?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
PCI: SR-IOV quirk for Intel 82576 NIC
[linux-2.6/mini2440.git] / fs / ocfs2 / namei.c
blob2220f93f668bc9d0b0cc5446efdb612ccfd1d4bb
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 *orphan_entry_bh = NULL;
1029 struct buffer_head *newfe_bh = NULL;
1030 struct buffer_head *old_inode_bh = NULL;
1031 struct buffer_head *insert_entry_bh = NULL;
1032 struct ocfs2_super *osb = NULL;
1033 u64 newfe_blkno, old_de_ino;
1034 handle_t *handle = NULL;
1035 struct buffer_head *old_dir_bh = NULL;
1036 struct buffer_head *new_dir_bh = NULL;
1037 nlink_t old_dir_nlink = old_dir->i_nlink;
1038 struct ocfs2_dinode *old_di;
1039 struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1040 struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1041 struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1042 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1043 struct ocfs2_dir_lookup_result target_insert = { NULL, };
1044
1045 /* At some point it might be nice to break this function up a
1046 * bit. */
1047
1048 mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1049 old_dir, old_dentry, new_dir, new_dentry,
1050 old_dentry->d_name.len, old_dentry->d_name.name,
1051 new_dentry->d_name.len, new_dentry->d_name.name);
1052
1053 osb = OCFS2_SB(old_dir->i_sb);
1054
1055 if (new_inode) {
1056 if (!igrab(new_inode))
1057 BUG();
1058 }
1059
1060 /* Assume a directory hierarchy thusly:
1061 * a/b/c
1062 * a/d
1063 * a,b,c, and d are all directories.
1064 *
1065 * from cwd of 'a' on both nodes:
1066 * node1: mv b/c d
1067 * node2: mv d b/c
1068 *
1069 * And that's why, just like the VFS, we need a file system
1070 * rename lock. */
1071 if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1072 status = ocfs2_rename_lock(osb);
1073 if (status < 0) {
1074 mlog_errno(status);
1075 goto bail;
1076 }
1077 rename_lock = 1;
1078 }
1079
1080 /* if old and new are the same, this'll just do one lock. */
1081 status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1082 &new_dir_bh, new_dir);
1083 if (status < 0) {
1084 mlog_errno(status);
1085 goto bail;
1086 }
1087 parents_locked = 1;
1088
1089 /* make sure both dirs have bhs
1090 * get an extra ref on old_dir_bh if old==new */
1091 if (!new_dir_bh) {
1092 if (old_dir_bh) {
1093 new_dir_bh = old_dir_bh;
1094 get_bh(new_dir_bh);
1095 } else {
1096 mlog(ML_ERROR, "no old_dir_bh!\n");
1097 status = -EIO;
1098 goto bail;
1099 }
1100 }
1101
1102 /*
1103 * Aside from allowing a meta data update, the locking here
1104 * also ensures that the downconvert thread on other nodes
1105 * won't have to concurrently downconvert the inode and the
1106 * dentry locks.
1107 */
1108 status = ocfs2_inode_lock(old_inode, &old_inode_bh, 1);
1109 if (status < 0) {
1110 if (status != -ENOENT)
1111 mlog_errno(status);
1112 goto bail;
1113 }
1114 old_child_locked = 1;
1115
1116 status = ocfs2_remote_dentry_delete(old_dentry);
1117 if (status < 0) {
1118 mlog_errno(status);
1119 goto bail;
1120 }
1121
1122 if (S_ISDIR(old_inode->i_mode)) {
1123 u64 old_inode_parent;
1124
1125 update_dot_dot = 1;
1126 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1127 old_inode,
1128 &old_inode_dot_dot_res);
1129 if (status) {
1130 status = -EIO;
1131 goto bail;
1132 }
1133
1134 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1135 status = -EIO;
1136 goto bail;
1137 }
1138
1139 if (!new_inode && new_dir != old_dir &&
1140 new_dir->i_nlink >= ocfs2_link_max(osb)) {
1141 status = -EMLINK;
1142 goto bail;
1143 }
1144 }
1145
1146 status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1147 old_dentry->d_name.len,
1148 &old_de_ino);
1149 if (status) {
1150 status = -ENOENT;
1151 goto bail;
1152 }
1153
1154 /*
1155 * Check for inode number is _not_ due to possible IO errors.
