?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ocfs2: Call refcount tree remove process properly.
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / ocfs2 / namei.c
blobc07217ad8796bf5dc74b00a38223763a7074d863
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_nested(dir, NULL, 0, OI_LS_PARENT);
122 if (status < 0) {
123 if (status != -ENOENT)
124 mlog_errno(status);
125 ret = ERR_PTR(status);
126 goto bail;
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, INODE_CACHE(dir),
379 parent_fe_bh,
380 OCFS2_JOURNAL_ACCESS_WRITE);
381 if (status < 0) {
382 mlog_errno(status);
383 goto leave;
384 }
385 ocfs2_add_links_count(dirfe, 1);
386 status = ocfs2_journal_dirty(handle, parent_fe_bh);
387 if (status < 0) {
388 mlog_errno(status);
389 goto leave;
390 }
391 inc_nlink(dir);
392 }
393
394 status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
395 meta_ac, data_ac);
396 if (status < 0) {
397 mlog_errno(status);
398 goto leave;
399 }
400
401 if (si.enable) {
402 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
403 meta_ac, data_ac);
404 if (status < 0) {
405 mlog_errno(status);
406 goto leave;
407 }
408 }
409
410 status = ocfs2_add_entry(handle, dentry, inode,
411 OCFS2_I(inode)->ip_blkno, parent_fe_bh,
412 &lookup);
413 if (status < 0) {
414 mlog_errno(status);
415 goto leave;
416 }
417
418 status = ocfs2_dentry_attach_lock(dentry, inode,
419 OCFS2_I(dir)->ip_blkno);
420 if (status) {
421 mlog_errno(status);
422 goto leave;
423 }
424
425 insert_inode_hash(inode);
426 dentry->d_op = &ocfs2_dentry_ops;
427 d_instantiate(dentry, inode);
428 status = 0;
429 leave:
430 if (status < 0 && did_quota_inode)
431 vfs_dq_free_inode(inode);
432 if (handle)
433 ocfs2_commit_trans(osb, handle);
434
435 ocfs2_inode_unlock(dir, 1);
436
437 if (status == -ENOSPC)
438 mlog(0, "Disk is full\n");
439
440 brelse(new_fe_bh);
441 brelse(parent_fe_bh);
442 kfree(si.name);
443 kfree(si.value);
444
445 ocfs2_free_dir_lookup_result(&lookup);
446
447 if ((status < 0) && inode) {
448 clear_nlink(inode);
449 iput(inode);
450 }
451
452 if (inode_ac)
453 ocfs2_free_alloc_context(inode_ac);
454
455 if (data_ac)
456 ocfs2_free_alloc_context(data_ac);
457
458 if (meta_ac)
459 ocfs2_free_alloc_context(meta_ac);
460
461 mlog_exit(status);
462
463 return status;
464 }
465
466 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
467 struct inode *dir,
468 struct inode *inode,
469 struct dentry *dentry,
470 dev_t dev,
471 struct buffer_head **new_fe_bh,
472 struct buffer_head *parent_fe_bh,
473 handle_t *handle,
474 struct ocfs2_alloc_context *inode_ac)
475 {
476 int status = 0;
477 struct ocfs2_dinode *fe = NULL;
478 struct ocfs2_extent_list *fel;
479 u64 fe_blkno = 0;
480 u16 suballoc_bit;
481 u16 feat;
482
483 mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry,
484 inode->i_mode, (unsigned long)dev, dentry->d_name.len,
485 dentry->d_name.name);
486
487 *new_fe_bh = NULL;
488
489 status = ocfs2_claim_new_inode(osb, handle, dir, parent_fe_bh,
490 inode_ac, &suballoc_bit, &fe_blkno);
491 if (status < 0) {
492 mlog_errno(status);
493 goto leave;
494 }
495
496 /* populate as many fields early on as possible - many of
497 * these are used by the support functions here and in
498 * callers. */
499 inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
500 OCFS2_I(inode)->ip_blkno = fe_blkno;
501 spin_lock(&osb->osb_lock);
502 inode->i_generation = osb->s_next_generation++;
503 spin_unlock(&osb->osb_lock);
504
505 *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
506 if (!*new_fe_bh) {
507 status = -EIO;
508 mlog_errno(status);
509 goto leave;
510 }
511 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
512
513 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
514 *new_fe_bh,
515 OCFS2_JOURNAL_ACCESS_CREATE);
516 if (status < 0) {
517 mlog_errno(status);
518 goto leave;
519 }
520
521 fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
522 memset(fe, 0, osb->sb->s_blocksize);
523
524 fe->i_generation = cpu_to_le32(inode->i_generation);
525 fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
526 fe->i_blkno = cpu_to_le64(fe_blkno);
527 fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
528 fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
529 fe->i_uid = cpu_to_le32(inode->i_uid);
530 fe->i_gid = cpu_to_le32(inode->i_gid);
531 fe->i_mode = cpu_to_le16(inode->i_mode);
532 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
533 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
534
535 ocfs2_set_links_count(fe, inode->i_nlink);
536
537 fe->i_last_eb_blk = 0;
538 strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
539 le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
540 fe->i_atime = fe->i_ctime = fe->i_mtime =
541 cpu_to_le64(CURRENT_TIME.tv_sec);
542 fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
543 cpu_to_le32(CURRENT_TIME.tv_nsec);
544 fe->i_dtime = 0;
545
546 /*
547 * If supported, directories start with inline data. If inline
548 * isn't supported, but indexing is, we start them as indexed.
