jbd: Journal block numbers can ever be only 32-bit use unsigned int for them
[linux/fpc-iii.git] / fs / ocfs2 / namei.c
blob8601f934010b80e37c528b326e9121c72213ea7b
1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
4 * namei.c
6 * Create and rename file, directory, symlinks
8 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
10 * Portions of this code from linux/fs/ext3/dir.c
12 * Copyright (C) 1992, 1993, 1994, 1995
13 * Remy Card (card@masi.ibp.fr)
14 * Laboratoire MASI - Institut Blaise pascal
15 * Universite Pierre et Marie Curie (Paris VI)
17 * from
19 * linux/fs/minix/dir.c
21 * Copyright (C) 1991, 1992 Linux Torvalds
23 * This program is free software; you can redistribute it and/or
24 * modify it under the terms of the GNU General Public
25 * License as published by the Free Software Foundation; either
26 * version 2 of the License, or (at your option) any later version.
28 * This program is distributed in the hope that it will be useful,
29 * but WITHOUT ANY WARRANTY; without even the implied warranty of
30 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
31 * General Public License for more details.
33 * You should have received a copy of the GNU General Public
34 * License along with this program; if not, write to the
35 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
36 * Boston, MA 021110-1307, USA.
39 #include <linux/fs.h>
40 #include <linux/types.h>
41 #include <linux/slab.h>
42 #include <linux/highmem.h>
43 #include <linux/quotaops.h>
45 #define MLOG_MASK_PREFIX ML_NAMEI
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"
67 #include "buffer_head_io.h"
69 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
70 struct inode *dir,
71 struct inode *inode,
72 struct dentry *dentry,
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 struct inode *inode,
82 char *name,
83 struct ocfs2_dir_lookup_result *lookup);
85 static int ocfs2_orphan_add(struct ocfs2_super *osb,
86 handle_t *handle,
87 struct inode *inode,
88 struct ocfs2_dinode *fe,
89 char *name,
90 struct ocfs2_dir_lookup_result *lookup,
91 struct inode *orphan_dir_inode);
93 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
94 handle_t *handle,
95 struct inode *inode,
96 const char *symname);
98 /* An orphan dir name is an 8 byte value, printed as a hex string */
99 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
101 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
102 struct nameidata *nd)
104 int status;
105 u64 blkno;
106 struct inode *inode = NULL;
107 struct dentry *ret;
108 struct ocfs2_inode_info *oi;
110 mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
111 dentry->d_name.len, dentry->d_name.name);
113 if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
114 ret = ERR_PTR(-ENAMETOOLONG);
115 goto bail;
118 mlog(0, "find name %.*s in directory %llu\n", dentry->d_name.len,
119 dentry->d_name.name, (unsigned long long)OCFS2_I(dir)->ip_blkno);
121 status = ocfs2_inode_lock_nested(dir, NULL, 0, OI_LS_PARENT);
122 if (status < 0) {
123 if (status != -ENOENT)
124 mlog_errno(status);
125 ret = ERR_PTR(status);
126 goto bail;
129 status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
130 dentry->d_name.len, &blkno);
131 if (status < 0)
132 goto bail_add;
134 inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
135 if (IS_ERR(inode)) {
136 ret = ERR_PTR(-EACCES);
137 goto bail_unlock;
140 oi = OCFS2_I(inode);
141 /* Clear any orphaned state... If we were able to look up the
142 * inode from a directory, it certainly can't be orphaned. We
143 * might have the bad state from a node which intended to
144 * orphan this inode but crashed before it could commit the
145 * unlink. */
146 spin_lock(&oi->ip_lock);
147 oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
148 spin_unlock(&oi->ip_lock);
150 bail_add:
151 dentry->d_op = &ocfs2_dentry_ops;
152 ret = d_splice_alias(inode, dentry);
154 if (inode) {
156 * If d_splice_alias() finds a DCACHE_DISCONNECTED
157 * dentry, it will d_move() it on top of ourse. The
158 * return value will indicate this however, so in
159 * those cases, we switch them around for the locking
160 * code.
162 * NOTE: This dentry already has ->d_op set from
163 * ocfs2_get_parent() and ocfs2_get_dentry()
165 if (ret)
166 dentry = ret;
168 status = ocfs2_dentry_attach_lock(dentry, inode,
169 OCFS2_I(dir)->ip_blkno);
170 if (status) {
171 mlog_errno(status);
172 ret = ERR_PTR(status);
173 goto bail_unlock;
177 bail_unlock:
178 /* Don't drop the cluster lock until *after* the d_add --
179 * unlink on another node will message us to remove that
180 * dentry under this lock so otherwise we can race this with
181 * the downconvert thread and have a stale dentry. */
182 ocfs2_inode_unlock(dir, 0);
184 bail:
186 mlog_exit_ptr(ret);
188 return ret;
191 static struct inode *ocfs2_get_init_inode(struct inode *dir, int mode)
193 struct inode *inode;
195 inode = new_inode(dir->i_sb);
196 if (!inode) {
197 mlog(ML_ERROR, "new_inode failed!\n");
198 return NULL;
201 /* populate as many fields early on as possible - many of
202 * these are used by the support functions here and in
203 * callers. */
204 if (S_ISDIR(mode))
205 inode->i_nlink = 2;
206 else
207 inode->i_nlink = 1;
208 inode->i_uid = current_fsuid();
209 if (dir->i_mode & S_ISGID) {
210 inode->i_gid = dir->i_gid;
211 if (S_ISDIR(mode))
212 mode |= S_ISGID;
213 } else
214 inode->i_gid = current_fsgid();
215 inode->i_mode = mode;
216 vfs_dq_init(inode);
217 return inode;
220 static int ocfs2_mknod(struct inode *dir,
221 struct dentry *dentry,
222 int mode,
223 dev_t dev)
225 int status = 0;
226 struct buffer_head *parent_fe_bh = NULL;
227 handle_t *handle = NULL;
228 struct ocfs2_super *osb;
229 struct ocfs2_dinode *dirfe;
230 struct buffer_head *new_fe_bh = NULL;
231 struct inode *inode = NULL;
232 struct ocfs2_alloc_context *inode_ac = NULL;
233 struct ocfs2_alloc_context *data_ac = NULL;
234 struct ocfs2_alloc_context *meta_ac = NULL;
235 int want_clusters = 0;
236 int want_meta = 0;
237 int xattr_credits = 0;
238 struct ocfs2_security_xattr_info si = {
239 .enable = 1,
241 int did_quota_inode = 0;
242 struct ocfs2_dir_lookup_result lookup = { NULL, };
244 mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
245 (unsigned long)dev, dentry->d_name.len,
246 dentry->d_name.name);
248 /* get our super block */
249 osb = OCFS2_SB(dir->i_sb);
251 status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
252 if (status < 0) {
253 if (status != -ENOENT)
254 mlog_errno(status);
255 return status;
258 if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
259 status = -EMLINK;
260 goto leave;
263 dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
264 if (!ocfs2_read_links_count(dirfe)) {
265 /* can't make a file in a deleted directory. */
266 status = -ENOENT;
267 goto leave;
270 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
