ocfs2: Store dir index records inline
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / ocfs2 / namei.c
blobf911edc8378bec3529464122444c30edbf865bec
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(dir, NULL, 0);
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)) {
259 status = -EMLINK;
260 goto leave;
263 dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
264 if (!dirfe->i_links_count) {
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 le16_add_cpu(&dirfe->i_links_count, 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, inode_ac, &suballoc_bit,
489 &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));
532 fe->i_links_count = cpu_to_le16(inode->i_nlink);
534 fe->i_last_eb_blk = 0;
535 strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
536 le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
537 fe->i_atime = fe->i_ctime = fe->i_mtime =
538 cpu_to_le64(CURRENT_TIME.tv_sec);
539 fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
540 cpu_to_le32(CURRENT_TIME.tv_nsec);
541 fe->i_dtime = 0;
544 * If supported, directories start with inline data. If inline
545 * isn't supported, but indexing is, we start them as indexed.
547 feat = le16_to_cpu(fe->i_dyn_features);
548 if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
549 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
551 fe->id2.i_data.id_count = cpu_to_le16(
552 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
553 } else {
554 fel = &fe->id2.i_list;
555 fel->l_tree_depth = 0;
556 fel->l_next_free_rec = 0;
557 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
560 status = ocfs2_journal_dirty(handle, *new_fe_bh);
561 if (status < 0) {
562 mlog_errno(status);
563 goto leave;
566 ocfs2_populate_inode(inode, fe, 1);
567 ocfs2_inode_set_new(osb, inode);
568 if (!ocfs2_mount_local(osb)) {
569 status = ocfs2_create_new_inode_locks(inode);
570 if (status < 0)
571 mlog_errno(status);
574 status = 0; /* error in ocfs2_create_new_inode_locks is not
575 * critical */
577 leave:
578 if (status < 0) {
579 if (*new_fe_bh) {
580 brelse(*new_fe_bh);
581 *new_fe_bh = NULL;
585 mlog_exit(status);
586 return status;
589 static int ocfs2_mkdir(struct inode *dir,
590 struct dentry *dentry,
591 int mode)
593 int ret;
595 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
596 dentry->d_name.len, dentry->d_name.name);
597 ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
598 mlog_exit(ret);
600 return ret;
603 static int ocfs2_create(struct inode *dir,
604 struct dentry *dentry,
605 int mode,
606 struct nameidata *nd)
608 int ret;
610 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
611 dentry->d_name.len, dentry->d_name.name);
612 ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
613 mlog_exit(ret);
615 return ret;
618 static int ocfs2_link(struct dentry *old_dentry,
619 struct inode *dir,
620 struct dentry *dentry)
622 handle_t *handle;
623 struct inode *inode = old_dentry->d_inode;
624 int err;
625 struct buffer_head *fe_bh = NULL;
626 struct buffer_head *parent_fe_bh = NULL;
627 struct ocfs2_dinode *fe = NULL;
628 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
629 struct ocfs2_dir_lookup_result lookup = { NULL, };
631 mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
632 old_dentry->d_name.len, old_dentry->d_name.name,
633 dentry->d_name.len, dentry->d_name.name);
635 if (S_ISDIR(inode->i_mode))
636 return -EPERM;
638 err = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
639 if (err < 0) {
640 if (err != -ENOENT)
641 mlog_errno(err);
642 return err;
645 if (!dir->i_nlink) {
646 err = -ENOENT;
647 goto out;
650 err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
651 dentry->d_name.len);
652 if (err)
653 goto out;
655 err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
656 dentry->d_name.name,
657 dentry->d_name.len, &lookup);
658 if (err < 0) {
659 mlog_errno(err);
660 goto out;
663 err = ocfs2_inode_lock(inode, &fe_bh, 1);
664 if (err < 0) {
665 if (err != -ENOENT)
666 mlog_errno(err);
667 goto out;
670 fe = (struct ocfs2_dinode *) fe_bh->b_data;
671 if (le16_to_cpu(fe->i_links_count) >= OCFS2_LINK_MAX) {
672 err = -EMLINK;
673 goto out_unlock_inode;
676 handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
677 if (IS_ERR(handle)) {
678 err = PTR_ERR(handle);
679 handle = NULL;
680 mlog_errno(err);
681 goto out_unlock_inode;
684 err = ocfs2_journal_access_di(handle, inode, fe_bh,
685 OCFS2_JOURNAL_ACCESS_WRITE);
686 if (err < 0) {
687 mlog_errno(err);
688 goto out_commit;
691 inc_nlink(inode);
692 inode->i_ctime = CURRENT_TIME;
693 fe->i_links_count = cpu_to_le16(inode->i_nlink);
694 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
695 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
697 err = ocfs2_journal_dirty(handle, fe_bh);
698 if (err < 0) {
699 le16_add_cpu(&fe->i_links_count, -1);
700 drop_nlink(inode);
701 mlog_errno(err);
702 goto out_commit;
705 err = ocfs2_add_entry(handle, dentry, inode,
706 OCFS2_I(inode)->ip_blkno,
707 parent_fe_bh, &lookup);
708 if (err) {
709 le16_add_cpu(&fe->i_links_count, -1);
710 drop_nlink(inode);
711 mlog_errno(err);
712 goto out_commit;
715 err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
716 if (err) {
717 mlog_errno(err);
718 goto out_commit;
721 atomic_inc(&inode->i_count);
722 dentry->d_op = &ocfs2_dentry_ops;
723 d_instantiate(dentry, inode);
725 out_commit:
726 ocfs2_commit_trans(osb, handle);
727 out_unlock_inode:
728 ocfs2_inode_unlock(inode, 1);
730 out:
731 ocfs2_inode_unlock(dir, 1);
733 brelse(fe_bh);
734 brelse(parent_fe_bh);
736 ocfs2_free_dir_lookup_result(&lookup);
738 mlog_exit(err);
740 return err;
744 * Takes and drops an exclusive lock on the given dentry. This will
745 * force other nodes to drop it.
