_dir_bh);
2324 return status;
2327 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2328 int ocfs2_orphan_del(struct ocfs2_super *osb,
2329 handle_t *handle,
2330 struct inode *orphan_dir_inode,
2331 struct inode *inode,
2332 struct buffer_head *orphan_dir_bh,
2333 bool dio)
2335 char name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2336 struct ocfs2_dinode *orphan_fe;
2337 int status = 0;
2338 struct ocfs2_dir_lookup_result lookup = { NULL, };
2340 if (dio) {
2341 status = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2342 OCFS2_DIO_ORPHAN_PREFIX);
2343 if (status != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2344 status = -EINVAL;
2345 mlog_errno(status);
2346 return status;
2349 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno,
2350 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2351 } else
2352 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2353 if (status < 0) {
2354 mlog_errno(status);
2355 goto leave;
2358 trace_ocfs2_orphan_del(
2359 (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2360 name, strlen(name));
2362 status = ocfs2_journal_access_di(handle,
2363 INODE_CACHE(orphan_dir_inode),
2364 orphan_dir_bh,
2365 OCFS2_JOURNAL_ACCESS_WRITE);
2366 if (status < 0) {
2367 mlog_errno(status);
2368 goto leave;
2371 /* find it's spot in the orphan directory */
2372 status = ocfs2_find_entry(name, strlen(name), orphan_dir_inode,
2373 &lookup);
2374 if (status) {
2375 mlog_errno(status);
2376 goto leave;
2379 /* remove it from the orphan directory */
2380 status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2381 if (status < 0) {
2382 mlog_errno(status);
2383 goto leave;
2386 /* do the i_nlink dance! :) */
2387 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2388 if (S_ISDIR(inode->i_mode))
2389 ocfs2_add_links_count(orphan_fe, -1);
2390 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2391 ocfs2_journal_dirty(handle, orphan_dir_bh);
2393 leave:
2394 ocfs2_free_dir_lookup_result(&lookup);
2396 if (status)
2397 mlog_errno(status);
2398 return status;
2402 * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to receive a newly
2403 * allocated file. This is different from the typical 'add to orphan dir'
2404 * operation in that the inode does not yet exist. This is a problem because
2405 * the orphan dir stringifies the inode block number to come up with it's
2406 * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2407 * problem. This function works around it by calling deeper into the orphan
2408 * and suballoc code than other callers. Use this only by necessity.
2409 * @dir: The directory which this inode will ultimately wind up under - not the
2410 * orphan dir!
2411 * @dir_bh: buffer_head the @dir inode block
2412 * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2413 * with the string to be used for orphan dirent. Pass back to the orphan dir
2414 * code.
2415 * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2416 * dir code.
2417 * @ret_di_blkno: block number where the new inode will be allocated.
2418 * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2419 * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2421 * Returns zero on success and the ret_orphan_dir, name and lookup
2422 * fields will be populated.
2424 * Returns non-zero on failure.
2426 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2427 struct buffer_head *dir_bh,
2428 char *orphan_name,
2429 struct inode **ret_orphan_dir,
2430 u64 *ret_di_blkno,
2431 struct ocfs2_dir_lookup_result *orphan_insert,
2432 struct ocfs2_alloc_context **ret_inode_ac)
2434 int ret;
2435 u64 di_blkno;
2436 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2437 struct inode *orphan_dir = NULL;
2438 struct buffer_head *orphan_dir_bh = NULL;
2439 struct ocfs2_alloc_context *inode_ac = NULL;
2441 ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2442 if (ret < 0) {
2443 mlog_errno(ret);
2444 return ret;
2447 /* reserve an inode spot */
2448 ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2449 if (ret < 0) {
2450 if (ret != -ENOSPC)
2451 mlog_errno(ret);
2452 goto out;
2455 ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2456 &di_blkno);
2457 if (ret) {
2458 mlog_errno(ret);
2459 goto out;
2462 ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2463 di_blkno, orphan_name, orphan_insert,
2464 false);
2465 if (ret < 0) {
2466 mlog_errno(ret);
2467 goto out;
2470 out:
2471 if (ret == 0) {
2472 *ret_orphan_dir = orphan_dir;
2473 *ret_di_blkno = di_blkno;
2474 *ret_inode_ac = inode_ac;
2476 * orphan_name and orphan_insert are already up to
2477 * date via prepare_orphan_dir
2479 } else {
2480 /* Unroll reserve_new_inode* */
2481 if (inode_ac)
2482 ocfs2_free_alloc_context(inode_ac);
2484 /* Unroll orphan dir locking */
2485 inode_unlock(orphan_dir);
2486 ocfs2_inode_unlock(orphan_dir, 1);
2487 iput(orphan_dir);
2490 brelse(orphan_dir_bh);
2492 return ret;
2495 int ocfs2_create_inode_in_orphan(struct inode *dir,
2496 int mode,
2497 struct inode **new_inode)
2499 int status, did_quota_inode = 0;
2500 struct inode *inode = NULL;
2501 struct inode *orphan_dir = NULL;
2502 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2503 struct ocfs2_dinode *di = NULL;
2504 handle_t *handle = NULL;
2505 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2506 struct buffer_head *parent_di_bh = NULL;
2507 struct buffer_head *new_di_bh = NULL;
2508 struct ocfs2_alloc_context *inode_ac = NULL;
2509 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2510 u64 uninitialized_var(di_blkno), suballoc_loc;
2511 u16 suballoc_bit;
2513 status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2514 if (status < 0) {
2515 if (status != -ENOENT)
2516 mlog_errno(status);
2517 return status;
2520 status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2521 orphan_name, &orphan_dir,
2522 &di_blkno, &orphan_insert, &inode_ac);
2523 if (status < 0) {
2524 if (status != -ENOSPC)
2525 mlog_errno(status);
2526 goto leave;
2529 inode = ocfs2_get_init_inode(dir, mode);
2530 if (IS_ERR(inode)) {
2531 status = PTR_ERR(inode);
2532 inode = NULL;
2533 mlog_errno(status);
2534 goto leave;
2537 handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2538 if (IS_ERR(handle)) {
2539 status = PTR_ERR(handle);
2540 handle = NULL;
2541 mlog_errno(status);
2542 goto leave;
2545 status = dquot_alloc_inode(inode);
2546 if (status)
2547 goto leave;
2548 did_quota_inode = 1;
2550 status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2551 &suballoc_loc,
2552 &suballoc_bit, di_blkno);
2553 if (status < 0) {
2554 mlog_errno(status);
2555 goto leave;
2558 clear_nlink(inode);
2559 /* do the real work now. */
2560 status = __ocfs2_mknod_locked(dir, inode,
2561 0, &new_di_bh, parent_di_bh, handle,
2562 inode_ac, di_blkno, suballoc_loc,
2563 suballoc_bit);
2564 if (status < 0) {
2565 mlog_errno(status);
2566 goto leave;
2569 di = (struct ocfs2_dinode *)new_di_bh->b_data;
2570 status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2571 &orphan_insert, orphan_dir, false);
2572 if (status < 0) {
2573 mlog_errno(status);
2574 goto leave;
2577 /* get open lock so that only nodes can't remove it from orphan dir. */
2578 status = ocfs2_open_lock(inode);
2579 if (status < 0)
2580 mlog_errno(status);
2582 insert_inode_hash(inode);
2583 leave:
2584 if (status < 0 && did_quota_inode)
2585 dquot_free_inode(inode);
2586 if (handle)
2587 ocfs2_commit_trans(osb, handle);
2589 if (orphan_dir) {
2590 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2591 ocfs2_inode_unlock(orphan_dir, 1);
2592 inode_unlock(orphan_dir);
2593 iput(orphan_dir);
2596 if ((status < 0) && inode) {
2597 clear_nlink(inode);
2598 iput(inode);
2601 if (inode_ac)
2602 ocfs2_free_alloc_context(inode_ac);
2604 brelse(new_di_bh);
2606 if (!status)
2607 *new_inode = inode;
2609 ocfs2_free_dir_lookup_result(&orphan_insert);
2611 ocfs2_inode_unlock(dir, 1);
2612 brelse(parent_di_bh);
2613 return status;
2616 int ocfs2_add_inode_to_orphan(struct ocfs2_super *osb,
2617 struct inode *inode)
2619 char orphan_name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2620 struct inode *orphan_dir_inode = NULL;
2621 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2622 struct buffer_head *di_bh = NULL;
2623 int status = 0;
2624 handle_t *handle = NULL;
2625 struct ocfs2_dinode *di = NULL;
2627 status = ocfs2_inode_lock(inode, &di_bh, 1);
2628 if (status < 0) {
2629 mlog_errno(status);
2630 goto bail;
2633 di = (struct ocfs2_dinode *) di_bh->b_data;
2635 * Another append dio crashed?
2636 * If so, manually recover it first.
2638 if (unlikely(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL))) {
2639 status = ocfs2_truncate_file(inode, di_bh, i_size_read(inode));
2640 if (status < 0) {
2641 if (status != -ENOSPC)
2642 mlog_errno(status);
2643 goto bail_unlock_inode;
2646 status = ocfs2_del_inode_from_orphan(osb, inode, di_bh, 0, 0);
2647 if (status < 0) {
2648 mlog_errno(status);
2649 goto bail_unlock_inode;
2653 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir_inode,
2654 OCFS2_I(inode)->ip_blkno,
2655 orphan_name,
2656 &orphan_insert,
2657 true);
2658 if (status < 0) {
2659 mlog_errno(status);
2660 goto bail_unlock_inode;
2663 handle = ocfs2_start_trans(osb,
2664 OCFS2_INODE_ADD_TO_ORPHAN_CREDITS);
2665 if (IS_ERR(handle)) {
2666 status = PTR_ERR(handle);
2667 goto bail_unlock_orphan;
2670 status = ocfs2_orphan_add(osb, handle, inode, di_bh, orphan_name,
2671 &orphan_insert, orphan_dir_inode, true);
2672 if (status)
2673 mlog_errno(status);
2675 ocfs2_commit_trans(osb, handle);
2677 bail_unlock_orphan:
2678 ocfs2_inode_unlock(orphan_dir_inode, 1);
2679 inode_unlock(orphan_dir_inode);
2680 iput(orphan_dir_inode);
2682 ocfs2_free_dir_lookup_result(&orphan_insert);
2684 bail_unlock_inode:
2685 ocfs2_inode_unlock(inode, 1);
2686 brelse(di_bh);
2688 bail:
2689 return status;
2692 int ocfs2_del_inode_from_orphan(struct ocfs2_super *osb,
2693 struct inode *inode, struct buffer_head *di_bh,
2694 int update_isize, loff_t end)
2696 struct inode *orphan_dir_inode = NULL;
2697 struct buffer_head *orphan_dir_bh = NULL;
2698 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2699 handle_t *handle = NULL;
2700 int status = 0;
2702 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2703 ORPHAN_DIR_SYSTEM_INODE,
2704 le16_to_cpu(di->i_dio_orphaned_slot));
2705 if (!orphan_dir_inode) {
2706 status = -ENOENT;
2707 mlog_errno(status);
2708 goto bail;
2711 inode_lock(orphan_dir_inode);
2712 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2713 if (status < 0) {
2714 inode_unlock(orphan_dir_inode);
2715 iput(orphan_dir_inode);
2716 mlog_errno(status);
2717 goto bail;
2720 handle = ocfs2_start_trans(osb,
2721 OCFS2_INODE_DEL_FROM_ORPHAN_CREDITS);
2722 if (IS_ERR(handle)) {
2723 status = PTR_ERR(handle);
2724 goto bail_unlock_orphan;
2727 BUG_ON(!(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)));
2729 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode,
2730 inode, orphan_dir_bh, true);
2731 if (status < 0) {
2732 mlog_errno(status);
2733 goto bail_commit;
2736 status = ocfs2_journal_access_di(handle,
2737 INODE_CACHE(inode),
2738 di_bh,
2739 OCFS2_JOURNAL_ACCESS_WRITE);
2740 if (status < 0) {
2741 mlog_errno(status);
2742 goto bail_commit;
2745 di->i_flags &= ~cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2746 di->i_dio_orphaned_slot = 0;
2748 if (update_isize) {
2749 status = ocfs2_set_inode_size(handle, inode, di_bh, end);
2750 if (status)
2751 mlog_errno(status);
2752 } else
2753 ocfs2_journal_dirty(handle, di_bh);
2755 bail_commit:
2756 ocfs2_commit_trans(osb, handle);
2758 bail_unlock_orphan:
2759 ocfs2_inode_unlock(orphan_dir_inode, 1);
2760 inode_unlock(orphan_dir_inode);
2761 brelse(orphan_dir_bh);
2762 iput(orphan_dir_inode);
2764 bail:
2765 return status;
2768 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2769 struct inode *inode,
2770 struct dentry *dentry)
2772 int status = 0;
2773 struct buffer_head *parent_di_bh = NULL;
2774 handle_t *handle = NULL;
2775 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2776 struct ocfs2_dinode *dir_di, *di;
2777 struct inode *orphan_dir_inode = NULL;
2778 struct buffer_head *orphan_dir_bh = NULL;
2779 struct buffer_head *di_bh = NULL;
2780 struct ocfs2_dir_lookup_result lookup = { NULL, };
2782 trace_ocfs2_mv_orphaned_inode_to_new(dir, dentry,
2783 dentry->d_name.len, dentry->d_name.name,
2784 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2785 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2787 status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2788 if (status < 0) {
2789 if (status != -ENOENT)
2790 mlog_errno(status);
2791 return status;
2794 dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2795 if (!dir_di->i_links_count) {
2796 /* can't make a file in a deleted directory. */
2797 status = -ENOENT;
2798 goto leave;
2801 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2802 dentry->d_name.len);
2803 if (status)
2804 goto leave;
2806 /* get a spot inside the dir. */
2807 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2808 dentry->d_name.name,
2809 dentry->d_name.len, &lookup);
2810 if (status < 0) {
2811 mlog_errno(status);
2812 goto leave;
2815 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2816 ORPHAN_DIR_SYSTEM_INODE,
2817 osb->slot_num);
2818 if (!orphan_dir_inode) {
2819 status = -ENOENT;
2820 mlog_errno(status);
2821 goto leave;
2824 inode_lock(orphan_dir_inode);
2826 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2827 if (status < 0) {
2828 mlog_errno(status);
2829 inode_unlock(orphan_dir_inode);
2830 iput(orphan_dir_inode);
2831 goto leave;
2834 status = ocfs2_read_inode_block(inode, &di_bh);
2835 if (status < 0) {
2836 mlog_errno(status);
2837 goto orphan_unlock;
2840 handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2841 if (IS_ERR(handle)) {
2842 status = PTR_ERR(handle);
2843 handle = NULL;
2844 mlog_errno(status);
2845 goto orphan_unlock;
2848 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2849 di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2850 if (status < 0) {
2851 mlog_errno(status);
2852 goto out_commit;
2855 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2856 orphan_dir_bh, false);
2857 if (status < 0) {
2858 mlog_errno(status);
2859 goto out_commit;
2862 di = (struct ocfs2_dinode *)di_bh->b_data;
2863 di->i_flags &= ~cpu_to_le32(OCFS2_ORPHANED_FL);
2864 di->i_orphaned_slot = 0;
2865 set_nlink(inode, 1);
2866 ocfs2_set_links_count(di, inode->i_nlink);
2867 ocfs2_update_inode_fsync_trans(handle, inode, 1);
2868 ocfs2_journal_dirty(handle, di_bh);
2870 status = ocfs2_add_entry(handle, dentry, inode,
2871 OCFS2_I(inode)->ip_blkno, parent_di_bh,
2872 &lookup);
2873 if (status < 0) {
2874 mlog_errno(status);
2875 goto out_commit;
2878 status = ocfs2_dentry_attach_lock(dentry, inode,
2879 OCFS2_I(dir)->ip_blkno);
2880 if (status) {
2881 mlog_errno(status);
2882 goto out_commit;
2885 d_instantiate(dentry, inode);
2886 status = 0;
2887 out_commit:
2888 ocfs2_commit_trans(osb, handle);
2889 orphan_unlock:
2890 ocfs2_inode_unlock(orphan_dir_inode, 1);
2891 inode_unlock(orphan_dir_inode);
2892 iput(orphan_dir_inode);
2893 leave:
2895 ocfs2_inode_unlock(dir, 1);
2897 brelse(di_bh);
2898 brelse(parent_di_bh);
2899 brelse(orphan_dir_bh);
2901 ocfs2_free_dir_lookup_result(&lookup);
2903 if (status)
2904 mlog_errno(status);
2906 return status;
2909 const struct inode_operations ocfs2_dir_iops = {
2910 .create = ocfs2_create,
2911 .lookup = ocfs2_lookup,
2912 .link = ocfs2_link,
2913 .unlink = ocfs2_unlink,
2914 .rmdir = ocfs2_unlink,
2915 .symlink = ocfs2_symlink,
2916 .mkdir = ocfs2_mkdir,
2917 .mknod = ocfs2_mknod,
2918 .rename = ocfs2_rename,
2919 .setattr = ocfs2_setattr,
2920 .getattr = ocfs2_getattr,
2921 .permission = ocfs2_permission,
2922 .listxattr = ocfs2_listxattr,
2923 .fiemap = ocfs2_fiemap,
2924 .get_acl = ocfs2_iop_get_acl,
2925 .set_acl = ocfs2_iop_set_acl,