271 dentry->d_name.len);
272 if (status)
273 goto leave;
275 /* get a spot inside the dir. */
276 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
277 dentry->d_name.name,
278 dentry->d_name.len, &lookup);
279 if (status < 0) {
280 mlog_errno(status);
281 goto leave;
284 /* reserve an inode spot */
285 status = ocfs2_reserve_new_inode(osb, &inode_ac);
286 if (status < 0) {
287 if (status != -ENOSPC)
288 mlog_errno(status);
289 goto leave;
292 inode = ocfs2_get_init_inode(dir, mode);
293 if (!inode) {
294 status = -ENOMEM;
295 mlog_errno(status);
296 goto leave;
299 /* get security xattr */
300 status = ocfs2_init_security_get(inode, dir, &si);
301 if (status) {
302 if (status == -EOPNOTSUPP)
303 si.enable = 0;
304 else {
305 mlog_errno(status);
306 goto leave;
310 /* calculate meta data/clusters for setting security and acl xattr */
311 status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
312 &si, &want_clusters,
313 &xattr_credits, &want_meta);
314 if (status < 0) {
315 mlog_errno(status);
316 goto leave;
319 /* Reserve a cluster if creating an extent based directory. */
320 if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
321 want_clusters += 1;
323 /* Dir indexing requires extra space as well */
324 if (ocfs2_supports_indexed_dirs(osb))
325 want_meta++;
328 status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
329 if (status < 0) {
330 if (status != -ENOSPC)
331 mlog_errno(status);
332 goto leave;
335 status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
336 if (status < 0) {
337 if (status != -ENOSPC)
338 mlog_errno(status);
339 goto leave;
342 handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
343 S_ISDIR(mode),
344 xattr_credits));
345 if (IS_ERR(handle)) {
346 status = PTR_ERR(handle);
347 handle = NULL;
348 mlog_errno(status);
349 goto leave;
352 /* We don't use standard VFS wrapper because we don't want vfs_dq_init
353 * to be called. */
354 if (sb_any_quota_active(osb->sb) &&
355 osb->sb->dq_op->alloc_inode(inode, 1) == NO_QUOTA) {
356 status = -EDQUOT;
357 goto leave;
359 did_quota_inode = 1;
361 /* do the real work now. */
362 status = ocfs2_mknod_locked(osb, dir, inode, dentry, dev,
363 &new_fe_bh, parent_fe_bh, handle,
364 inode_ac);
365 if (status < 0) {
366 mlog_errno(status);
367 goto leave;
370 if (S_ISDIR(mode)) {
371 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
372 new_fe_bh, data_ac, meta_ac);
373 if (status < 0) {
374 mlog_errno(status);
375 goto leave;
378 status = ocfs2_journal_access_di(handle, dir, parent_fe_bh,
379 OCFS2_JOURNAL_ACCESS_WRITE);
380 if (status < 0) {
381 mlog_errno(status);
382 goto leave;
384 ocfs2_add_links_count(dirfe, 1);
385 status = ocfs2_journal_dirty(handle, parent_fe_bh);
386 if (status < 0) {
387 mlog_errno(status);
388 goto leave;
390 inc_nlink(dir);
393 status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
394 meta_ac, data_ac);
395 if (status < 0) {
396 mlog_errno(status);
397 goto leave;
400 if (si.enable) {
401 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
402 meta_ac, data_ac);
403 if (status < 0) {
404 mlog_errno(status);
405 goto leave;
409 status = ocfs2_add_entry(handle, dentry, inode,
410 OCFS2_I(inode)->ip_blkno, parent_fe_bh,
411 &lookup);
412 if (status < 0) {
413 mlog_errno(status);
414 goto leave;
417 status = ocfs2_dentry_attach_lock(dentry, inode,
418 OCFS2_I(dir)->ip_blkno);
419 if (status) {
420 mlog_errno(status);
421 goto leave;
424 insert_inode_hash(inode);
425 dentry->d_op = &ocfs2_dentry_ops;
426 d_instantiate(dentry, inode);
427 status = 0;
428 leave:
429 if (status < 0 && did_quota_inode)
430 vfs_dq_free_inode(inode);
431 if (handle)
432 ocfs2_commit_trans(osb, handle);
434 ocfs2_inode_unlock(dir, 1);
436 if (status == -ENOSPC)
437 mlog(0, "Disk is full\n");
439 brelse(new_fe_bh);
440 brelse(parent_fe_bh);
441 kfree(si.name);
442 kfree(si.value);
444 ocfs2_free_dir_lookup_result(&lookup);
446 if ((status < 0) && inode) {
447 clear_nlink(inode);
448 iput(inode);
451 if (inode_ac)
452 ocfs2_free_alloc_context(inode_ac);
454 if (data_ac)
455 ocfs2_free_alloc_context(data_ac);
457 if (meta_ac)
458 ocfs2_free_alloc_context(meta_ac);
460 mlog_exit(status);
462 return status;
465 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
466 struct inode *dir,
467 struct inode *inode,
468 struct dentry *dentry,
469 dev_t dev,
470 struct buffer_head **new_fe_bh,
471 struct buffer_head *parent_fe_bh,
472 handle_t *handle,
473 struct ocfs2_alloc_context *inode_ac)
475 int status = 0;
476 struct ocfs2_dinode *fe = NULL;
477 struct ocfs2_extent_list *fel;
478 u64 fe_blkno = 0;
479 u16 suballoc_bit;
480 u16 feat;
482 mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry,
483 inode->i_mode, (unsigned long)dev, dentry->d_name.len,
484 dentry->d_name.name);
486 *new_fe_bh = NULL;
488 status = ocfs2_claim_new_inode(osb, handle, dir, parent_fe_bh,
489 inode_ac, &suballoc_bit, &fe_blkno);
490 if (status < 0) {
491 mlog_errno(status);
492 goto leave;
495 /* populate as many fields early on as possible - many of
496 * these are used by the support functions here and in
497 * callers. */
498 inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
499 OCFS2_I(inode)->ip_blkno = fe_blkno;
500 spin_lock(&osb->osb_lock);
501 inode->i_generation = osb->s_next_generation++;
502 spin_unlock(&osb->osb_lock);
504 *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
505 if (!*new_fe_bh) {
506 status = -EIO;
507 mlog_errno(status);
508 goto leave;
510 ocfs2_set_new_buffer_uptodate(inode, *new_fe_bh);
512 status = ocfs2_journal_access_di(handle, inode, *new_fe_bh,
513 OCFS2_JOURNAL_ACCESS_CREATE);
514 if (status < 0) {
515 mlog_errno(status);
516 goto leave;
519 fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
520 memset(fe, 0, osb->sb->s_blocksize);
522 fe->i_generation = cpu_to_le32(inode->i_generation);
523 fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
524 fe->i_blkno = cpu_to_le64(fe_blkno);
525 fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
526 fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
527 fe->i_uid = cpu_to_le32(inode->i_uid);
528 fe->i_gid = cpu_to_le32(inode->i_gid);
529 fe->i_mode = cpu_to_le16(inode->i_mode);
530 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
531 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
533 ocfs2_set_links_count(fe, inode->i_nlink);
535 fe->i_last_eb_blk = 0;
536 strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
537 le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
538 fe->i_atime = fe->i_ctime = fe->i_mtime =
539 cpu_to_le64(CURRENT_TIME.tv_sec);
540 fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
541 cpu_to_le32(CURRENT_TIME.tv_nsec);
542 fe->i_dtime = 0;
545 * If supported, directories start with inline data. If inline
546 * isn't supported, but indexing is, we start them as indexed.