747 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
749 int ret;
751 ret = ocfs2_dentry_lock(dentry, 1);
752 if (ret)
753 mlog_errno(ret);
754 else
755 ocfs2_dentry_unlock(dentry, 1);
757 return ret;
760 static inline int inode_is_unlinkable(struct inode *inode)
762 if (S_ISDIR(inode->i_mode)) {
763 if (inode->i_nlink == 2)
764 return 1;
765 return 0;
768 if (inode->i_nlink == 1)
769 return 1;
770 return 0;
773 static int ocfs2_unlink(struct inode *dir,
774 struct dentry *dentry)
776 int status;
777 int child_locked = 0;
778 struct inode *inode = dentry->d_inode;
779 struct inode *orphan_dir = NULL;
780 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
781 u64 blkno;
782 struct ocfs2_dinode *fe = NULL;
783 struct buffer_head *fe_bh = NULL;
784 struct buffer_head *parent_node_bh = NULL;
785 handle_t *handle = NULL;
786 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
787 struct ocfs2_dir_lookup_result lookup = { NULL, };
788 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
790 mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
791 dentry->d_name.len, dentry->d_name.name);
793 BUG_ON(dentry->d_parent->d_inode != dir);
795 mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
797 if (inode == osb->root_inode) {
798 mlog(0, "Cannot delete the root directory\n");
799 return -EPERM;
802 status = ocfs2_inode_lock(dir, &parent_node_bh, 1);
803 if (status < 0) {
804 if (status != -ENOENT)
805 mlog_errno(status);
806 return status;
809 status = ocfs2_find_files_on_disk(dentry->d_name.name,
810 dentry->d_name.len, &blkno, dir,
811 &lookup);
812 if (status < 0) {
813 if (status != -ENOENT)
814 mlog_errno(status);
815 goto leave;
818 if (OCFS2_I(inode)->ip_blkno != blkno) {
819 status = -ENOENT;
821 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
822 (unsigned long long)OCFS2_I(inode)->ip_blkno,
823 (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
824 goto leave;
827 status = ocfs2_inode_lock(inode, &fe_bh, 1);
828 if (status < 0) {
829 if (status != -ENOENT)
830 mlog_errno(status);
831 goto leave;
833 child_locked = 1;
835 if (S_ISDIR(inode->i_mode)) {
836 if (!ocfs2_empty_dir(inode)) {
837 status = -ENOTEMPTY;
838 goto leave;
839 } else if (inode->i_nlink != 2) {
840 status = -ENOTEMPTY;
841 goto leave;
845 status = ocfs2_remote_dentry_delete(dentry);
846 if (status < 0) {
847 /* This remote delete should succeed under all normal
848 * circumstances. */
849 mlog_errno(status);
850 goto leave;
853 if (inode_is_unlinkable(inode)) {
854 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir, inode,
855 orphan_name, &orphan_insert);
856 if (status < 0) {
857 mlog_errno(status);
858 goto leave;
862 handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
863 if (IS_ERR(handle)) {
864 status = PTR_ERR(handle);
865 handle = NULL;
866 mlog_errno(status);
867 goto leave;
870 status = ocfs2_journal_access_di(handle, inode, fe_bh,
871 OCFS2_JOURNAL_ACCESS_WRITE);
872 if (status < 0) {
873 mlog_errno(status);
874 goto leave;
877 fe = (struct ocfs2_dinode *) fe_bh->b_data;
879 if (inode_is_unlinkable(inode)) {
880 status = ocfs2_orphan_add(osb, handle, inode, fe, orphan_name,
881 &orphan_insert, orphan_dir);
882 if (status < 0) {
883 mlog_errno(status);
884 goto leave;
888 /* delete the name from the parent dir */
889 status = ocfs2_delete_entry(handle, dir, &lookup);
890 if (status < 0) {
891 mlog_errno(status);
892 goto leave;
895 if (S_ISDIR(inode->i_mode))
896 drop_nlink(inode);
897 drop_nlink(inode);
898 fe->i_links_count = cpu_to_le16(inode->i_nlink);
900 status = ocfs2_journal_dirty(handle, fe_bh);
901 if (status < 0) {
902 mlog_errno(status);
903 goto leave;
906 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
907 if (S_ISDIR(inode->i_mode))
908 drop_nlink(dir);
910 status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
911 if (status < 0) {
912 mlog_errno(status);
913 if (S_ISDIR(inode->i_mode))
914 inc_nlink(dir);
917 leave:
918 if (handle)
919 ocfs2_commit_trans(osb, handle);
921 if (child_locked)
922 ocfs2_inode_unlock(inode, 1);
924 ocfs2_inode_unlock(dir, 1);
926 if (orphan_dir) {
927 /* This was locked for us in ocfs2_prepare_orphan_dir() */
928 ocfs2_inode_unlock(orphan_dir, 1);
929 mutex_unlock(&orphan_dir->i_mutex);
930 iput(orphan_dir);
933 brelse(fe_bh);
934 brelse(parent_node_bh);
936 ocfs2_free_dir_lookup_result(&orphan_insert);
937 ocfs2_free_dir_lookup_result(&lookup);
939 mlog_exit(status);
941 return status;
945 * The only place this should be used is rename!
