?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
kmemleak: Add the slub memory allocation/freeing hooks
[linux-2.6/mini2440.git] / fs / ocfs2 / namei.c
blob33464c6b60a2ca91252af9d4f588398379089c04
1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * namei.c
5 *
6 * Create and rename file, directory, symlinks
7 *
8 * Copyright (C) 2002, 2004 Oracle. All rights reserved.
9 *
10 * Portions of this code from linux/fs/ext3/dir.c
11 *
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)
16 *
17 * from
18 *
19 * linux/fs/minix/dir.c
20 *
21 * Copyright (C) 1991, 1992 Linux Torvalds
22 *
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.
27 *
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.
32 *
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.
37 */
38
39 #include <linux/fs.h>
40 #include <linux/types.h>
41 #include <linux/slab.h>
42 #include <linux/highmem.h>
43 #include <linux/quotaops.h>
44
45 #define MLOG_MASK_PREFIX ML_NAMEI
46 #include <cluster/masklog.h>
47
48 #include "ocfs2.h"
49
50 #include "alloc.h"
51 #include "dcache.h"
52 #include "dir.h"
53 #include "dlmglue.h"
54 #include "extent_map.h"
55 #include "file.h"
56 #include "inode.h"
57 #include "journal.h"
58 #include "namei.h"
59 #include "suballoc.h"
60 #include "super.h"
61 #include "symlink.h"
62 #include "sysfile.h"
63 #include "uptodate.h"
64 #include "xattr.h"
65 #include "acl.h"
66
67 #include "buffer_head_io.h"
68
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);
78
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);
84
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);
92
93 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
94 handle_t *handle,
95 struct inode *inode,
96 const char *symname);
97
98 /* An orphan dir name is an 8 byte value, printed as a hex string */
99 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
100
101 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
102 struct nameidata *nd)
103 {
104 int status;
105 u64 blkno;
106 struct inode *inode = NULL;
107 struct dentry *ret;
108 struct ocfs2_inode_info *oi;
109
110 mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
111 dentry->d_name.len, dentry->d_name.name);
112
113 if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
114 ret = ERR_PTR(-ENAMETOOLONG);
115 goto bail;
116 }
117
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);
120
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;
127 }
128
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;
133
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;
138 }
139
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);
149
150 bail_add:
151 dentry->d_op = &ocfs2_dentry_ops;
152 ret = d_splice_alias(inode, dentry);
153
154 if (inode) {
155 /*
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.
161 *
162 * NOTE: This dentry already has ->d_op set from
163 * ocfs2_get_parent() and ocfs2_get_dentry()
164 */
165 if (ret)
166 dentry = ret;
167
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;
174 }
175 }
176
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);
183
184 bail:
185
186 mlog_exit_ptr(ret);
187
188 return ret;
189 }
190
191 static struct inode *ocfs2_get_init_inode(struct inode *dir, int mode)
192 {
193 struct inode *inode;
194
195 inode = new_inode(dir->i_sb);
196 if (!inode) {
197 mlog(ML_ERROR, "new_inode failed!\n");
198 return NULL;
199 }
200
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;
218 }
219
220 static int ocfs2_mknod(struct inode *dir,
221 struct dentry *dentry,
222 int mode,
223 dev_t dev)
224 {
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,
240 };
241 int did_quota_inode = 0;
242 struct ocfs2_dir_lookup_result lookup = { NULL, };
243
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);
247
248 /* get our super block */
249 osb = OCFS2_SB(dir->i_sb);
250
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;
256 }
257
258 if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
259 status = -EMLINK;
260 goto leave;
261 }
262
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;
268 }
269
270 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
271 dentry->d_name.len);
272 if (status)
273 goto leave;
274
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;
282 }
283
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;
290 }
291
292 inode = ocfs2_get_init_inode(dir, mode);
293 if (!inode) {
294 status = -ENOMEM;
295 mlog_errno(status);
296 goto leave;
297 }
298
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;
307 }
308 }
309
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;
317 }
318
319 /* Reserve a cluster if creating an extent based directory. */
320 if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
321 want_clusters += 1;
322
323 /* Dir indexing requires extra space as well */
324 if (ocfs2_supports_indexed_dirs(osb))
325 want_meta++;
326 }
327
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;
333 }
334
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;
340 }
341
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;
350 }
351
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;
358 }
359 did_quota_inode = 1;
360
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;
368 }
369
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;
376 }
377
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;
383 }
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;
389 }
390 inc_nlink(dir);
391 }
392
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;
398 }
399
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;
406 }
407 }
408
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;
415 }
416
417 status = ocfs2_dentry_attach_lock(dentry, inode,
418 OCFS2_I(dir)->ip_blkno);
419 if (status) {
420 mlog_errno(status);
421 goto leave;
422 }
423
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);
433
434 ocfs2_inode_unlock(dir, 1);
435
436 if (status == -ENOSPC)
437 mlog(0, "Disk is full\n");
438
439 brelse(new_fe_bh);
440 brelse(parent_fe_bh);
441 kfree(si.name);
442 kfree(si.value);
443
444 ocfs2_free_dir_lookup_result(&lookup);
445
446 if ((status < 0) && inode) {
447 clear_nlink(inode);
448 iput(inode);
449 }
450
451 if (inode_ac)
452 ocfs2_free_alloc_context(inode_ac);
453
454 if (data_ac)
455 ocfs2_free_alloc_context(data_ac);
456
457 if (meta_ac)
458 ocfs2_free_alloc_context(meta_ac);
459
460 mlog_exit(status);
461
462 return status;
463 }
464
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)
474 {
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;
481
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);
485
486 *new_fe_bh = NULL;
487
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;
493 }
494
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);
503
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;
509 }
510 ocfs2_set_new_buffer_uptodate(inode, *new_fe_bh);
511
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;
517 }
518
519 fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
520 memset(fe, 0, osb->sb->s_blocksize);
521
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
533 ocfs2_set_links_count(fe, inode->i_nlink);
534
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;
543
544 /*
545 * If supported, directories start with inline data. If inline
546 * isn't supported, but indexing is, we start them as indexed.