548 feat = le16_to_cpu(fe->i_dyn_features);
549 if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
550 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
552 fe->id2.i_data.id_count = cpu_to_le16(
553 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
554 } else {
555 fel = &fe->id2.i_list;
556 fel->l_tree_depth = 0;
557 fel->l_next_free_rec = 0;
558 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
561 status = ocfs2_journal_dirty(handle, *new_fe_bh);
562 if (status < 0) {
563 mlog_errno(status);
564 goto leave;
567 ocfs2_populate_inode(inode, fe, 1);
568 ocfs2_inode_set_new(osb, inode);
569 if (!ocfs2_mount_local(osb)) {
570 status = ocfs2_create_new_inode_locks(inode);
571 if (status < 0)
572 mlog_errno(status);
575 status = 0; /* error in ocfs2_create_new_inode_locks is not
576 * critical */
578 leave:
579 if (status < 0) {
580 if (*new_fe_bh) {
581 brelse(*new_fe_bh);
582 *new_fe_bh = NULL;
586 mlog_exit(status);
587 return status;
590 static int ocfs2_mkdir(struct inode *dir,
591 struct dentry *dentry,
592 int mode)
594 int ret;
596 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
597 dentry->d_name.len, dentry->d_name.name);
598 ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
599 mlog_exit(ret);
601 return ret;
604 static int ocfs2_create(struct inode *dir,
605 struct dentry *dentry,
606 int mode,
607 struct nameidata *nd)
609 int ret;
611 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
612 dentry->d_name.len, dentry->d_name.name);
613 ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
614 mlog_exit(ret);
616 return ret;
619 static int ocfs2_link(struct dentry *old_dentry,
620 struct inode *dir,
621 struct dentry *dentry)
623 handle_t *handle;
624 struct inode *inode = old_dentry->d_inode;
625 int err;
626 struct buffer_head *fe_bh = NULL;
627 struct buffer_head *parent_fe_bh = NULL;
628 struct ocfs2_dinode *fe = NULL;
629 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
630 struct ocfs2_dir_lookup_result lookup = { NULL, };
632 mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
633 old_dentry->d_name.len, old_dentry->d_name.name,
634 dentry->d_name.len, dentry->d_name.name);
636 if (S_ISDIR(inode->i_mode))
637 return -EPERM;
639 err = ocfs2_inode_lock_nested(dir, &parent_fe_bh, 1, OI_LS_PARENT);
640 if (err < 0) {
641 if (err != -ENOENT)
642 mlog_errno(err);
643 return err;
646 if (!dir->i_nlink) {
647 err = -ENOENT;
648 goto out;
651 err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
652 dentry->d_name.len);
653 if (err)
654 goto out;
656 err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
657 dentry->d_name.name,
658 dentry->d_name.len, &lookup);
659 if (err < 0) {
660 mlog_errno(err);
661 goto out;
664 err = ocfs2_inode_lock(inode, &fe_bh, 1);
665 if (err < 0) {
666 if (err != -ENOENT)
667 mlog_errno(err);
668 goto out;
671 fe = (struct ocfs2_dinode *) fe_bh->b_data;
672 if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
673 err = -EMLINK;
674 goto out_unlock_inode;
677 handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
678 if (IS_ERR(handle)) {
679 err = PTR_ERR(handle);
680 handle = NULL;
681 mlog_errno(err);
682 goto out_unlock_inode;
685 err = ocfs2_journal_access_di(handle, inode, fe_bh,
686 OCFS2_JOURNAL_ACCESS_WRITE);
687 if (err < 0) {
688 mlog_errno(err);
689 goto out_commit;
692 inc_nlink(inode);
693 inode->i_ctime = CURRENT_TIME;
694 ocfs2_set_links_count(fe, inode->i_nlink);
695 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
696 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
698 err = ocfs2_journal_dirty(handle, fe_bh);
699 if (err < 0) {
700 ocfs2_add_links_count(fe, -1);
701 drop_nlink(inode);
702 mlog_errno(err);
703 goto out_commit;
706 err = ocfs2_add_entry(handle, dentry, inode,
707 OCFS2_I(inode)->ip_blkno,
708 parent_fe_bh, &lookup);
709 if (err) {
710 ocfs2_add_links_count(fe, -1);
711 drop_nlink(inode);
712 mlog_errno(err);
713 goto out_commit;
716 err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
717 if (err) {
718 mlog_errno(err);
719 goto out_commit;
722 atomic_inc(&inode->i_count);
723 dentry->d_op = &ocfs2_dentry_ops;
724 d_instantiate(dentry, inode);
726 out_commit:
727 ocfs2_commit_trans(osb, handle);
728 out_unlock_inode:
729 ocfs2_inode_unlock(inode, 1);
731 out:
732 ocfs2_inode_unlock(dir, 1);
734 brelse(fe_bh);
735 brelse(parent_fe_bh);
737 ocfs2_free_dir_lookup_result(&lookup);
739 mlog_exit(err);
741 return err;
745 * Takes and drops an exclusive lock on the given dentry. This will
746 * force other nodes to drop it.