946 * if they have the same id, then the 1st one is the only one locked.
948 static int ocfs2_double_lock(struct ocfs2_super *osb,
949 struct buffer_head **bh1,
950 struct inode *inode1,
951 struct buffer_head **bh2,
952 struct inode *inode2)
954 int status;
955 struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
956 struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
957 struct buffer_head **tmpbh;
958 struct inode *tmpinode;
960 mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
961 (unsigned long long)oi1->ip_blkno,
962 (unsigned long long)oi2->ip_blkno);
964 if (*bh1)
965 *bh1 = NULL;
966 if (*bh2)
967 *bh2 = NULL;
969 /* we always want to lock the one with the lower lockid first. */
970 if (oi1->ip_blkno != oi2->ip_blkno) {
971 if (oi1->ip_blkno < oi2->ip_blkno) {
972 /* switch id1 and id2 around */
973 mlog(0, "switching them around...\n");
974 tmpbh = bh2;
975 bh2 = bh1;
976 bh1 = tmpbh;
978 tmpinode = inode2;
979 inode2 = inode1;
980 inode1 = tmpinode;
982 /* lock id2 */
983 status = ocfs2_inode_lock(inode2, bh2, 1);
984 if (status < 0) {
985 if (status != -ENOENT)
986 mlog_errno(status);
987 goto bail;
991 /* lock id1 */
992 status = ocfs2_inode_lock(inode1, bh1, 1);
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 *orphan_entry_bh = NULL;
1031 struct buffer_head *newfe_bh = NULL;
1032 struct buffer_head *old_inode_bh = NULL;
1033 struct buffer_head *insert_entry_bh = NULL;
1034 struct ocfs2_super *osb = NULL;
1035 u64 newfe_blkno, old_de_ino;
1036 handle_t *handle = NULL;
1037 struct buffer_head *old_dir_bh = NULL;
1038 struct buffer_head *new_dir_bh = NULL;
1039 nlink_t old_dir_nlink = old_dir->i_nlink;
1040 struct ocfs2_dinode *old_di;
1041 struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1042 struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1043 struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1044 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1045 struct ocfs2_dir_lookup_result target_insert = { NULL, };
1047 /* At some point it might be nice to break this function up a
1048 * bit. */
1050 mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1051 old_dir, old_dentry, new_dir, new_dentry,
1052 old_dentry->d_name.len, old_dentry->d_name.name,
1053 new_dentry->d_name.len, new_dentry->d_name.name);
1055 osb = OCFS2_SB(old_dir->i_sb);
1057 if (new_inode) {
1058 if (!igrab(new_inode))
1059 BUG();
1062 /* Assume a directory hierarchy thusly:
1063 * a/b/c
1064 * a/d
1065 * a,b,c, and d are all directories.
1067 * from cwd of 'a' on both nodes:
1068 * node1: mv b/c d
1069 * node2: mv d b/c
1071 * And that's why, just like the VFS, we need a file system
1072 * rename lock. */
1073 if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1074 status = ocfs2_rename_lock(osb);
1075 if (status < 0) {
1076 mlog_errno(status);
1077 goto bail;
1079 rename_lock = 1;
1082 /* if old and new are the same, this'll just do one lock. */
1083 status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1084 &new_dir_bh, new_dir);
1085 if (status < 0) {
1086 mlog_errno(status);
1087 goto bail;
1089 parents_locked = 1;
1091 /* make sure both dirs have bhs
1092 * get an extra ref on old_dir_bh if old==new */
1093 if (!new_dir_bh) {
1094 if (old_dir_bh) {
1095 new_dir_bh = old_dir_bh;
1096 get_bh(new_dir_bh);
1097 } else {
1098 mlog(ML_ERROR, "no old_dir_bh!\n");
1099 status = -EIO;
1100 goto bail;
1105 * Aside from allowing a meta data update, the locking here
1106 * also ensures that the downconvert thread on other nodes
1107 * won't have to concurrently downconvert the inode and the
1108 * dentry locks.
