ext4: Optimize ext4 DIO overwrites
[linux/fpc-iii.git] / fs / ocfs2 / namei.c
blob8ea51cf27b9707a25468fdf7a96029f666d52197
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* -*- mode: c; c-basic-offset: 8; -*-
3 * vim: noexpandtab sw=8 ts=8 sts=0:
5 * namei.c
7 * Create and rename file, directory, symlinks
9 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
11 * Portions of this code from linux/fs/ext3/dir.c
13 * Copyright (C) 1992, 1993, 1994, 1995
14 * Remy Card (card@masi.ibp.fr)
15 * Laboratoire MASI - Institut Blaise pascal
16 * Universite Pierre et Marie Curie (Paris VI)
18 * from
20 * linux/fs/minix/dir.c
22 * Copyright (C) 1991, 1992 Linux Torvalds
25 #include <linux/fs.h>
26 #include <linux/types.h>
27 #include <linux/slab.h>
28 #include <linux/highmem.h>
29 #include <linux/quotaops.h>
30 #include <linux/iversion.h>
32 #include <cluster/masklog.h>
34 #include "ocfs2.h"
36 #include "alloc.h"
37 #include "dcache.h"
38 #include "dir.h"
39 #include "dlmglue.h"
40 #include "extent_map.h"
41 #include "file.h"
42 #include "inode.h"
43 #include "journal.h"
44 #include "namei.h"
45 #include "suballoc.h"
46 #include "super.h"
47 #include "symlink.h"
48 #include "sysfile.h"
49 #include "uptodate.h"
50 #include "xattr.h"
51 #include "acl.h"
52 #include "ocfs2_trace.h"
54 #include "buffer_head_io.h"
56 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
57 struct inode *dir,
58 struct inode *inode,
59 dev_t dev,
60 struct buffer_head **new_fe_bh,
61 struct buffer_head *parent_fe_bh,
62 handle_t *handle,
63 struct ocfs2_alloc_context *inode_ac);
65 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
66 struct inode **ret_orphan_dir,
67 u64 blkno,
68 char *name,
69 struct ocfs2_dir_lookup_result *lookup,
70 bool dio);
72 static int ocfs2_orphan_add(struct ocfs2_super *osb,
73 handle_t *handle,
74 struct inode *inode,
75 struct buffer_head *fe_bh,
76 char *name,
77 struct ocfs2_dir_lookup_result *lookup,
78 struct inode *orphan_dir_inode,
79 bool dio);
81 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
82 handle_t *handle,
83 struct inode *inode,
84 const char *symname);
86 static int ocfs2_double_lock(struct ocfs2_super *osb,
87 struct buffer_head **bh1,
88 struct inode *inode1,
89 struct buffer_head **bh2,
90 struct inode *inode2,
91 int rename);
93 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2);
94 /* An orphan dir name is an 8 byte value, printed as a hex string */
95 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
97 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
98 unsigned int flags)
100 int status;
101 u64 blkno;
102 struct inode *inode = NULL;
103 struct dentry *ret;
104 struct ocfs2_inode_info *oi;
106 trace_ocfs2_lookup(dir, dentry, dentry->d_name.len,
107 dentry->d_name.name,
108 (unsigned long long)OCFS2_I(dir)->ip_blkno, 0);
110 if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
111 ret = ERR_PTR(-ENAMETOOLONG);
112 goto bail;
115 status = ocfs2_inode_lock_nested(dir, NULL, 0, OI_LS_PARENT);
116 if (status < 0) {
117 if (status != -ENOENT)
118 mlog_errno(status);
119 ret = ERR_PTR(status);
120 goto bail;
123 status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
124 dentry->d_name.len, &blkno);
125 if (status < 0)
126 goto bail_add;
128 inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
129 if (IS_ERR(inode)) {
130 ret = ERR_PTR(-EACCES);
131 goto bail_unlock;
134 oi = OCFS2_I(inode);
135 /* Clear any orphaned state... If we were able to look up the
136 * inode from a directory, it certainly can't be orphaned. We
137 * might have the bad state from a node which intended to
138 * orphan this inode but crashed before it could commit the
139 * unlink. */
140 spin_lock(&oi->ip_lock);
141 oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
142 spin_unlock(&oi->ip_lock);
144 bail_add:
145 ret = d_splice_alias(inode, dentry);
147 if (inode) {
149 * If d_splice_alias() finds a DCACHE_DISCONNECTED
150 * dentry, it will d_move() it on top of ourse. The
151 * return value will indicate this however, so in
152 * those cases, we switch them around for the locking
153 * code.
155 * NOTE: This dentry already has ->d_op set from
156 * ocfs2_get_parent() and ocfs2_get_dentry()
158 if (!IS_ERR_OR_NULL(ret))
159 dentry = ret;
161 status = ocfs2_dentry_attach_lock(dentry, inode,
162 OCFS2_I(dir)->ip_blkno);
163 if (status) {
164 mlog_errno(status);
165 ret = ERR_PTR(status);
166 goto bail_unlock;
168 } else
169 ocfs2_dentry_attach_gen(dentry);
171 bail_unlock:
172 /* Don't drop the cluster lock until *after* the d_add --
173 * unlink on another node will message us to remove that
174 * dentry under this lock so otherwise we can race this with
175 * the downconvert thread and have a stale dentry. */
176 ocfs2_inode_unlock(dir, 0);
178 bail:
180 trace_ocfs2_lookup_ret(ret);
182 return ret;
185 static struct inode *ocfs2_get_init_inode(struct inode *dir, umode_t mode)
187 struct inode *inode;
188 int status;
190 inode = new_inode(dir->i_sb);
191 if (!inode) {
192 mlog(ML_ERROR, "new_inode failed!\n");
193 return ERR_PTR(-ENOMEM);
196 /* populate as many fields early on as possible - many of
197 * these are used by the support functions here and in
198 * callers. */
199 if (S_ISDIR(mode))
200 set_nlink(inode, 2);
201 inode_init_owner(inode, dir, mode);
202 status = dquot_initialize(inode);
203 if (status)
204 return ERR_PTR(status);
206 return inode;
209 static void ocfs2_cleanup_add_entry_failure(struct ocfs2_super *osb,
210 struct dentry *dentry, struct inode *inode)
212 struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
214 ocfs2_simple_drop_lockres(osb, &dl->dl_lockres);
215 ocfs2_lock_res_free(&dl->dl_lockres);
216 BUG_ON(dl->dl_count != 1);
217 spin_lock(&dentry_attach_lock);
218 dentry->d_fsdata = NULL;
219 spin_unlock(&dentry_attach_lock);
220 kfree(dl);
221 iput(inode);
224 static int ocfs2_mknod(struct inode *dir,
225 struct dentry *dentry,
226 umode_t mode,
227 dev_t dev)
229 int status = 0;
230 struct buffer_head *parent_fe_bh = NULL;
231 handle_t *handle = NULL;
232 struct ocfs2_super *osb;
233 struct ocfs2_dinode *dirfe;
234 struct buffer_head *new_fe_bh = NULL;
235 struct inode *inode = NULL;
236 struct ocfs2_alloc_context *inode_ac = NULL;
237 struct ocfs2_alloc_context *data_ac = NULL;
238 struct ocfs2_alloc_context *meta_ac = NULL;
239 int want_clusters = 0;
240 int want_meta = 0;
241 int xattr_credits = 0;
242 struct ocfs2_security_xattr_info si = {
243 .enable = 1,
245 int did_quota_inode = 0;
246 struct ocfs2_dir_lookup_result lookup = { NULL, };
247 sigset_t oldset;
248 int did_block_signals = 0;
249 struct ocfs2_dentry_lock *dl = NULL;
251 trace_ocfs2_mknod(dir, dentry, dentry->d_name.len, dentry->d_name.name,
252 (unsigned long long)OCFS2_I(dir)->ip_blkno,
253 (unsigned long)dev, mode);
255 status = dquot_initialize(dir);
256 if (status) {
257 mlog_errno(status);
258 return status;
261 /* get our super block */
262 osb = OCFS2_SB(dir->i_sb);
264 status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
265 if (status < 0) {
266 if (status != -ENOENT)
267 mlog_errno(status);
268 return status;
271 if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
272 status = -EMLINK;
273 goto leave;
276 dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
277 if (!ocfs2_read_links_count(dirfe)) {
278 /* can't make a file in a deleted directory. */
279 status = -ENOENT;
280 goto leave;
283 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
284 dentry->d_name.len);
285 if (status)
286 goto leave;
288 /* get a spot inside the dir. */
289 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
290 dentry->d_name.name,
291 dentry->d_name.len, &lookup);
292 if (status < 0) {
293 mlog_errno(status);
294 goto leave;
297 /* reserve an inode spot */
298 status = ocfs2_reserve_new_inode(osb, &inode_ac);
299 if (status < 0) {
300 if (status != -ENOSPC)
301 mlog_errno(status);
302 goto leave;
305 inode = ocfs2_get_init_inode(dir, mode);
306 if (IS_ERR(inode)) {
307 status = PTR_ERR(inode);
308 inode = NULL;
309 mlog_errno(status);
310 goto leave;
313 /* get security xattr */
314 status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
315 if (status) {
316 if (status == -EOPNOTSUPP)
317 si.enable = 0;
318 else {
319 mlog_errno(status);
320 goto leave;
324 /* calculate meta data/clusters for setting security and acl xattr */
325 status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
326 &si, &want_clusters,
327 &xattr_credits, &want_meta);
328 if (status < 0) {
329 mlog_errno(status);
330 goto leave;
333 /* Reserve a cluster if creating an extent based directory. */
334 if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
335 want_clusters += 1;
337 /* Dir indexing requires extra space as well */
338 if (ocfs2_supports_indexed_dirs(osb))
339 want_meta++;
342 status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
343 if (status < 0) {
344 if (status != -ENOSPC)
345 mlog_errno(status);
346 goto leave;
349 status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
350 if (status < 0) {
351 if (status != -ENOSPC)
352 mlog_errno(status);
353 goto leave;
356 handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
357 S_ISDIR(mode),
358 xattr_credits));
359 if (IS_ERR(handle)) {
360 status = PTR_ERR(handle);
361 handle = NULL;
362 mlog_errno(status);
363 goto leave;
366 /* Starting to change things, restart is no longer possible. */
367 ocfs2_block_signals(&oldset);
368 did_block_signals = 1;
370 status = dquot_alloc_inode(inode);
371 if (status)
372 goto leave;
373 did_quota_inode = 1;
375 /* do the real work now. */
376 status = ocfs2_mknod_locked(osb, dir, inode, dev,
377 &new_fe_bh, parent_fe_bh, handle,
378 inode_ac);
379 if (status < 0) {
380 mlog_errno(status);
381 goto leave;
384 if (S_ISDIR(mode)) {
385 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
386 new_fe_bh, data_ac, meta_ac);
387 if (status < 0) {
388 mlog_errno(status);
389 goto leave;
392 status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
393 parent_fe_bh,
394 OCFS2_JOURNAL_ACCESS_WRITE);
395 if (status < 0) {
396 mlog_errno(status);
397 goto leave;
399 ocfs2_add_links_count(dirfe, 1);
400 ocfs2_journal_dirty(handle, parent_fe_bh);
401 inc_nlink(dir);
404 status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
405 meta_ac, data_ac);
407 if (status < 0) {
408 mlog_errno(status);
409 goto leave;
412 if (si.enable) {
413 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
414 meta_ac, data_ac);
415 if (status < 0) {
416 mlog_errno(status);
417 goto leave;
422 * Do this before adding the entry to the directory. We add
423 * also set d_op after success so that ->d_iput() will cleanup
424 * the dentry lock even if ocfs2_add_entry() fails below.