1156 * We might rmdir the source, keep it as pwd of some process
1157 * and merrily kill the link to whatever was created under the
1158 * same name. Goodbye sticky bit ;-<
1159 */
1160 if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1161 status = -ENOENT;
1162 goto bail;
1163 }
1164
1165 /* check if the target already exists (in which case we need
1166 * to delete it */
1167 status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1168 new_dentry->d_name.len,
1169 &newfe_blkno, new_dir,
1170 &target_lookup_res);
1171 /* The only error we allow here is -ENOENT because the new
1172 * file not existing is perfectly valid. */
1173 if ((status < 0) && (status != -ENOENT)) {
1174 /* If we cannot find the file specified we should just */
1175 /* return the error... */
1176 mlog_errno(status);
1177 goto bail;
1178 }
1179 if (status == 0)
1180 target_exists = 1;
1181
1182 if (!target_exists && new_inode) {
1183 /*
1184 * Target was unlinked by another node while we were
1185 * waiting to get to ocfs2_rename(). There isn't
1186 * anything we can do here to help the situation, so
1187 * bubble up the appropriate error.
1188 */
1189 status = -ENOENT;
1190 goto bail;
1191 }
1192
1193 /* In case we need to overwrite an existing file, we blow it
1194 * away first */
1195 if (target_exists) {
1196 /* VFS didn't think there existed an inode here, but
1197 * someone else in the cluster must have raced our
1198 * rename to create one. Today we error cleanly, in
1199 * the future we should consider calling iget to build
1200 * a new struct inode for this entry. */
1201 if (!new_inode) {
1202 status = -EACCES;
1203
1204 mlog(0, "We found an inode for name %.*s but VFS "
1205 "didn't give us one.\n", new_dentry->d_name.len,
1206 new_dentry->d_name.name);
1207 goto bail;
1208 }
1209
1210 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1211 status = -EACCES;
1212
1213 mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1214 (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1215 (unsigned long long)newfe_blkno,
1216 OCFS2_I(new_inode)->ip_flags);
1217 goto bail;
1218 }
1219
1220 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1221 if (status < 0) {
1222 if (status != -ENOENT)
1223 mlog_errno(status);
1224 goto bail;
1225 }
1226 new_child_locked = 1;
1227
1228 status = ocfs2_remote_dentry_delete(new_dentry);
1229 if (status < 0) {
1230 mlog_errno(status);
1231 goto bail;
1232 }
1233
1234 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1235
1236 mlog(0, "aha rename over existing... new_blkno=%llu "
1237 "newfebh=%p bhblocknr=%llu\n",
1238 (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1239 (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1240
1241 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1242 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1243 new_inode,
1244 orphan_name,
1245 &orphan_insert);
1246 if (status < 0) {
1247 mlog_errno(status);
1248 goto bail;
1249 }
1250 }
1251 } else {
1252 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1253
1254 status = ocfs2_check_dir_for_entry(new_dir,
1255 new_dentry->d_name.name,
1256 new_dentry->d_name.len);
1257 if (status)
1258 goto bail;
1259
1260 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1261 new_dentry->d_name.name,
1262 new_dentry->d_name.len,
1263 &target_insert);
1264 if (status < 0) {
1265 mlog_errno(status);
1266 goto bail;
1267 }
1268 }
1269
1270 handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1271 if (IS_ERR(handle)) {