547 */
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);
551
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));
559 }
560
561 status = ocfs2_journal_dirty(handle, *new_fe_bh);
562 if (status < 0) {
563 mlog_errno(status);
564 goto leave;
565 }
566
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);
573 }
574
575 status = 0; /* error in ocfs2_create_new_inode_locks is not
576 * critical */
577
578 leave:
579 if (status < 0) {
580 if (*new_fe_bh) {
581 brelse(*new_fe_bh);
582 *new_fe_bh = NULL;
583 }
584 }
585
586 mlog_exit(status);
587 return status;
588 }
589
590 static int ocfs2_mkdir(struct inode *dir,
591 struct dentry *dentry,
592 int mode)
593 {
594 int ret;
595
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);
600
601 return ret;
602 }
603
604 static int ocfs2_create(struct inode *dir,
605 struct dentry *dentry,
606 int mode,
607 struct nameidata *nd)
608 {
609 int ret;
610
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);
615
616 return ret;
617 }
618
619 static int ocfs2_link(struct dentry *old_dentry,
620 struct inode *dir,
621 struct dentry *dentry)
622 {
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, };
631
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);
635
636 if (S_ISDIR(inode->i_mode))
637 return -EPERM;
638
639 err = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
640 if (err < 0) {
641 if (err != -ENOENT)
642 mlog_errno(err);
643 return err;
644 }
645
646 if (!dir->i_nlink) {
647 err = -ENOENT;
648 goto out;
649 }
650
651 err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
652 dentry->d_name.len);
653 if (err)
654 goto out;
655
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;
662 }
663
664 err = ocfs2_inode_lock(inode, &fe_bh, 1);
665 if (err < 0) {
666 if (err != -ENOENT)
667 mlog_errno(err);
668 goto out;
669 }
670
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;
675 }
676
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;
683 }
684
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;
690 }
691
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);
697
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;
704 }
705
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;
714 }
715
716 err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
717 if (err) {
718 mlog_errno(err);
719 goto out_commit;
720 }
721
722 atomic_inc(&inode->i_count);
723 dentry->d_op = &ocfs2_dentry_ops;
724 d_instantiate(dentry, inode);
725
726 out_commit:
727 ocfs2_commit_trans(osb, handle);
728 out_unlock_inode:
729 ocfs2_inode_unlock(inode, 1);
730
731 out:
732 ocfs2_inode_unlock(dir, 1);
733
734 brelse(fe_bh);
735 brelse(parent_fe_bh);
736
737 ocfs2_free_dir_lookup_result(&lookup);
738
739 mlog_exit(err);
740
741 return err;
742 }
743
744 /*
745 * Takes and drops an exclusive lock on the given dentry. This will
746 * force other nodes to drop it.
747 */
748 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
749 {
750 int ret;
751
752 ret = ocfs2_dentry_lock(dentry, 1);
753 if (ret)
754 mlog_errno(ret);
755 else
756 ocfs2_dentry_unlock(dentry, 1);
757
758 return ret;
759 }
760
761 static inline int inode_is_unlinkable(struct inode *inode)
762 {
763 if (S_ISDIR(inode->i_mode)) {
764 if (inode->i_nlink == 2)
765 return 1;
766 return 0;
767 }
768
769 if (inode->i_nlink == 1)
770 return 1;
771 return 0;
772 }
773
774 static int ocfs2_unlink(struct inode *dir,
775 struct dentry *dentry)
776 {
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, };
790
791 mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
792 dentry->d_name.len, dentry->d_name.name);
793
794 BUG_ON(dentry->d_parent->d_inode != dir);
795
796 mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
797
798 if (inode == osb->root_inode) {
799 mlog(0, "Cannot delete the root directory\n");
800 return -EPERM;
801 }
802
803 status = ocfs2_inode_lock(dir, &parent_node_bh, 1);
804 if (status < 0) {
805 if (status != -ENOENT)
806 mlog_errno(status);
807 return status;
808 }
809
810 status = ocfs2_find_files_on_disk(dentry->d_name.name,
811 dentry->d_name.len, &blkno, dir,
812 &lookup);
813 if (status < 0) {
814 if (status != -ENOENT)
815 mlog_errno(status);
816 goto leave;
817 }
818
819 if (OCFS2_I(inode)->ip_blkno != blkno) {
820 status = -ENOENT;
821
822 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
823 (unsigned long long)OCFS2_I(inode)->ip_blkno,
824 (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
825 goto leave;
826 }
827
828 status = ocfs2_inode_lock(inode, &fe_bh, 1);
829 if (status < 0) {
830 if (status != -ENOENT)
831 mlog_errno(status);
832 goto leave;
833 }
834 child_locked = 1;
835
836 if (S_ISDIR(inode->i_mode)) {
837 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
838 status = -ENOTEMPTY;
839 goto leave;
840 }
841 }
842
843 status = ocfs2_remote_dentry_delete(dentry);
844 if (status < 0) {
845 /* This remote delete should succeed under all normal
846 * circumstances. */
847 mlog_errno(status);
848 goto leave;
849 }
850
851 if (inode_is_unlinkable(inode)) {
852 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir, inode,
853 orphan_name, &orphan_insert);
854 if (status < 0) {
855 mlog_errno(status);
856 goto leave;
857 }
858 }
859
860 handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
861 if (IS_ERR(handle)) {
862 status = PTR_ERR(handle);