426 status = ocfs2_dentry_attach_lock(dentry, inode,
427 OCFS2_I(dir)->ip_blkno);
428 if (status) {
429 mlog_errno(status);
430 goto leave;
433 dl = dentry->d_fsdata;
435 status = ocfs2_add_entry(handle, dentry, inode,
436 OCFS2_I(inode)->ip_blkno, parent_fe_bh,
437 &lookup);
438 if (status < 0) {
439 mlog_errno(status);
440 goto leave;
443 insert_inode_hash(inode);
444 d_instantiate(dentry, inode);
445 status = 0;
446 leave:
447 if (status < 0 && did_quota_inode)
448 dquot_free_inode(inode);
449 if (handle)
450 ocfs2_commit_trans(osb, handle);
452 ocfs2_inode_unlock(dir, 1);
453 if (did_block_signals)
454 ocfs2_unblock_signals(&oldset);
456 brelse(new_fe_bh);
457 brelse(parent_fe_bh);
458 kfree(si.value);
460 ocfs2_free_dir_lookup_result(&lookup);
462 if (inode_ac)
463 ocfs2_free_alloc_context(inode_ac);
465 if (data_ac)
466 ocfs2_free_alloc_context(data_ac);
468 if (meta_ac)
469 ocfs2_free_alloc_context(meta_ac);
472 * We should call iput after the i_mutex of the bitmap been
473 * unlocked in ocfs2_free_alloc_context, or the
474 * ocfs2_delete_inode will mutex_lock again.
476 if ((status < 0) && inode) {
477 if (dl)
478 ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
480 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
481 clear_nlink(inode);
482 iput(inode);
485 if (status)
486 mlog_errno(status);
488 return status;
491 static int __ocfs2_mknod_locked(struct inode *dir,
492 struct inode *inode,
493 dev_t dev,
494 struct buffer_head **new_fe_bh,
495 struct buffer_head *parent_fe_bh,
496 handle_t *handle,
497 struct ocfs2_alloc_context *inode_ac,
498 u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
500 int status = 0;
501 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
502 struct ocfs2_dinode *fe = NULL;
503 struct ocfs2_extent_list *fel;
504 u16 feat;
505 struct ocfs2_inode_info *oi = OCFS2_I(inode);
506 struct timespec64 ts;
508 *new_fe_bh = NULL;
510 /* populate as many fields early on as possible - many of
511 * these are used by the support functions here and in
512 * callers. */
513 inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
514 oi->ip_blkno = fe_blkno;
515 spin_lock(&osb->osb_lock);
516 inode->i_generation = osb->s_next_generation++;
517 spin_unlock(&osb->osb_lock);
519 *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
520 if (!*new_fe_bh) {
521 status = -ENOMEM;
522 mlog_errno(status);
523 goto leave;
525 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
527 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
528 *new_fe_bh,
529 OCFS2_JOURNAL_ACCESS_CREATE);
530 if (status < 0) {
531 mlog_errno(status);
532 goto leave;
535 fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
536 memset(fe, 0, osb->sb->s_blocksize);
538 fe->i_generation = cpu_to_le32(inode->i_generation);
539 fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
540 fe->i_blkno = cpu_to_le64(fe_blkno);
541 fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
542 fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
543 fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
544 fe->i_uid = cpu_to_le32(i_uid_read(inode));
545 fe->i_gid = cpu_to_le32(i_gid_read(inode));
546 fe->i_mode = cpu_to_le16(inode->i_mode);
547 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
548 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
550 ocfs2_set_links_count(fe, inode->i_nlink);
552 fe->i_last_eb_blk = 0;
553 strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
554 fe->i_flags |= cpu_to_le32(OCFS2_VALID_FL);
555 ktime_get_real_ts64(&ts);
556 fe->i_atime = fe->i_ctime = fe->i_mtime =
557 cpu_to_le64(ts.tv_sec);
558 fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
559 cpu_to_le32(ts.tv_nsec);
560 fe->i_dtime = 0;
563 * If supported, directories start with inline data. If inline
564 * isn't supported, but indexing is, we start them as indexed.
566 feat = le16_to_cpu(fe->i_dyn_features);
567 if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
568 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
570 fe->id2.i_data.id_count = cpu_to_le16(
571 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
572 } else {
573 fel = &fe->id2.i_list;
574 fel->l_tree_depth = 0;
575 fel->l_next_free_rec = 0;
576 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
579 ocfs2_journal_dirty(handle, *new_fe_bh);
581 ocfs2_populate_inode(inode, fe, 1);
582 ocfs2_ci_set_new(osb, INODE_CACHE(inode));
583 if (!ocfs2_mount_local(osb)) {
584 status = ocfs2_create_new_inode_locks(inode);
585 if (status < 0)
586 mlog_errno(status);
589 oi->i_sync_tid = handle->h_transaction->t_tid;
590 oi->i_datasync_tid = handle->h_transaction->t_tid;
592 leave:
593 if (status < 0) {
594 if (*new_fe_bh) {
595 brelse(*new_fe_bh);
596 *new_fe_bh = NULL;
600 if (status)
601 mlog_errno(status);
602 return status;
605 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
606 struct inode *dir,
607 struct inode *inode,
608 dev_t dev,
609 struct buffer_head **new_fe_bh,
610 struct buffer_head *parent_fe_bh,
611 handle_t *handle,
612 struct ocfs2_alloc_context *inode_ac)
614 int status = 0;
615 u64 suballoc_loc, fe_blkno = 0;
616 u16 suballoc_bit;
618 *new_fe_bh = NULL;
620 status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
621 inode_ac, &suballoc_loc,
622 &suballoc_bit, &fe_blkno);
623 if (status < 0) {
624 mlog_errno(status);
625 return status;
628 status = __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
629 parent_fe_bh, handle, inode_ac,
630 fe_blkno, suballoc_loc, suballoc_bit);
631 if (status < 0) {
632 u64 bg_blkno = ocfs2_which_suballoc_group(fe_blkno, suballoc_bit);
633 int tmp = ocfs2_free_suballoc_bits(handle, inode_ac->ac_inode,
634 inode_ac->ac_bh, suballoc_bit, bg_blkno, 1);
635 if (tmp)
636 mlog_errno(tmp);
639 return status;
642 static int ocfs2_mkdir(struct inode *dir,
643 struct dentry *dentry,
644 umode_t mode)
646 int ret;
648 trace_ocfs2_mkdir(dir, dentry, dentry->d_name.len, dentry->d_name.name,
649 OCFS2_I(dir)->ip_blkno, mode);
650 ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
651 if (ret)
652 mlog_errno(ret);
654 return ret;
657 static int ocfs2_create(struct inode *dir,
658 struct dentry *dentry,
659 umode_t mode,
660 bool excl)
662 int ret;
664 trace_ocfs2_create(dir, dentry, dentry->d_name.len, dentry->d_name.name,
665 (unsigned long long)OCFS2_I(dir)->ip_blkno, mode);
666 ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
667 if (ret)
668 mlog_errno(ret);
670 return ret;
673 static int ocfs2_link(struct dentry *old_dentry,
674 struct inode *dir,
675 struct dentry *dentry)
677 handle_t *handle;
678 struct inode *inode = d_inode(old_dentry);
679 struct inode *old_dir = d_inode(old_dentry->d_parent);
680 int err;
681 struct buffer_head *fe_bh = NULL;
682 struct buffer_head *old_dir_bh = NULL;
683 struct buffer_head *parent_fe_bh = NULL;
684 struct ocfs2_dinode *fe = NULL;
685 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
686 struct ocfs2_dir_lookup_result lookup = { NULL, };
687 sigset_t oldset;
688 u64 old_de_ino;
690 trace_ocfs2_link((unsigned long long)OCFS2_I(inode)->ip_blkno,
691 old_dentry->d_name.len, old_dentry->d_name.name,
692 dentry->d_name.len, dentry->d_name.name);
694 if (S_ISDIR(inode->i_mode))
695 return -EPERM;
697 err = dquot_initialize(dir);
698 if (err) {
699 mlog_errno(err);
700 return err;
703 err = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
704 &parent_fe_bh, dir, 0);
705 if (err < 0) {
706 if (err != -ENOENT)
707 mlog_errno(err);
708 return err;
711 /* make sure both dirs have bhs
712 * get an extra ref on old_dir_bh if old==new */
713 if (!parent_fe_bh) {
714 if (old_dir_bh) {
715 parent_fe_bh = old_dir_bh;
716 get_bh(parent_fe_bh);
717 } else {
718 mlog(ML_ERROR, "%s: no old_dir_bh!\n", osb->uuid_str);
719 err = -EIO;
720 goto out;
724 if (!dir->i_nlink) {
725 err = -ENOENT;
726 goto out;
729 err = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
730 old_dentry->d_name.len, &old_de_ino);
731 if (err) {
732 err = -ENOENT;
733 goto out;
737 * Check whether another node removed the source inode while we
738 * were in the vfs.