549 */
550 feat = le16_to_cpu(fe->i_dyn_features);
551 if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
552 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
553
554 fe->id2.i_data.id_count = cpu_to_le16(
555 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
556 } else {
557 fel = &fe->id2.i_list;
558 fel->l_tree_depth = 0;
559 fel->l_next_free_rec = 0;
560 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
561 }
562
563 status = ocfs2_journal_dirty(handle, *new_fe_bh);
564 if (status < 0) {
565 mlog_errno(status);
566 goto leave;
567 }
568
569 ocfs2_populate_inode(inode, fe, 1);
570 ocfs2_ci_set_new(osb, INODE_CACHE(inode));
571 if (!ocfs2_mount_local(osb)) {
572 status = ocfs2_create_new_inode_locks(inode);
573 if (status < 0)
574 mlog_errno(status);
575 }
576
577 status = 0; /* error in ocfs2_create_new_inode_locks is not
578 * critical */
579
580 leave:
581 if (status < 0) {
582 if (*new_fe_bh) {
583 brelse(*new_fe_bh);
584 *new_fe_bh = NULL;
585 }
586 }
587
588 mlog_exit(status);
589 return status;
590 }
591
592 static int ocfs2_mkdir(struct inode *dir,
593 struct dentry *dentry,
594 int mode)
595 {
596 int ret;
597
598 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
599 dentry->d_name.len, dentry->d_name.name);
600 ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
601 mlog_exit(ret);
602
603 return ret;
604 }
605
606 static int ocfs2_create(struct inode *dir,
607 struct dentry *dentry,
608 int mode,
609 struct nameidata *nd)
610 {
611 int ret;
612
613 mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
614 dentry->d_name.len, dentry->d_name.name);
615 ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
616 mlog_exit(ret);
617
618 return ret;
619 }
620
621 static int ocfs2_link(struct dentry *old_dentry,
622 struct inode *dir,
623 struct dentry *dentry)
624 {
625 handle_t *handle;
626 struct inode *inode = old_dentry->d_inode;
627 int err;
628 struct buffer_head *fe_bh = NULL;
629 struct buffer_head *parent_fe_bh = NULL;
630 struct ocfs2_dinode *fe = NULL;
631 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
632 struct ocfs2_dir_lookup_result lookup = { NULL, };
633
634 mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
635 old_dentry->d_name.len, old_dentry->d_name.name,
636 dentry->d_name.len, dentry->d_name.name);
637
638 if (S_ISDIR(inode->i_mode))
639 return -EPERM;
640
641 err = ocfs2_inode_lock_nested(dir, &parent_fe_bh, 1, OI_LS_PARENT);
642 if (err < 0) {
643 if (err != -ENOENT)
644 mlog_errno(err);
645 return err;
646 }
647
648 if (!dir->i_nlink) {
649 err = -ENOENT;
650 goto out;
651 }
652
653 err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
654 dentry->d_name.len);
655 if (err)
656 goto out;
657
658 err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
659 dentry->d_name.name,
660 dentry->d_name.len, &lookup);
661 if (err < 0) {
662 mlog_errno(err);
663 goto out;
664 }
665
666 err = ocfs2_inode_lock(inode, &fe_bh, 1);
667 if (err < 0) {
668 if (err != -ENOENT)
669 mlog_errno(err);
670 goto out;
671 }
672
673 fe = (struct ocfs2_dinode *) fe_bh->b_data;
674 if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
675 err = -EMLINK;
676 goto out_unlock_inode;
677 }
678
679 handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
680 if (IS_ERR(handle)) {
681 err = PTR_ERR(handle);
682 handle = NULL;
683 mlog_errno(err);
684 goto out_unlock_inode;
685 }
686
687 err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
688 OCFS2_JOURNAL_ACCESS_WRITE);
689 if (err < 0) {
690 mlog_errno(err);
691 goto out_commit;
692 }
693
694 inc_nlink(inode);
695 inode->i_ctime = CURRENT_TIME;
696 ocfs2_set_links_count(fe, inode->i_nlink);
697 fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
698 fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
699
700 err = ocfs2_journal_dirty(handle, fe_bh);
701 if (err < 0) {
702 ocfs2_add_links_count(fe, -1);
703 drop_nlink(inode);
704 mlog_errno(err);
705 goto out_commit;
706 }
707
708 err = ocfs2_add_entry(handle, dentry, inode,
709 OCFS2_I(inode)->ip_blkno,
710 parent_fe_bh, &lookup);
711 if (err) {
712 ocfs2_add_links_count(fe, -1);
713 drop_nlink(inode);
714 mlog_errno(err);
715 goto out_commit;
716 }
717
718 err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
719 if (err) {
720 mlog_errno(err);
721 goto out_commit;
722 }
723
724 atomic_inc(&inode->i_count);
725 dentry->d_op = &ocfs2_dentry_ops;
726 d_instantiate(dentry, inode);
727
728 out_commit:
729 ocfs2_commit_trans(osb, handle);
730 out_unlock_inode:
731 ocfs2_inode_unlock(inode, 1);
732
733 out:
734 ocfs2_inode_unlock(dir, 1);
735
736 brelse(fe_bh);
737 brelse(parent_fe_bh);
738
739 ocfs2_free_dir_lookup_result(&lookup);
740
741 mlog_exit(err);
742
743 return err;
744 }
745
746 /*
747 * Takes and drops an exclusive lock on the given dentry. This will
748 * force other nodes to drop it.