863 handle = NULL;
864 mlog_errno(status);
865 goto leave;
866 }
867
868 status = ocfs2_journal_access_di(handle, inode, fe_bh,
869 OCFS2_JOURNAL_ACCESS_WRITE);
870 if (status < 0) {
871 mlog_errno(status);
872 goto leave;
873 }
874
875 fe = (struct ocfs2_dinode *) fe_bh->b_data;
876
877 if (inode_is_unlinkable(inode)) {
878 status = ocfs2_orphan_add(osb, handle, inode, fe, orphan_name,
879 &orphan_insert, orphan_dir);
880 if (status < 0) {
881 mlog_errno(status);
882 goto leave;
883 }
884 }
885
886 /* delete the name from the parent dir */
887 status = ocfs2_delete_entry(handle, dir, &lookup);
888 if (status < 0) {
889 mlog_errno(status);
890 goto leave;
891 }
892
893 if (S_ISDIR(inode->i_mode))
894 drop_nlink(inode);
895 drop_nlink(inode);
896 ocfs2_set_links_count(fe, inode->i_nlink);
897
898 status = ocfs2_journal_dirty(handle, fe_bh);
899 if (status < 0) {
900 mlog_errno(status);
901 goto leave;
902 }
903
904 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
905 if (S_ISDIR(inode->i_mode))
906 drop_nlink(dir);
907
908 status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
909 if (status < 0) {
910 mlog_errno(status);
911 if (S_ISDIR(inode->i_mode))
912 inc_nlink(dir);
913 }
914
915 leave:
916 if (handle)
917 ocfs2_commit_trans(osb, handle);
918
919 if (child_locked)
920 ocfs2_inode_unlock(inode, 1);
921
922 ocfs2_inode_unlock(dir, 1);
923
924 if (orphan_dir) {
925 /* This was locked for us in ocfs2_prepare_orphan_dir() */
926 ocfs2_inode_unlock(orphan_dir, 1);
927 mutex_unlock(&orphan_dir->i_mutex);
928 iput(orphan_dir);
929 }
930
931 brelse(fe_bh);
932 brelse(parent_node_bh);
933
934 ocfs2_free_dir_lookup_result(&orphan_insert);
935 ocfs2_free_dir_lookup_result(&lookup);
936
937 mlog_exit(status);
938
939 return status;
940 }
941
942 /*
943 * The only place this should be used is rename!
944 * if they have the same id, then the 1st one is the only one locked.
945 */
946 static int ocfs2_double_lock(struct ocfs2_super *osb,
947 struct buffer_head **bh1,
948 struct inode *inode1,
949 struct buffer_head **bh2,
950 struct inode *inode2)
951 {
952 int status;
953 struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
954 struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
955 struct buffer_head **tmpbh;
956 struct inode *tmpinode;
957
958 mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
959 (unsigned long long)oi1->ip_blkno,
960 (unsigned long long)oi2->ip_blkno);
961
962 if (*bh1)
963 *bh1 = NULL;
964 if (*bh2)
965 *bh2 = NULL;
966
967 /* we always want to lock the one with the lower lockid first. */
968 if (oi1->ip_blkno != oi2->ip_blkno) {
969 if (oi1->ip_blkno < oi2->ip_blkno) {
970 /* switch id1 and id2 around */
971 mlog(0, "switching them around...\n");
972 tmpbh = bh2;
973 bh2 = bh1;
974 bh1 = tmpbh;
975
976 tmpinode = inode2;
977 inode2 = inode1;
978 inode1 = tmpinode;
979 }
980 /* lock id2 */
981 status = ocfs2_inode_lock(inode2, bh2, 1);
982 if (status < 0) {
983 if (status != -ENOENT)
984 mlog_errno(status);
985 goto bail;
986 }
987 }
988
989 /* lock id1 */
990 status = ocfs2_inode_lock(inode1, bh1, 1);
991 if (status < 0) {
992 /*
993 * An error return must mean that no cluster locks
994 * were held on function exit.
995 */
996 if (oi1->ip_blkno != oi2->ip_blkno)
997 ocfs2_inode_unlock(inode2, 1);
998
999 if (status != -ENOENT)
1000 mlog_errno(status);
1001 }
1002
1003 bail:
1004 mlog_exit(status);
1005 return status;
1006 }
1007
1008 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1009 {
1010 ocfs2_inode_unlock(inode1, 1);
1011
1012 if (inode1 != inode2)
1013 ocfs2_inode_unlock(inode2, 1);
1014 }
1015
1016 static int ocfs2_rename(struct inode *old_dir,
1017 struct dentry *old_dentry,
1018 struct inode *new_dir,
1019 struct dentry *new_dentry)
1020 {
1021 int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1022 int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1023 struct inode *old_inode = old_dentry->d_inode;
1024 struct inode *new_inode = new_dentry->d_inode;
1025 struct inode *orphan_dir = NULL;
1026 struct ocfs2_dinode *newfe = NULL;
1027 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1028 struct buffer_head *newfe_bh = NULL;
1029 struct buffer_head *old_inode_bh = NULL;
1030 struct ocfs2_super *osb = NULL;
1031 u64 newfe_blkno, old_de_ino;
1032 handle_t *handle = NULL;
1033 struct buffer_head *old_dir_bh = NULL;
1034 struct buffer_head *new_dir_bh = NULL;
1035 nlink_t old_dir_nlink = old_dir->i_nlink;
1036 struct ocfs2_dinode *old_di;
1037 struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1038 struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1039 struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1040 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1041 struct ocfs2_dir_lookup_result target_insert = { NULL, };
1042
1043 /* At some point it might be nice to break this function up a
1044 * bit. */
1045
1046 mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1047 old_dir, old_dentry, new_dir, new_dentry,
1048 old_dentry->d_name.len, old_dentry->d_name.name,
1049 new_dentry->d_name.len, new_dentry->d_name.name);
1050
1051 osb = OCFS2_SB(old_dir->i_sb);
1052
1053 if (new_inode) {
1054 if (!igrab(new_inode))
1055 BUG();
1056 }
1057
1058 /* Assume a directory hierarchy thusly:
1059 * a/b/c
1060 * a/d
1061 * a,b,c, and d are all directories.