1272 status = PTR_ERR(handle);
1273 handle = NULL;
1274 mlog_errno(status);
1275 goto bail;
1276 }
1277
1278 if (target_exists) {
1279 if (S_ISDIR(new_inode->i_mode)) {
1280 if (new_inode->i_nlink != 2 ||
1281 !ocfs2_empty_dir(new_inode)) {
1282 status = -ENOTEMPTY;
1283 goto bail;
1284 }
1285 }
1286 status = ocfs2_journal_access_di(handle, new_inode, newfe_bh,
1287 OCFS2_JOURNAL_ACCESS_WRITE);
1288 if (status < 0) {
1289 mlog_errno(status);
1290 goto bail;
1291 }
1292
1293 if (S_ISDIR(new_inode->i_mode) ||
1294 (ocfs2_read_links_count(newfe) == 1)) {
1295 status = ocfs2_orphan_add(osb, handle, new_inode,
1296 newfe, orphan_name,
1297 &orphan_insert, orphan_dir);
1298 if (status < 0) {
1299 mlog_errno(status);
1300 goto bail;
1301 }
1302 }
1303
1304 /* change the dirent to point to the correct inode */
1305 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1306 old_inode);
1307 if (status < 0) {
1308 mlog_errno(status);
1309 goto bail;
1310 }
1311 new_dir->i_version++;
1312
1313 if (S_ISDIR(new_inode->i_mode))
1314 ocfs2_set_links_count(newfe, 0);
1315 else
1316 ocfs2_add_links_count(newfe, -1);
1317
1318 status = ocfs2_journal_dirty(handle, newfe_bh);
1319 if (status < 0) {
1320 mlog_errno(status);
1321 goto bail;
1322 }
1323 } else {
1324 /* if the name was not found in new_dir, add it now */
1325 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1326 OCFS2_I(old_inode)->ip_blkno,
1327 new_dir_bh, &target_insert);
1328 }
1329
1330 old_inode->i_ctime = CURRENT_TIME;
1331 mark_inode_dirty(old_inode);
1332
1333 status = ocfs2_journal_access_di(handle, old_inode, old_inode_bh,
1334 OCFS2_JOURNAL_ACCESS_WRITE);
1335 if (status >= 0) {
1336 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1337
1338 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1339 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1340
1341 status = ocfs2_journal_dirty(handle, old_inode_bh);
1342 if (status < 0)
1343 mlog_errno(status);
1344 } else
1345 mlog_errno(status);
1346
1347 /*
1348 * Now that the name has been added to new_dir, remove the old name.
1349 *
1350 * We don't keep any directory entry context around until now
1351 * because the insert might have changed the type of directory
1352 * we're dealing with.
1353 */
1354 status = ocfs2_find_entry(old_dentry->d_name.name,
1355 old_dentry->d_name.len, old_dir,
1356 &old_entry_lookup);
1357 if (status)
1358 goto bail;
1359
1360 status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1361 if (status < 0) {
1362 mlog_errno(status);
1363 goto bail;
1364 }
1365
1366 if (new_inode) {
1367 new_inode->i_nlink--;
1368 new_inode->i_ctime = CURRENT_TIME;
1369 }
1370 old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1371
1372 if (update_dot_dot) {
1373 status = ocfs2_update_entry(old_inode, handle,
1374 &old_inode_dot_dot_res, new_dir);
1375 old_dir->i_nlink--;
1376 if (new_inode) {
1377 new_inode->i_nlink--;
1378 } else {
1379 inc_nlink(new_dir);
1380 mark_inode_dirty(new_dir);
1381 }
1382 }
1383 mark_inode_dirty(old_dir);
1384 ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1385 if (new_inode) {
1386 mark_inode_dirty(new_inode);
1387 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1388 }
1389
1390 if (old_dir != new_dir) {
1391 /* Keep the same times on both directories.*/
1392 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1393
1394 /*
1395 * This will also pick up the i_nlink change from the
1396 * block above.