1110 status = ocfs2_inode_lock(old_inode, &old_inode_bh, 1);
1111 if (status < 0) {
1112 if (status != -ENOENT)
1113 mlog_errno(status);
1114 goto bail;
1116 old_child_locked = 1;
1118 status = ocfs2_remote_dentry_delete(old_dentry);
1119 if (status < 0) {
1120 mlog_errno(status);
1121 goto bail;
1124 if (S_ISDIR(old_inode->i_mode)) {
1125 u64 old_inode_parent;
1127 update_dot_dot = 1;
1128 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1129 old_inode,
1130 &old_inode_dot_dot_res);
1131 if (status) {
1132 status = -EIO;
1133 goto bail;
1136 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1137 status = -EIO;
1138 goto bail;
1141 if (!new_inode && new_dir != old_dir &&
1142 new_dir->i_nlink >= OCFS2_LINK_MAX) {
1143 status = -EMLINK;
1144 goto bail;
1148 status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1149 old_dentry->d_name.len,
1150 &old_de_ino);
1151 if (status) {
1152 status = -ENOENT;
1153 goto bail;
1157 * Check for inode number is _not_ due to possible IO errors.
1158 * We might rmdir the source, keep it as pwd of some process
1159 * and merrily kill the link to whatever was created under the
1160 * same name. Goodbye sticky bit ;-<
1162 if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1163 status = -ENOENT;
1164 goto bail;
1167 /* check if the target already exists (in which case we need
1168 * to delete it */
1169 status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1170 new_dentry->d_name.len,
1171 &newfe_blkno, new_dir,
1172 &target_lookup_res);
1173 /* The only error we allow here is -ENOENT because the new
1174 * file not existing is perfectly valid. */
1175 if ((status < 0) && (status != -ENOENT)) {
1176 /* If we cannot find the file specified we should just */
1177 /* return the error... */
1178 mlog_errno(status);
1179 goto bail;
1181 if (status == 0)
1182 target_exists = 1;
1184 if (!target_exists && new_inode) {
1186 * Target was unlinked by another node while we were
1187 * waiting to get to ocfs2_rename(). There isn't
1188 * anything we can do here to help the situation, so
1189 * bubble up the appropriate error.
1191 status = -ENOENT;
1192 goto bail;
1195 /* In case we need to overwrite an existing file, we blow it
1196 * away first */
1197 if (target_exists) {
1198 /* VFS didn't think there existed an inode here, but
1199 * someone else in the cluster must have raced our
1200 * rename to create one. Today we error cleanly, in
1201 * the future we should consider calling iget to build
1202 * a new struct inode for this entry. */
1203 if (!new_inode) {
1204 status = -EACCES;
1206 mlog(0, "We found an inode for name %.*s but VFS "
1207 "didn't give us one.\n", new_dentry->d_name.len,
1208 new_dentry->d_name.name);
1209 goto bail;
1212 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1213 status = -EACCES;
1215 mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1216 (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1217 (unsigned long long)newfe_blkno,
1218 OCFS2_I(new_inode)->ip_flags);
1219 goto bail;
1222 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1223 if (status < 0) {
1224 if (status != -ENOENT)
1225 mlog_errno(status);
1226 goto bail;
1228 new_child_locked = 1;
1230 status = ocfs2_remote_dentry_delete(new_dentry);
1231 if (status < 0) {
1232 mlog_errno(status);
1233 goto bail;
1236 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1238 mlog(0, "aha rename over existing... new_blkno=%llu "
1239 "newfebh=%p bhblocknr=%llu\n",
1240 (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1241 (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1243 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1244 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1245 new_inode,
1246 orphan_name,
1247 &orphan_insert);
1248 if (status < 0) {
1249 mlog_errno(status);
1250 goto bail;
1253 } else {
1254 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1256 status = ocfs2_check_dir_for_entry(new_dir,
1257 new_dentry->d_name.name,
1258 new_dentry->d_name.len);
1259 if (status)
1260 goto bail;
1262 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1263 new_dentry->d_name.name,
1264 new_dentry->d_name.len,
1265 &target_insert);
1266 if (status < 0) {
1267 mlog_errno(status);
1268 goto bail;
1272 handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1273 if (IS_ERR(handle)) {
1274 status = PTR_ERR(handle);
1275 handle = NULL;
1276 mlog_errno(status);
1277 goto bail;
1280 if (target_exists) {
1281 if (S_ISDIR(new_inode->i_mode)) {
1282 if (!ocfs2_empty_dir(new_inode) ||
1283 new_inode->i_nlink != 2) {
1284 status = -ENOTEMPTY;
1285 goto bail;
1288 status = ocfs2_journal_access_di(handle, new_inode, newfe_bh,
1289 OCFS2_JOURNAL_ACCESS_WRITE);
1290 if (status < 0) {
1291 mlog_errno(status);
1292 goto bail;
1295 if (S_ISDIR(new_inode->i_mode) ||
1296 (newfe->i_links_count == cpu_to_le16(1))){
1297 status = ocfs2_orphan_add(osb, handle, new_inode,
1298 newfe, orphan_name,
1299 &orphan_insert, orphan_dir);
1300 if (status < 0) {
1301 mlog_errno(status);
1302 goto bail;
1306 /* change the dirent to point to the correct inode */
1307 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1308 old_inode);
1309 if (status < 0) {
1310 mlog_errno(status);
1311 goto bail;
1313 new_dir->i_version++;
1315 if (S_ISDIR(new_inode->i_mode))
1316 newfe->i_links_count = 0;
1317 else
1318 le16_add_cpu(&newfe->i_links_count, -1);
1320 status = ocfs2_journal_dirty(handle, newfe_bh);
1321 if (status < 0) {
1322 mlog_errno(status);
1323 goto bail;
1325 } else {
1326 /* if the name was not found in new_dir, add it now */
1327 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1328 OCFS2_I(old_inode)->ip_blkno,
1329 new_dir_bh, &target_insert);
1332 old_inode->i_ctime = CURRENT_TIME;
1333 mark_inode_dirty(old_inode);
1335 status = ocfs2_journal_access_di(handle, old_inode, old_inode_bh,
1336 OCFS2_JOURNAL_ACCESS_WRITE);
1337 if (status >= 0) {
1338 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1340 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1341 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1343 status = ocfs2_journal_dirty(handle, old_inode_bh);
1344 if (status < 0)
1345 mlog_errno(status);
1346 } else
1347 mlog_errno(status);
1350 * Now that the name has been added to new_dir, remove the old name.