1062 *
1063 * from cwd of 'a' on both nodes:
1064 * node1: mv b/c d
1065 * node2: mv d b/c
1066 *
1067 * And that's why, just like the VFS, we need a file system
1068 * rename lock. */
1069 if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1070 status = ocfs2_rename_lock(osb);
1071 if (status < 0) {
1072 mlog_errno(status);
1073 goto bail;
1074 }
1075 rename_lock = 1;
1076 }
1077
1078 /* if old and new are the same, this'll just do one lock. */
1079 status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1080 &new_dir_bh, new_dir);
1081 if (status < 0) {
1082 mlog_errno(status);
1083 goto bail;
1084 }
1085 parents_locked = 1;
1086
1087 /* make sure both dirs have bhs
1088 * get an extra ref on old_dir_bh if old==new */
1089 if (!new_dir_bh) {
1090 if (old_dir_bh) {
1091 new_dir_bh = old_dir_bh;
1092 get_bh(new_dir_bh);
1093 } else {
1094 mlog(ML_ERROR, "no old_dir_bh!\n");
1095 status = -EIO;
1096 goto bail;
1097 }
1098 }
1099
1100 /*
1101 * Aside from allowing a meta data update, the locking here
1102 * also ensures that the downconvert thread on other nodes
1103 * won't have to concurrently downconvert the inode and the
1104 * dentry locks.
1105 */
1106 status = ocfs2_inode_lock(old_inode, &old_inode_bh, 1);
1107 if (status < 0) {
1108 if (status != -ENOENT)
1109 mlog_errno(status);
1110 goto bail;
1111 }
1112 old_child_locked = 1;
1113
1114 status = ocfs2_remote_dentry_delete(old_dentry);
1115 if (status < 0) {
1116 mlog_errno(status);
1117 goto bail;
1118 }
1119
1120 if (S_ISDIR(old_inode->i_mode)) {
1121 u64 old_inode_parent;
1122
1123 update_dot_dot = 1;
1124 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1125 old_inode,
1126 &old_inode_dot_dot_res);
1127 if (status) {
1128 status = -EIO;
1129 goto bail;
1130 }
1131
1132 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1133 status = -EIO;
1134 goto bail;
1135 }
1136
1137 if (!new_inode && new_dir != old_dir &&
1138 new_dir->i_nlink >= ocfs2_link_max(osb)) {
1139 status = -EMLINK;
1140 goto bail;
1141 }
1142 }
1143
1144 status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1145 old_dentry->d_name.len,
1146 &old_de_ino);
1147 if (status) {
1148 status = -ENOENT;
1149 goto bail;
1150 }
1151
1152 /*
1153 * Check for inode number is _not_ due to possible IO errors.
1154 * We might rmdir the source, keep it as pwd of some process
1155 * and merrily kill the link to whatever was created under the
1156 * same name. Goodbye sticky bit ;-<
1157 */
1158 if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1159 status = -ENOENT;
1160 goto bail;
1161 }
1162
1163 /* check if the target already exists (in which case we need
1164 * to delete it */
1165 status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1166 new_dentry->d_name.len,
1167 &newfe_blkno, new_dir,
1168 &target_lookup_res);
1169 /* The only error we allow here is -ENOENT because the new
1170 * file not existing is perfectly valid. */
1171 if ((status < 0) && (status != -ENOENT)) {
1172 /* If we cannot find the file specified we should just */
1173 /* return the error... */
1174 mlog_errno(status);
1175 goto bail;
1176 }
1177 if (status == 0)
1178 target_exists = 1;
1179
1180 if (!target_exists && new_inode) {
1181 /*
1182 * Target was unlinked by another node while we were
1183 * waiting to get to ocfs2_rename(). There isn't
1184 * anything we can do here to help the situation, so
1185 * bubble up the appropriate error.
1186 */
1187 status = -ENOENT;
1188 goto bail;
1189 }
1190
1191 /* In case we need to overwrite an existing file, we blow it
1192 * away first */
1193 if (target_exists) {
1194 /* VFS didn't think there existed an inode here, but
1195 * someone else in the cluster must have raced our
1196 * rename to create one. Today we error cleanly, in
1197 * the future we should consider calling iget to build
1198 * a new struct inode for this entry. */
1199 if (!new_inode) {
1200 status = -EACCES;
1201
1202 mlog(0, "We found an inode for name %.*s but VFS "
1203 "didn't give us one.\n", new_dentry->d_name.len,
1204 new_dentry->d_name.name);
1205 goto bail;
1206 }
1207
1208 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1209 status = -EACCES;
1210
1211 mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1212 (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1213 (unsigned long long)newfe_blkno,
1214 OCFS2_I(new_inode)->ip_flags);
1215 goto bail;
1216 }
1217
1218 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1219 if (status < 0) {
1220 if (status != -ENOENT)
1221 mlog_errno(status);
1222 goto bail;
1223 }
1224 new_child_locked = 1;
1225
1226 status = ocfs2_remote_dentry_delete(new_dentry);
1227 if (status < 0) {
1228 mlog_errno(status);
1229 goto bail;
1230 }
1231
1232 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1233
1234 mlog(0, "aha rename over existing... new_blkno=%llu "
1235 "newfebh=%p bhblocknr=%llu\n",
1236 (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1237 (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1238
1239 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1240 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1241 new_inode,
1242 orphan_name,
1243 &orphan_insert);
1244 if (status < 0) {
1245 mlog_errno(status);
1246 goto bail;
1247 }
1248 }
1249 } else {
1250 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1251
1252 status = ocfs2_check_dir_for_entry(new_dir,
1253 new_dentry->d_name.name,
1254 new_dentry->d_name.len);
1255 if (status)
1256 goto bail;
1257
1258 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1259 new_dentry->d_name.name,
1260 new_dentry->d_name.len,
1261 &target_insert);