1397 */
1398 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1399 }
1400
1401 if (old_dir_nlink != old_dir->i_nlink) {
1402 if (!old_dir_bh) {
1403 mlog(ML_ERROR, "need to change nlink for old dir "
1404 "%llu from %d to %d but bh is NULL!\n",
1405 (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1406 (int)old_dir_nlink, old_dir->i_nlink);
1407 } else {
1408 struct ocfs2_dinode *fe;
1409 status = ocfs2_journal_access_di(handle, old_dir,
1410 old_dir_bh,
1411 OCFS2_JOURNAL_ACCESS_WRITE);
1412 fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1413 ocfs2_set_links_count(fe, old_dir->i_nlink);
1414 status = ocfs2_journal_dirty(handle, old_dir_bh);
1415 }
1416 }
1417 ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1418 status = 0;
1419 bail:
1420 if (rename_lock)
1421 ocfs2_rename_unlock(osb);
1422
1423 if (handle)
1424 ocfs2_commit_trans(osb, handle);
1425
1426 if (parents_locked)
1427 ocfs2_double_unlock(old_dir, new_dir);
1428
1429 if (old_child_locked)
1430 ocfs2_inode_unlock(old_inode, 1);
1431
1432 if (new_child_locked)
1433 ocfs2_inode_unlock(new_inode, 1);
1434
1435 if (orphan_dir) {
1436 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1437 ocfs2_inode_unlock(orphan_dir, 1);
1438 mutex_unlock(&orphan_dir->i_mutex);
1439 iput(orphan_dir);
1440 }
1441
1442 if (new_inode)
1443 sync_mapping_buffers(old_inode->i_mapping);
1444
1445 if (new_inode)
1446 iput(new_inode);
1447
1448 ocfs2_free_dir_lookup_result(&target_lookup_res);
1449 ocfs2_free_dir_lookup_result(&old_entry_lookup);
1450 ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1451 ocfs2_free_dir_lookup_result(&orphan_insert);
1452 ocfs2_free_dir_lookup_result(&target_insert);
1453
1454 brelse(newfe_bh);
1455 brelse(old_inode_bh);
1456 brelse(old_dir_bh);
1457 brelse(new_dir_bh);
1458 brelse(orphan_entry_bh);
1459 brelse(insert_entry_bh);
1460
1461 mlog_exit(status);
1462
1463 return status;
1464 }
1465
1466 /*
1467 * we expect i_size = strlen(symname). Copy symname into the file
1468 * data, including the null terminator.
1469 */
1470 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1471 handle_t *handle,
1472 struct inode *inode,
1473 const char *symname)
1474 {
1475 struct buffer_head **bhs = NULL;
1476 const char *c;
1477 struct super_block *sb = osb->sb;
1478 u64 p_blkno, p_blocks;
1479 int virtual, blocks, status, i, bytes_left;
1480
1481 bytes_left = i_size_read(inode) + 1;
1482 /* we can't trust i_blocks because we're actually going to
1483 * write i_size + 1 bytes. */
1484 blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1485
1486 mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1487 (unsigned long long)inode->i_blocks,
1488 i_size_read(inode), blocks);
1489
1490 /* Sanity check -- make sure we're going to fit. */
1491 if (bytes_left >
1492 ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1493 status = -EIO;
1494 mlog_errno(status);
1495 goto bail;
1496 }
1497
1498 bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1499 if (!bhs) {
1500 status = -ENOMEM;
1501 mlog_errno(status);
1502 goto bail;
1503 }
1504
1505 status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1506 NULL);
1507 if (status < 0) {
1508 mlog_errno(status);
1509 goto bail;
1510 }
1511
1512 /* links can never be larger than one cluster so we know this
1513 * is all going to be contiguous, but do a sanity check
1514 * anyway. */
1515 if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1516 status = -EIO;