749 */
750 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
751 {
752 int ret;
753
754 ret = ocfs2_dentry_lock(dentry, 1);
755 if (ret)
756 mlog_errno(ret);
757 else
758 ocfs2_dentry_unlock(dentry, 1);
759
760 return ret;
761 }
762
763 static inline int inode_is_unlinkable(struct inode *inode)
764 {
765 if (S_ISDIR(inode->i_mode)) {
766 if (inode->i_nlink == 2)
767 return 1;
768 return 0;
769 }
770
771 if (inode->i_nlink == 1)
772 return 1;
773 return 0;
774 }
775
776 static int ocfs2_unlink(struct inode *dir,
777 struct dentry *dentry)
778 {
779 int status;
780 int child_locked = 0;
781 struct inode *inode = dentry->d_inode;
782 struct inode *orphan_dir = NULL;
783 struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
784 u64 blkno;
785 struct ocfs2_dinode *fe = NULL;
786 struct buffer_head *fe_bh = NULL;
787 struct buffer_head *parent_node_bh = NULL;
788 handle_t *handle = NULL;
789 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
790 struct ocfs2_dir_lookup_result lookup = { NULL, };
791 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
792
793 mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
794 dentry->d_name.len, dentry->d_name.name);
795
796 BUG_ON(dentry->d_parent->d_inode != dir);
797
798 mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
799
800 if (inode == osb->root_inode) {
801 mlog(0, "Cannot delete the root directory\n");
802 return -EPERM;
803 }
804
805 status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
806 OI_LS_PARENT);
807 if (status < 0) {
808 if (status != -ENOENT)
809 mlog_errno(status);
810 return status;
811 }
812
813 status = ocfs2_find_files_on_disk(dentry->d_name.name,
814 dentry->d_name.len, &blkno, dir,
815 &lookup);
816 if (status < 0) {
817 if (status != -ENOENT)
818 mlog_errno(status);
819 goto leave;
820 }
821
822 if (OCFS2_I(inode)->ip_blkno != blkno) {
823 status = -ENOENT;
824
825 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
826 (unsigned long long)OCFS2_I(inode)->ip_blkno,
827 (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
828 goto leave;
829 }
830
831 status = ocfs2_inode_lock(inode, &fe_bh, 1);
832 if (status < 0) {
833 if (status != -ENOENT)
834 mlog_errno(status);
835 goto leave;
836 }
837 child_locked = 1;
838
839 if (S_ISDIR(inode->i_mode)) {
840 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
841 status = -ENOTEMPTY;
842 goto leave;
843 }
844 }
845
846 status = ocfs2_remote_dentry_delete(dentry);
847 if (status < 0) {
848 /* This remote delete should succeed under all normal
849 * circumstances. */
850 mlog_errno(status);
851 goto leave;
852 }
853
854 if (inode_is_unlinkable(inode)) {
855 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir, inode,
856 orphan_name, &orphan_insert);
857 if (status < 0) {
858 mlog_errno(status);
859 goto leave;
860 }
861 }
862
863 handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
864 if (IS_ERR(handle)) {
865 status = PTR_ERR(handle);
866 handle = NULL;
867 mlog_errno(status);
868 goto leave;
869 }
870
871 status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
872 OCFS2_JOURNAL_ACCESS_WRITE);
873 if (status < 0) {
874 mlog_errno(status);
875 goto leave;
876 }
877
878 fe = (struct ocfs2_dinode *) fe_bh->b_data;
879
880 if (inode_is_unlinkable(inode)) {
881 status = ocfs2_orphan_add(osb, handle, inode, fe, orphan_name,
882 &orphan_insert, orphan_dir);
883 if (status < 0) {
884 mlog_errno(status);
885 goto leave;
886 }
887 }
888
889 /* delete the name from the parent dir */
890 status = ocfs2_delete_entry(handle, dir, &lookup);
891 if (status < 0) {
892 mlog_errno(status);
893 goto leave;
894 }
895
896 if (S_ISDIR(inode->i_mode))
897 drop_nlink(inode);
898 drop_nlink(inode);
899 ocfs2_set_links_count(fe, inode->i_nlink);
900
901 status = ocfs2_journal_dirty(handle, fe_bh);
902 if (status < 0) {
903 mlog_errno(status);
904 goto leave;
905 }
906
907 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
908 if (S_ISDIR(inode->i_mode))
909 drop_nlink(dir);
910
911 status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
912 if (status < 0) {
913 mlog_errno(status);
914 if (S_ISDIR(inode->i_mode))
915 inc_nlink(dir);
916 }
917
918 leave:
919 if (handle)
920 ocfs2_commit_trans(osb, handle);
921
922 if (child_locked)
923 ocfs2_inode_unlock(inode, 1);
924
925 ocfs2_inode_unlock(dir, 1);
926
927 if (orphan_dir) {
928 /* This was locked for us in ocfs2_prepare_orphan_dir() */
929 ocfs2_inode_unlock(orphan_dir, 1);
930 mutex_unlock(&orphan_dir->i_mutex);
931 iput(orphan_dir);
932 }
933
934 brelse(fe_bh);
935 brelse(parent_node_bh);
936
937 ocfs2_free_dir_lookup_result(&orphan_insert);
938 ocfs2_free_dir_lookup_result(&lookup);
939
940 mlog_exit(status);
941
942 return status;
943 }
944
945 /*
946 * The only place this should be used is rename!
947 * if they have the same id, then the 1st one is the only one locked.
948 */
949 static int ocfs2_double_lock(struct ocfs2_super *osb,
950 struct buffer_head **bh1,
951 struct inode *inode1,
952 struct buffer_head **bh2,
953 struct inode *inode2)
954 {
955 int status;
956 struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
957 struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
958 struct buffer_head **tmpbh;
959 struct inode *tmpinode;
960
961 mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
962 (unsigned long long)oi1->ip_blkno,
963 (unsigned long long)oi2->ip_blkno);
964
965 if (*bh1)
966 *bh1 = NULL;
967 if (*bh2)
968 *bh2 = NULL;
969
970 /* we always want to lock the one with the lower lockid first. */
971 if (oi1->ip_blkno != oi2->ip_blkno) {
972 if (oi1->ip_blkno < oi2->ip_blkno) {
973 /* switch id1 and id2 around */
974 mlog(0, "switching them around...\n");
975 tmpbh = bh2;
976 bh2 = bh1;
977 bh1 = tmpbh;
978
979 tmpinode = inode2;
980 inode2 = inode1;
981 inode1 = tmpinode;
982 }
983 /* lock id2 */
984 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
985 OI_LS_RENAME1);
986 if (status < 0) {
987 if (status != -ENOENT)
988 mlog_errno(status);
989 goto bail;
990 }
991 }
992
993 /* lock id1 */
994 status = ocfs2_inode_lock_nested(inode1, bh1, 1, OI_LS_RENAME2);
995 if (status < 0) {
996 /*
997 * An error return must mean that no cluster locks
998 * were held on function exit.