1352 * We don't keep any directory entry context around until now
1353 * because the insert might have changed the type of directory
1354 * we're dealing with.
1356 status = ocfs2_find_entry(old_dentry->d_name.name,
1357 old_dentry->d_name.len, old_dir,
1358 &old_entry_lookup);
1359 if (status)
1360 goto bail;
1362 status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1363 if (status < 0) {
1364 mlog_errno(status);
1365 goto bail;
1368 if (new_inode) {
1369 new_inode->i_nlink--;
1370 new_inode->i_ctime = CURRENT_TIME;
1372 old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1374 if (update_dot_dot) {
1375 status = ocfs2_update_entry(old_inode, handle,
1376 &old_inode_dot_dot_res, new_dir);
1377 old_dir->i_nlink--;
1378 if (new_inode) {
1379 new_inode->i_nlink--;
1380 } else {
1381 inc_nlink(new_dir);
1382 mark_inode_dirty(new_dir);
1385 mark_inode_dirty(old_dir);
1386 ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1387 if (new_inode) {
1388 mark_inode_dirty(new_inode);
1389 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1392 if (old_dir != new_dir) {
1393 /* Keep the same times on both directories.*/
1394 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1397 * This will also pick up the i_nlink change from the
1398 * block above.
1400 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1403 if (old_dir_nlink != old_dir->i_nlink) {
1404 if (!old_dir_bh) {
1405 mlog(ML_ERROR, "need to change nlink for old dir "
1406 "%llu from %d to %d but bh is NULL!\n",
1407 (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1408 (int)old_dir_nlink, old_dir->i_nlink);
1409 } else {
1410 struct ocfs2_dinode *fe;
1411 status = ocfs2_journal_access_di(handle, old_dir,
1412 old_dir_bh,
1413 OCFS2_JOURNAL_ACCESS_WRITE);
1414 fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1415 fe->i_links_count = cpu_to_le16(old_dir->i_nlink);
1416 status = ocfs2_journal_dirty(handle, old_dir_bh);
1420 ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1421 status = 0;
1422 bail:
1423 if (rename_lock)
1424 ocfs2_rename_unlock(osb);
1426 if (handle)
1427 ocfs2_commit_trans(osb, handle);
1429 if (parents_locked)
1430 ocfs2_double_unlock(old_dir, new_dir);
1432 if (old_child_locked)
1433 ocfs2_inode_unlock(old_inode, 1);
1435 if (new_child_locked)
1436 ocfs2_inode_unlock(new_inode, 1);
1438 if (orphan_dir) {
1439 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1440 ocfs2_inode_unlock(orphan_dir, 1);
1441 mutex_unlock(&orphan_dir->i_mutex);
1442 iput(orphan_dir);
1445 if (new_inode)
1446 sync_mapping_buffers(old_inode->i_mapping);
1448 if (new_inode)
1449 iput(new_inode);
1451 ocfs2_free_dir_lookup_result(&target_lookup_res);
1452 ocfs2_free_dir_lookup_result(&old_entry_lookup);
1453 ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1454 ocfs2_free_dir_lookup_result(&orphan_insert);
1455 ocfs2_free_dir_lookup_result(&target_insert);
1457 brelse(newfe_bh);
1458 brelse(old_inode_bh);
1459 brelse(old_dir_bh);
1460 brelse(new_dir_bh);
1461 brelse(orphan_entry_bh);
1462 brelse(insert_entry_bh);
1464 mlog_exit(status);
1466 return status;
1470 * we expect i_size = strlen(symname). Copy symname into the file
1471 * data, including the null terminator.