1517 mlog_errno(status);
1518 goto bail;
1519 }
1520
1521 virtual = 0;
1522 while(bytes_left > 0) {
1523 c = &symname[virtual * sb->s_blocksize];
1524
1525 bhs[virtual] = sb_getblk(sb, p_blkno);
1526 if (!bhs[virtual]) {
1527 status = -ENOMEM;
1528 mlog_errno(status);
1529 goto bail;
1530 }
1531 ocfs2_set_new_buffer_uptodate(inode, bhs[virtual]);
1532
1533 status = ocfs2_journal_access(handle, inode, bhs[virtual],
1534 OCFS2_JOURNAL_ACCESS_CREATE);
1535 if (status < 0) {
1536 mlog_errno(status);
1537 goto bail;
1538 }
1539
1540 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1541
1542 memcpy(bhs[virtual]->b_data, c,
1543 (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1544 bytes_left);
1545
1546 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1547 if (status < 0) {
1548 mlog_errno(status);
1549 goto bail;
1550 }
1551
1552 virtual++;
1553 p_blkno++;
1554 bytes_left -= sb->s_blocksize;
1555 }
1556
1557 status = 0;
1558 bail:
1559
1560 if (bhs) {
1561 for(i = 0; i < blocks; i++)
1562 brelse(bhs[i]);
1563 kfree(bhs);
1564 }
1565
1566 mlog_exit(status);
1567 return status;
1568 }
1569
1570 static int ocfs2_symlink(struct inode *dir,
1571 struct dentry *dentry,
1572 const char *symname)
1573 {
1574 int status, l, credits;
1575 u64 newsize;
1576 struct ocfs2_super *osb = NULL;
1577 struct inode *inode = NULL;
1578 struct super_block *sb;
1579 struct buffer_head *new_fe_bh = NULL;
1580 struct buffer_head *parent_fe_bh = NULL;
1581 struct ocfs2_dinode *fe = NULL;
1582 struct ocfs2_dinode *dirfe;
1583 handle_t *handle = NULL;
1584 struct ocfs2_alloc_context *inode_ac = NULL;
1585 struct ocfs2_alloc_context *data_ac = NULL;
1586 struct ocfs2_alloc_context *xattr_ac = NULL;
1587 int want_clusters = 0;
1588 int xattr_credits = 0;
1589 struct ocfs2_security_xattr_info si = {
1590 .enable = 1,
1591 };
1592 int did_quota = 0, did_quota_inode = 0;
1593 struct ocfs2_dir_lookup_result lookup = { NULL, };
1594
1595 mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1596 dentry, symname, dentry->d_name.len, dentry->d_name.name);
1597
1598 sb = dir->i_sb;
1599 osb = OCFS2_SB(sb);
1600
1601 l = strlen(symname) + 1;
1602
1603 credits = ocfs2_calc_symlink_credits(sb);
1604
1605 /* lock the parent directory */
1606 status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1607 if (status < 0) {
1608 if (status != -ENOENT)
1609 mlog_errno(status);
1610 return status;
1611 }
1612
1613 dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1614 if (!ocfs2_read_links_count(dirfe)) {
1615 /* can't make a file in a deleted directory. */
1616 status = -ENOENT;
1617 goto bail;
1618 }
1619
1620 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1621 dentry->d_name.len);
1622 if (status)
1623 goto bail;
1624
1625 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1626 dentry->d_name.name,
1627 dentry->d_name.len, &lookup);
1628 if (status < 0) {
1629 mlog_errno(status);
1630 goto bail;
1631 }
1632
1633 status = ocfs2_reserve_new_inode(osb, &inode_ac);
1634 if (status < 0) {
1635 if (status != -ENOSPC)
1636 mlog_errno(status);
1637 goto bail;
1638 }
1639
1640 inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1641 if (!inode) {
1642 status = -ENOMEM;
1643 mlog_errno(status);
1644 goto bail;
1645 }
1646
1647 /* get security xattr */
1648 status = ocfs2_init_security_get(inode, dir, &si);
1649 if (status) {
1650 if (status == -EOPNOTSUPP)
1651 si.enable = 0;
1652 else {
1653 mlog_errno(status);