740 if (old_de_ino != OCFS2_I(inode)->ip_blkno) {
741 err = -ENOENT;
742 goto out;
745 err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
746 dentry->d_name.len);
747 if (err)
748 goto out;
750 err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
751 dentry->d_name.name,
752 dentry->d_name.len, &lookup);
753 if (err < 0) {
754 mlog_errno(err);
755 goto out;
758 err = ocfs2_inode_lock(inode, &fe_bh, 1);
759 if (err < 0) {
760 if (err != -ENOENT)
761 mlog_errno(err);
762 goto out;
765 fe = (struct ocfs2_dinode *) fe_bh->b_data;
766 if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
767 err = -EMLINK;
768 goto out_unlock_inode;
771 handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
772 if (IS_ERR(handle)) {
773 err = PTR_ERR(handle);
774 handle = NULL;
775 mlog_errno(err);
776 goto out_unlock_inode;
779 /* Starting to change things, restart is no longer possible. */
780 ocfs2_block_signals(&oldset);
782 err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
783 OCFS2_JOURNAL_ACCESS_WRITE);
784 if (err < 0) {
785 mlog_errno(err);
786 goto out_commit;
789 inc_nlink(inode);
790 inode->i_ctime = current_time(inode);
791 ocfs2_set_links_count(fe, inode->i_nlink);
792 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
793 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
794 ocfs2_journal_dirty(handle, fe_bh);
796 err = ocfs2_add_entry(handle, dentry, inode,
797 OCFS2_I(inode)->ip_blkno,
798 parent_fe_bh, &lookup);
799 if (err) {
800 ocfs2_add_links_count(fe, -1);
801 drop_nlink(inode);
802 mlog_errno(err);
803 goto out_commit;
806 err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
807 if (err) {
808 mlog_errno(err);
809 goto out_commit;
812 ihold(inode);
813 d_instantiate(dentry, inode);
815 out_commit:
816 ocfs2_commit_trans(osb, handle);
817 ocfs2_unblock_signals(&oldset);
818 out_unlock_inode:
819 ocfs2_inode_unlock(inode, 1);
821 out:
822 ocfs2_double_unlock(old_dir, dir);
824 brelse(fe_bh);
825 brelse(parent_fe_bh);
826 brelse(old_dir_bh);
828 ocfs2_free_dir_lookup_result(&lookup);
830 if (err)
831 mlog_errno(err);
833 return err;
837 * Takes and drops an exclusive lock on the given dentry. This will
838 * force other nodes to drop it.
840 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
842 int ret;
844 ret = ocfs2_dentry_lock(dentry, 1);
845 if (ret)
846 mlog_errno(ret);
847 else
848 ocfs2_dentry_unlock(dentry, 1);
850 return ret;
853 static inline int ocfs2_inode_is_unlinkable(struct inode *inode)
855 if (S_ISDIR(inode->i_mode)) {
856 if (inode->i_nlink == 2)
857 return 1;
858 return 0;
861 if (inode->i_nlink == 1)
862 return 1;
863 return 0;
866 static int ocfs2_unlink(struct inode *dir,
867 struct dentry *dentry)
869 int status;
870 int child_locked = 0;
871 bool is_unlinkable = false;
872 struct inode *inode = d_inode(dentry);
873 struct inode *orphan_dir = NULL;
874 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
875 u64 blkno;
876 struct ocfs2_dinode *fe = NULL;
877 struct buffer_head *fe_bh = NULL;
878 struct buffer_head *parent_node_bh = NULL;
879 handle_t *handle = NULL;
880 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
881 struct ocfs2_dir_lookup_result lookup = { NULL, };
882 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
884 trace_ocfs2_unlink(dir, dentry, dentry->d_name.len,
885 dentry->d_name.name,
886 (unsigned long long)OCFS2_I(dir)->ip_blkno,
887 (unsigned long long)OCFS2_I(inode)->ip_blkno);
889 status = dquot_initialize(dir);
890 if (status) {
891 mlog_errno(status);
892 return status;
895 BUG_ON(d_inode(dentry->d_parent) != dir);
897 if (inode == osb->root_inode)
898 return -EPERM;
900 status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
901 OI_LS_PARENT);
902 if (status < 0) {
903 if (status != -ENOENT)
904 mlog_errno(status);
905 return status;
908 status = ocfs2_find_files_on_disk(dentry->d_name.name,
909 dentry->d_name.len, &blkno, dir,
910 &lookup);
911 if (status < 0) {
912 if (status != -ENOENT)
913 mlog_errno(status);
914 goto leave;
917 if (OCFS2_I(inode)->ip_blkno != blkno) {
918 status = -ENOENT;
920 trace_ocfs2_unlink_noent(
921 (unsigned long long)OCFS2_I(inode)->ip_blkno,
922 (unsigned long long)blkno,
923 OCFS2_I(inode)->ip_flags);
924 goto leave;
927 status = ocfs2_inode_lock(inode, &fe_bh, 1);
928 if (status < 0) {
929 if (status != -ENOENT)
930 mlog_errno(status);
931 goto leave;
933 child_locked = 1;
935 if (S_ISDIR(inode->i_mode)) {
936 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
937 status = -ENOTEMPTY;
938 goto leave;
942 status = ocfs2_remote_dentry_delete(dentry);
943 if (status < 0) {
944 /* This remote delete should succeed under all normal
945 * circumstances. */
946 mlog_errno(status);
947 goto leave;
950 if (ocfs2_inode_is_unlinkable(inode)) {
951 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
952 OCFS2_I(inode)->ip_blkno,
953 orphan_name, &orphan_insert,
954 false);
955 if (status < 0) {
956 mlog_errno(status);
957 goto leave;
959 is_unlinkable = true;
962 handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
963 if (IS_ERR(handle)) {
964 status = PTR_ERR(handle);
965 handle = NULL;
966 mlog_errno(status);
967 goto leave;
970 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
971 OCFS2_JOURNAL_ACCESS_WRITE);
972 if (status < 0) {
973 mlog_errno(status);
974 goto leave;
977 fe = (struct ocfs2_dinode *) fe_bh->b_data;
979 /* delete the name from the parent dir */
980 status = ocfs2_delete_entry(handle, dir, &lookup);
981 if (status < 0) {
982 mlog_errno(status);
983 goto leave;
986 if (S_ISDIR(inode->i_mode))
987 drop_nlink(inode);
988 drop_nlink(inode);
989 ocfs2_set_links_count(fe, inode->i_nlink);
990 ocfs2_journal_dirty(handle, fe_bh);
992 dir->i_ctime = dir->i_mtime = current_time(dir);
993 if (S_ISDIR(inode->i_mode))
994 drop_nlink(dir);
996 status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
997 if (status < 0) {
998 mlog_errno(status);
999 if (S_ISDIR(inode->i_mode))
1000 inc_nlink(dir);
1001 goto leave;
1004 if (is_unlinkable) {
1005 status = ocfs2_orphan_add(osb, handle, inode, fe_bh,
1006 orphan_name, &orphan_insert, orphan_dir, false);
1007 if (status < 0)
1008 mlog_errno(status);
1011 leave:
1012 if (handle)
1013 ocfs2_commit_trans(osb, handle);
1015 if (orphan_dir) {
1016 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1017 ocfs2_inode_unlock(orphan_dir, 1);
1018 inode_unlock(orphan_dir);
1019 iput(orphan_dir);
1022 if (child_locked)
1023 ocfs2_inode_unlock(inode, 1);
1025 ocfs2_inode_unlock(dir, 1);
1027 brelse(fe_bh);
1028 brelse(parent_node_bh);
1030 ocfs2_free_dir_lookup_result(&orphan_insert);
1031 ocfs2_free_dir_lookup_result(&lookup);
1033 if (status && (status != -ENOTEMPTY) && (status != -ENOENT))
1034 mlog_errno(status);
1036 return status;
1039 static int ocfs2_check_if_ancestor(struct ocfs2_super *osb,
1040 u64 src_inode_no, u64 dest_inode_no)
1042 int ret = 0, i = 0;
1043 u64 parent_inode_no = 0;
1044 u64 child_inode_no = src_inode_no;
1045 struct inode *child_inode;
1047 #define MAX_LOOKUP_TIMES 32
1048 while (1) {
1049 child_inode = ocfs2_iget(osb, child_inode_no, 0, 0);
1050 if (IS_ERR(child_inode)) {
1051 ret = PTR_ERR(child_inode);
1052 break;
1055 ret = ocfs2_inode_lock(child_inode, NULL, 0);
1056 if (ret < 0) {
1057 iput(child_inode);
1058 if (ret != -ENOENT)
1059 mlog_errno(ret);
1060 break;
1063 ret = ocfs2_lookup_ino_from_name(child_inode, "..", 2,
1064 &parent_inode_no);
1065 ocfs2_inode_unlock(child_inode, 0);
1066 iput(child_inode);
1067 if (ret < 0) {
1068 ret = -ENOENT;
1069 break;
1072 if (parent_inode_no == dest_inode_no) {
1073 ret = 1;
1074 break;
1077 if (parent_inode_no == osb->root_inode->i_ino) {
1078 ret = 0;
1079 break;
1082 child_inode_no = parent_inode_no;
1084 if (++i >= MAX_LOOKUP_TIMES) {
1085 mlog(ML_NOTICE, "max lookup times reached, filesystem "
1086 "may have nested directories, "
1087 "src inode: %llu, dest inode: %llu.\n",
1088 (unsigned long long)src_inode_no,
1089 (unsigned long long)dest_inode_no);
1090 ret = 0;
1091 break;
1095 return ret;
1099 * The only place this should be used is rename and link!
1100 * if they have the same id, then the 1st one is the only one locked.
1102 static int ocfs2_double_lock(struct ocfs2_super *osb,
1103 struct buffer_head **bh1,
1104 struct inode *inode1,
1105 struct buffer_head **bh2,
1106 struct inode *inode2,
1107 int rename)
1109 int status;
1110 int inode1_is_ancestor, inode2_is_ancestor;
1111 struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
1112 struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
1114 trace_ocfs2_double_lock((unsigned long long)oi1->ip_blkno,
1115 (unsigned long long)oi2->ip_blkno);
1117 if (*bh1)
1118 *bh1 = NULL;
1119 if (*bh2)
1120 *bh2 = NULL;
1122 /* we always want to lock the one with the lower lockid first.