1473 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1474 handle_t *handle,
1475 struct inode *inode,
1476 const char *symname)
1478 struct buffer_head **bhs = NULL;
1479 const char *c;
1480 struct super_block *sb = osb->sb;
1481 u64 p_blkno, p_blocks;
1482 int virtual, blocks, status, i, bytes_left;
1484 bytes_left = i_size_read(inode) + 1;
1485 /* we can't trust i_blocks because we're actually going to
1486 * write i_size + 1 bytes. */
1487 blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1489 mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1490 (unsigned long long)inode->i_blocks,
1491 i_size_read(inode), blocks);
1493 /* Sanity check -- make sure we're going to fit. */
1494 if (bytes_left >
1495 ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1496 status = -EIO;
1497 mlog_errno(status);
1498 goto bail;
1501 bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1502 if (!bhs) {
1503 status = -ENOMEM;
1504 mlog_errno(status);
1505 goto bail;
1508 status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1509 NULL);
1510 if (status < 0) {
1511 mlog_errno(status);
1512 goto bail;
1515 /* links can never be larger than one cluster so we know this
1516 * is all going to be contiguous, but do a sanity check
1517 * anyway. */
1518 if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1519 status = -EIO;
1520 mlog_errno(status);
1521 goto bail;
1524 virtual = 0;
1525 while(bytes_left > 0) {
1526 c = &symname[virtual * sb->s_blocksize];
1528 bhs[virtual] = sb_getblk(sb, p_blkno);
1529 if (!bhs[virtual]) {
1530 status = -ENOMEM;
1531 mlog_errno(status);
1532 goto bail;
1534 ocfs2_set_new_buffer_uptodate(inode, bhs[virtual]);
1536 status = ocfs2_journal_access(handle, inode, bhs[virtual],
1537 OCFS2_JOURNAL_ACCESS_CREATE);
1538 if (status < 0) {
1539 mlog_errno(status);
1540 goto bail;
1543 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1545 memcpy(bhs[virtual]->b_data, c,
1546 (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1547 bytes_left);
1549 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1550 if (status < 0) {
1551 mlog_errno(status);
1552 goto bail;
1555 virtual++;
1556 p_blkno++;
1557 bytes_left -= sb->s_blocksize;
1560 status = 0;
1561 bail:
1563 if (bhs) {
1564 for(i = 0; i < blocks; i++)
1565 brelse(bhs[i]);
1566 kfree(bhs);
1569 mlog_exit(status);
1570 return status;
1573 static int ocfs2_symlink(struct inode *dir,
1574 struct dentry *dentry,
1575 const char *symname)
1577 int status, l, credits;
1578 u64 newsize;
1579 struct ocfs2_super *osb = NULL;
1580 struct inode *inode = NULL;
1581 struct super_block *sb;
1582 struct buffer_head *new_fe_bh = NULL;
1583 struct buffer_head *parent_fe_bh = NULL;
1584 struct ocfs2_dinode *fe = NULL;
1585 struct ocfs2_dinode *dirfe;
1586 handle_t *handle = NULL;
1587 struct ocfs2_alloc_context *inode_ac = NULL;
1588 struct ocfs2_alloc_context *data_ac = NULL;
1589 struct ocfs2_alloc_context *xattr_ac = NULL;
1590 int want_clusters = 0;
1591 int xattr_credits = 0;
1592 struct ocfs2_security_xattr_info si = {
1593 .enable = 1,
1595 int did_quota = 0, did_quota_inode = 0;
1596 struct ocfs2_dir_lookup_result lookup = { NULL, };
1598 mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1599 dentry, symname, dentry->d_name.len, dentry->d_name.name);
1601 sb = dir->i_sb;
1602 osb = OCFS2_SB(sb);
1604 l = strlen(symname) + 1;
1606 credits = ocfs2_calc_symlink_credits(sb);
1608 /* lock the parent directory */
1609 status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1610 if (status < 0) {
1611 if (status != -ENOENT)
1612 mlog_errno(status);
1613 return status;
1616 dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1617 if (!dirfe->i_links_count) {
1618 /* can't make a file in a deleted directory. */
1619 status = -ENOENT;
1620 goto bail;
1623 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1624 dentry->d_name.len);
1625 if (status)
1626 goto bail;
1628 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1629 dentry->d_name.name,
1630 dentry->d_name.len, &lookup);
1631 if (status < 0) {
1632 mlog_errno(status);
1633 goto bail;
1636 status = ocfs2_reserve_new_inode(osb, &inode_ac);
1637 if (status < 0) {
1638 if (status != -ENOSPC)
1639 mlog_errno(status);
1640 goto bail;
1643 inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1644 if (!inode) {
1645 status = -ENOMEM;
1646 mlog_errno(status);
1647 goto bail;
1650 /* get security xattr */
1651 status = ocfs2_init_security_get(inode, dir, &si);
1652 if (status) {
1653 if (status == -EOPNOTSUPP)
1654 si.enable = 0;
1655 else {
1656 mlog_errno(status);
1657 goto bail;
1661 /* calculate meta data/clusters for setting security xattr */
1662 if (si.enable) {
1663 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1664 &xattr_credits, &xattr_ac);