1654 goto bail;
1655 }
1656 }
1657
1658 /* calculate meta data/clusters for setting security xattr */
1659 if (si.enable) {
1660 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1661 &xattr_credits, &xattr_ac);
1662 if (status < 0) {
1663 mlog_errno(status);
1664 goto bail;
1665 }
1666 }
1667
1668 /* don't reserve bitmap space for fast symlinks. */
1669 if (l > ocfs2_fast_symlink_chars(sb))
1670 want_clusters += 1;
1671
1672 status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1673 if (status < 0) {
1674 if (status != -ENOSPC)
1675 mlog_errno(status);
1676 goto bail;
1677 }
1678
1679 handle = ocfs2_start_trans(osb, credits + xattr_credits);
1680 if (IS_ERR(handle)) {
1681 status = PTR_ERR(handle);
1682 handle = NULL;
1683 mlog_errno(status);
1684 goto bail;
1685 }
1686
1687 /* We don't use standard VFS wrapper because we don't want vfs_dq_init
1688 * to be called. */
1689 if (sb_any_quota_active(osb->sb) &&
1690 osb->sb->dq_op->alloc_inode(inode, 1) == NO_QUOTA) {
1691 status = -EDQUOT;
1692 goto bail;
1693 }
1694 did_quota_inode = 1;
1695
1696 status = ocfs2_mknod_locked(osb, dir, inode, dentry,
1697 0, &new_fe_bh, parent_fe_bh, handle,
1698 inode_ac);
1699 if (status < 0) {
1700 mlog_errno(status);
1701 goto bail;
1702 }
1703
1704 fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1705 inode->i_rdev = 0;
1706 newsize = l - 1;
1707 if (l > ocfs2_fast_symlink_chars(sb)) {
1708 u32 offset = 0;
1709
1710 inode->i_op = &ocfs2_symlink_inode_operations;
1711 if (vfs_dq_alloc_space_nodirty(inode,
1712 ocfs2_clusters_to_bytes(osb->sb, 1))) {
1713 status = -EDQUOT;
1714 goto bail;
1715 }
1716 did_quota = 1;
1717 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1718 new_fe_bh,
1719 handle, data_ac, NULL,
1720 NULL);
1721 if (status < 0) {
1722 if (status != -ENOSPC && status != -EINTR) {
1723 mlog(ML_ERROR,
1724 "Failed to extend file to %llu\n",
1725 (unsigned long long)newsize);
1726 mlog_errno(status);
1727 status = -ENOSPC;
1728 }
1729 goto bail;
1730 }
1731 i_size_write(inode, newsize);
1732 inode->i_blocks = ocfs2_inode_sector_count(inode);
1733 } else {
1734 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1735 memcpy((char *) fe->id2.i_symlink, symname, l);
1736 i_size_write(inode, newsize);
1737 inode->i_blocks = 0;
1738 }
1739
1740 status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1741 if (status < 0) {
1742 mlog_errno(status);
1743 goto bail;
1744 }
1745
1746 if (!ocfs2_inode_is_fast_symlink(inode)) {
1747 status = ocfs2_create_symlink_data(osb, handle, inode,
1748 symname);
1749 if (status < 0) {
1750 mlog_errno(status);
1751 goto bail;
1752 }
1753 }
1754
1755 if (si.enable) {
1756 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1757 xattr_ac, data_ac);
1758 if (status < 0) {
1759 mlog_errno(status);
1760 goto bail;
1761 }
1762 }
1763
1764 status = ocfs2_add_entry(handle, dentry, inode,
1765 le64_to_cpu(fe->i_blkno), parent_fe_bh,
1766 &lookup);
1767 if (status < 0) {
1768 mlog_errno(status);
1769 goto bail;
1770 }
1771
1772 status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1773 if (status) {
1774 mlog_errno(status);
1775 goto bail;
1776 }
1777
1778 insert_inode_hash(inode);
1779 dentry->d_op = &ocfs2_dentry_ops;
1780 d_instantiate(dentry, inode);
1781 bail:
1782 if (status < 0 && did_quota)
1783 vfs_dq_free_space_nodirty(inode,
1784 ocfs2_clusters_to_bytes(osb->sb, 1));
1785 if (status < 0 && did_quota_inode)
1786 vfs_dq_free_inode(inode);