999 */
1000 if (oi1->ip_blkno != oi2->ip_blkno)
1001 ocfs2_inode_unlock(inode2, 1);
1002
1003 if (status != -ENOENT)
1004 mlog_errno(status);
1005 }
1006
1007 bail:
1008 mlog_exit(status);
1009 return status;
1010 }
1011
1012 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1013 {
1014 ocfs2_inode_unlock(inode1, 1);
1015
1016 if (inode1 != inode2)
1017 ocfs2_inode_unlock(inode2, 1);
1018 }
1019
1020 static int ocfs2_rename(struct inode *old_dir,
1021 struct dentry *old_dentry,
1022 struct inode *new_dir,
1023 struct dentry *new_dentry)
1024 {
1025 int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1026 int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1027 struct inode *old_inode = old_dentry->d_inode;
1028 struct inode *new_inode = new_dentry->d_inode;
1029 struct inode *orphan_dir = NULL;
1030 struct ocfs2_dinode *newfe = NULL;
1031 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1032 struct buffer_head *newfe_bh = NULL;
1033 struct buffer_head *old_inode_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, };
1046
1047 /* At some point it might be nice to break this function up a
1048 * bit. */
1049
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);
1054
1055 osb = OCFS2_SB(old_dir->i_sb);
1056
1057 if (new_inode) {
1058 if (!igrab(new_inode))
1059 BUG();
1060 }
1061
1062 /* Assume a directory hierarchy thusly:
1063 * a/b/c
1064 * a/d
1065 * a,b,c, and d are all directories.
1066 *
1067 * from cwd of 'a' on both nodes:
1068 * node1: mv b/c d
1069 * node2: mv d b/c
1070 *
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;
1078 }
1079 rename_lock = 1;
1080 }
1081
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;
1088 }
1089 parents_locked = 1;
1090
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;
1101 }
1102 }
1103
1104 /*
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.
1109 */
1110 status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1111 OI_LS_PARENT);
1112 if (status < 0) {
1113 if (status != -ENOENT)
1114 mlog_errno(status);
1115 goto bail;
1116 }
1117 old_child_locked = 1;
1118
1119 status = ocfs2_remote_dentry_delete(old_dentry);
1120 if (status < 0) {
1121 mlog_errno(status);
1122 goto bail;
1123 }
1124
1125 if (S_ISDIR(old_inode->i_mode)) {
1126 u64 old_inode_parent;
1127
1128 update_dot_dot = 1;
1129 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1130 old_inode,
1131 &old_inode_dot_dot_res);
1132 if (status) {
1133 status = -EIO;
1134 goto bail;
1135 }
1136
1137 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1138 status = -EIO;
1139 goto bail;
1140 }
1141
1142 if (!new_inode && new_dir != old_dir &&
1143 new_dir->i_nlink >= ocfs2_link_max(osb)) {
1144 status = -EMLINK;
1145 goto bail;
1146 }
1147 }
1148
1149 status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1150 old_dentry->d_name.len,
1151 &old_de_ino);
1152 if (status) {
1153 status = -ENOENT;
1154 goto bail;
1155 }
1156
1157 /*
1158 * Check for inode number is _not_ due to possible IO errors.
1159 * We might rmdir the source, keep it as pwd of some process
1160 * and merrily kill the link to whatever was created under the
1161 * same name. Goodbye sticky bit ;-<
1162 */
1163 if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1164 status = -ENOENT;
1165 goto bail;
1166 }
1167
1168 /* check if the target already exists (in which case we need
1169 * to delete it */
1170 status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1171 new_dentry->d_name.len,
1172 &newfe_blkno, new_dir,
1173 &target_lookup_res);
1174 /* The only error we allow here is -ENOENT because the new
1175 * file not existing is perfectly valid. */
1176 if ((status < 0) && (status != -ENOENT)) {
1177 /* If we cannot find the file specified we should just */
1178 /* return the error... */
1179 mlog_errno(status);
1180 goto bail;
1181 }
1182 if (status == 0)
1183 target_exists = 1;
1184
1185 if (!target_exists && new_inode) {
1186 /*
1187 * Target was unlinked by another node while we were
1188 * waiting to get to ocfs2_rename(). There isn't
1189 * anything we can do here to help the situation, so
1190 * bubble up the appropriate error.
1191 */
1192 status = -ENOENT;
1193 goto bail;
1194 }
1195
1196 /* In case we need to overwrite an existing file, we blow it
1197 * away first */
1198 if (target_exists) {
1199 /* VFS didn't think there existed an inode here, but
1200 * someone else in the cluster must have raced our
1201 * rename to create one. Today we error cleanly, in
1202 * the future we should consider calling iget to build
1203 * a new struct inode for this entry. */
1204 if (!new_inode) {
1205 status = -EACCES;
1206
1207 mlog(0, "We found an inode for name %.*s but VFS "
1208 "didn't give us one.\n", new_dentry->d_name.len,
1209 new_dentry->d_name.name);
1210 goto bail;
1211 }
1212
1213 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1214 status = -EACCES;
1215
1216 mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1217 (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1218 (unsigned long long)newfe_blkno,
1219 OCFS2_I(new_inode)->ip_flags);
1220 goto bail;
1221 }
1222
1223 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1224 if (status < 0) {
1225 if (status != -ENOENT)
1226 mlog_errno(status);
1227 goto bail;
1228 }
1229 new_child_locked = 1;
1230
1231 status = ocfs2_remote_dentry_delete(new_dentry);
1232 if (status < 0) {
1233 mlog_errno(status);
1234 goto bail;
1235 }
1236
1237 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1238
1239 mlog(0, "aha rename over existing... new_blkno=%llu "
1240 "newfebh=%p bhblocknr=%llu\n",
1241 (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1242 (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1243
1244 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1245 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1246 new_inode,
1247 orphan_name,
1248 &orphan_insert);
1249 if (status < 0) {
1250 mlog_errno(status);
1251 goto bail;
1252 }
1253 }
1254 } else {