748 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
750 int ret;
752 ret = ocfs2_dentry_lock(dentry, 1);
753 if (ret)
754 mlog_errno(ret);
755 else
756 ocfs2_dentry_unlock(dentry, 1);
758 return ret;
761 static inline int inode_is_unlinkable(struct inode *inode)
763 if (S_ISDIR(inode->i_mode)) {
764 if (inode->i_nlink == 2)
765 return 1;
766 return 0;
769 if (inode->i_nlink == 1)
770 return 1;
771 return 0;
774 static int ocfs2_unlink(struct inode *dir,
775 struct dentry *dentry)
777 int status;
778 int child_locked = 0;
779 struct inode *inode = dentry->d_inode;
780 struct inode *orphan_dir = NULL;
781 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
782 u64 blkno;
783 struct ocfs2_dinode *fe = NULL;
784 struct buffer_head *fe_bh = NULL;
785 struct buffer_head *parent_node_bh = NULL;
786 handle_t *handle = NULL;
787 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
788 struct ocfs2_dir_lookup_result lookup = { NULL, };
789 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
791 mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
792 dentry->d_name.len, dentry->d_name.name);
794 BUG_ON(dentry->d_parent->d_inode != dir);
796 mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
798 if (inode == osb->root_inode) {
799 mlog(0, "Cannot delete the root directory\n");
800 return -EPERM;
803 status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
804 OI_LS_PARENT);
805 if (status < 0) {
806 if (status != -ENOENT)
807 mlog_errno(status);
808 return status;
811 status = ocfs2_find_files_on_disk(dentry->d_name.name,
812 dentry->d_name.len, &blkno, dir,
813 &lookup);
814 if (status < 0) {
815 if (status != -ENOENT)
816 mlog_errno(status);
817 goto leave;
820 if (OCFS2_I(inode)->ip_blkno != blkno) {
821 status = -ENOENT;
823 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
824 (unsigned long long)OCFS2_I(inode)->ip_blkno,
825 (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
826 goto leave;
829 status = ocfs2_inode_lock(inode, &fe_bh, 1);
830 if (status < 0) {
831 if (status != -ENOENT)
832 mlog_errno(status);
833 goto leave;
835 child_locked = 1;
837 if (S_ISDIR(inode->i_mode)) {
838 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
839 status = -ENOTEMPTY;
840 goto leave;
844 status = ocfs2_remote_dentry_delete(dentry);
845 if (status < 0) {
846 /* This remote delete should succeed under all normal
847 * circumstances. */
848 mlog_errno(status);
849 goto leave;
852 if (inode_is_unlinkable(inode)) {
853 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir, inode,
854 orphan_name, &orphan_insert);
855 if (status < 0) {
856 mlog_errno(status);
857 goto leave;
861 handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
862 if (IS_ERR(handle)) {
863 status = PTR_ERR(handle);
864 handle = NULL;
865 mlog_errno(status);
866 goto leave;
869 status = ocfs2_journal_access_di(handle, inode, fe_bh,
870 OCFS2_JOURNAL_ACCESS_WRITE);
871 if (status < 0) {
872 mlog_errno(status);
873 goto leave;
876 fe = (struct ocfs2_dinode *) fe_bh->b_data;
878 if (inode_is_unlinkable(inode)) {
879 status = ocfs2_orphan_add(osb, handle, inode, fe, orphan_name,
880 &orphan_insert, orphan_dir);
881 if (status < 0) {
882 mlog_errno(status);
883 goto leave;
887 /* delete the name from the parent dir */
888 status = ocfs2_delete_entry(handle, dir, &lookup);
889 if (status < 0) {
890 mlog_errno(status);
891 goto leave;
894 if (S_ISDIR(inode->i_mode))
895 drop_nlink(inode);
896 drop_nlink(inode);
897 ocfs2_set_links_count(fe, inode->i_nlink);
899 status = ocfs2_journal_dirty(handle, fe_bh);
900 if (status < 0) {
901 mlog_errno(status);
902 goto leave;
905 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
906 if (S_ISDIR(inode->i_mode))
907 drop_nlink(dir);
909 status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
910 if (status < 0) {
911 mlog_errno(status);
912 if (S_ISDIR(inode->i_mode))
913 inc_nlink(dir);
916 leave:
917 if (handle)
918 ocfs2_commit_trans(osb, handle);
920 if (child_locked)
921 ocfs2_inode_unlock(inode, 1);
923 ocfs2_inode_unlock(dir, 1);
925 if (orphan_dir) {
926 /* This was locked for us in ocfs2_prepare_orphan_dir() */
927 ocfs2_inode_unlock(orphan_dir, 1);
928 mutex_unlock(&orphan_dir->i_mutex);
929 iput(orphan_dir);
932 brelse(fe_bh);
933 brelse(parent_node_bh);
935 ocfs2_free_dir_lookup_result(&orphan_insert);
936 ocfs2_free_dir_lookup_result(&lookup);
938 mlog_exit(status);
940 return status;
944 * The only place this should be used is rename!
945 * if they have the same id, then the 1st one is the only one locked.
947 static int ocfs2_double_lock(struct ocfs2_super *osb,
948 struct buffer_head **bh1,
949 struct inode *inode1,
950 struct buffer_head **bh2,
951 struct inode *inode2)
953 int status;
954 struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
955 struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
956 struct buffer_head **tmpbh;
957 struct inode *tmpinode;
959 mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
960 (unsigned long long)oi1->ip_blkno,
961 (unsigned long long)oi2->ip_blkno);
963 if (*bh1)
964 *bh1 = NULL;
965 if (*bh2)
966 *bh2 = NULL;
968 /* we always want to lock the one with the lower lockid first. */
969 if (oi1->ip_blkno != oi2->ip_blkno) {
970 if (oi1->ip_blkno < oi2->ip_blkno) {
971 /* switch id1 and id2 around */
972 mlog(0, "switching them around...\n");
973 tmpbh = bh2;
974 bh2 = bh1;
975 bh1 = tmpbh;
977 tmpinode = inode2;
978 inode2 = inode1;
979 inode1 = tmpinode;
981 /* lock id2 */
982 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
983 OI_LS_RENAME1);
984 if (status < 0) {
985 if (status != -ENOENT)
986 mlog_errno(status);
987 goto bail;
991 /* lock id1 */
992 status = ocfs2_inode_lock_nested(inode1, bh1, 1, OI_LS_RENAME2);
993 if (status < 0) {
995 * An error return must mean that no cluster locks
996 * were held on function exit.
998 if (oi1->ip_blkno != oi2->ip_blkno)
999 ocfs2_inode_unlock(inode2, 1);
1001 if (status != -ENOENT)
1002 mlog_errno(status);
1005 bail:
1006 mlog_exit(status);
1007 return status;
1010 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1012 ocfs2_inode_unlock(inode1, 1);
1014 if (inode1 != inode2)
1015 ocfs2_inode_unlock(inode2, 1);
1018 static int ocfs2_rename(struct inode *old_dir,
1019 struct dentry *old_dentry,
1020 struct inode *new_dir,
1021 struct dentry *new_dentry)
1023 int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1024 int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1025 struct inode *old_inode = old_dentry->d_inode;
1026 struct inode *new_inode = new_dentry->d_inode;
1027 struct inode *orphan_dir = NULL;
1028 struct ocfs2_dinode *newfe = NULL;
1029 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1030 struct buffer_head *newfe_bh = NULL;
1031 struct buffer_head *old_inode_bh = NULL;
1032 struct ocfs2_super *osb = NULL;
1033 u64 newfe_blkno, old_de_ino;
1034 handle_t *handle = NULL;
1035 struct buffer_head *old_dir_bh = NULL;
1036 struct buffer_head *new_dir_bh = NULL;
1037 nlink_t old_dir_nlink = old_dir->i_nlink;
1038 struct ocfs2_dinode *old_di;
1039 struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1040 struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1041 struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1042 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1043 struct ocfs2_dir_lookup_result target_insert = { NULL, };
1045 /* At some point it might be nice to break this function up a
1046 * bit. */
1048 mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1049 old_dir, old_dentry, new_dir, new_dentry,
1050 old_dentry->d_name.len, old_dentry->d_name.name,
1051 new_dentry->d_name.len, new_dentry->d_name.name);
1053 osb = OCFS2_SB(old_dir->i_sb);
1055 if (new_inode) {
1056 if (!igrab(new_inode))
1057 BUG();
1060 /* Assume a directory hierarchy thusly:
1061 * a/b/c
1062 * a/d
1063 * a,b,c, and d are all directories.