1665 if (status < 0) {
1666 mlog_errno(status);
1667 goto bail;
1671 /* don't reserve bitmap space for fast symlinks. */
1672 if (l > ocfs2_fast_symlink_chars(sb))
1673 want_clusters += 1;
1675 status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1676 if (status < 0) {
1677 if (status != -ENOSPC)
1678 mlog_errno(status);
1679 goto bail;
1682 handle = ocfs2_start_trans(osb, credits + xattr_credits);
1683 if (IS_ERR(handle)) {
1684 status = PTR_ERR(handle);
1685 handle = NULL;
1686 mlog_errno(status);
1687 goto bail;
1690 /* We don't use standard VFS wrapper because we don't want vfs_dq_init
1691 * to be called. */
1692 if (sb_any_quota_active(osb->sb) &&
1693 osb->sb->dq_op->alloc_inode(inode, 1) == NO_QUOTA) {
1694 status = -EDQUOT;
1695 goto bail;
1697 did_quota_inode = 1;
1699 status = ocfs2_mknod_locked(osb, dir, inode, dentry,
1700 0, &new_fe_bh, parent_fe_bh, handle,
1701 inode_ac);
1702 if (status < 0) {
1703 mlog_errno(status);
1704 goto bail;
1707 fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1708 inode->i_rdev = 0;
1709 newsize = l - 1;
1710 if (l > ocfs2_fast_symlink_chars(sb)) {
1711 u32 offset = 0;
1713 inode->i_op = &ocfs2_symlink_inode_operations;
1714 if (vfs_dq_alloc_space_nodirty(inode,
1715 ocfs2_clusters_to_bytes(osb->sb, 1))) {
1716 status = -EDQUOT;
1717 goto bail;
1719 did_quota = 1;
1720 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1721 new_fe_bh,
1722 handle, data_ac, NULL,
1723 NULL);
1724 if (status < 0) {
1725 if (status != -ENOSPC && status != -EINTR) {
1726 mlog(ML_ERROR,
1727 "Failed to extend file to %llu\n",
1728 (unsigned long long)newsize);
1729 mlog_errno(status);
1730 status = -ENOSPC;
1732 goto bail;
1734 i_size_write(inode, newsize);
1735 inode->i_blocks = ocfs2_inode_sector_count(inode);
1736 } else {
1737 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1738 memcpy((char *) fe->id2.i_symlink, symname, l);
1739 i_size_write(inode, newsize);
1740 inode->i_blocks = 0;
1743 status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1744 if (status < 0) {
1745 mlog_errno(status);
1746 goto bail;
1749 if (!ocfs2_inode_is_fast_symlink(inode)) {
1750 status = ocfs2_create_symlink_data(osb, handle, inode,
1751 symname);
1752 if (status < 0) {
1753 mlog_errno(status);
1754 goto bail;
1758 if (si.enable) {
1759 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1760 xattr_ac, data_ac);
1761 if (status < 0) {
1762 mlog_errno(status);
1763 goto bail;
1767 status = ocfs2_add_entry(handle, dentry, inode,
1768 le64_to_cpu(fe->i_blkno), parent_fe_bh,
1769 &lookup);
1770 if (status < 0) {
1771 mlog_errno(status);
1772 goto bail;
1775 status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1776 if (status) {
1777 mlog_errno(status);
1778 goto bail;
1781 insert_inode_hash(inode);
1782 dentry->d_op = &ocfs2_dentry_ops;
1783 d_instantiate(dentry, inode);
1784 bail:
1785 if (status < 0 && did_quota)
1786 vfs_dq_free_space_nodirty(inode,
1787 ocfs2_clusters_to_bytes(osb->sb, 1));
1788 if (status < 0 && did_quota_inode)
1789 vfs_dq_free_inode(inode);
1790 if (handle)
1791 ocfs2_commit_trans(osb, handle);
1793 ocfs2_inode_unlock(dir, 1);
1795 brelse(new_fe_bh);
1796 brelse(parent_fe_bh);
1797 kfree(si.name);
1798 kfree(si.value);
1799 ocfs2_free_dir_lookup_result(&lookup);
1800 if (inode_ac)
1801 ocfs2_free_alloc_context(inode_ac);
1802 if (data_ac)
1803 ocfs2_free_alloc_context(data_ac);
1804 if (xattr_ac)
1805 ocfs2_free_alloc_context(xattr_ac);
1806 if ((status < 0) && inode) {
1807 clear_nlink(inode);
1808 iput(inode);
1811 mlog_exit(status);
1813 return status;
1816 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1818 int status, namelen;
1820 mlog_entry_void();
1822 namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1823 (long long)blkno);
1824 if (namelen <= 0) {
1825 if (namelen)
1826 status = namelen;
1827 else
1828 status = -EINVAL;
1829 mlog_errno(status);
1830 goto bail;
1832 if (namelen != OCFS2_ORPHAN_NAMELEN) {
1833 status = -EINVAL;
1834 mlog_errno(status);
1835 goto bail;
1838 mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
1839 namelen);
1841 status = 0;
1842 bail:
1843 mlog_exit(status);
1844 return status;
1847 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1848 struct inode **ret_orphan_dir,
1849 struct inode *inode,
1850 char *name,
1851 struct ocfs2_dir_lookup_result *lookup)
1853 struct inode *orphan_dir_inode;
1854 struct buffer_head *orphan_dir_bh = NULL;
1855 int status = 0;
1857 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1858 if (status < 0) {
1859 mlog_errno(status);
1860 return status;
1863 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1864 ORPHAN_DIR_SYSTEM_INODE,
1865 osb->slot_num);
1866 if (!orphan_dir_inode) {
1867 status = -ENOENT;