1787 if (handle)
1788 ocfs2_commit_trans(osb, handle);
1789
1790 ocfs2_inode_unlock(dir, 1);
1791
1792 brelse(new_fe_bh);
1793 brelse(parent_fe_bh);
1794 kfree(si.name);
1795 kfree(si.value);
1796 ocfs2_free_dir_lookup_result(&lookup);
1797 if (inode_ac)
1798 ocfs2_free_alloc_context(inode_ac);
1799 if (data_ac)
1800 ocfs2_free_alloc_context(data_ac);
1801 if (xattr_ac)
1802 ocfs2_free_alloc_context(xattr_ac);
1803 if ((status < 0) && inode) {
1804 clear_nlink(inode);
1805 iput(inode);
1806 }
1807
1808 mlog_exit(status);
1809
1810 return status;
1811 }
1812
1813 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1814 {
1815 int status, namelen;
1816
1817 mlog_entry_void();
1818
1819 namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1820 (long long)blkno);
1821 if (namelen <= 0) {
1822 if (namelen)
1823 status = namelen;
1824 else
1825 status = -EINVAL;
1826 mlog_errno(status);
1827 goto bail;
1828 }
1829 if (namelen != OCFS2_ORPHAN_NAMELEN) {
1830 status = -EINVAL;
1831 mlog_errno(status);
1832 goto bail;
1833 }
1834
1835 mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
1836 namelen);
1837
1838 status = 0;
1839 bail:
1840 mlog_exit(status);
1841 return status;
1842 }
1843
1844 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1845 struct inode **ret_orphan_dir,
1846 struct inode *inode,
1847 char *name,
1848 struct ocfs2_dir_lookup_result *lookup)
1849 {
1850 struct inode *orphan_dir_inode;
1851 struct buffer_head *orphan_dir_bh = NULL;
1852 int status = 0;
1853
1854 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1855 if (status < 0) {
1856 mlog_errno(status);
1857 return status;
1858 }
1859
1860 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1861 ORPHAN_DIR_SYSTEM_INODE,
1862 osb->slot_num);
1863 if (!orphan_dir_inode) {
1864 status = -ENOENT;
1865 mlog_errno(status);
1866 return status;
1867 }
1868
1869 mutex_lock(&orphan_dir_inode->i_mutex);
1870
1871 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1872 if (status < 0) {
1873 mlog_errno(status);
1874 goto leave;
1875 }
1876
1877 status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1878 orphan_dir_bh, name,
1879 OCFS2_ORPHAN_NAMELEN, lookup);
1880 if (status < 0) {
1881 ocfs2_inode_unlock(orphan_dir_inode, 1);
1882
1883 mlog_errno(status);
1884 goto leave;
1885 }
1886
1887 *ret_orphan_dir = orphan_dir_inode;
1888
1889 leave:
1890 if (status) {
1891 mutex_unlock(&orphan_dir_inode->i_mutex);
1892 iput(orphan_dir_inode);
1893 }
1894
1895 brelse(orphan_dir_bh);
1896
1897 mlog_exit(status);
1898 return status;
1899 }
1900
1901 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1902 handle_t *handle,
1903 struct inode *inode,
1904 struct ocfs2_dinode *fe,
1905 char *name,
1906 struct ocfs2_dir_lookup_result *lookup,
1907 struct inode *orphan_dir_inode)
1908 {
1909 struct buffer_head *orphan_dir_bh = NULL;
1910 int status = 0;
1911 struct ocfs2_dinode *orphan_fe;
1912
1913 mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
1914
1915 status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
1916 if (status < 0) {
1917 mlog_errno(status);
1918 goto leave;
1919 }
1920
1921 status = ocfs2_journal_access_di(handle, orphan_dir_inode, orphan_dir_bh,
1922 OCFS2_JOURNAL_ACCESS_WRITE);
1923 if (status < 0) {
1924 mlog_errno(status);
1925 goto leave;
1926 }
1927
1928 /* we're a cluster, and nlink can change on disk from
1929 * underneath us... */
1930 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1931 if (S_ISDIR(inode->i_mode))
1932 ocfs2_add_links_count(orphan_fe, 1);
1933 orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
1934
1935 status = ocfs2_journal_dirty(handle, orphan_dir_bh);
1936 if (status < 0) {
1937 mlog_errno(status);
1938 goto leave;
1939 }
1940
1941 status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
1942 OCFS2_ORPHAN_NAMELEN, inode,
1943 OCFS2_I(inode)->ip_blkno,
1944 orphan_dir_bh, lookup);
1945 if (status < 0) {
1946 mlog_errno(status);
1947 goto leave;
1948 }
1949
1950 le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
1951
1952 /* Record which orphan dir our inode now resides
1953 * in. delete_inode will use this to determine which orphan
1954 * dir to lock. */
1955 fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
1956
1957 mlog(0, "Inode %llu orphaned in slot %d\n",
1958 (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
1959
1960 leave:
1961 brelse(orphan_dir_bh);
1962
1963 mlog_exit(status);
1964 return status;
1965 }
1966
1967 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
1968 int ocfs2_orphan_del(struct ocfs2_super *osb,
1969 handle_t *handle,
1970 struct inode *orphan_dir_inode,
1971 struct inode *inode,
1972 struct buffer_head *orphan_dir_bh)
1973 {
1974 char name[OCFS2_ORPHAN_NAMELEN + 1];
1975 struct ocfs2_dinode *orphan_fe;
1976 int status = 0;
1977 struct ocfs2_dir_lookup_result lookup = { NULL, };
1978
1979 mlog_entry_void();
1980
1981 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1982 if (status < 0) {
1983 mlog_errno(status);
1984 goto leave;
1985 }
1986
1987 mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
1988 name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
1989 OCFS2_ORPHAN_NAMELEN);
1990
1991 /* find it's spot in the orphan directory */
1992 status = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN, orphan_dir_inode,
1993 &lookup);
1994 if (status) {
1995 mlog_errno(status);
1996 goto leave;
1997 }
1998
1999 /* remove it from the orphan directory */
2000 status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2001 if (status < 0) {
2002 mlog_errno(status);
2003 goto leave;
2004 }
2005
2006 status = ocfs2_journal_access_di(handle,orphan_dir_inode, orphan_dir_bh,
2007 OCFS2_JOURNAL_ACCESS_WRITE);
2008 if (status < 0) {
2009 mlog_errno(status);
2010 goto leave;
2011 }
2012
2013 /* do the i_nlink dance! :) */
2014 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2015 if (S_ISDIR(inode->i_mode))
2016 ocfs2_add_links_count(orphan_fe, -1);
2017 orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
2018
2019 status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2020 if (status < 0) {
2021 mlog_errno(status);
2022 goto leave;
2023 }
2024
2025 leave:
2026 ocfs2_free_dir_lookup_result(&lookup);
2027
2028 mlog_exit(status);
2029 return status;
2030 }
2031
2032 const struct inode_operations ocfs2_dir_iops = {
2033 .create = ocfs2_create,
2034 .lookup = ocfs2_lookup,
2035 .link = ocfs2_link,
2036 .unlink = ocfs2_unlink,
2037 .rmdir = ocfs2_unlink,
2038 .symlink = ocfs2_symlink,
2039 .mkdir = ocfs2_mkdir,
2040 .mknod = ocfs2_mknod,
2041 .rename = ocfs2_rename,
2042 .setattr = ocfs2_setattr,
2043 .getattr = ocfs2_getattr,
2044 .permission = ocfs2_permission,
2045 .setxattr = generic_setxattr,
2046 .getxattr = generic_getxattr,
2047 .listxattr = ocfs2_listxattr,
2048 .removexattr = generic_removexattr,
2049 };
Support for the Chinese Samsung S3C2440 based development boards