1255 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1256
1257 status = ocfs2_check_dir_for_entry(new_dir,
1258 new_dentry->d_name.name,
1259 new_dentry->d_name.len);
1260 if (status)
1261 goto bail;
1262
1263 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1264 new_dentry->d_name.name,
1265 new_dentry->d_name.len,
1266 &target_insert);
1267 if (status < 0) {
1268 mlog_errno(status);
1269 goto bail;
1270 }
1271 }
1272
1273 handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1274 if (IS_ERR(handle)) {
1275 status = PTR_ERR(handle);
1276 handle = NULL;
1277 mlog_errno(status);
1278 goto bail;
1279 }
1280
1281 if (target_exists) {
1282 if (S_ISDIR(new_inode->i_mode)) {
1283 if (new_inode->i_nlink != 2 ||
1284 !ocfs2_empty_dir(new_inode)) {
1285 status = -ENOTEMPTY;
1286 goto bail;
1287 }
1288 }
1289 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1290 newfe_bh,
1291 OCFS2_JOURNAL_ACCESS_WRITE);
1292 if (status < 0) {
1293 mlog_errno(status);
1294 goto bail;
1295 }
1296
1297 if (S_ISDIR(new_inode->i_mode) ||
1298 (ocfs2_read_links_count(newfe) == 1)) {
1299 status = ocfs2_orphan_add(osb, handle, new_inode,
1300 newfe, orphan_name,
1301 &orphan_insert, orphan_dir);
1302 if (status < 0) {
1303 mlog_errno(status);
1304 goto bail;
1305 }
1306 }
1307
1308 /* change the dirent to point to the correct inode */
1309 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1310 old_inode);
1311 if (status < 0) {
1312 mlog_errno(status);
1313 goto bail;
1314 }
1315 new_dir->i_version++;
1316
1317 if (S_ISDIR(new_inode->i_mode))
1318 ocfs2_set_links_count(newfe, 0);
1319 else
1320 ocfs2_add_links_count(newfe, -1);
1321
1322 status = ocfs2_journal_dirty(handle, newfe_bh);
1323 if (status < 0) {
1324 mlog_errno(status);
1325 goto bail;
1326 }
1327 } else {
1328 /* if the name was not found in new_dir, add it now */
1329 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1330 OCFS2_I(old_inode)->ip_blkno,
1331 new_dir_bh, &target_insert);
1332 }
1333
1334 old_inode->i_ctime = CURRENT_TIME;
1335 mark_inode_dirty(old_inode);
1336
1337 status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1338 old_inode_bh,
1339 OCFS2_JOURNAL_ACCESS_WRITE);
1340 if (status >= 0) {
1341 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1342
1343 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1344 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1345
1346 status = ocfs2_journal_dirty(handle, old_inode_bh);
1347 if (status < 0)
1348 mlog_errno(status);
1349 } else
1350 mlog_errno(status);
1351
1352 /*
1353 * Now that the name has been added to new_dir, remove the old name.
1354 *
1355 * We don't keep any directory entry context around until now
1356 * because the insert might have changed the type of directory
1357 * we're dealing with.
1358 */
1359 status = ocfs2_find_entry(old_dentry->d_name.name,
1360 old_dentry->d_name.len, old_dir,
1361 &old_entry_lookup);
1362 if (status)
1363 goto bail;
1364
1365 status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1366 if (status < 0) {
1367 mlog_errno(status);
1368 goto bail;
1369 }
1370
1371 if (new_inode) {
1372 new_inode->i_nlink--;
1373 new_inode->i_ctime = CURRENT_TIME;
1374 }
1375 old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1376
1377 if (update_dot_dot) {
1378 status = ocfs2_update_entry(old_inode, handle,
1379 &old_inode_dot_dot_res, new_dir);
1380 old_dir->i_nlink--;
1381 if (new_inode) {
1382 new_inode->i_nlink--;
1383 } else {
1384 inc_nlink(new_dir);
1385 mark_inode_dirty(new_dir);
1386 }
1387 }
1388 mark_inode_dirty(old_dir);
1389 ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1390 if (new_inode) {
1391 mark_inode_dirty(new_inode);
1392 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1393 }
1394
1395 if (old_dir != new_dir) {
1396 /* Keep the same times on both directories.*/
1397 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1398
1399 /*
1400 * This will also pick up the i_nlink change from the
1401 * block above.
1402 */
1403 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1404 }
1405
1406 if (old_dir_nlink != old_dir->i_nlink) {
1407 if (!old_dir_bh) {
1408 mlog(ML_ERROR, "need to change nlink for old dir "
1409 "%llu from %d to %d but bh is NULL!\n",
1410 (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1411 (int)old_dir_nlink, old_dir->i_nlink);
1412 } else {
1413 struct ocfs2_dinode *fe;
1414 status = ocfs2_journal_access_di(handle,
1415 INODE_CACHE(old_dir),
1416 old_dir_bh,
1417 OCFS2_JOURNAL_ACCESS_WRITE);
1418 fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1419 ocfs2_set_links_count(fe, old_dir->i_nlink);
1420 status = ocfs2_journal_dirty(handle, old_dir_bh);
1421 }
1422 }
1423 ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1424 status = 0;
1425 bail:
1426 if (rename_lock)
1427 ocfs2_rename_unlock(osb);
1428
1429 if (handle)
1430 ocfs2_commit_trans(osb, handle);
1431
1432 if (parents_locked)
1433 ocfs2_double_unlock(old_dir, new_dir);
1434
1435 if (old_child_locked)
1436 ocfs2_inode_unlock(old_inode, 1);
1437
1438 if (new_child_locked)
1439 ocfs2_inode_unlock(new_inode, 1);
1440
1441 if (orphan_dir) {
1442 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1443 ocfs2_inode_unlock(orphan_dir, 1);
1444 mutex_unlock(&orphan_dir->i_mutex);
1445 iput(orphan_dir);
1446 }
1447
1448 if (new_inode)
1449 sync_mapping_buffers(old_inode->i_mapping);
1450
1451 if (new_inode)
1452 iput(new_inode);
1453
1454 ocfs2_free_dir_lookup_result(&target_lookup_res);
1455 ocfs2_free_dir_lookup_result(&old_entry_lookup);
1456 ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1457 ocfs2_free_dir_lookup_result(&orphan_insert);
1458 ocfs2_free_dir_lookup_result(&target_insert);
1459
1460 brelse(newfe_bh);
1461 brelse(old_inode_bh);
1462 brelse(old_dir_bh);
1463 brelse(new_dir_bh);
1464
1465 mlog_exit(status);
1466
1467 return status;
1468 }
1469
1470 /*
1471 * we expect i_size = strlen(symname). Copy symname into the file
1472 * data, including the null terminator.