1065 * from cwd of 'a' on both nodes:
1066 * node1: mv b/c d
1067 * node2: mv d b/c
1069 * And that's why, just like the VFS, we need a file system
1070 * rename lock. */
1071 if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1072 status = ocfs2_rename_lock(osb);
1073 if (status < 0) {
1074 mlog_errno(status);
1075 goto bail;
1077 rename_lock = 1;
1080 /* if old and new are the same, this'll just do one lock. */
1081 status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1082 &new_dir_bh, new_dir);
1083 if (status < 0) {
1084 mlog_errno(status);
1085 goto bail;
1087 parents_locked = 1;
1089 /* make sure both dirs have bhs
1090 * get an extra ref on old_dir_bh if old==new */
1091 if (!new_dir_bh) {
1092 if (old_dir_bh) {
1093 new_dir_bh = old_dir_bh;
1094 get_bh(new_dir_bh);
1095 } else {
1096 mlog(ML_ERROR, "no old_dir_bh!\n");
1097 status = -EIO;
1098 goto bail;
1103 * Aside from allowing a meta data update, the locking here
1104 * also ensures that the downconvert thread on other nodes
1105 * won't have to concurrently downconvert the inode and the
1106 * dentry locks.
1108 status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1109 OI_LS_PARENT);
1110 if (status < 0) {
1111 if (status != -ENOENT)
1112 mlog_errno(status);
1113 goto bail;
1115 old_child_locked = 1;
1117 status = ocfs2_remote_dentry_delete(old_dentry);
1118 if (status < 0) {
1119 mlog_errno(status);
1120 goto bail;
1123 if (S_ISDIR(old_inode->i_mode)) {
1124 u64 old_inode_parent;
1126 update_dot_dot = 1;
1127 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1128 old_inode,
1129 &old_inode_dot_dot_res);
1130 if (status) {
1131 status = -EIO;
1132 goto bail;
1135 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1136 status = -EIO;
1137 goto bail;
1140 if (!new_inode && new_dir != old_dir &&
1141 new_dir->i_nlink >= ocfs2_link_max(osb)) {
1142 status = -EMLINK;
1143 goto bail;
1147 status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1148 old_dentry->d_name.len,
1149 &old_de_ino);
1150 if (status) {
1151 status = -ENOENT;
1152 goto bail;
1156 * Check for inode number is _not_ due to possible IO errors.
1157 * We might rmdir the source, keep it as pwd of some process
1158 * and merrily kill the link to whatever was created under the
1159 * same name. Goodbye sticky bit ;-<
1161 if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1162 status = -ENOENT;
1163 goto bail;
1166 /* check if the target already exists (in which case we need
1167 * to delete it */
1168 status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1169 new_dentry->d_name.len,
1170 &newfe_blkno, new_dir,
1171 &target_lookup_res);
1172 /* The only error we allow here is -ENOENT because the new
1173 * file not existing is perfectly valid. */
1174 if ((status < 0) && (status != -ENOENT)) {
1175 /* If we cannot find the file specified we should just */
1176 /* return the error... */
1177 mlog_errno(status);
1178 goto bail;
1180 if (status == 0)
1181 target_exists = 1;
1183 if (!target_exists && new_inode) {
1185 * Target was unlinked by another node while we were
1186 * waiting to get to ocfs2_rename(). There isn't
1187 * anything we can do here to help the situation, so
1188 * bubble up the appropriate error.
1190 status = -ENOENT;
1191 goto bail;
1194 /* In case we need to overwrite an existing file, we blow it
1195 * away first */
1196 if (target_exists) {
1197 /* VFS didn't think there existed an inode here, but
1198 * someone else in the cluster must have raced our
1199 * rename to create one. Today we error cleanly, in
1200 * the future we should consider calling iget to build
1201 * a new struct inode for this entry. */
1202 if (!new_inode) {
1203 status = -EACCES;
1205 mlog(0, "We found an inode for name %.*s but VFS "
1206 "didn't give us one.\n", new_dentry->d_name.len,
1207 new_dentry->d_name.name);
1208 goto bail;
1211 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1212 status = -EACCES;
1214 mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1215 (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1216 (unsigned long long)newfe_blkno,
1217 OCFS2_I(new_inode)->ip_flags);
1218 goto bail;
1221 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1222 if (status < 0) {
1223 if (status != -ENOENT)
1224 mlog_errno(status);
1225 goto bail;
1227 new_child_locked = 1;
1229 status = ocfs2_remote_dentry_delete(new_dentry);
1230 if (status < 0) {
1231 mlog_errno(status);
1232 goto bail;
1235 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1237 mlog(0, "aha rename over existing... new_blkno=%llu "
1238 "newfebh=%p bhblocknr=%llu\n",
1239 (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1240 (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1242 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1243 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1244 new_inode,
1245 orphan_name,
1246 &orphan_insert);
1247 if (status < 0) {
1248 mlog_errno(status);
1249 goto bail;
1252 } else {
1253 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1255 status = ocfs2_check_dir_for_entry(new_dir,
1256 new_dentry->d_name.name,
1257 new_dentry->d_name.len);
1258 if (status)
1259 goto bail;
1261 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1262 new_dentry->d_name.name,
1263 new_dentry->d_name.len,
1264 &target_insert);
1265 if (status < 0) {
1266 mlog_errno(status);
1267 goto bail;
1271 handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1272 if (IS_ERR(handle)) {
1273 status = PTR_ERR(handle);
1274 handle = NULL;
1275 mlog_errno(status);
1276 goto bail;
1279 if (target_exists) {
1280 if (S_ISDIR(new_inode->i_mode)) {
1281 if (new_inode->i_nlink != 2 ||
1282 !ocfs2_empty_dir(new_inode)) {
1283 status = -ENOTEMPTY;
1284 goto bail;
1287 status = ocfs2_journal_access_di(handle, new_inode, newfe_bh,
1288 OCFS2_JOURNAL_ACCESS_WRITE);
1289 if (status < 0) {
1290 mlog_errno(status);
1291 goto bail;
1294 if (S_ISDIR(new_inode->i_mode) ||
1295 (ocfs2_read_links_count(newfe) == 1)) {
1296 status = ocfs2_orphan_add(osb, handle, new_inode,
1297 newfe, orphan_name,
1298 &orphan_insert, orphan_dir);
1299 if (status < 0) {
1300 mlog_errno(status);
1301 goto bail;
1305 /* change the dirent to point to the correct inode */
1306 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1307 old_inode);
1308 if (status < 0) {
1309 mlog_errno(status);
1310 goto bail;
1312 new_dir->i_version++;
1314 if (S_ISDIR(new_inode->i_mode))
1315 ocfs2_set_links_count(newfe, 0);
1316 else
1317 ocfs2_add_links_count(newfe, -1);
1319 status = ocfs2_journal_dirty(handle, newfe_bh);
1320 if (status < 0) {
1321 mlog_errno(status);
1322 goto bail;
1324 } else {
1325 /* if the name was not found in new_dir, add it now */
1326 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1327 OCFS2_I(old_inode)->ip_blkno,
1328 new_dir_bh, &target_insert);
1331 old_inode->i_ctime = CURRENT_TIME;
1332 mark_inode_dirty(old_inode);
1334 status = ocfs2_journal_access_di(handle, old_inode, old_inode_bh,
1335 OCFS2_JOURNAL_ACCESS_WRITE);
1336 if (status >= 0) {
1337 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1339 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1340 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1342 status = ocfs2_journal_dirty(handle, old_inode_bh);
1343 if (status < 0)
1344 mlog_errno(status);
1345 } else
1346 mlog_errno(status);
1349 * Now that the name has been added to new_dir, remove the old name.