1262 if (status < 0) {
1263 mlog_errno(status);
1264 goto bail;
1265 }
1266 }
1267
1268 handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1269 if (IS_ERR(handle)) {
1270 status = PTR_ERR(handle);
1271 handle = NULL;
1272 mlog_errno(status);
1273 goto bail;
1274 }
1275
1276 if (target_exists) {
1277 if (S_ISDIR(new_inode->i_mode)) {
1278 if (new_inode->i_nlink != 2 ||
1279 !ocfs2_empty_dir(new_inode)) {
1280 status = -ENOTEMPTY;
1281 goto bail;
1282 }
1283 }
1284 status = ocfs2_journal_access_di(handle, new_inode, newfe_bh,
1285 OCFS2_JOURNAL_ACCESS_WRITE);
1286 if (status < 0) {
1287 mlog_errno(status);
1288 goto bail;
1289 }
1290
1291 if (S_ISDIR(new_inode->i_mode) ||
1292 (ocfs2_read_links_count(newfe) == 1)) {
1293 status = ocfs2_orphan_add(osb, handle, new_inode,
1294 newfe, orphan_name,
1295 &orphan_insert, orphan_dir);
1296 if (status < 0) {
1297 mlog_errno(status);
1298 goto bail;
1299 }
1300 }
1301
1302 /* change the dirent to point to the correct inode */
1303 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1304 old_inode);
1305 if (status < 0) {
1306 mlog_errno(status);
1307 goto bail;
1308 }
1309 new_dir->i_version++;
1310
1311 if (S_ISDIR(new_inode->i_mode))
1312 ocfs2_set_links_count(newfe, 0);
1313 else
1314 ocfs2_add_links_count(newfe, -1);
1315
1316 status = ocfs2_journal_dirty(handle, newfe_bh);
1317 if (status < 0) {
1318 mlog_errno(status);
1319 goto bail;
1320 }
1321 } else {
1322 /* if the name was not found in new_dir, add it now */
1323 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1324 OCFS2_I(old_inode)->ip_blkno,
1325 new_dir_bh, &target_insert);
1326 }
1327
1328 old_inode->i_ctime = CURRENT_TIME;
1329 mark_inode_dirty(old_inode);
1330
1331 status = ocfs2_journal_access_di(handle, old_inode, old_inode_bh,
1332 OCFS2_JOURNAL_ACCESS_WRITE);
1333 if (status >= 0) {
1334 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1335
1336 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1337 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1338
1339 status = ocfs2_journal_dirty(handle, old_inode_bh);
1340 if (status < 0)
1341 mlog_errno(status);
1342 } else
1343 mlog_errno(status);
1344
1345 /*
1346 * Now that the name has been added to new_dir, remove the old name.
1347 *
1348 * We don't keep any directory entry context around until now
1349 * because the insert might have changed the type of directory
1350 * we're dealing with.
1351 */
1352 status = ocfs2_find_entry(old_dentry->d_name.name,
1353 old_dentry->d_name.len, old_dir,
1354 &old_entry_lookup);
1355 if (status)
1356 goto bail;
1357
1358 status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1359 if (status < 0) {
1360 mlog_errno(status);
1361 goto bail;
1362 }
1363
1364 if (new_inode) {
1365 new_inode->i_nlink--;
1366 new_inode->i_ctime = CURRENT_TIME;
1367 }
1368 old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1369
1370 if (update_dot_dot) {
1371 status = ocfs2_update_entry(old_inode, handle,
1372 &old_inode_dot_dot_res, new_dir);
1373 old_dir->i_nlink--;
1374 if (new_inode) {
1375 new_inode->i_nlink--;
1376 } else {
1377 inc_nlink(new_dir);
1378 mark_inode_dirty(new_dir);
1379 }
1380 }
1381 mark_inode_dirty(old_dir);
1382 ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1383 if (new_inode) {
1384 mark_inode_dirty(new_inode);
1385 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1386 }
1387
1388 if (old_dir != new_dir) {
1389 /* Keep the same times on both directories.*/
1390 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1391
1392 /*
1393 * This will also pick up the i_nlink change from the
1394 * block above.
1395 */
1396 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1397 }
1398
1399 if (old_dir_nlink != old_dir->i_nlink) {
1400 if (!old_dir_bh) {
1401 mlog(ML_ERROR, "need to change nlink for old dir "
1402 "%llu from %d to %d but bh is NULL!\n",
1403 (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1404 (int)old_dir_nlink, old_dir->i_nlink);
1405 } else {
1406 struct ocfs2_dinode *fe;
1407 status = ocfs2_journal_access_di(handle, old_dir,
1408 old_dir_bh,
1409 OCFS2_JOURNAL_ACCESS_WRITE);
1410 fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1411 ocfs2_set_links_count(fe, old_dir->i_nlink);
1412 status = ocfs2_journal_dirty(handle, old_dir_bh);
1413 }
1414 }
1415 ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1416 status = 0;
1417 bail:
1418 if (rename_lock)
1419 ocfs2_rename_unlock(osb);
1420
1421 if (handle)
1422 ocfs2_commit_trans(osb, handle);
1423
1424 if (parents_locked)
1425 ocfs2_double_unlock(old_dir, new_dir);
1426
1427 if (old_child_locked)
1428 ocfs2_inode_unlock(old_inode, 1);
1429
1430 if (new_child_locked)
1431 ocfs2_inode_unlock(new_inode, 1);
1432
1433 if (orphan_dir) {
1434 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1435 ocfs2_inode_unlock(orphan_dir, 1);
1436 mutex_unlock(&orphan_dir->i_mutex);
1437 iput(orphan_dir);
1438 }
1439
1440 if (new_inode)
1441 sync_mapping_buffers(old_inode->i_mapping);
1442
1443 if (new_inode)
1444 iput(new_inode);
1445
1446 ocfs2_free_dir_lookup_result(&target_lookup_res);
1447 ocfs2_free_dir_lookup_result(&old_entry_lookup);
1448 ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1449 ocfs2_free_dir_lookup_result(&orphan_insert);
1450 ocfs2_free_dir_lookup_result(&target_insert);
1451
1452 brelse(newfe_bh);
1453 brelse(old_inode_bh);
1454 brelse(old_dir_bh);
1455 brelse(new_dir_bh);
1456
1457 mlog_exit(status);
1458
1459 return status;
1460 }
1461
1462 /*
1463 * we expect i_size = strlen(symname). Copy symname into the file
1464 * data, including the null terminator.
1465 */
1466 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1467 handle_t *handle,
1468 struct inode *inode,
1469 const char *symname)
1470 {
1471 struct buffer_head **bhs = NULL;
1472 const char *c;
1473 struct super_block *sb = osb->sb;
1474 u64 p_blkno, p_blocks;
1475 int virtual, blocks, status, i, bytes_left;
1476
1477 bytes_left = i_size_read(inode) + 1;
1478 /* we can't trust i_blocks because we're actually going to
1479 * write i_size + 1 bytes. */
1480 blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1481
1482 mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1483 (unsigned long long)inode->i_blocks,
1484 i_size_read(inode), blocks);
1485
1486 /* Sanity check -- make sure we're going to fit. */
1487 if (bytes_left >
1488 ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1489 status = -EIO;
1490 mlog_errno(status);
1491 goto bail;
1492 }
1493
1494 bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1495 if (!bhs) {
1496 status = -ENOMEM;
1497 mlog_errno(status);
1498 goto bail;
1499 }
1500
1501 status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1502 NULL);
1503 if (status < 0) {
1504 mlog_errno(status);
1505 goto bail;
1506 }
1507
1508 /* links can never be larger than one cluster so we know this
1509 * is all going to be contiguous, but do a sanity check
1510 * anyway. */
1511 if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1512 status = -EIO;
1513 mlog_errno(status);
1514 goto bail;
1515 }
1516
1517 virtual = 0;
1518 while(bytes_left > 0) {
1519 c = &symname[virtual * sb->s_blocksize];
1520
1521 bhs[virtual] = sb_getblk(sb, p_blkno);
1522 if (!bhs[virtual]) {
1523 status = -ENOMEM;
1524 mlog_errno(status);
1525 goto bail;
1526 }
1527 ocfs2_set_new_buffer_uptodate(inode, bhs[virtual]);
1528
1529 status = ocfs2_journal_access(handle, inode, bhs[virtual],
1530 OCFS2_JOURNAL_ACCESS_CREATE);
1531 if (status < 0) {
1532 mlog_errno(status);
1533 goto bail;
1534 }
1535
1536 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1537
1538 memcpy(bhs[virtual]->b_data, c,
1539 (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1540 bytes_left);
1541
1542 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1543 if (status < 0) {
1544 mlog_errno(status);
1545 goto bail;
1546 }
1547
1548 virtual++;
1549 p_blkno++;
1550 bytes_left -= sb->s_blocksize;
1551 }
1552
1553 status = 0;
1554 bail:
1555
1556 if (bhs) {
1557 for(i = 0; i < blocks; i++)
1558 brelse(bhs[i]);
1559 kfree(bhs);
1560 }
1561
1562 mlog_exit(status);
1563 return status;
1564 }
1565
1566 static int ocfs2_symlink(struct inode *dir,
1567 struct dentry *dentry,
1568 const char *symname)
1569 {
1570 int status, l, credits;
1571 u64 newsize;
1572 struct ocfs2_super *osb = NULL;
1573 struct inode *inode = NULL;
1574 struct super_block *sb;
1575 struct buffer_head *new_fe_bh = NULL;
1576 struct buffer_head *parent_fe_bh = NULL;
1577 struct ocfs2_dinode *fe = NULL;
1578 struct ocfs2_dinode *dirfe;
1579 handle_t *handle = NULL;
1580 struct ocfs2_alloc_context *inode_ac = NULL;
1581 struct ocfs2_alloc_context *data_ac = NULL;
1582 struct ocfs2_alloc_context *xattr_ac = NULL;
1583 int want_clusters = 0;
1584 int xattr_credits = 0;
1585 struct ocfs2_security_xattr_info si = {
1586 .enable = 1,
1587 };
1588 int did_quota = 0, did_quota_inode = 0;
1589 struct ocfs2_dir_lookup_result lookup = { NULL, };
1590
1591 mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1592 dentry, symname, dentry->d_name.len, dentry->d_name.name);
1593
1594 sb = dir->i_sb;
1595 osb = OCFS2_SB(sb);
1596
1597 l = strlen(symname) + 1;
1598
1599 credits = ocfs2_calc_symlink_credits(sb);
1600
1601 /* lock the parent directory */
1602 status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1603 if (status < 0) {
1604 if (status != -ENOENT)
1605 mlog_errno(status);
1606 return status;
1607 }
1608
1609 dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1610 if (!ocfs2_read_links_count(dirfe)) {
1611 /* can't make a file in a deleted directory. */
1612 status = -ENOENT;
1613 goto bail;
1614 }
1615
1616 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1617 dentry->d_name.len);
1618 if (status)
1619 goto bail;
1620
1621 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1622 dentry->d_name.name,
1623 dentry->d_name.len, &lookup);
1624 if (status < 0) {
1625 mlog_errno(status);
1626 goto bail;
1627 }
1628
1629 status = ocfs2_reserve_new_inode(osb, &inode_ac);
1630 if (status < 0) {
1631 if (status != -ENOSPC)
1632 mlog_errno(status);
1633 goto bail;
1634 }
1635
1636 inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1637 if (!inode) {
1638 status = -ENOMEM;
1639 mlog_errno(status);
1640 goto bail;
1641 }
1642
1643 /* get security xattr */
1644 status = ocfs2_init_security_get(inode, dir, &si);
1645 if (status) {
1646 if (status == -EOPNOTSUPP)
1647 si.enable = 0;
1648 else {
1649 mlog_errno(status);
1650 goto bail;
1651 }
1652 }
1653
1654 /* calculate meta data/clusters for setting security xattr */
1655 if (si.enable) {
1656 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1657 &xattr_credits, &xattr_ac);
1658 if (status < 0) {
1659 mlog_errno(status);
1660 goto bail;
1661 }
1662 }
1663
1664 /* don't reserve bitmap space for fast symlinks. */
1665 if (l > ocfs2_fast_symlink_chars(sb))
1666 want_clusters += 1;
1667
1668 status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1669 if (status < 0) {
1670 if (status != -ENOSPC)
1671 mlog_errno(status);
1672 goto bail;
1673 }
1674
1675 handle = ocfs2_start_trans(osb, credits + xattr_credits);
1676 if (IS_ERR(handle)) {
1677 status = PTR_ERR(handle);
1678 handle = NULL;
1679 mlog_errno(status);
1680 goto bail;
1681 }
1682
1683 /* We don't use standard VFS wrapper because we don't want vfs_dq_init
1684 * to be called. */
1685 if (sb_any_quota_active(osb->sb) &&
1686 osb->sb->dq_op->alloc_inode(inode, 1) == NO_QUOTA) {
1687 status = -EDQUOT;
1688 goto bail;
1689 }
1690 did_quota_inode = 1;
1691
1692 status = ocfs2_mknod_locked(osb, dir, inode, dentry,
1693 0, &new_fe_bh, parent_fe_bh, handle,
1694 inode_ac);
1695 if (status < 0) {
1696 mlog_errno(status);
1697 goto bail;
1698 }
1699
1700 fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1701 inode->i_rdev = 0;
1702 newsize = l - 1;
1703 if (l > ocfs2_fast_symlink_chars(sb)) {
1704 u32 offset = 0;
1705
1706 inode->i_op = &ocfs2_symlink_inode_operations;
1707 if (vfs_dq_alloc_space_nodirty(inode,
1708 ocfs2_clusters_to_bytes(osb->sb, 1))) {
1709 status = -EDQUOT;
1710 goto bail;
1711 }
1712 did_quota = 1;
1713 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1714 new_fe_bh,
1715 handle, data_ac, NULL,
1716 NULL);
1717 if (status < 0) {
1718 if (status != -ENOSPC && status != -EINTR) {
1719 mlog(ML_ERROR,
1720 "Failed to extend file to %llu\n",
1721 (unsigned long long)newsize);
1722 mlog_errno(status);
1723 status = -ENOSPC;
1724 }
1725 goto bail;
1726 }
1727 i_size_write(inode, newsize);
1728 inode->i_blocks = ocfs2_inode_sector_count(inode);
1729 } else {
1730 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1731 memcpy((char *) fe->id2.i_symlink, symname, l);
1732 i_size_write(inode, newsize);
1733 inode->i_blocks = 0;
1734 }
1735
1736 status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1737 if (status < 0) {
1738 mlog_errno(status);
1739 goto bail;
1740 }
1741
1742 if (!ocfs2_inode_is_fast_symlink(inode)) {
1743 status = ocfs2_create_symlink_data(osb, handle, inode,
1744 symname);
1745 if (status < 0) {
1746 mlog_errno(status);
1747 goto bail;
1748 }
1749 }
1750
1751 if (si.enable) {
1752 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1753 xattr_ac, data_ac);
1754 if (status < 0) {
1755 mlog_errno(status);
1756 goto bail;
1757 }
1758 }
1759
1760 status = ocfs2_add_entry(handle, dentry, inode,
1761 le64_to_cpu(fe->i_blkno), parent_fe_bh,
1762 &lookup);
1763 if (status < 0) {
1764 mlog_errno(status);
1765 goto bail;
1766 }
1767
1768 status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1769 if (status) {
1770 mlog_errno(status);
1771 goto bail;
1772 }
1773
1774 insert_inode_hash(inode);
1775 dentry->d_op = &ocfs2_dentry_ops;
1776 d_instantiate(dentry, inode);
1777 bail:
1778 if (status < 0 && did_quota)
1779 vfs_dq_free_space_nodirty(inode,
1780 ocfs2_clusters_to_bytes(osb->sb, 1));
1781 if (status < 0 && did_quota_inode)
1782 vfs_dq_free_inode(inode);
1783 if (handle)
1784 ocfs2_commit_trans(osb, handle);
1785
1786 ocfs2_inode_unlock(dir, 1);
1787
1788 brelse(new_fe_bh);
1789 brelse(parent_fe_bh);
1790 kfree(si.name);
1791 kfree(si.value);
1792 ocfs2_free_dir_lookup_result(&lookup);
1793 if (inode_ac)
1794 ocfs2_free_alloc_context(inode_ac);
1795 if (data_ac)
1796 ocfs2_free_alloc_context(data_ac);
1797 if (xattr_ac)
1798 ocfs2_free_alloc_context(xattr_ac);
1799 if ((status < 0) && inode) {
1800 clear_nlink(inode);
1801 iput(inode);
1802 }
1803
1804 mlog_exit(status);
1805
1806 return status;
1807 }
1808
1809 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1810 {
1811 int status, namelen;
1812
1813 mlog_entry_void();
1814
1815 namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1816 (long long)blkno);
1817 if (namelen <= 0) {
1818 if (namelen)
1819 status = namelen;
1820 else
1821 status = -EINVAL;
1822 mlog_errno(status);
1823 goto bail;
1824 }
1825 if (namelen != OCFS2_ORPHAN_NAMELEN) {
1826 status = -EINVAL;
1827 mlog_errno(status);
1828 goto bail;
1829 }
1830
1831 mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
1832 namelen);
1833
1834 status = 0;
1835 bail:
1836 mlog_exit(status);
1837 return status;
1838 }
1839
1840 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1841 struct inode **ret_orphan_dir,
1842 struct inode *inode,
1843 char *name,
1844 struct ocfs2_dir_lookup_result *lookup)
1845 {
1846 struct inode *orphan_dir_inode;
1847 struct buffer_head *orphan_dir_bh = NULL;
1848 int status = 0;
1849
1850 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1851 if (status < 0) {
1852 mlog_errno(status);
1853 return status;
1854 }
1855
1856 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1857 ORPHAN_DIR_SYSTEM_INODE,
1858 osb->slot_num);
1859 if (!orphan_dir_inode) {
1860 status = -ENOENT;
1861 mlog_errno(status);