1868 mlog_errno(status);
1869 return status;
1872 mutex_lock(&orphan_dir_inode->i_mutex);
1874 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1875 if (status < 0) {
1876 mlog_errno(status);
1877 goto leave;
1880 status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1881 orphan_dir_bh, name,
1882 OCFS2_ORPHAN_NAMELEN, lookup);
1883 if (status < 0) {
1884 ocfs2_inode_unlock(orphan_dir_inode, 1);
1886 mlog_errno(status);
1887 goto leave;
1890 *ret_orphan_dir = orphan_dir_inode;
1892 leave:
1893 if (status) {
1894 mutex_unlock(&orphan_dir_inode->i_mutex);
1895 iput(orphan_dir_inode);
1898 brelse(orphan_dir_bh);
1900 mlog_exit(status);
1901 return status;
1904 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1905 handle_t *handle,
1906 struct inode *inode,
1907 struct ocfs2_dinode *fe,
1908 char *name,
1909 struct ocfs2_dir_lookup_result *lookup,
1910 struct inode *orphan_dir_inode)
1912 struct buffer_head *orphan_dir_bh = NULL;
1913 int status = 0;
1914 struct ocfs2_dinode *orphan_fe;
1916 mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
1918 status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
1919 if (status < 0) {
1920 mlog_errno(status);
1921 goto leave;
1924 status = ocfs2_journal_access_di(handle, orphan_dir_inode, orphan_dir_bh,
1925 OCFS2_JOURNAL_ACCESS_WRITE);
1926 if (status < 0) {
1927 mlog_errno(status);
1928 goto leave;
1931 /* we're a cluster, and nlink can change on disk from
1932 * underneath us... */
1933 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1934 if (S_ISDIR(inode->i_mode))
1935 le16_add_cpu(&orphan_fe->i_links_count, 1);
1936 orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
1938 status = ocfs2_journal_dirty(handle, orphan_dir_bh);
1939 if (status < 0) {
1940 mlog_errno(status);
1941 goto leave;
1944 status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
1945 OCFS2_ORPHAN_NAMELEN, inode,
1946 OCFS2_I(inode)->ip_blkno,
1947 orphan_dir_bh, lookup);
1948 if (status < 0) {
1949 mlog_errno(status);
1950 goto leave;
1953 le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
1955 /* Record which orphan dir our inode now resides
1956 * in. delete_inode will use this to determine which orphan
1957 * dir to lock. */
1958 fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
1960 mlog(0, "Inode %llu orphaned in slot %d\n",
1961 (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
1963 leave:
1964 brelse(orphan_dir_bh);
1966 mlog_exit(status);
1967 return status;
1970 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
1971 int ocfs2_orphan_del(struct ocfs2_super *osb,
1972 handle_t *handle,
1973 struct inode *orphan_dir_inode,
1974 struct inode *inode,
1975 struct buffer_head *orphan_dir_bh)
1977 char name[OCFS2_ORPHAN_NAMELEN + 1];
1978 struct ocfs2_dinode *orphan_fe;
1979 int status = 0;
1980 struct ocfs2_dir_lookup_result lookup = { NULL, };
1982 mlog_entry_void();
1984 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1985 if (status < 0) {
1986 mlog_errno(status);
1987 goto leave;
1990 mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
1991 name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
1992 OCFS2_ORPHAN_NAMELEN);
1994 /* find it's spot in the orphan directory */
1995 status = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN, orphan_dir_inode,
1996 &lookup);
1997 if (status) {
1998 mlog_errno(status);
1999 goto leave;
2002 /* remove it from the orphan directory */
2003 status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2004 if (status < 0) {
2005 mlog_errno(status);
2006 goto leave;
2009 status = ocfs2_journal_access_di(handle,orphan_dir_inode, orphan_dir_bh,
2010 OCFS2_JOURNAL_ACCESS_WRITE);
2011 if (status < 0) {
2012 mlog_errno(status);
2013 goto leave;
2016 /* do the i_nlink dance! :) */
2017 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2018 if (S_ISDIR(inode->i_mode))
2019 le16_add_cpu(&orphan_fe->i_links_count, -1);
2020 orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
2022 status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2023 if (status < 0) {
2024 mlog_errno(status);
2025 goto leave;
2028 leave:
2029 ocfs2_free_dir_lookup_result(&lookup);
2031 mlog_exit(status);
2032 return status;
2035 const struct inode_operations ocfs2_dir_iops = {
2036 .create = ocfs2_create,
2037 .lookup = ocfs2_lookup,
2038 .link = ocfs2_link,
2039 .unlink = ocfs2_unlink,
2040 .rmdir = ocfs2_unlink,
2041 .symlink = ocfs2_symlink,
2042 .mkdir = ocfs2_mkdir,
2043 .mknod = ocfs2_mknod,
2044 .rename = ocfs2_rename,
2045 .setattr = ocfs2_setattr,
2046 .getattr = ocfs2_getattr,
2047 .permission = ocfs2_permission,
2048 .setxattr = generic_setxattr,
2049 .getxattr = generic_getxattr,
2050 .listxattr = ocfs2_listxattr,
2051 .removexattr = generic_removexattr,