1351 * We don't keep any directory entry context around until now
1352 * because the insert might have changed the type of directory
1353 * we're dealing with.
1355 status = ocfs2_find_entry(old_dentry->d_name.name,
1356 old_dentry->d_name.len, old_dir,
1357 &old_entry_lookup);
1358 if (status)
1359 goto bail;
1361 status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1362 if (status < 0) {
1363 mlog_errno(status);
1364 goto bail;
1367 if (new_inode) {
1368 new_inode->i_nlink--;
1369 new_inode->i_ctime = CURRENT_TIME;
1371 old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1373 if (update_dot_dot) {
1374 status = ocfs2_update_entry(old_inode, handle,
1375 &old_inode_dot_dot_res, new_dir);
1376 old_dir->i_nlink--;
1377 if (new_inode) {
1378 new_inode->i_nlink--;
1379 } else {
1380 inc_nlink(new_dir);
1381 mark_inode_dirty(new_dir);
1384 mark_inode_dirty(old_dir);
1385 ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1386 if (new_inode) {
1387 mark_inode_dirty(new_inode);
1388 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1391 if (old_dir != new_dir) {
1392 /* Keep the same times on both directories.*/
1393 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1396 * This will also pick up the i_nlink change from the
1397 * block above.
1399 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1402 if (old_dir_nlink != old_dir->i_nlink) {
1403 if (!old_dir_bh) {
1404 mlog(ML_ERROR, "need to change nlink for old dir "
1405 "%llu from %d to %d but bh is NULL!\n",
1406 (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1407 (int)old_dir_nlink, old_dir->i_nlink);
1408 } else {
1409 struct ocfs2_dinode *fe;
1410 status = ocfs2_journal_access_di(handle, old_dir,
1411 old_dir_bh,
1412 OCFS2_JOURNAL_ACCESS_WRITE);
1413 fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1414 ocfs2_set_links_count(fe, old_dir->i_nlink);
1415 status = ocfs2_journal_dirty(handle, old_dir_bh);
1418 ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1419 status = 0;
1420 bail:
1421 if (rename_lock)
1422 ocfs2_rename_unlock(osb);
1424 if (handle)
1425 ocfs2_commit_trans(osb, handle);
1427 if (parents_locked)
1428 ocfs2_double_unlock(old_dir, new_dir);
1430 if (old_child_locked)
1431 ocfs2_inode_unlock(old_inode, 1);
1433 if (new_child_locked)
1434 ocfs2_inode_unlock(new_inode, 1);
1436 if (orphan_dir) {
1437 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1438 ocfs2_inode_unlock(orphan_dir, 1);
1439 mutex_unlock(&orphan_dir->i_mutex);
1440 iput(orphan_dir);
1443 if (new_inode)
1444 sync_mapping_buffers(old_inode->i_mapping);
1446 if (new_inode)
1447 iput(new_inode);
1449 ocfs2_free_dir_lookup_result(&target_lookup_res);
1450 ocfs2_free_dir_lookup_result(&old_entry_lookup);
1451 ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1452 ocfs2_free_dir_lookup_result(&orphan_insert);
1453 ocfs2_free_dir_lookup_result(&target_insert);
1455 brelse(newfe_bh);
1456 brelse(old_inode_bh);
1457 brelse(old_dir_bh);
1458 brelse(new_dir_bh);
1460 mlog_exit(status);
1462 return status;
1466 * we expect i_size = strlen(symname). Copy symname into the file
1467 * data, including the null terminator.