1862 return status;
1863 }
1864
1865 mutex_lock(&orphan_dir_inode->i_mutex);
1866
1867 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1868 if (status < 0) {
1869 mlog_errno(status);
1870 goto leave;
1871 }
1872
1873 status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1874 orphan_dir_bh, name,
1875 OCFS2_ORPHAN_NAMELEN, lookup);
1876 if (status < 0) {
1877 ocfs2_inode_unlock(orphan_dir_inode, 1);
1878
1879 mlog_errno(status);
1880 goto leave;
1881 }
1882
1883 *ret_orphan_dir = orphan_dir_inode;
1884
1885 leave:
1886 if (status) {
1887 mutex_unlock(&orphan_dir_inode->i_mutex);
1888 iput(orphan_dir_inode);
1889 }
1890
1891 brelse(orphan_dir_bh);
1892
1893 mlog_exit(status);
1894 return status;
1895 }
1896
1897 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1898 handle_t *handle,
1899 struct inode *inode,
1900 struct ocfs2_dinode *fe,
1901 char *name,
1902 struct ocfs2_dir_lookup_result *lookup,
1903 struct inode *orphan_dir_inode)
1904 {
1905 struct buffer_head *orphan_dir_bh = NULL;
1906 int status = 0;
1907 struct ocfs2_dinode *orphan_fe;
1908
1909 mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
1910
1911 status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
1912 if (status < 0) {
1913 mlog_errno(status);
1914 goto leave;
1915 }
1916
1917 status = ocfs2_journal_access_di(handle, orphan_dir_inode, orphan_dir_bh,
1918 OCFS2_JOURNAL_ACCESS_WRITE);
1919 if (status < 0) {
1920 mlog_errno(status);
1921 goto leave;
1922 }
1923
1924 /* we're a cluster, and nlink can change on disk from
1925 * underneath us... */
1926 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1927 if (S_ISDIR(inode->i_mode))
1928 ocfs2_add_links_count(orphan_fe, 1);
1929 orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
1930
1931 status = ocfs2_journal_dirty(handle, orphan_dir_bh);
1932 if (status < 0) {
1933 mlog_errno(status);
1934 goto leave;
1935 }
1936
1937 status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
1938 OCFS2_ORPHAN_NAMELEN, inode,
1939 OCFS2_I(inode)->ip_blkno,
1940 orphan_dir_bh, lookup);
1941 if (status < 0) {
1942 mlog_errno(status);
1943 goto leave;
1944 }
1945
1946 le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
1947
1948 /* Record which orphan dir our inode now resides
1949 * in. delete_inode will use this to determine which orphan
1950 * dir to lock. */
1951 fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
1952
1953 mlog(0, "Inode %llu orphaned in slot %d\n",
1954 (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
1955
1956 leave:
1957 brelse(orphan_dir_bh);
1958
1959 mlog_exit(status);
1960 return status;
1961 }
1962
1963 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
1964 int ocfs2_orphan_del(struct ocfs2_super *osb,
1965 handle_t *handle,
1966 struct inode *orphan_dir_inode,
1967 struct inode *inode,
1968 struct buffer_head *orphan_dir_bh)
1969 {
1970 char name[OCFS2_ORPHAN_NAMELEN + 1];
1971 struct ocfs2_dinode *orphan_fe;
1972 int status = 0;
1973 struct ocfs2_dir_lookup_result lookup = { NULL, };
1974
1975 mlog_entry_void();
1976
1977 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1978 if (status < 0) {
1979 mlog_errno(status);
1980 goto leave;
1981 }
1982
1983 mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
1984 name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
1985 OCFS2_ORPHAN_NAMELEN);
1986
1987 /* find it's spot in the orphan directory */
1988 status = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN, orphan_dir_inode,
1989 &lookup);
1990 if (status) {
1991 mlog_errno(status);
1992 goto leave;
1993 }
1994
1995 /* remove it from the orphan directory */
1996 status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
1997 if (status < 0) {
1998 mlog_errno(status);
1999 goto leave;
2000 }
2001
2002 status = ocfs2_journal_access_di(handle,orphan_dir_inode, orphan_dir_bh,
2003 OCFS2_JOURNAL_ACCESS_WRITE);
2004 if (status < 0) {
2005 mlog_errno(status);
2006 goto leave;
2007 }
2008
2009 /* do the i_nlink dance! :) */
2010 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2011 if (S_ISDIR(inode->i_mode))
2012 ocfs2_add_links_count(orphan_fe, -1);
2013 orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
2014
2015 status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2016 if (status < 0) {
2017 mlog_errno(status);
2018 goto leave;
2019 }
2020
2021 leave:
2022 ocfs2_free_dir_lookup_result(&lookup);
2023
2024 mlog_exit(status);
2025 return status;
2026 }
2027
2028 const struct inode_operations ocfs2_dir_iops = {
2029 .create = ocfs2_create,
2030 .lookup = ocfs2_lookup,
2031 .link = ocfs2_link,
2032 .unlink = ocfs2_unlink,
2033 .rmdir = ocfs2_unlink,
2034 .symlink = ocfs2_symlink,
2035 .mkdir = ocfs2_mkdir,
2036 .mknod = ocfs2_mknod,
2037 .rename = ocfs2_rename,
2038 .setattr = ocfs2_setattr,
2039 .getattr = ocfs2_getattr,
2040 .permission = ocfs2_permission,
2041 .setxattr = generic_setxattr,
2042 .getxattr = generic_getxattr,
2043 .listxattr = ocfs2_listxattr,
2044 .removexattr = generic_removexattr,
2045 };
Support for the Chinese Samsung S3C2440 based development boards