1473 */
1474 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1475 handle_t *handle,
1476 struct inode *inode,
1477 const char *symname)
1478 {
1479 struct buffer_head **bhs = NULL;
1480 const char *c;
1481 struct super_block *sb = osb->sb;
1482 u64 p_blkno, p_blocks;
1483 int virtual, blocks, status, i, bytes_left;
1484
1485 bytes_left = i_size_read(inode) + 1;
1486 /* we can't trust i_blocks because we're actually going to
1487 * write i_size + 1 bytes. */
1488 blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1489
1490 mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1491 (unsigned long long)inode->i_blocks,
1492 i_size_read(inode), blocks);
1493
1494 /* Sanity check -- make sure we're going to fit. */
1495 if (bytes_left >
1496 ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1497 status = -EIO;
1498 mlog_errno(status);
1499 goto bail;
1500 }
1501
1502 bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1503 if (!bhs) {
1504 status = -ENOMEM;
1505 mlog_errno(status);
1506 goto bail;
1507 }
1508
1509 status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1510 NULL);
1511 if (status < 0) {
1512 mlog_errno(status);
1513 goto bail;
1514 }
1515
1516 /* links can never be larger than one cluster so we know this
1517 * is all going to be contiguous, but do a sanity check
1518 * anyway. */
1519 if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1520 status = -EIO;
1521 mlog_errno(status);
1522 goto bail;
1523 }
1524
1525 virtual = 0;
1526 while(bytes_left > 0) {
1527 c = &symname[virtual * sb->s_blocksize];
1528
1529 bhs[virtual] = sb_getblk(sb, p_blkno);
1530 if (!bhs[virtual]) {
1531 status = -ENOMEM;
1532 mlog_errno(status);
1533 goto bail;
1534 }
1535 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1536 bhs[virtual]);
1537
1538 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1539 bhs[virtual],
1540 OCFS2_JOURNAL_ACCESS_CREATE);
1541 if (status < 0) {
1542 mlog_errno(status);
1543 goto bail;
1544 }
1545
1546 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1547
1548 memcpy(bhs[virtual]->b_data, c,
1549 (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1550 bytes_left);
1551
1552 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1553 if (status < 0) {
1554 mlog_errno(status);
1555 goto bail;
1556 }
1557
1558 virtual++;
1559 p_blkno++;
1560 bytes_left -= sb->s_blocksize;
1561 }
1562
1563 status = 0;
1564 bail:
1565
1566 if (bhs) {
1567 for(i = 0; i < blocks; i++)
1568 brelse(bhs[i]);
1569 kfree(bhs);
1570 }
1571
1572 mlog_exit(status);
1573 return status;
1574 }
1575
1576 static int ocfs2_symlink(struct inode *dir,
1577 struct dentry *dentry,
1578 const char *symname)
1579 {
1580 int status, l, credits;
1581 u64 newsize;
1582 struct ocfs2_super *osb = NULL;
1583 struct inode *inode = NULL;
1584 struct super_block *sb;
1585 struct buffer_head *new_fe_bh = NULL;
1586 struct buffer_head *parent_fe_bh = NULL;
1587 struct ocfs2_dinode *fe = NULL;
1588 struct ocfs2_dinode *dirfe;
1589 handle_t *handle = NULL;
1590 struct ocfs2_alloc_context *inode_ac = NULL;
1591 struct ocfs2_alloc_context *data_ac = NULL;
1592 struct ocfs2_alloc_context *xattr_ac = NULL;
1593 int want_clusters = 0;
1594 int xattr_credits = 0;
1595 struct ocfs2_security_xattr_info si = {
1596 .enable = 1,
1597 };
1598 int did_quota = 0, did_quota_inode = 0;
1599 struct ocfs2_dir_lookup_result lookup = { NULL, };
1600
1601 mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1602 dentry, symname, dentry->d_name.len, dentry->d_name.name);
1603
1604 sb = dir->i_sb;
1605 osb = OCFS2_SB(sb);
1606
1607 l = strlen(symname) + 1;
1608
1609 credits = ocfs2_calc_symlink_credits(sb);
1610
1611 /* lock the parent directory */
1612 status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1613 if (status < 0) {
1614 if (status != -ENOENT)
1615 mlog_errno(status);
1616 return status;
1617 }
1618
1619 dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1620 if (!ocfs2_read_links_count(dirfe)) {
1621 /* can't make a file in a deleted directory. */
1622 status = -ENOENT;
1623 goto bail;
1624 }
1625
1626 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1627 dentry->d_name.len);
1628 if (status)
1629 goto bail;
1630
1631 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1632 dentry->d_name.name,
1633 dentry->d_name.len, &lookup);
1634 if (status < 0) {
1635 mlog_errno(status);
1636 goto bail;
1637 }
1638
1639 status = ocfs2_reserve_new_inode(osb, &inode_ac);
1640 if (status < 0) {
1641 if (status != -ENOSPC)
1642 mlog_errno(status);
1643 goto bail;
1644 }
1645
1646 inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1647 if (!inode) {
1648 status = -ENOMEM;
1649 mlog_errno(status);
1650 goto bail;
1651 }
1652
1653 /* get security xattr */
1654 status = ocfs2_init_security_get(inode, dir, &si);
1655 if (status) {
1656 if (status == -EOPNOTSUPP)
1657 si.enable = 0;
1658 else {
1659 mlog_errno(status);
1660 goto bail;
1661 }
1662 }
1663
1664 /* calculate meta data/clusters for setting security xattr */
1665 if (si.enable) {
1666 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1667 &xattr_credits, &xattr_ac);