1469 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1470 handle_t *handle,
1471 struct inode *inode,
1472 const char *symname)
1474 struct buffer_head **bhs = NULL;
1475 const char *c;
1476 struct super_block *sb = osb->sb;
1477 u64 p_blkno, p_blocks;
1478 int virtual, blocks, status, i, bytes_left;
1480 bytes_left = i_size_read(inode) + 1;
1481 /* we can't trust i_blocks because we're actually going to
1482 * write i_size + 1 bytes. */
1483 blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1485 mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1486 (unsigned long long)inode->i_blocks,
1487 i_size_read(inode), blocks);
1489 /* Sanity check -- make sure we're going to fit. */
1490 if (bytes_left >
1491 ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1492 status = -EIO;
1493 mlog_errno(status);
1494 goto bail;
1497 bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1498 if (!bhs) {
1499 status = -ENOMEM;
1500 mlog_errno(status);
1501 goto bail;
1504 status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1505 NULL);
1506 if (status < 0) {
1507 mlog_errno(status);
1508 goto bail;
1511 /* links can never be larger than one cluster so we know this
1512 * is all going to be contiguous, but do a sanity check
1513 * anyway. */
1514 if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1515 status = -EIO;
1516 mlog_errno(status);
1517 goto bail;
1520 virtual = 0;
1521 while(bytes_left > 0) {
1522 c = &symname[virtual * sb->s_blocksize];
1524 bhs[virtual] = sb_getblk(sb, p_blkno);
1525 if (!bhs[virtual]) {
1526 status = -ENOMEM;
1527 mlog_errno(status);
1528 goto bail;
1530 ocfs2_set_new_buffer_uptodate(inode, bhs[virtual]);
1532 status = ocfs2_journal_access(handle, inode, bhs[virtual],
1533 OCFS2_JOURNAL_ACCESS_CREATE);
1534 if (status < 0) {
1535 mlog_errno(status);
1536 goto bail;
1539 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1541 memcpy(bhs[virtual]->b_data, c,
1542 (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1543 bytes_left);
1545 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1546 if (status < 0) {
1547 mlog_errno(status);
1548 goto bail;
1551 virtual++;
1552 p_blkno++;
1553 bytes_left -= sb->s_blocksize;
1556 status = 0;
1557 bail:
1559 if (bhs) {
1560 for(i = 0; i < blocks; i++)
1561 brelse(bhs[i]);
1562 kfree(bhs);
1565 mlog_exit(status);
1566 return status;
1569 static int ocfs2_symlink(struct inode *dir,
1570 struct dentry *dentry,
1571 const char *symname)
1573 int status, l, credits;
1574 u64 newsize;
1575 struct ocfs2_super *osb = NULL;
1576 struct inode *inode = NULL;
1577 struct super_block *sb;
1578 struct buffer_head *new_fe_bh = NULL;
1579 struct buffer_head *parent_fe_bh = NULL;
1580 struct ocfs2_dinode *fe = NULL;
1581 struct ocfs2_dinode *dirfe;
1582 handle_t *handle = NULL;
1583 struct ocfs2_alloc_context *inode_ac = NULL;
1584 struct ocfs2_alloc_context *data_ac = NULL;
1585 struct ocfs2_alloc_context *xattr_ac = NULL;
1586 int want_clusters = 0;
1587 int xattr_credits = 0;
1588 struct ocfs2_security_xattr_info si = {
1589 .enable = 1,
1591 int did_quota = 0, did_quota_inode = 0;
1592 struct ocfs2_dir_lookup_result lookup = { NULL, };
1594 mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1595 dentry, symname, dentry->d_name.len, dentry->d_name.name);
1597 sb = dir->i_sb;
1598 osb = OCFS2_SB(sb);
1600 l = strlen(symname) + 1;
1602 credits = ocfs2_calc_symlink_credits(sb);
1604 /* lock the parent directory */
1605 status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1606 if (status < 0) {
1607 if (status != -ENOENT)
1608 mlog_errno(status);
1609 return status;
1612 dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1613 if (!ocfs2_read_links_count(dirfe)) {
1614 /* can't make a file in a deleted directory. */
1615 status = -ENOENT;
1616 goto bail;
1619 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1620 dentry->d_name.len);
1621 if (status)
1622 goto bail;
1624 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1625 dentry->d_name.name,
1626 dentry->d_name.len, &lookup);
1627 if (status < 0) {
1628 mlog_errno(status);
1629 goto bail;
1632 status = ocfs2_reserve_new_inode(osb, &inode_ac);
1633 if (status < 0) {
1634 if (status != -ENOSPC)
1635 mlog_errno(status);
1636 goto bail;
1639 inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1640 if (!inode) {
1641 status = -ENOMEM;
1642 mlog_errno(status);
1643 goto bail;
1646 /* get security xattr */
1647 status = ocfs2_init_security_get(inode, dir, &si);
1648 if (status) {
1649 if (status == -EOPNOTSUPP)
1650 si.enable = 0;
1651 else {
1652 mlog_errno(status);
1653 goto bail;
1657 /* calculate meta data/clusters for setting security xattr */
1658 if (si.enable) {
1659 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1660 &xattr_credits, &xattr_ac);
1661 if (status < 0) {
1662 mlog_errno(status);
1663 goto bail;
1667 /* don't reserve bitmap space for fast symlinks. */
1668 if (l > ocfs2_fast_symlink_chars(sb))
1669 want_clusters += 1;
1671 status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1672 if (status < 0) {
1673 if (status != -ENOSPC)
1674 mlog_errno(status);
1675 goto bail;
1678 handle = ocfs2_start_trans(osb, credits + xattr_credits);
1679 if (IS_ERR(handle)) {
1680 status = PTR_ERR(handle);
1681 handle = NULL;
1682 mlog_errno(status);
1683 goto bail;
1686 /* We don't use standard VFS wrapper because we don't want vfs_dq_init
1687 * to be called. */
1688 if (sb_any_quota_active(osb->sb) &&
1689 osb->sb->dq_op->alloc_inode(inode, 1) == NO_QUOTA) {
1690 status = -EDQUOT;
1691 goto bail;
1693 did_quota_inode = 1;
1695 status = ocfs2_mknod_locked(osb, dir, inode, dentry,
1696 0, &new_fe_bh, parent_fe_bh, handle,
1697 inode_ac);
1698 if (status < 0) {
1699 mlog_errno(status);
1700 goto bail;
1703 fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1704 inode->i_rdev = 0;
1705 newsize = l - 1;
1706 if (l > ocfs2_fast_symlink_chars(sb)) {
1707 u32 offset = 0;
1709 inode->i_op = &ocfs2_symlink_inode_operations;
1710 if (vfs_dq_alloc_space_nodirty(inode,
1711 ocfs2_clusters_to_bytes(osb->sb, 1))) {
1712 status = -EDQUOT;
1713 goto bail;
1715 did_quota = 1;
1716 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1717 new_fe_bh,
1718 handle, data_ac, NULL,
1719 NULL);
1720 if (status < 0) {
1721 if (status != -ENOSPC && status != -EINTR) {
1722 mlog(ML_ERROR,
1723 "Failed to extend file to %llu\n",
1724 (unsigned long long)newsize);
1725 mlog_errno(status);
1726 status = -ENOSPC;
1728 goto bail;
1730 i_size_write(inode, newsize);
1731 inode->i_blocks = ocfs2_inode_sector_count(inode);
1732 } else {
1733 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1734 memcpy((char *) fe->id2.i_symlink, symname, l);
1735 i_size_write(inode, newsize);
1736 inode->i_blocks = 0;
1739 status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1740 if (status < 0) {
1741 mlog_errno(status);
1742 goto bail;
1745 if (!ocfs2_inode_is_fast_symlink(inode)) {
1746 status = ocfs2_create_symlink_data(osb, handle, inode,
1747 symname);
1748 if (status < 0) {
1749 mlog_errno(status);
1750 goto bail;
1754 if (si.enable) {
1755 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1756 xattr_ac, data_ac);
1757 if (status < 0) {
1758 mlog_errno(status);
1759 goto bail;
1763 status = ocfs2_add_entry(handle, dentry, inode,
1764 le64_to_cpu(fe->i_blkno), parent_fe_bh,
1765 &lookup);
1766 if (status < 0) {
1767 mlog_errno(status);
1768 goto bail;
1771 status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1772 if (status) {
1773 mlog_errno(status);
1774 goto bail;
1777 insert_inode_hash(inode);
1778 dentry->d_op = &ocfs2_dentry_ops;
1779 d_instantiate(dentry, inode);
1780 bail:
1781 if (status < 0 && did_quota)
1782 vfs_dq_free_space_nodirty(inode,
1783 ocfs2_clusters_to_bytes(osb->sb, 1));
1784 if (status < 0 && did_quota_inode)
1785 vfs_dq_free_inode(inode);
1786 if (handle)
1787 ocfs2_commit_trans(osb, handle);
1789 ocfs2_inode_unlock(dir, 1);
1791 brelse(new_fe_bh);
1792 brelse(parent_fe_bh);
1793 kfree(si.name);
1794 kfree(si.value);
1795 ocfs2_free_dir_lookup_result(&lookup);
1796 if (inode_ac)
1797 ocfs2_free_alloc_context(inode_ac);
1798 if (data_ac)
1799 ocfs2_free_alloc_context(data_ac);
1800 if (xattr_ac)
1801 ocfs2_free_alloc_context(xattr_ac);
1802 if ((status < 0) && inode) {
1803 clear_nlink(inode);
1804 iput(inode);
1807 mlog_exit(status);
1809 return status;
1812 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1814 int status, namelen;
1816 mlog_entry_void();
1818 namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1819 (long long)blkno);
1820 if (namelen <= 0) {
1821 if (namelen)
1822 status = namelen;
1823 else
1824 status = -EINVAL;
1825 mlog_errno(status);
1826 goto bail;
1828 if (namelen != OCFS2_ORPHAN_NAMELEN) {
1829 status = -EINVAL;
1830 mlog_errno(status);
1831 goto bail;
1834 mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
1835 namelen);
1837 status = 0;
1838 bail:
1839 mlog_exit(status);
1840 return status;
1843 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1844 struct inode **ret_orphan_dir,
1845 struct inode *inode,
1846 char *name,
1847 struct ocfs2_dir_lookup_result *lookup)
1849 struct inode *orphan_dir_inode;
1850 struct buffer_head *orphan_dir_bh = NULL;
1851 int status = 0;
1853 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1854 if (status < 0) {
1855 mlog_errno(status);
1856 return status;
1859 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1860 ORPHAN_DIR_SYSTEM_INODE,
1861 osb->slot_num);
1862 if (!orphan_dir_inode) {
1863 status = -ENOENT;
1864 mlog_errno(status);
1865 return status;
1868 mutex_lock(&orphan_dir_inode->i_mutex);
1870 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1871 if (status < 0) {
1872 mlog_errno(status);
1873 goto leave;
1876 status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1877 orphan_dir_bh, name,
1878 OCFS2_ORPHAN_NAMELEN, lookup);
1879 if (status < 0) {
1880 ocfs2_inode_unlock(orphan_dir_inode, 1);
1882 mlog_errno(status);
1883 goto leave;
1886 *ret_orphan_dir = orphan_dir_inode;
1888 leave:
1889 if (status) {
1890 mutex_unlock(&orphan_dir_inode->i_mutex);
1891 iput(orphan_dir_inode);
1894 brelse(orphan_dir_bh);
1896 mlog_exit(status);
1897 return status;
1900 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1901 handle_t *handle,
1902 struct inode *inode,
1903 struct ocfs2_dinode *fe,
1904 char *name,
1905 struct ocfs2_dir_lookup_result *lookup,
1906 struct inode *orphan_dir_inode)
1908 struct buffer_head *orphan_dir_bh = NULL;
1909 int status = 0;
1910 struct ocfs2_dinode *orphan_fe;
1912 mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
1914 status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
1915 if (status < 0) {
1916 mlog_errno(status);
1917 goto leave;
1920 status = ocfs2_journal_access_di(handle, orphan_dir_inode, orphan_dir_bh,
1921 OCFS2_JOURNAL_ACCESS_WRITE);
1922 if (status < 0) {
1923 mlog_errno(status);
1924 goto leave;
1927 /* we're a cluster, and nlink can change on disk from
1928 * underneath us... */
1929 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1930 if (S_ISDIR(inode->i_mode))
1931 ocfs2_add_links_count(orphan_fe, 1);
1932 orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
1934 status = ocfs2_journal_dirty(handle, orphan_dir_bh);
1935 if (status < 0) {
1936 mlog_errno(status);
1937 goto leave;
1940 status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
1941 OCFS2_ORPHAN_NAMELEN, inode,
1942 OCFS2_I(inode)->ip_blkno,
1943 orphan_dir_bh, lookup);
1944 if (status < 0) {
1945 mlog_errno(status);
1946 goto leave;
1949 le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
1951 /* Record which orphan dir our inode now resides
1952 * in. delete_inode will use this to determine which orphan
1953 * dir to lock. */
1954 fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
1956 mlog(0, "Inode %llu orphaned in slot %d\n",
1957 (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
1959 leave:
1960 brelse(orphan_dir_bh);
1962 mlog_exit(status);
1963 return status;
1966 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
1967 int ocfs2_orphan_del(struct ocfs2_super *osb,
1968 handle_t *handle,
1969 struct inode *orphan_dir_inode,
1970 struct inode *inode,
1971 struct buffer_head *orphan_dir_bh)
1973 char name[OCFS2_ORPHAN_NAMELEN + 1];
1974 struct ocfs2_dinode *orphan_fe;
1975 int status = 0;
1976 struct ocfs2_dir_lookup_result lookup = { NULL, };
1978 mlog_entry_void();
1980 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1981 if (status < 0) {
1982 mlog_errno(status);
1983 goto leave;
1986 mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
1987 name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
1988 OCFS2_ORPHAN_NAMELEN);
1990 /* find it's spot in the orphan directory */
1991 status = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN, orphan_dir_inode,
1992 &lookup);
1993 if (status) {
1994 mlog_errno(status);
1995 goto leave;
1998 /* remove it from the orphan directory */
1999 status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2000 if (status < 0) {
2001 mlog_errno(status);
2002 goto leave;
2005 status = ocfs2_journal_access_di(handle,orphan_dir_inode, orphan_dir_bh,
2006 OCFS2_JOURNAL_ACCESS_WRITE);
2007 if (status < 0) {
2008 mlog_errno(status);
2009 goto leave;
2012 /* do the i_nlink dance! :) */
2013 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2014 if (S_ISDIR(inode->i_mode))
2015 ocfs2_add_links_count(orphan_fe, -1);
2016 orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
2018 status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2019 if (status < 0) {
2020 mlog_errno(status);
2021 goto leave;
2024 leave:
2025 ocfs2_free_dir_lookup_result(&lookup);
2027 mlog_exit(status);
2028 return status;
2031 const struct inode_operations ocfs2_dir_iops = {
2032 .create = ocfs2_create,
2033 .lookup = ocfs2_lookup,
2034 .link = ocfs2_link,
2035 .unlink = ocfs2_unlink,
2036 .rmdir = ocfs2_unlink,
2037 .symlink = ocfs2_symlink,
2038 .mkdir = ocfs2_mkdir,
2039 .mknod = ocfs2_mknod,
2040 .rename = ocfs2_rename,
2041 .setattr = ocfs2_setattr,
2042 .getattr = ocfs2_getattr,
2043 .permission = ocfs2_permission,
2044 .setxattr = generic_setxattr,
2045 .getxattr = generic_getxattr,
2046 .listxattr = ocfs2_listxattr,
2047 .removexattr = generic_removexattr,