1123 * and if they are nested, we lock ancestor first */
1124 if (oi1->ip_blkno != oi2->ip_blkno) {
1125 inode1_is_ancestor = ocfs2_check_if_ancestor(osb, oi2->ip_blkno,
1126 oi1->ip_blkno);
1127 if (inode1_is_ancestor < 0) {
1128 status = inode1_is_ancestor;
1129 goto bail;
1132 inode2_is_ancestor = ocfs2_check_if_ancestor(osb, oi1->ip_blkno,
1133 oi2->ip_blkno);
1134 if (inode2_is_ancestor < 0) {
1135 status = inode2_is_ancestor;
1136 goto bail;
1139 if ((inode1_is_ancestor == 1) ||
1140 (oi1->ip_blkno < oi2->ip_blkno &&
1141 inode2_is_ancestor == 0)) {
1142 /* switch id1 and id2 around */
1143 swap(bh2, bh1);
1144 swap(inode2, inode1);
1146 /* lock id2 */
1147 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
1148 rename == 1 ? OI_LS_RENAME1 : OI_LS_PARENT);
1149 if (status < 0) {
1150 if (status != -ENOENT)
1151 mlog_errno(status);
1152 goto bail;
1156 /* lock id1 */
1157 status = ocfs2_inode_lock_nested(inode1, bh1, 1,
1158 rename == 1 ? OI_LS_RENAME2 : OI_LS_PARENT);
1159 if (status < 0) {
1161 * An error return must mean that no cluster locks
1162 * were held on function exit.
1164 if (oi1->ip_blkno != oi2->ip_blkno) {
1165 ocfs2_inode_unlock(inode2, 1);
1166 brelse(*bh2);
1167 *bh2 = NULL;
1170 if (status != -ENOENT)
1171 mlog_errno(status);
1174 trace_ocfs2_double_lock_end(
1175 (unsigned long long)oi1->ip_blkno,
1176 (unsigned long long)oi2->ip_blkno);
1178 bail:
1179 if (status)
1180 mlog_errno(status);
1181 return status;
1184 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1186 ocfs2_inode_unlock(inode1, 1);
1188 if (inode1 != inode2)
1189 ocfs2_inode_unlock(inode2, 1);
1192 static int ocfs2_rename(struct inode *old_dir,
1193 struct dentry *old_dentry,
1194 struct inode *new_dir,
1195 struct dentry *new_dentry,
1196 unsigned int flags)
1198 int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1199 int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1200 struct inode *old_inode = d_inode(old_dentry);
1201 struct inode *new_inode = d_inode(new_dentry);
1202 struct inode *orphan_dir = NULL;
1203 struct ocfs2_dinode *newfe = NULL;
1204 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1205 struct buffer_head *newfe_bh = NULL;
1206 struct buffer_head *old_inode_bh = NULL;
1207 struct ocfs2_super *osb = NULL;
1208 u64 newfe_blkno, old_de_ino;
1209 handle_t *handle = NULL;
1210 struct buffer_head *old_dir_bh = NULL;
1211 struct buffer_head *new_dir_bh = NULL;
1212 u32 old_dir_nlink = old_dir->i_nlink;
1213 struct ocfs2_dinode *old_di;
1214 struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1215 struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1216 struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1217 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1218 struct ocfs2_dir_lookup_result target_insert = { NULL, };
1219 bool should_add_orphan = false;
1221 if (flags)
1222 return -EINVAL;
1224 /* At some point it might be nice to break this function up a
1225 * bit. */
1227 trace_ocfs2_rename(old_dir, old_dentry, new_dir, new_dentry,
1228 old_dentry->d_name.len, old_dentry->d_name.name,
1229 new_dentry->d_name.len, new_dentry->d_name.name);
1231 status = dquot_initialize(old_dir);
1232 if (status) {
1233 mlog_errno(status);
1234 goto bail;
1236 status = dquot_initialize(new_dir);
1237 if (status) {
1238 mlog_errno(status);
1239 goto bail;
1242 osb = OCFS2_SB(old_dir->i_sb);
1244 if (new_inode) {
1245 if (!igrab(new_inode))
1246 BUG();
1249 /* Assume a directory hierarchy thusly:
1250 * a/b/c
1251 * a/d
1252 * a,b,c, and d are all directories.
1254 * from cwd of 'a' on both nodes:
1255 * node1: mv b/c d
1256 * node2: mv d b/c
1258 * And that's why, just like the VFS, we need a file system
1259 * rename lock. */
1260 if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1261 status = ocfs2_rename_lock(osb);
1262 if (status < 0) {
1263 mlog_errno(status);
1264 goto bail;
1266 rename_lock = 1;
1268 /* here we cannot guarantee the inodes haven't just been
1269 * changed, so check if they are nested again */
1270 status = ocfs2_check_if_ancestor(osb, new_dir->i_ino,
1271 old_inode->i_ino);
1272 if (status < 0) {
1273 mlog_errno(status);
1274 goto bail;
1275 } else if (status == 1) {
1276 status = -EPERM;
1277 trace_ocfs2_rename_not_permitted(
1278 (unsigned long long)old_inode->i_ino,
1279 (unsigned long long)new_dir->i_ino);
1280 goto bail;
1284 /* if old and new are the same, this'll just do one lock. */
1285 status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1286 &new_dir_bh, new_dir, 1);
1287 if (status < 0) {
1288 mlog_errno(status);
1289 goto bail;
1291 parents_locked = 1;
1293 if (!new_dir->i_nlink) {
1294 status = -EACCES;
1295 goto bail;
1298 /* make sure both dirs have bhs
1299 * get an extra ref on old_dir_bh if old==new */
1300 if (!new_dir_bh) {
1301 if (old_dir_bh) {
1302 new_dir_bh = old_dir_bh;
1303 get_bh(new_dir_bh);
1304 } else {
1305 mlog(ML_ERROR, "no old_dir_bh!\n");
1306 status = -EIO;
1307 goto bail;
1312 * Aside from allowing a meta data update, the locking here
1313 * also ensures that the downconvert thread on other nodes
1314 * won't have to concurrently downconvert the inode and the
1315 * dentry locks.
1317 status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1318 OI_LS_PARENT);
1319 if (status < 0) {
1320 if (status != -ENOENT)
1321 mlog_errno(status);
1322 goto bail;
1324 old_child_locked = 1;
1326 status = ocfs2_remote_dentry_delete(old_dentry);
1327 if (status < 0) {
1328 mlog_errno(status);
1329 goto bail;
1332 if (S_ISDIR(old_inode->i_mode)) {
1333 u64 old_inode_parent;
1335 update_dot_dot = 1;
1336 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1337 old_inode,
1338 &old_inode_dot_dot_res);
1339 if (status) {
1340 status = -EIO;
1341 goto bail;
1344 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1345 status = -EIO;
1346 goto bail;
1349 if (!new_inode && new_dir != old_dir &&
1350 new_dir->i_nlink >= ocfs2_link_max(osb)) {
1351 status = -EMLINK;
1352 goto bail;
1356 status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1357 old_dentry->d_name.len,
1358 &old_de_ino);
1359 if (status) {
1360 status = -ENOENT;
1361 goto bail;
1365 * Check for inode number is _not_ due to possible IO errors.
1366 * We might rmdir the source, keep it as pwd of some process
1367 * and merrily kill the link to whatever was created under the
1368 * same name. Goodbye sticky bit ;-<
1370 if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1371 status = -ENOENT;
1372 goto bail;
1375 /* check if the target already exists (in which case we need
1376 * to delete it */
1377 status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1378 new_dentry->d_name.len,
1379 &newfe_blkno, new_dir,
1380 &target_lookup_res);
1381 /* The only error we allow here is -ENOENT because the new
1382 * file not existing is perfectly valid. */
1383 if ((status < 0) && (status != -ENOENT)) {
1384 /* If we cannot find the file specified we should just */
1385 /* return the error... */
1386 mlog_errno(status);
1387 goto bail;
1389 if (status == 0)
1390 target_exists = 1;
1392 if (!target_exists && new_inode) {
1394 * Target was unlinked by another node while we were
1395 * waiting to get to ocfs2_rename(). There isn't
1396 * anything we can do here to help the situation, so
1397 * bubble up the appropriate error.
1399 status = -ENOENT;
1400 goto bail;
1403 /* In case we need to overwrite an existing file, we blow it
1404 * away first */
1405 if (target_exists) {
1406 /* VFS didn't think there existed an inode here, but
1407 * someone else in the cluster must have raced our
1408 * rename to create one. Today we error cleanly, in
1409 * the future we should consider calling iget to build
1410 * a new struct inode for this entry. */
1411 if (!new_inode) {
1412 status = -EACCES;
1414 trace_ocfs2_rename_target_exists(new_dentry->d_name.len,
1415 new_dentry->d_name.name);
1416 goto bail;
1419 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1420 status = -EACCES;
1422 trace_ocfs2_rename_disagree(
1423 (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1424 (unsigned long long)newfe_blkno,
1425 OCFS2_I(new_inode)->ip_flags);
1426 goto bail;
1429 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1430 if (status < 0) {
1431 if (status != -ENOENT)
1432 mlog_errno(status);
1433 goto bail;
1435 new_child_locked = 1;
1437 status = ocfs2_remote_dentry_delete(new_dentry);
1438 if (status < 0) {
1439 mlog_errno(status);
1440 goto bail;
1443 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1445 trace_ocfs2_rename_over_existing(
1446 (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1447 (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1449 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1450 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1451 OCFS2_I(new_inode)->ip_blkno,
1452 orphan_name, &orphan_insert,
1453 false);
1454 if (status < 0) {
1455 mlog_errno(status);
1456 goto bail;
1458 should_add_orphan = true;
1460 } else {
1461 BUG_ON(d_inode(new_dentry->d_parent) != new_dir);
1463 status = ocfs2_check_dir_for_entry(new_dir,
1464 new_dentry->d_name.name,
1465 new_dentry->d_name.len);
1466 if (status)
1467 goto bail;
1469 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1470 new_dentry->d_name.name,
1471 new_dentry->d_name.len,
1472 &target_insert);
1473 if (status < 0) {
1474 mlog_errno(status);
1475 goto bail;
1479 handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1480 if (IS_ERR(handle)) {
1481 status = PTR_ERR(handle);
1482 handle = NULL;
1483 mlog_errno(status);
1484 goto bail;
1487 if (target_exists) {
1488 if (S_ISDIR(new_inode->i_mode)) {
1489 if (new_inode->i_nlink != 2 ||
1490 !ocfs2_empty_dir(new_inode)) {
1491 status = -ENOTEMPTY;
1492 goto bail;
1495 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1496 newfe_bh,
1497 OCFS2_JOURNAL_ACCESS_WRITE);
1498 if (status < 0) {
1499 mlog_errno(status);
1500 goto bail;
1503 /* change the dirent to point to the correct inode */
1504 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1505 old_inode);
1506 if (status < 0) {
1507 mlog_errno(status);
1508 goto bail;
1510 inode_inc_iversion(new_dir);
1512 if (S_ISDIR(new_inode->i_mode))
1513 ocfs2_set_links_count(newfe, 0);
1514 else
1515 ocfs2_add_links_count(newfe, -1);
1516 ocfs2_journal_dirty(handle, newfe_bh);
1517 if (should_add_orphan) {
1518 status = ocfs2_orphan_add(osb, handle, new_inode,
1519 newfe_bh, orphan_name,
1520 &orphan_insert, orphan_dir, false);
1521 if (status < 0) {
1522 mlog_errno(status);
1523 goto bail;
1526 } else {
1527 /* if the name was not found in new_dir, add it now */
1528 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1529 OCFS2_I(old_inode)->ip_blkno,
1530 new_dir_bh, &target_insert);
1533 old_inode->i_ctime = current_time(old_inode);
1534 mark_inode_dirty(old_inode);
1536 status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1537 old_inode_bh,
1538 OCFS2_JOURNAL_ACCESS_WRITE);
1539 if (status >= 0) {
1540 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1542 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1543 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1544 ocfs2_journal_dirty(handle, old_inode_bh);
1545 } else
1546 mlog_errno(status);
1549 * Now that the name has been added to new_dir, remove the old name.
1551 * We don't keep any directory entry context around until now
1552 * because the insert might have changed the type of directory
1553 * we're dealing with.
1555 status = ocfs2_find_entry(old_dentry->d_name.name,
1556 old_dentry->d_name.len, old_dir,
1557 &old_entry_lookup);
1558 if (status) {
1559 if (!is_journal_aborted(osb->journal->j_journal)) {
1560 ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1561 "is not deleted.",
1562 new_dentry->d_name.len, new_dentry->d_name.name,
1563 old_dentry->d_name.len, old_dentry->d_name.name);
1565 goto bail;
1568 status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1569 if (status < 0) {
1570 mlog_errno(status);
1571 if (!is_journal_aborted(osb->journal->j_journal)) {
1572 ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1573 "is not deleted.",
1574 new_dentry->d_name.len, new_dentry->d_name.name,
1575 old_dentry->d_name.len, old_dentry->d_name.name);
1577 goto bail;
1580 if (new_inode) {
1581 drop_nlink(new_inode);
1582 new_inode->i_ctime = current_time(new_inode);
1584 old_dir->i_ctime = old_dir->i_mtime = current_time(old_dir);
1586 if (update_dot_dot) {
1587 status = ocfs2_update_entry(old_inode, handle,
1588 &old_inode_dot_dot_res, new_dir);
1589 drop_nlink(old_dir);
1590 if (new_inode) {
1591 drop_nlink(new_inode);
1592 } else {
1593 inc_nlink(new_dir);
1594 mark_inode_dirty(new_dir);
1597 mark_inode_dirty(old_dir);
1598 ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1599 if (new_inode) {
1600 mark_inode_dirty(new_inode);
1601 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1604 if (old_dir != new_dir) {
1605 /* Keep the same times on both directories.*/
1606 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1609 * This will also pick up the i_nlink change from the
1610 * block above.
1612 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1615 if (old_dir_nlink != old_dir->i_nlink) {
1616 if (!old_dir_bh) {
1617 mlog(ML_ERROR, "need to change nlink for old dir "
1618 "%llu from %d to %d but bh is NULL!\n",
1619 (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1620 (int)old_dir_nlink, old_dir->i_nlink);
1621 } else {
1622 struct ocfs2_dinode *fe;
1623 status = ocfs2_journal_access_di(handle,
1624 INODE_CACHE(old_dir),
1625 old_dir_bh,
1626 OCFS2_JOURNAL_ACCESS_WRITE);
1627 fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1628 ocfs2_set_links_count(fe, old_dir->i_nlink);
1629 ocfs2_journal_dirty(handle, old_dir_bh);
1632 ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1633 status = 0;
1634 bail:
1635 if (handle)
1636 ocfs2_commit_trans(osb, handle);
1638 if (orphan_dir) {
1639 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1640 ocfs2_inode_unlock(orphan_dir, 1);
1641 inode_unlock(orphan_dir);
1642 iput(orphan_dir);
1645 if (new_child_locked)
1646 ocfs2_inode_unlock(new_inode, 1);
1648 if (old_child_locked)
1649 ocfs2_inode_unlock(old_inode, 1);
1651 if (parents_locked)
1652 ocfs2_double_unlock(old_dir, new_dir);
1654 if (rename_lock)
1655 ocfs2_rename_unlock(osb);
1657 if (new_inode)
1658 sync_mapping_buffers(old_inode->i_mapping);
1660 iput(new_inode);
1662 ocfs2_free_dir_lookup_result(&target_lookup_res);
1663 ocfs2_free_dir_lookup_result(&old_entry_lookup);
1664 ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1665 ocfs2_free_dir_lookup_result(&orphan_insert);
1666 ocfs2_free_dir_lookup_result(&target_insert);
1668 brelse(newfe_bh);
1669 brelse(old_inode_bh);
1670 brelse(old_dir_bh);
1671 brelse(new_dir_bh);
1673 if (status)
1674 mlog_errno(status);
1676 return status;
1680 * we expect i_size = strlen(symname). Copy symname into the file
1681 * data, including the null terminator.
1683 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1684 handle_t *handle,
1685 struct inode *inode,
1686 const char *symname)
1688 struct buffer_head **bhs = NULL;
1689 const char *c;
1690 struct super_block *sb = osb->sb;
1691 u64 p_blkno, p_blocks;
1692 int virtual, blocks, status, i, bytes_left;
1694 bytes_left = i_size_read(inode) + 1;
1695 /* we can't trust i_blocks because we're actually going to
1696 * write i_size + 1 bytes. */
1697 blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1699 trace_ocfs2_create_symlink_data((unsigned long long)inode->i_blocks,
1700 i_size_read(inode), blocks);
1702 /* Sanity check -- make sure we're going to fit. */
1703 if (bytes_left >
1704 ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1705 status = -EIO;
1706 mlog_errno(status);
1707 goto bail;
1710 bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1711 if (!bhs) {
1712 status = -ENOMEM;
1713 mlog_errno(status);
1714 goto bail;
1717 status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1718 NULL);
1719 if (status < 0) {
1720 mlog_errno(status);
1721 goto bail;
1724 /* links can never be larger than one cluster so we know this
1725 * is all going to be contiguous, but do a sanity check
1726 * anyway. */
1727 if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1728 status = -EIO;
1729 mlog_errno(status);
1730 goto bail;
1733 virtual = 0;
1734 while(bytes_left > 0) {
1735 c = &symname[virtual * sb->s_blocksize];
1737 bhs[virtual] = sb_getblk(sb, p_blkno);
1738 if (!bhs[virtual]) {
1739 status = -ENOMEM;
1740 mlog_errno(status);
1741 goto bail;
1743 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1744 bhs[virtual]);
1746 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1747 bhs[virtual],
1748 OCFS2_JOURNAL_ACCESS_CREATE);
1749 if (status < 0) {
1750 mlog_errno(status);
1751 goto bail;
1754 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1756 memcpy(bhs[virtual]->b_data, c,
1757 (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1758 bytes_left);
1760 ocfs2_journal_dirty(handle, bhs[virtual]);
1762 virtual++;
1763 p_blkno++;
1764 bytes_left -= sb->s_blocksize;
1767 status = 0;
1768 bail:
1770 if (bhs) {
1771 for(i = 0; i < blocks; i++)
1772 brelse(bhs[i]);
1773 kfree(bhs);
1776 if (status)
1777 mlog_errno(status);
1778 return status;
1781 static int ocfs2_symlink(struct inode *dir,
1782 struct dentry *dentry,
1783 const char *symname)
1785 int status, l, credits;
1786 u64 newsize;
1787 struct ocfs2_super *osb = NULL;
1788 struct inode *inode = NULL;
1789 struct super_block *sb;
1790 struct buffer_head *new_fe_bh = NULL;
1791 struct buffer_head *parent_fe_bh = NULL;
1792 struct ocfs2_dinode *fe = NULL;
1793 struct ocfs2_dinode *dirfe;
1794 handle_t *handle = NULL;
1795 struct ocfs2_alloc_context *inode_ac = NULL;
1796 struct ocfs2_alloc_context *data_ac = NULL;
1797 struct ocfs2_alloc_context *xattr_ac = NULL;
1798 int want_clusters = 0;
1799 int xattr_credits = 0;
1800 struct ocfs2_security_xattr_info si = {
1801 .enable = 1,
1803 int did_quota = 0, did_quota_inode = 0;
1804 struct ocfs2_dir_lookup_result lookup = { NULL, };
1805 sigset_t oldset;
1806 int did_block_signals = 0;
1807 struct ocfs2_dentry_lock *dl = NULL;
1809 trace_ocfs2_symlink_begin(dir, dentry, symname,
1810 dentry->d_name.len, dentry->d_name.name);
1812 status = dquot_initialize(dir);
1813 if (status) {
1814 mlog_errno(status);
1815 goto bail;
1818 sb = dir->i_sb;
1819 osb = OCFS2_SB(sb);
1821 l = strlen(symname) + 1;
1823 credits = ocfs2_calc_symlink_credits(sb);
1825 /* lock the parent directory */
1826 status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1827 if (status < 0) {
1828 if (status != -ENOENT)
1829 mlog_errno(status);
1830 return status;
1833 dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1834 if (!ocfs2_read_links_count(dirfe)) {
1835 /* can't make a file in a deleted directory. */
1836 status = -ENOENT;
1837 goto bail;
1840 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1841 dentry->d_name.len);
1842 if (status)
1843 goto bail;
1845 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1846 dentry->d_name.name,
1847 dentry->d_name.len, &lookup);
1848 if (status < 0) {
1849 mlog_errno(status);
1850 goto bail;
1853 status = ocfs2_reserve_new_inode(osb, &inode_ac);
1854 if (status < 0) {
1855 if (status != -ENOSPC)
1856 mlog_errno(status);
1857 goto bail;
1860 inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1861 if (IS_ERR(inode)) {
1862 status = PTR_ERR(inode);
1863 inode = NULL;
1864 mlog_errno(status);
1865 goto bail;
1868 /* get security xattr */
1869 status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
1870 if (status) {
1871 if (status == -EOPNOTSUPP)
1872 si.enable = 0;
1873 else {
1874 mlog_errno(status);
1875 goto bail;
1879 /* calculate meta data/clusters for setting security xattr */
1880 if (si.enable) {
1881 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1882 &xattr_credits, &xattr_ac);
1883 if (status < 0) {
1884 mlog_errno(status);
1885 goto bail;
1889 /* don't reserve bitmap space for fast symlinks. */
1890 if (l > ocfs2_fast_symlink_chars(sb))
1891 want_clusters += 1;
1893 status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1894 if (status < 0) {
1895 if (status != -ENOSPC)
1896 mlog_errno(status);
1897 goto bail;
1900 handle = ocfs2_start_trans(osb, credits + xattr_credits);
1901 if (IS_ERR(handle)) {
1902 status = PTR_ERR(handle);
1903 handle = NULL;
1904 mlog_errno(status);
1905 goto bail;
1908 /* Starting to change things, restart is no longer possible. */
1909 ocfs2_block_signals(&oldset);
1910 did_block_signals = 1;
1912 status = dquot_alloc_inode(inode);
1913 if (status)
1914 goto bail;
1915 did_quota_inode = 1;
1917 trace_ocfs2_symlink_create(dir, dentry, dentry->d_name.len,
1918 dentry->d_name.name,
1919 (unsigned long long)OCFS2_I(dir)->ip_blkno,
1920 inode->i_mode);
1922 status = ocfs2_mknod_locked(osb, dir, inode,
1923 0, &new_fe_bh, parent_fe_bh, handle,
1924 inode_ac);
1925 if (status < 0) {
1926 mlog_errno(status);
1927 goto bail;
1930 fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1931 inode->i_rdev = 0;
1932 newsize = l - 1;
1933 inode->i_op = &ocfs2_symlink_inode_operations;
1934 inode_nohighmem(inode);
1935 if (l > ocfs2_fast_symlink_chars(sb)) {
1936 u32 offset = 0;
1938 status = dquot_alloc_space_nodirty(inode,
1939 ocfs2_clusters_to_bytes(osb->sb, 1));
1940 if (status)
1941 goto bail;
1942 did_quota = 1;
1943 inode->i_mapping->a_ops = &ocfs2_aops;
1944 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1945 new_fe_bh,
1946 handle, data_ac, NULL,
1947 NULL);
1948 if (status < 0) {
1949 if (status != -ENOSPC && status != -EINTR) {
1950 mlog(ML_ERROR,
1951 "Failed to extend file to %llu\n",
1952 (unsigned long long)newsize);
1953 mlog_errno(status);
1954 status = -ENOSPC;
1956 goto bail;
1958 i_size_write(inode, newsize);
1959 inode->i_blocks = ocfs2_inode_sector_count(inode);
1960 } else {
1961 inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
1962 memcpy((char *) fe->id2.i_symlink, symname, l);
1963 i_size_write(inode, newsize);
1964 inode->i_blocks = 0;
1967 status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1968 if (status < 0) {
1969 mlog_errno(status);
1970 goto bail;
1973 if (!ocfs2_inode_is_fast_symlink(inode)) {
1974 status = ocfs2_create_symlink_data(osb, handle, inode,
1975 symname);
1976 if (status < 0) {
1977 mlog_errno(status);
1978 goto bail;
1982 if (si.enable) {
1983 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1984 xattr_ac, data_ac);
1985 if (status < 0) {
1986 mlog_errno(status);
1987 goto bail;
1992 * Do this before adding the entry to the directory. We add
1993 * also set d_op after success so that ->d_iput() will cleanup
1994 * the dentry lock even if ocfs2_add_entry() fails below.
1996 status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1997 if (status) {
1998 mlog_errno(status);
1999 goto bail;
2002 dl = dentry->d_fsdata;
2004 status = ocfs2_add_entry(handle, dentry, inode,
2005 le64_to_cpu(fe->i_blkno), parent_fe_bh,
2006 &lookup);
2007 if (status < 0) {
2008 mlog_errno(status);
2009 goto bail;
2012 insert_inode_hash(inode);
2013 d_instantiate(dentry, inode);
2014 bail:
2015 if (status < 0 && did_quota)
2016 dquot_free_space_nodirty(inode,
2017 ocfs2_clusters_to_bytes(osb->sb, 1));
2018 if (status < 0 && did_quota_inode)
2019 dquot_free_inode(inode);
2020 if (handle)
2021 ocfs2_commit_trans(osb, handle);
2023 ocfs2_inode_unlock(dir, 1);
2024 if (did_block_signals)
2025 ocfs2_unblock_signals(&oldset);
2027 brelse(new_fe_bh);
2028 brelse(parent_fe_bh);
2029 kfree(si.value);
2030 ocfs2_free_dir_lookup_result(&lookup);
2031 if (inode_ac)
2032 ocfs2_free_alloc_context(inode_ac);
2033 if (data_ac)
2034 ocfs2_free_alloc_context(data_ac);
2035 if (xattr_ac)
2036 ocfs2_free_alloc_context(xattr_ac);
2037 if ((status < 0) && inode) {
2038 if (dl)
2039 ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
2041 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
2042 clear_nlink(inode);
2043 iput(inode);
2046 if (status)
2047 mlog_errno(status);
2049 return status;
2052 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2054 int status, namelen;
2056 namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2057 (long long)blkno);
2058 if (namelen <= 0) {
2059 if (namelen)
2060 status = namelen;
2061 else
2062 status = -EINVAL;
2063 mlog_errno(status);
2064 goto bail;
2066 if (namelen != OCFS2_ORPHAN_NAMELEN) {
2067 status = -EINVAL;
2068 mlog_errno(status);
2069 goto bail;
2072 trace_ocfs2_blkno_stringify(blkno, name, namelen);
2074 status = 0;
2075 bail:
2076 if (status < 0)
2077 mlog_errno(status);
2078 return status;
2081 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
2082 struct inode **ret_orphan_dir,
2083 struct buffer_head **ret_orphan_dir_bh)
2085 struct inode *orphan_dir_inode;
2086 struct buffer_head *orphan_dir_bh = NULL;
2087 int ret = 0;
2089 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2090 ORPHAN_DIR_SYSTEM_INODE,
2091 osb->slot_num);
2092 if (!orphan_dir_inode) {
2093 ret = -ENOENT;
2094 mlog_errno(ret);
2095 return ret;
2098 inode_lock(orphan_dir_inode);
2100 ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2101 if (ret < 0) {
2102 inode_unlock(orphan_dir_inode);
2103 iput(orphan_dir_inode);
2105 mlog_errno(ret);
2106 return ret;
2109 *ret_orphan_dir = orphan_dir_inode;
2110 *ret_orphan_dir_bh = orphan_dir_bh;
2112 return 0;
2115 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
2116 struct buffer_head *orphan_dir_bh,
2117 u64 blkno,
2118 char *name,
2119 struct ocfs2_dir_lookup_result *lookup,
2120 bool dio)
2122 int ret;
2123 struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
2124 int namelen = dio ?
2125 (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2126 OCFS2_ORPHAN_NAMELEN;
2128 if (dio) {
2129 ret = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2130 OCFS2_DIO_ORPHAN_PREFIX);
2131 if (ret != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2132 ret = -EINVAL;
2133 mlog_errno(ret);
2134 return ret;
2137 ret = ocfs2_blkno_stringify(blkno,
2138 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2139 } else
2140 ret = ocfs2_blkno_stringify(blkno, name);
2141 if (ret < 0) {
2142 mlog_errno(ret);
2143 return ret;
2146 ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2147 orphan_dir_bh, name,
2148 namelen, lookup);
2149 if (ret < 0) {
2150 mlog_errno(ret);
2151 return ret;
2154 return 0;
2158 * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
2159 * insertion of an orphan.
2160 * @osb: ocfs2 file system
2161 * @ret_orphan_dir: Orphan dir inode - returned locked!
2162 * @blkno: Actual block number of the inode to be inserted into orphan dir.
2163 * @lookup: dir lookup result, to be passed back into functions like
2164 * ocfs2_orphan_add
2166 * Returns zero on success and the ret_orphan_dir, name and lookup
2167 * fields will be populated.
2169 * Returns non-zero on failure.
2171 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2172 struct inode **ret_orphan_dir,
2173 u64 blkno,
2174 char *name,
2175 struct ocfs2_dir_lookup_result *lookup,
2176 bool dio)
2178 struct inode *orphan_dir_inode = NULL;
2179 struct buffer_head *orphan_dir_bh = NULL;
2180 int ret = 0;
2182 ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
2183 &orphan_dir_bh);
2184 if (ret < 0) {
2185 mlog_errno(ret);
2186 return ret;
2189 ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
2190 blkno, name, lookup, dio);
2191 if (ret < 0) {
2192 mlog_errno(ret);
2193 goto out;
2196 *ret_orphan_dir = orphan_dir_inode;
2198 out:
2199 brelse(orphan_dir_bh);
2201 if (ret) {
2202 ocfs2_inode_unlock(orphan_dir_inode, 1);
2203 inode_unlock(orphan_dir_inode);
2204 iput(orphan_dir_inode);
2207 if (ret)
2208 mlog_errno(ret);
2209 return ret;
2212 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2213 handle_t *handle,
2214 struct inode *inode,
2215 struct buffer_head *fe_bh,
2216 char *name,
2217 struct ocfs2_dir_lookup_result *lookup,
2218 struct inode *orphan_dir_inode,
2219 bool dio)
2221 struct buffer_head *orphan_dir_bh = NULL;
2222 int status = 0;
2223 struct ocfs2_dinode *orphan_fe;
2224 struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
2225 int namelen = dio ?
2226 (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2227 OCFS2_ORPHAN_NAMELEN;
2229 trace_ocfs2_orphan_add_begin(
2230 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2232 status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2233 if (status < 0) {
2234 mlog_errno(status);
2235 goto leave;
2238 status = ocfs2_journal_access_di(handle,
2239 INODE_CACHE(orphan_dir_inode),
2240 orphan_dir_bh,
2241 OCFS2_JOURNAL_ACCESS_WRITE);
2242 if (status < 0) {
2243 mlog_errno(status);
2244 goto leave;
2248 * We're going to journal the change of i_flags and i_orphaned_slot.
2249 * It's safe anyway, though some callers may duplicate the journaling.
2250 * Journaling within the func just make the logic look more
2251 * straightforward.
2253 status = ocfs2_journal_access_di(handle,
2254 INODE_CACHE(inode),
2255 fe_bh,
2256 OCFS2_JOURNAL_ACCESS_WRITE);
2257 if (status < 0) {
2258 mlog_errno(status);
2259 goto leave;
2262 /* we're a cluster, and nlink can change on disk from
2263 * underneath us... */
2264 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2265 if (S_ISDIR(inode->i_mode))
2266 ocfs2_add_links_count(orphan_fe, 1);
2267 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2268 ocfs2_journal_dirty(handle, orphan_dir_bh);
2270 status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2271 namelen, inode,
2272 OCFS2_I(inode)->ip_blkno,
2273 orphan_dir_bh, lookup);
2274 if (status < 0) {
2275 mlog_errno(status);
2276 goto rollback;
2279 if (dio) {
2280 /* Update flag OCFS2_DIO_ORPHANED_FL and record the orphan
2281 * slot.
2283 fe->i_flags |= cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2284 fe->i_dio_orphaned_slot = cpu_to_le16(osb->slot_num);
2285 } else {
2286 fe->i_flags |= cpu_to_le32(OCFS2_ORPHANED_FL);
2287 OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2289 /* Record which orphan dir our inode now resides
2290 * in. delete_inode will use this to determine which orphan
2291 * dir to lock. */
2292 fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2295 ocfs2_journal_dirty(handle, fe_bh);
2297 trace_ocfs2_orphan_add_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
2298 osb->slot_num);
2300 rollback:
2301 if (status < 0) {
2302 if (S_ISDIR(inode->i_mode))
2303 ocfs2_add_links_count(orphan_fe, -1);
2304 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2307 leave:
2308 brelse(orphan_dir_bh);
2310 return status;
2313 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2314 int ocfs2_orphan_del(struct ocfs2_super *osb,
2315 handle_t *handle,
2316 struct inode *orphan_dir_inode,
2317 struct inode *inode,
2318 struct buffer_head *orphan_dir_bh,
2319 bool dio)
2321 char name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2322 struct ocfs2_dinode *orphan_fe;
2323 int status = 0;
2324 struct ocfs2_dir_lookup_result lookup = { NULL, };
2326 if (dio) {
2327 status = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2328 OCFS2_DIO_ORPHAN_PREFIX);
2329 if (status != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2330 status = -EINVAL;
2331 mlog_errno(status);
2332 return status;
2335 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno,
2336 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2337 } else
2338 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2339 if (status < 0) {
2340 mlog_errno(status);
2341 goto leave;
2344 trace_ocfs2_orphan_del(
2345 (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2346 name, strlen(name));
2348 status = ocfs2_journal_access_di(handle,
2349 INODE_CACHE(orphan_dir_inode),
2350 orphan_dir_bh,
2351 OCFS2_JOURNAL_ACCESS_WRITE);
2352 if (status < 0) {
2353 mlog_errno(status);
2354 goto leave;
2357 /* find it's spot in the orphan directory */
2358 status = ocfs2_find_entry(name, strlen(name), orphan_dir_inode,
2359 &lookup);
2360 if (status) {
2361 mlog_errno(status);
2362 goto leave;
2365 /* remove it from the orphan directory */
2366 status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2367 if (status < 0) {
2368 mlog_errno(status);
2369 goto leave;
2372 /* do the i_nlink dance! :) */
2373 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2374 if (S_ISDIR(inode->i_mode))
2375 ocfs2_add_links_count(orphan_fe, -1);
2376 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2377 ocfs2_journal_dirty(handle, orphan_dir_bh);
2379 leave:
2380 ocfs2_free_dir_lookup_result(&lookup);
2382 if (status)
2383 mlog_errno(status);
2384 return status;
2388 * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to receive a newly
2389 * allocated file. This is different from the typical 'add to orphan dir'
2390 * operation in that the inode does not yet exist. This is a problem because
2391 * the orphan dir stringifies the inode block number to come up with it's
2392 * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2393 * problem. This function works around it by calling deeper into the orphan
2394 * and suballoc code than other callers. Use this only by necessity.
2395 * @dir: The directory which this inode will ultimately wind up under - not the
2396 * orphan dir!
2397 * @dir_bh: buffer_head the @dir inode block
2398 * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2399 * with the string to be used for orphan dirent. Pass back to the orphan dir
2400 * code.
2401 * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2402 * dir code.
2403 * @ret_di_blkno: block number where the new inode will be allocated.
2404 * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2405 * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2407 * Returns zero on success and the ret_orphan_dir, name and lookup
2408 * fields will be populated.
2410 * Returns non-zero on failure.
2412 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2413 struct buffer_head *dir_bh,
2414 char *orphan_name,
2415 struct inode **ret_orphan_dir,
2416 u64 *ret_di_blkno,
2417 struct ocfs2_dir_lookup_result *orphan_insert,
2418 struct ocfs2_alloc_context **ret_inode_ac)
2420 int ret;
2421 u64 di_blkno;
2422 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2423 struct inode *orphan_dir = NULL;
2424 struct buffer_head *orphan_dir_bh = NULL;
2425 struct ocfs2_alloc_context *inode_ac = NULL;
2427 ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2428 if (ret < 0) {
2429 mlog_errno(ret);
2430 return ret;
2433 /* reserve an inode spot */
2434 ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2435 if (ret < 0) {
2436 if (ret != -ENOSPC)
2437 mlog_errno(ret);
2438 goto out;
2441 ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2442 &di_blkno);
2443 if (ret) {
2444 mlog_errno(ret);
2445 goto out;
2448 ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2449 di_blkno, orphan_name, orphan_insert,
2450 false);
2451 if (ret < 0) {
2452 mlog_errno(ret);
2453 goto out;
2456 out:
2457 if (ret == 0) {
2458 *ret_orphan_dir = orphan_dir;
2459 *ret_di_blkno = di_blkno;
2460 *ret_inode_ac = inode_ac;
2462 * orphan_name and orphan_insert are already up to
2463 * date via prepare_orphan_dir
2465 } else {
2466 /* Unroll reserve_new_inode* */
2467 if (inode_ac)
2468 ocfs2_free_alloc_context(inode_ac);
2470 /* Unroll orphan dir locking */
2471 inode_unlock(orphan_dir);
2472 ocfs2_inode_unlock(orphan_dir, 1);
2473 iput(orphan_dir);
2476 brelse(orphan_dir_bh);
2478 return ret;
2481 int ocfs2_create_inode_in_orphan(struct inode *dir,
2482 int mode,
2483 struct inode **new_inode)
2485 int status, did_quota_inode = 0;
2486 struct inode *inode = NULL;
2487 struct inode *orphan_dir = NULL;
2488 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2489 handle_t *handle = NULL;
2490 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2491 struct buffer_head *parent_di_bh = NULL;
2492 struct buffer_head *new_di_bh = NULL;
2493 struct ocfs2_alloc_context *inode_ac = NULL;
2494 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2495 u64 uninitialized_var(di_blkno), suballoc_loc;
2496 u16 suballoc_bit;
2498 status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2499 if (status < 0) {
2500 if (status != -ENOENT)
2501 mlog_errno(status);
2502 return status;
2505 status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2506 orphan_name, &orphan_dir,
2507 &di_blkno, &orphan_insert, &inode_ac);
2508 if (status < 0) {
2509 if (status != -ENOSPC)
2510 mlog_errno(status);
2511 goto leave;
2514 inode = ocfs2_get_init_inode(dir, mode);
2515 if (IS_ERR(inode)) {
2516 status = PTR_ERR(inode);
2517 inode = NULL;
2518 mlog_errno(status);
2519 goto leave;
2522 handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2523 if (IS_ERR(handle)) {
2524 status = PTR_ERR(handle);
2525 handle = NULL;
2526 mlog_errno(status);
2527 goto leave;
2530 status = dquot_alloc_inode(inode);
2531 if (status)
2532 goto leave;
2533 did_quota_inode = 1;
2535 status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2536 &suballoc_loc,
2537 &suballoc_bit, di_blkno);
2538 if (status < 0) {
2539 mlog_errno(status);
2540 goto leave;
2543 clear_nlink(inode);
2544 /* do the real work now. */
2545 status = __ocfs2_mknod_locked(dir, inode,
2546 0, &new_di_bh, parent_di_bh, handle,
2547 inode_ac, di_blkno, suballoc_loc,
2548 suballoc_bit);
2549 if (status < 0) {
2550 mlog_errno(status);
2551 goto leave;
2554 status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2555 &orphan_insert, orphan_dir, false);
2556 if (status < 0) {
2557 mlog_errno(status);
2558 goto leave;
2561 /* get open lock so that only nodes can't remove it from orphan dir. */
2562 status = ocfs2_open_lock(inode);
2563 if (status < 0)
2564 mlog_errno(status);
2566 insert_inode_hash(inode);
2567 leave:
2568 if (status < 0 && did_quota_inode)
2569 dquot_free_inode(inode);
2570 if (handle)
2571 ocfs2_commit_trans(osb, handle);
2573 if (orphan_dir) {
2574 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2575 ocfs2_inode_unlock(orphan_dir, 1);
2576 inode_unlock(orphan_dir);
2577 iput(orphan_dir);
2580 if ((status < 0) && inode) {
2581 clear_nlink(inode);
2582 iput(inode);
2585 if (inode_ac)
2586 ocfs2_free_alloc_context(inode_ac);
2588 brelse(new_di_bh);
2590 if (!status)
2591 *new_inode = inode;
2593 ocfs2_free_dir_lookup_result(&orphan_insert);
2595 ocfs2_inode_unlock(dir, 1);
2596 brelse(parent_di_bh);
2597 return status;
2600 int ocfs2_add_inode_to_orphan(struct ocfs2_super *osb,
2601 struct inode *inode)
2603 char orphan_name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2604 struct inode *orphan_dir_inode = NULL;
2605 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2606 struct buffer_head *di_bh = NULL;
2607 int status = 0;
2608 handle_t *handle = NULL;
2609 struct ocfs2_dinode *di = NULL;
2611 status = ocfs2_inode_lock(inode, &di_bh, 1);
2612 if (status < 0) {
2613 mlog_errno(status);
2614 goto bail;
2617 di = (struct ocfs2_dinode *) di_bh->b_data;
2619 * Another append dio crashed?
2620 * If so, manually recover it first.
2622 if (unlikely(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL))) {
2623 status = ocfs2_truncate_file(inode, di_bh, i_size_read(inode));
2624 if (status < 0) {
2625 if (status != -ENOSPC)
2626 mlog_errno(status);
2627 goto bail_unlock_inode;
2630 status = ocfs2_del_inode_from_orphan(osb, inode, di_bh, 0, 0);
2631 if (status < 0) {
2632 mlog_errno(status);
2633 goto bail_unlock_inode;
2637 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir_inode,
2638 OCFS2_I(inode)->ip_blkno,
2639 orphan_name,
2640 &orphan_insert,
2641 true);
2642 if (status < 0) {
2643 mlog_errno(status);
2644 goto bail_unlock_inode;
2647 handle = ocfs2_start_trans(osb,
2648 OCFS2_INODE_ADD_TO_ORPHAN_CREDITS);
2649 if (IS_ERR(handle)) {
2650 status = PTR_ERR(handle);
2651 goto bail_unlock_orphan;
2654 status = ocfs2_orphan_add(osb, handle, inode, di_bh, orphan_name,
2655 &orphan_insert, orphan_dir_inode, true);
2656 if (status)
2657 mlog_errno(status);
2659 ocfs2_commit_trans(osb, handle);
2661 bail_unlock_orphan:
2662 ocfs2_inode_unlock(orphan_dir_inode, 1);
2663 inode_unlock(orphan_dir_inode);
2664 iput(orphan_dir_inode);
2666 ocfs2_free_dir_lookup_result(&orphan_insert);
2668 bail_unlock_inode:
2669 ocfs2_inode_unlock(inode, 1);
2670 brelse(di_bh);
2672 bail:
2673 return status;
2676 int ocfs2_del_inode_from_orphan(struct ocfs2_super *osb,
2677 struct inode *inode, struct buffer_head *di_bh,
2678 int update_isize, loff_t end)
2680 struct inode *orphan_dir_inode = NULL;
2681 struct buffer_head *orphan_dir_bh = NULL;
2682 struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2683 handle_t *handle = NULL;
2684 int status = 0;
2686 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2687 ORPHAN_DIR_SYSTEM_INODE,
2688 le16_to_cpu(di->i_dio_orphaned_slot));
2689 if (!orphan_dir_inode) {
2690 status = -ENOENT;
2691 mlog_errno(status);
2692 goto bail;
2695 inode_lock(orphan_dir_inode);
2696 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2697 if (status < 0) {
2698 inode_unlock(orphan_dir_inode);
2699 iput(orphan_dir_inode);
2700 mlog_errno(status);
2701 goto bail;
2704 handle = ocfs2_start_trans(osb,
2705 OCFS2_INODE_DEL_FROM_ORPHAN_CREDITS);
2706 if (IS_ERR(handle)) {
2707 status = PTR_ERR(handle);
2708 goto bail_unlock_orphan;
2711 BUG_ON(!(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)));
2713 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode,
2714 inode, orphan_dir_bh, true);
2715 if (status < 0) {
2716 mlog_errno(status);
2717 goto bail_commit;
2720 status = ocfs2_journal_access_di(handle,
2721 INODE_CACHE(inode),
2722 di_bh,
2723 OCFS2_JOURNAL_ACCESS_WRITE);
2724 if (status < 0) {
2725 mlog_errno(status);
2726 goto bail_commit;
2729 di->i_flags &= ~cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2730 di->i_dio_orphaned_slot = 0;
2732 if (update_isize) {
2733 status = ocfs2_set_inode_size(handle, inode, di_bh, end);
2734 if (status)
2735 mlog_errno(status);
2736 } else
2737 ocfs2_journal_dirty(handle, di_bh);
2739 bail_commit:
2740 ocfs2_commit_trans(osb, handle);
2742 bail_unlock_orphan:
2743 ocfs2_inode_unlock(orphan_dir_inode, 1);
2744 inode_unlock(orphan_dir_inode);
2745 brelse(orphan_dir_bh);
2746 iput(orphan_dir_inode);
2748 bail:
2749 return status;
2752 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2753 struct inode *inode,
2754 struct dentry *dentry)
2756 int status = 0;
2757 struct buffer_head *parent_di_bh = NULL;
2758 handle_t *handle = NULL;
2759 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2760 struct ocfs2_dinode *dir_di, *di;
2761 struct inode *orphan_dir_inode = NULL;
2762 struct buffer_head *orphan_dir_bh = NULL;
2763 struct buffer_head *di_bh = NULL;
2764 struct ocfs2_dir_lookup_result lookup = { NULL, };
2766 trace_ocfs2_mv_orphaned_inode_to_new(dir, dentry,
2767 dentry->d_name.len, dentry->d_name.name,
2768 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2769 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2771 status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2772 if (status < 0) {
2773 if (status != -ENOENT)
2774 mlog_errno(status);
2775 return status;
2778 dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2779 if (!dir_di->i_links_count) {
2780 /* can't make a file in a deleted directory. */
2781 status = -ENOENT;
2782 goto leave;
2785 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2786 dentry->d_name.len);
2787 if (status)
2788 goto leave;
2790 /* get a spot inside the dir. */
2791 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2792 dentry->d_name.name,
2793 dentry->d_name.len, &lookup);
2794 if (status < 0) {
2795 mlog_errno(status);
2796 goto leave;
2799 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2800 ORPHAN_DIR_SYSTEM_INODE,
2801 osb->slot_num);
2802 if (!orphan_dir_inode) {
2803 status = -ENOENT;
2804 mlog_errno(status);
2805 goto leave;
2808 inode_lock(orphan_dir_inode);
2810 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2811 if (status < 0) {
2812 mlog_errno(status);
2813 inode_unlock(orphan_dir_inode);
2814 iput(orphan_dir_inode);
2815 goto leave;
2818 status = ocfs2_read_inode_block(inode, &di_bh);
2819 if (status < 0) {
2820 mlog_errno(status);
2821 goto orphan_unlock;
2824 handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2825 if (IS_ERR(handle)) {
2826 status = PTR_ERR(handle);
2827 handle = NULL;
2828 mlog_errno(status);
2829 goto orphan_unlock;
2832 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2833 di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2834 if (status < 0) {
2835 mlog_errno(status);
2836 goto out_commit;
2839 status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2840 orphan_dir_bh, false);
2841 if (status < 0) {
2842 mlog_errno(status);
2843 goto out_commit;
2846 di = (struct ocfs2_dinode *)di_bh->b_data;
2847 di->i_flags &= ~cpu_to_le32(OCFS2_ORPHANED_FL);
2848 di->i_orphaned_slot = 0;
2849 set_nlink(inode, 1);
2850 ocfs2_set_links_count(di, inode->i_nlink);
2851 ocfs2_update_inode_fsync_trans(handle, inode, 1);
2852 ocfs2_journal_dirty(handle, di_bh);
2854 status = ocfs2_add_entry(handle, dentry, inode,
2855 OCFS2_I(inode)->ip_blkno, parent_di_bh,
2856 &lookup);
2857 if (status < 0) {
2858 mlog_errno(status);
2859 goto out_commit;
2862 status = ocfs2_dentry_attach_lock(dentry, inode,
2863 OCFS2_I(dir)->ip_blkno);
2864 if (status) {
2865 mlog_errno(status);
2866 goto out_commit;
2869 d_instantiate(dentry, inode);
2870 status = 0;
2871 out_commit:
2872 ocfs2_commit_trans(osb, handle);
2873 orphan_unlock:
2874 ocfs2_inode_unlock(orphan_dir_inode, 1);
2875 inode_unlock(orphan_dir_inode);
2876 iput(orphan_dir_inode);
2877 leave:
2879 ocfs2_inode_unlock(dir, 1);
2881 brelse(di_bh);
2882 brelse(parent_di_bh);
2883 brelse(orphan_dir_bh);
2885 ocfs2_free_dir_lookup_result(&lookup);
2887 if (status)
2888 mlog_errno(status);
2890 return status;
2893 const struct inode_operations ocfs2_dir_iops = {
2894 .create = ocfs2_create,
2895 .lookup = ocfs2_lookup,
2896 .link = ocfs2_link,
2897 .unlink = ocfs2_unlink,
2898 .rmdir = ocfs2_unlink,
2899 .symlink = ocfs2_symlink,
2900 .mkdir = ocfs2_mkdir,
2901 .mknod = ocfs2_mknod,
2902 .rename = ocfs2_rename,
2903 .setattr = ocfs2_setattr,
2904 .getattr = ocfs2_getattr,
2905 .permission = ocfs2_permission,
2906 .listxattr = ocfs2_listxattr,
2907 .fiemap = ocfs2_fiemap,
2908 .get_acl = ocfs2_iop_get_acl,
2909 .set_acl = ocfs2_iop_set_acl,