1668 if (status < 0) {
1669 mlog_errno(status);
1670 goto bail;
1671 }
1672 }
1673
1674 /* don't reserve bitmap space for fast symlinks. */
1675 if (l > ocfs2_fast_symlink_chars(sb))
1676 want_clusters += 1;
1677
1678 status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1679 if (status < 0) {
1680 if (status != -ENOSPC)
1681 mlog_errno(status);
1682 goto bail;
1683 }
1684
1685 handle = ocfs2_start_trans(osb, credits + xattr_credits);
1686 if (IS_ERR(handle)) {
1687 status = PTR_ERR(handle);
1688 handle = NULL;
1689 mlog_errno(status);
1690 goto bail;
1691 }
1692
1693 /* We don't use standard VFS wrapper because we don't want vfs_dq_init
1694 * to be called. */
1695 if (sb_any_quota_active(osb->sb) &&
1696 osb->sb->dq_op->alloc_inode(inode, 1) == NO_QUOTA) {
1697 status = -EDQUOT;
1698 goto bail;
1699 }
1700 did_quota_inode = 1;
1701
1702 status = ocfs2_mknod_locked(osb, dir, inode, dentry,
1703 0, &new_fe_bh, parent_fe_bh, handle,
1704 inode_ac);
1705 if (status < 0) {
1706 mlog_errno(status);
1707 goto bail;
1708 }
1709
1710 fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1711 inode->i_rdev = 0;
1712 newsize = l - 1;
1713 if (l > ocfs2_fast_symlink_chars(sb)) {
1714 u32 offset = 0;
1715
1716 inode->i_op = &ocfs2_symlink_inode_operations;
1717 if (vfs_dq_alloc_space_nodirty(inode,
1718 ocfs2_clusters_to_bytes(osb->sb, 1))) {
1719 status = -EDQUOT;
1720 goto bail;
1721 }
1722 did_quota = 1;
1723 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1724 new_fe_bh,
1725 handle, data_ac, NULL,
1726 NULL);
1727 if (status < 0) {
1728 if (status != -ENOSPC && status != -EINTR) {
1729 mlog(ML_ERROR,
1730 "Failed to extend file to %llu\n",
1731 (unsigned long long)newsize);
1732 mlog_errno(status);
1733 status = -ENOSPC;
1734 }
1735 goto bail;
1736 }
1737 i_size_write(inode, newsize);
1738 inode->i_blocks = ocfs2_inode_sector_count(inode);
1739 } else {
1740 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1741 memcpy((char *) fe->id2.i_symlink, symname, l);
1742 i_size_write(inode, newsize);
1743 inode->i_blocks = 0;
1744 }
1745
1746 status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1747 if (status < 0) {
1748 mlog_errno(status);
1749 goto bail;
1750 }
1751
1752 if (!ocfs2_inode_is_fast_symlink(inode)) {
1753 status = ocfs2_create_symlink_data(osb, handle, inode,
1754 symname);
1755 if (status < 0) {
1756 mlog_errno(status);
1757 goto bail;
1758 }
1759 }
1760
1761 if (si.enable) {
1762 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1763 xattr_ac, data_ac);
1764 if (status < 0) {
1765 mlog_errno(status);
1766 goto bail;
1767 }
1768 }
1769
1770 status = ocfs2_add_entry(handle, dentry, inode,
1771 le64_to_cpu(fe->i_blkno), parent_fe_bh,
1772 &lookup);
1773 if (status < 0) {
1774 mlog_errno(status);
1775 goto bail;
1776 }
1777
1778 status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1779 if (status) {
1780 mlog_errno(status);
1781 goto bail;
1782 }
1783
1784 insert_inode_hash(inode);
1785 dentry->d_op = &ocfs2_dentry_ops;
1786 d_instantiate(dentry, inode);
1787 bail:
1788 if (status < 0 && did_quota)
1789 vfs_dq_free_space_nodirty(inode,
1790 ocfs2_clusters_to_bytes(osb->sb, 1));
1791 if (status < 0 && did_quota_inode)
1792 vfs_dq_free_inode(inode);
1793 if (handle)
1794 ocfs2_commit_trans(osb, handle);
1795
1796 ocfs2_inode_unlock(dir, 1);
1797
1798 brelse(new_fe_bh);
1799 brelse(parent_fe_bh);
1800 kfree(si.name);
1801 kfree(si.value);
1802 ocfs2_free_dir_lookup_result(&lookup);
1803 if (inode_ac)
1804 ocfs2_free_alloc_context(inode_ac);
1805 if (data_ac)
1806 ocfs2_free_alloc_context(data_ac);
1807 if (xattr_ac)
1808 ocfs2_free_alloc_context(xattr_ac);
1809 if ((status < 0) && inode) {
1810 clear_nlink(inode);
1811 iput(inode);
1812 }
1813
1814 mlog_exit(status);
1815
1816 return status;
1817 }
1818
1819 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1820 {
1821 int status, namelen;
1822
1823 mlog_entry_void();
1824
1825 namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1826 (long long)blkno);
1827 if (namelen <= 0) {
1828 if (namelen)
1829 status = namelen;
1830 else
1831 status = -EINVAL;
1832 mlog_errno(status);
1833 goto bail;
1834 }
1835 if (namelen != OCFS2_ORPHAN_NAMELEN) {
1836 status = -EINVAL;
1837 mlog_errno(status);
1838 goto bail;
1839 }
1840
1841 mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
1842 namelen);
1843
1844 status = 0;
1845 bail:
1846 mlog_exit(status);
1847 return status;
1848 }
1849
1850 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1851 struct inode **ret_orphan_dir,
1852 struct inode *inode,
1853 char *name,
1854 struct ocfs2_dir_lookup_result *lookup)
1855 {
1856 struct inode *orphan_dir_inode;
1857 struct buffer_head *orphan_dir_bh = NULL;
1858 int status = 0;
1859
1860 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1861 if (status < 0) {
1862 mlog_errno(status);
1863 return status;
1864 }
1865
1866 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1867 ORPHAN_DIR_SYSTEM_INODE,
1868 osb->slot_num);
1869 if (!orphan_dir_inode) {
1870 status = -ENOENT;
1871 mlog_errno(status);
1872 return status;
1873 }
1874
1875 mutex_lock(&orphan_dir_inode->i_mutex);
1876
1877 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1878 if (status < 0) {
1879 mlog_errno(status);
1880 goto leave;
1881 }
1882
1883 status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1884 orphan_dir_bh, name,
1885 OCFS2_ORPHAN_NAMELEN, lookup);
1886 if (status < 0) {
1887 ocfs2_inode_unlock(orphan_dir_inode, 1);
1888
1889 mlog_errno(status);
1890 goto leave;
1891 }
1892
1893 *ret_orphan_dir = orphan_dir_inode;
1894
1895 leave:
1896 if (status) {
1897 mutex_unlock(&orphan_dir_inode->i_mutex);
1898 iput(orphan_dir_inode);
1899 }
1900
1901 brelse(orphan_dir_bh);
1902
1903 mlog_exit(status);
1904 return status;
1905 }
1906
1907 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1908 handle_t *handle,
1909 struct inode *inode,
1910 struct ocfs2_dinode *fe,
1911 char *name,
1912 struct ocfs2_dir_lookup_result *lookup,
1913 struct inode *orphan_dir_inode)
1914 {
1915 struct buffer_head *orphan_dir_bh = NULL;
1916 int status = 0;
1917 struct ocfs2_dinode *orphan_fe;
1918
1919 mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
1920
1921 status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
1922 if (status < 0) {
1923 mlog_errno(status);
1924 goto leave;
1925 }
1926
1927 status = ocfs2_journal_access_di(handle,
1928 INODE_CACHE(orphan_dir_inode),
1929 orphan_dir_bh,
1930 OCFS2_JOURNAL_ACCESS_WRITE);
1931 if (status < 0) {
1932 mlog_errno(status);
1933 goto leave;
1934 }
1935
1936 /* we're a cluster, and nlink can change on disk from
1937 * underneath us... */
1938 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1939 if (S_ISDIR(inode->i_mode))
1940 ocfs2_add_links_count(orphan_fe, 1);
1941 orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
1942
1943 status = ocfs2_journal_dirty(handle, orphan_dir_bh);
1944 if (status < 0) {
1945 mlog_errno(status);
1946 goto leave;
1947 }
1948
1949 status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
1950 OCFS2_ORPHAN_NAMELEN, inode,
1951 OCFS2_I(inode)->ip_blkno,
1952 orphan_dir_bh, lookup);
1953 if (status < 0) {
1954 mlog_errno(status);
1955 goto leave;
1956 }
1957
1958 le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
1959
1960 /* Record which orphan dir our inode now resides
1961 * in. delete_inode will use this to determine which orphan
1962 * dir to lock. */
1963 fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
1964
1965 mlog(0, "Inode %llu orphaned in slot %d\n",
1966 (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
1967
1968 leave:
1969 brelse(orphan_dir_bh);
1970
1971 mlog_exit(status);
1972 return status;
1973 }
1974
1975 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
1976 int ocfs2_orphan_del(struct ocfs2_super *osb,
1977 handle_t *handle,
1978 struct inode *orphan_dir_inode,
1979 struct inode *inode,
1980 struct buffer_head *orphan_dir_bh)
1981 {
1982 char name[OCFS2_ORPHAN_NAMELEN + 1];
1983 struct ocfs2_dinode *orphan_fe;
1984 int status = 0;
1985 struct ocfs2_dir_lookup_result lookup = { NULL, };
1986
1987 mlog_entry_void();
1988
1989 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1990 if (status < 0) {
1991 mlog_errno(status);
1992 goto leave;
1993 }
1994
1995 mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
1996 name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
1997 OCFS2_ORPHAN_NAMELEN);
1998
1999 /* find it's spot in the orphan directory */
2000 status = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN, orphan_dir_inode,
2001 &lookup);
2002 if (status) {
2003 mlog_errno(status);
2004 goto leave;
2005 }
2006
2007 /* remove it from the orphan directory */
2008 status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2009 if (status < 0) {
2010 mlog_errno(status);
2011 goto leave;
2012 }
2013
2014 status = ocfs2_journal_access_di(handle,
2015 INODE_CACHE(orphan_dir_inode),
2016 orphan_dir_bh,
2017 OCFS2_JOURNAL_ACCESS_WRITE);
2018 if (status < 0) {
2019 mlog_errno(status);
2020 goto leave;
2021 }
2022
2023 /* do the i_nlink dance! :) */
2024 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2025 if (S_ISDIR(inode->i_mode))
2026 ocfs2_add_links_count(orphan_fe, -1);
2027 orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
2028
2029 status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2030 if (status < 0) {
2031 mlog_errno(status);
2032 goto leave;
2033 }
2034
2035 leave:
2036 ocfs2_free_dir_lookup_result(&lookup);
2037
2038 mlog_exit(status);
2039 return status;
2040 }
2041
2042 const struct inode_operations ocfs2_dir_iops = {
2043 .create = ocfs2_create,
2044 .lookup = ocfs2_lookup,
2045 .link = ocfs2_link,
2046 .unlink = ocfs2_unlink,
2047 .rmdir = ocfs2_unlink,
2048 .symlink = ocfs2_symlink,
2049 .mkdir = ocfs2_mkdir,
2050 .mknod = ocfs2_mknod,
2051 .rename = ocfs2_rename,
2052 .setattr = ocfs2_setattr,
2053 .getattr = ocfs2_getattr,
2054 .permission = ocfs2_permission,
2055 .setxattr = generic_setxattr,
2056 .getxattr = generic_getxattr,
2057 .listxattr = ocfs2_listxattr,
2058 .removexattr = generic_removexattr,
2059 };
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree