]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - fs/ocfs2/namei.c
projects / mirror_ubuntu-zesty-kernel.git / blob
? search:


67113cfddc9ba7be5d981adfa07b84359e3448f4
[mirror_ubuntu-zesty-kernel.git] / fs / ocfs2 / namei.c
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, inode_ac, &suballoc_bit,
489 &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 (!ocfs2_empty_dir(inode)) {
838 status = -ENOTEMPTY;
839 goto leave;
840 } else if (inode->i_nlink != 2) {
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, 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(inode2, bh2, 1);
985 if (status < 0) {
986 if (status != -ENOENT)
987 mlog_errno(status);
988 goto bail;
989 }
990 }
991
992 /* lock id1 */
993 status = ocfs2_inode_lock(inode1, bh1, 1);
994 if (status < 0) {
995 /*
996 * An error return must mean that no cluster locks
997 * were held on function exit.
998 */
999 if (oi1->ip_blkno != oi2->ip_blkno)
1000 ocfs2_inode_unlock(inode2, 1);
1001
1002 if (status != -ENOENT)
1003 mlog_errno(status);
1004 }
1005
1006 bail:
1007 mlog_exit(status);
1008 return status;
1009 }
1010
1011 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1012 {
1013 ocfs2_inode_unlock(inode1, 1);
1014
1015 if (inode1 != inode2)
1016 ocfs2_inode_unlock(inode2, 1);
1017 }
1018
1019 static int ocfs2_rename(struct inode *old_dir,
1020 struct dentry *old_dentry,
1021 struct inode *new_dir,
1022 struct dentry *new_dentry)
1023 {
1024 int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1025 int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1026 struct inode *old_inode = old_dentry->d_inode;
1027 struct inode *new_inode = new_dentry->d_inode;
1028 struct inode *orphan_dir = NULL;
1029 struct ocfs2_dinode *newfe = NULL;
1030 char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1031 struct buffer_head *orphan_entry_bh = NULL;
1032 struct buffer_head *newfe_bh = NULL;
1033 struct buffer_head *old_inode_bh = NULL;
1034 struct buffer_head *insert_entry_bh = NULL;
1035 struct ocfs2_super *osb = NULL;
1036 u64 newfe_blkno, old_de_ino;
1037 handle_t *handle = NULL;
1038 struct buffer_head *old_dir_bh = NULL;
1039 struct buffer_head *new_dir_bh = NULL;
1040 nlink_t old_dir_nlink = old_dir->i_nlink;
1041 struct ocfs2_dinode *old_di;
1042 struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1043 struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1044 struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1045 struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1046 struct ocfs2_dir_lookup_result target_insert = { NULL, };
1047
1048 /* At some point it might be nice to break this function up a
1049 * bit. */
1050
1051 mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1052 old_dir, old_dentry, new_dir, new_dentry,
1053 old_dentry->d_name.len, old_dentry->d_name.name,
1054 new_dentry->d_name.len, new_dentry->d_name.name);
1055
1056 osb = OCFS2_SB(old_dir->i_sb);
1057
1058 if (new_inode) {
1059 if (!igrab(new_inode))
1060 BUG();
1061 }
1062
1063 /* Assume a directory hierarchy thusly:
1064 * a/b/c
1065 * a/d
1066 * a,b,c, and d are all directories.
1067 *
1068 * from cwd of 'a' on both nodes:
1069 * node1: mv b/c d
1070 * node2: mv d b/c
1071 *
1072 * And that's why, just like the VFS, we need a file system
1073 * rename lock. */
1074 if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1075 status = ocfs2_rename_lock(osb);
1076 if (status < 0) {
1077 mlog_errno(status);
1078 goto bail;
1079 }
1080 rename_lock = 1;
1081 }
1082
1083 /* if old and new are the same, this'll just do one lock. */
1084 status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1085 &new_dir_bh, new_dir);
1086 if (status < 0) {
1087 mlog_errno(status);
1088 goto bail;
1089 }
1090 parents_locked = 1;
1091
1092 /* make sure both dirs have bhs
1093 * get an extra ref on old_dir_bh if old==new */
1094 if (!new_dir_bh) {
1095 if (old_dir_bh) {
1096 new_dir_bh = old_dir_bh;
1097 get_bh(new_dir_bh);
1098 } else {
1099 mlog(ML_ERROR, "no old_dir_bh!\n");
1100 status = -EIO;
1101 goto bail;
1102 }
1103 }
1104
1105 /*
1106 * Aside from allowing a meta data update, the locking here
1107 * also ensures that the downconvert thread on other nodes
1108 * won't have to concurrently downconvert the inode and the
1109 * dentry locks.
1110 */
1111 status = ocfs2_inode_lock(old_inode, &old_inode_bh, 1);
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 (!ocfs2_empty_dir(new_inode) ||
1284 new_inode->i_nlink != 2) {
1285 status = -ENOTEMPTY;
1286 goto bail;
1287 }
1288 }
1289 status = ocfs2_journal_access_di(handle, new_inode, newfe_bh,
1290 OCFS2_JOURNAL_ACCESS_WRITE);
1291 if (status < 0) {
1292 mlog_errno(status);
1293 goto bail;
1294 }
1295
1296 if (S_ISDIR(new_inode->i_mode) ||
1297 (ocfs2_read_links_count(newfe) == 1)) {
1298 status = ocfs2_orphan_add(osb, handle, new_inode,
1299 newfe, orphan_name,
1300 &orphan_insert, orphan_dir);
1301 if (status < 0) {
1302 mlog_errno(status);
1303 goto bail;
1304 }
1305 }
1306
1307 /* change the dirent to point to the correct inode */
1308 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1309 old_inode);
1310 if (status < 0) {
1311 mlog_errno(status);
1312 goto bail;
1313 }
1314 new_dir->i_version++;
1315
1316 if (S_ISDIR(new_inode->i_mode))
1317 ocfs2_set_links_count(newfe, 0);
1318 else
1319 ocfs2_add_links_count(newfe, -1);
1320
1321 status = ocfs2_journal_dirty(handle, newfe_bh);
1322 if (status < 0) {
1323 mlog_errno(status);
1324 goto bail;
1325 }
1326 } else {
1327 /* if the name was not found in new_dir, add it now */
1328 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1329 OCFS2_I(old_inode)->ip_blkno,
1330 new_dir_bh, &target_insert);
1331 }
1332
1333 old_inode->i_ctime = CURRENT_TIME;
1334 mark_inode_dirty(old_inode);
1335
1336 status = ocfs2_journal_access_di(handle, old_inode, old_inode_bh,
1337 OCFS2_JOURNAL_ACCESS_WRITE);
1338 if (status >= 0) {
1339 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1340
1341 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1342 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1343
1344 status = ocfs2_journal_dirty(handle, old_inode_bh);
1345 if (status < 0)
1346 mlog_errno(status);
1347 } else
1348 mlog_errno(status);
1349
1350 /*
1351 * Now that the name has been added to new_dir, remove the old name.
1352 *
1353 * We don't keep any directory entry context around until now
1354 * because the insert might have changed the type of directory
1355 * we're dealing with.
1356 */
1357 status = ocfs2_find_entry(old_dentry->d_name.name,
1358 old_dentry->d_name.len, old_dir,
1359 &old_entry_lookup);
1360 if (status)
1361 goto bail;
1362
1363 status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1364 if (status < 0) {
1365 mlog_errno(status);
1366 goto bail;
1367 }
1368
1369 if (new_inode) {
1370 new_inode->i_nlink--;
1371 new_inode->i_ctime = CURRENT_TIME;
1372 }
1373 old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1374
1375 if (update_dot_dot) {
1376 status = ocfs2_update_entry(old_inode, handle,
1377 &old_inode_dot_dot_res, new_dir);
1378 old_dir->i_nlink--;
1379 if (new_inode) {
1380 new_inode->i_nlink--;
1381 } else {
1382 inc_nlink(new_dir);
1383 mark_inode_dirty(new_dir);
1384 }
1385 }
1386 mark_inode_dirty(old_dir);
1387 ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1388 if (new_inode) {
1389 mark_inode_dirty(new_inode);
1390 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1391 }
1392
1393 if (old_dir != new_dir) {
1394 /* Keep the same times on both directories.*/
1395 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1396
1397 /*
1398 * This will also pick up the i_nlink change from the
1399 * block above.
1400 */
1401 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1402 }
1403
1404 if (old_dir_nlink != old_dir->i_nlink) {
1405 if (!old_dir_bh) {
1406 mlog(ML_ERROR, "need to change nlink for old dir "
1407 "%llu from %d to %d but bh is NULL!\n",
1408 (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1409 (int)old_dir_nlink, old_dir->i_nlink);
1410 } else {
1411 struct ocfs2_dinode *fe;
1412 status = ocfs2_journal_access_di(handle, old_dir,
1413 old_dir_bh,
1414 OCFS2_JOURNAL_ACCESS_WRITE);
1415 fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1416 ocfs2_set_links_count(fe, old_dir->i_nlink);
1417 status = ocfs2_journal_dirty(handle, old_dir_bh);
1418 }
1419 }
1420 ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1421 status = 0;
1422 bail:
1423 if (rename_lock)
1424 ocfs2_rename_unlock(osb);
1425
1426 if (handle)
1427 ocfs2_commit_trans(osb, handle);
1428
1429 if (parents_locked)
1430 ocfs2_double_unlock(old_dir, new_dir);
1431
1432 if (old_child_locked)
1433 ocfs2_inode_unlock(old_inode, 1);
1434
1435 if (new_child_locked)
1436 ocfs2_inode_unlock(new_inode, 1);
1437
1438 if (orphan_dir) {
1439 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1440 ocfs2_inode_unlock(orphan_dir, 1);
1441 mutex_unlock(&orphan_dir->i_mutex);
1442 iput(orphan_dir);
1443 }
1444
1445 if (new_inode)
1446 sync_mapping_buffers(old_inode->i_mapping);
1447
1448 if (new_inode)
1449 iput(new_inode);
1450
1451 ocfs2_free_dir_lookup_result(&target_lookup_res);
1452 ocfs2_free_dir_lookup_result(&old_entry_lookup);
1453 ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1454 ocfs2_free_dir_lookup_result(&orphan_insert);
1455 ocfs2_free_dir_lookup_result(&target_insert);
1456
1457 brelse(newfe_bh);
1458 brelse(old_inode_bh);
1459 brelse(old_dir_bh);
1460 brelse(new_dir_bh);
1461 brelse(orphan_entry_bh);
1462 brelse(insert_entry_bh);
1463
1464 mlog_exit(status);
1465
1466 return status;
1467 }
1468
1469 /*
1470 * we expect i_size = strlen(symname). Copy symname into the file
1471 * data, including the null terminator.
1472 */
1473 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1474 handle_t *handle,
1475 struct inode *inode,
1476 const char *symname)
1477 {
1478 struct buffer_head **bhs = NULL;
1479 const char *c;
1480 struct super_block *sb = osb->sb;
1481 u64 p_blkno, p_blocks;
1482 int virtual, blocks, status, i, bytes_left;
1483
1484 bytes_left = i_size_read(inode) + 1;
1485 /* we can't trust i_blocks because we're actually going to
1486 * write i_size + 1 bytes. */
1487 blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1488
1489 mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1490 (unsigned long long)inode->i_blocks,
1491 i_size_read(inode), blocks);
1492
1493 /* Sanity check -- make sure we're going to fit. */
1494 if (bytes_left >
1495 ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1496 status = -EIO;
1497 mlog_errno(status);
1498 goto bail;
1499 }
1500
1501 bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1502 if (!bhs) {
1503 status = -ENOMEM;
1504 mlog_errno(status);
1505 goto bail;
1506 }
1507
1508 status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1509 NULL);
1510 if (status < 0) {
1511 mlog_errno(status);
1512 goto bail;
1513 }
1514
1515 /* links can never be larger than one cluster so we know this
1516 * is all going to be contiguous, but do a sanity check
1517 * anyway. */
1518 if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1519 status = -EIO;
1520 mlog_errno(status);
1521 goto bail;
1522 }
1523
1524 virtual = 0;
1525 while(bytes_left > 0) {
1526 c = &symname[virtual * sb->s_blocksize];
1527
1528 bhs[virtual] = sb_getblk(sb, p_blkno);
1529 if (!bhs[virtual]) {
1530 status = -ENOMEM;
1531 mlog_errno(status);
1532 goto bail;
1533 }
1534 ocfs2_set_new_buffer_uptodate(inode, bhs[virtual]);
1535
1536 status = ocfs2_journal_access(handle, inode, bhs[virtual],
1537 OCFS2_JOURNAL_ACCESS_CREATE);
1538 if (status < 0) {
1539 mlog_errno(status);
1540 goto bail;
1541 }
1542
1543 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1544
1545 memcpy(bhs[virtual]->b_data, c,
1546 (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1547 bytes_left);
1548
1549 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1550 if (status < 0) {
1551 mlog_errno(status);
1552 goto bail;
1553 }
1554
1555 virtual++;
1556 p_blkno++;
1557 bytes_left -= sb->s_blocksize;
1558 }
1559
1560 status = 0;
1561 bail:
1562
1563 if (bhs) {
1564 for(i = 0; i < blocks; i++)
1565 brelse(bhs[i]);
1566 kfree(bhs);
1567 }
1568
1569 mlog_exit(status);
1570 return status;
1571 }
1572
1573 static int ocfs2_symlink(struct inode *dir,
1574 struct dentry *dentry,
1575 const char *symname)
1576 {
1577 int status, l, credits;
1578 u64 newsize;
1579 struct ocfs2_super *osb = NULL;
1580 struct inode *inode = NULL;
1581 struct super_block *sb;
1582 struct buffer_head *new_fe_bh = NULL;
1583 struct buffer_head *parent_fe_bh = NULL;
1584 struct ocfs2_dinode *fe = NULL;
1585 struct ocfs2_dinode *dirfe;
1586 handle_t *handle = NULL;
1587 struct ocfs2_alloc_context *inode_ac = NULL;
1588 struct ocfs2_alloc_context *data_ac = NULL;
1589 struct ocfs2_alloc_context *xattr_ac = NULL;
1590 int want_clusters = 0;
1591 int xattr_credits = 0;
1592 struct ocfs2_security_xattr_info si = {
1593 .enable = 1,
1594 };
1595 int did_quota = 0, did_quota_inode = 0;
1596 struct ocfs2_dir_lookup_result lookup = { NULL, };
1597
1598 mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1599 dentry, symname, dentry->d_name.len, dentry->d_name.name);
1600
1601 sb = dir->i_sb;
1602 osb = OCFS2_SB(sb);
1603
1604 l = strlen(symname) + 1;
1605
1606 credits = ocfs2_calc_symlink_credits(sb);
1607
1608 /* lock the parent directory */
1609 status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1610 if (status < 0) {
1611 if (status != -ENOENT)
1612 mlog_errno(status);
1613 return status;
1614 }
1615
1616 dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1617 if (!ocfs2_read_links_count(dirfe)) {
1618 /* can't make a file in a deleted directory. */
1619 status = -ENOENT;
1620 goto bail;
1621 }
1622
1623 status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1624 dentry->d_name.len);
1625 if (status)
1626 goto bail;
1627
1628 status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1629 dentry->d_name.name,
1630 dentry->d_name.len, &lookup);
1631 if (status < 0) {
1632 mlog_errno(status);
1633 goto bail;
1634 }
1635
1636 status = ocfs2_reserve_new_inode(osb, &inode_ac);
1637 if (status < 0) {
1638 if (status != -ENOSPC)
1639 mlog_errno(status);
1640 goto bail;
1641 }
1642
1643 inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1644 if (!inode) {
1645 status = -ENOMEM;
1646 mlog_errno(status);
1647 goto bail;
1648 }
1649
1650 /* get security xattr */
1651 status = ocfs2_init_security_get(inode, dir, &si);
1652 if (status) {
1653 if (status == -EOPNOTSUPP)
1654 si.enable = 0;
1655 else {
1656 mlog_errno(status);
1657 goto bail;
1658 }
1659 }
1660
1661 /* calculate meta data/clusters for setting security xattr */
1662 if (si.enable) {
1663 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1664 &xattr_credits, &xattr_ac);
1665 if (status < 0) {
1666 mlog_errno(status);
1667 goto bail;
1668 }
1669 }
1670
1671 /* don't reserve bitmap space for fast symlinks. */
1672 if (l > ocfs2_fast_symlink_chars(sb))
1673 want_clusters += 1;
1674
1675 status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1676 if (status < 0) {
1677 if (status != -ENOSPC)
1678 mlog_errno(status);
1679 goto bail;
1680 }
1681
1682 handle = ocfs2_start_trans(osb, credits + xattr_credits);
1683 if (IS_ERR(handle)) {
1684 status = PTR_ERR(handle);
1685 handle = NULL;
1686 mlog_errno(status);
1687 goto bail;
1688 }
1689
1690 /* We don't use standard VFS wrapper because we don't want vfs_dq_init
1691 * to be called. */
1692 if (sb_any_quota_active(osb->sb) &&
1693 osb->sb->dq_op->alloc_inode(inode, 1) == NO_QUOTA) {
1694 status = -EDQUOT;
1695 goto bail;
1696 }
1697 did_quota_inode = 1;
1698
1699 status = ocfs2_mknod_locked(osb, dir, inode, dentry,
1700 0, &new_fe_bh, parent_fe_bh, handle,
1701 inode_ac);
1702 if (status < 0) {
1703 mlog_errno(status);
1704 goto bail;
1705 }
1706
1707 fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1708 inode->i_rdev = 0;
1709 newsize = l - 1;
1710 if (l > ocfs2_fast_symlink_chars(sb)) {
1711 u32 offset = 0;
1712
1713 inode->i_op = &ocfs2_symlink_inode_operations;
1714 if (vfs_dq_alloc_space_nodirty(inode,
1715 ocfs2_clusters_to_bytes(osb->sb, 1))) {
1716 status = -EDQUOT;
1717 goto bail;
1718 }
1719 did_quota = 1;
1720 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1721 new_fe_bh,
1722 handle, data_ac, NULL,
1723 NULL);
1724 if (status < 0) {
1725 if (status != -ENOSPC && status != -EINTR) {
1726 mlog(ML_ERROR,
1727 "Failed to extend file to %llu\n",
1728 (unsigned long long)newsize);
1729 mlog_errno(status);
1730 status = -ENOSPC;
1731 }
1732 goto bail;
1733 }
1734 i_size_write(inode, newsize);
1735 inode->i_blocks = ocfs2_inode_sector_count(inode);
1736 } else {
1737 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1738 memcpy((char *) fe->id2.i_symlink, symname, l);
1739 i_size_write(inode, newsize);
1740 inode->i_blocks = 0;
1741 }
1742
1743 status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1744 if (status < 0) {
1745 mlog_errno(status);
1746 goto bail;
1747 }
1748
1749 if (!ocfs2_inode_is_fast_symlink(inode)) {
1750 status = ocfs2_create_symlink_data(osb, handle, inode,
1751 symname);
1752 if (status < 0) {
1753 mlog_errno(status);
1754 goto bail;
1755 }
1756 }
1757
1758 if (si.enable) {
1759 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1760 xattr_ac, data_ac);
1761 if (status < 0) {
1762 mlog_errno(status);
1763 goto bail;
1764 }
1765 }
1766
1767 status = ocfs2_add_entry(handle, dentry, inode,
1768 le64_to_cpu(fe->i_blkno), parent_fe_bh,
1769 &lookup);
1770 if (status < 0) {
1771 mlog_errno(status);
1772 goto bail;
1773 }
1774
1775 status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1776 if (status) {
1777 mlog_errno(status);
1778 goto bail;
1779 }
1780
1781 insert_inode_hash(inode);
1782 dentry->d_op = &ocfs2_dentry_ops;
1783 d_instantiate(dentry, inode);
1784 bail:
1785 if (status < 0 && did_quota)
1786 vfs_dq_free_space_nodirty(inode,
1787 ocfs2_clusters_to_bytes(osb->sb, 1));
1788 if (status < 0 && did_quota_inode)
1789 vfs_dq_free_inode(inode);
1790 if (handle)
1791 ocfs2_commit_trans(osb, handle);
1792
1793 ocfs2_inode_unlock(dir, 1);
1794
1795 brelse(new_fe_bh);
1796 brelse(parent_fe_bh);
1797 kfree(si.name);
1798 kfree(si.value);
1799 ocfs2_free_dir_lookup_result(&lookup);
1800 if (inode_ac)
1801 ocfs2_free_alloc_context(inode_ac);
1802 if (data_ac)
1803 ocfs2_free_alloc_context(data_ac);
1804 if (xattr_ac)
1805 ocfs2_free_alloc_context(xattr_ac);
1806 if ((status < 0) && inode) {
1807 clear_nlink(inode);
1808 iput(inode);
1809 }
1810
1811 mlog_exit(status);
1812
1813 return status;
1814 }
1815
1816 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1817 {
1818 int status, namelen;
1819
1820 mlog_entry_void();
1821
1822 namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1823 (long long)blkno);
1824 if (namelen <= 0) {
1825 if (namelen)
1826 status = namelen;
1827 else
1828 status = -EINVAL;
1829 mlog_errno(status);
1830 goto bail;
1831 }
1832 if (namelen != OCFS2_ORPHAN_NAMELEN) {
1833 status = -EINVAL;
1834 mlog_errno(status);
1835 goto bail;
1836 }
1837
1838 mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
1839 namelen);
1840
1841 status = 0;
1842 bail:
1843 mlog_exit(status);
1844 return status;
1845 }
1846
1847 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1848 struct inode **ret_orphan_dir,
1849 struct inode *inode,
1850 char *name,
1851 struct ocfs2_dir_lookup_result *lookup)
1852 {
1853 struct inode *orphan_dir_inode;
1854 struct buffer_head *orphan_dir_bh = NULL;
1855 int status = 0;
1856
1857 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1858 if (status < 0) {
1859 mlog_errno(status);
1860 return status;
1861 }
1862
1863 orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1864 ORPHAN_DIR_SYSTEM_INODE,
1865 osb->slot_num);
1866 if (!orphan_dir_inode) {
1867 status = -ENOENT;
1868 mlog_errno(status);
1869 return status;
1870 }
1871
1872 mutex_lock(&orphan_dir_inode->i_mutex);
1873
1874 status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1875 if (status < 0) {
1876 mlog_errno(status);
1877 goto leave;
1878 }
1879
1880 status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1881 orphan_dir_bh, name,
1882 OCFS2_ORPHAN_NAMELEN, lookup);
1883 if (status < 0) {
1884 ocfs2_inode_unlock(orphan_dir_inode, 1);
1885
1886 mlog_errno(status);
1887 goto leave;
1888 }
1889
1890 *ret_orphan_dir = orphan_dir_inode;
1891
1892 leave:
1893 if (status) {
1894 mutex_unlock(&orphan_dir_inode->i_mutex);
1895 iput(orphan_dir_inode);
1896 }
1897
1898 brelse(orphan_dir_bh);
1899
1900 mlog_exit(status);
1901 return status;
1902 }
1903
1904 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1905 handle_t *handle,
1906 struct inode *inode,
1907 struct ocfs2_dinode *fe,
1908 char *name,
1909 struct ocfs2_dir_lookup_result *lookup,
1910 struct inode *orphan_dir_inode)
1911 {
1912 struct buffer_head *orphan_dir_bh = NULL;
1913 int status = 0;
1914 struct ocfs2_dinode *orphan_fe;
1915
1916 mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
1917
1918 status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
1919 if (status < 0) {
1920 mlog_errno(status);
1921 goto leave;
1922 }
1923
1924 status = ocfs2_journal_access_di(handle, orphan_dir_inode, orphan_dir_bh,
1925 OCFS2_JOURNAL_ACCESS_WRITE);
1926 if (status < 0) {
1927 mlog_errno(status);
1928 goto leave;
1929 }
1930
1931 /* we're a cluster, and nlink can change on disk from
1932 * underneath us... */
1933 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1934 if (S_ISDIR(inode->i_mode))
1935 ocfs2_add_links_count(orphan_fe, 1);
1936 orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
1937
1938 status = ocfs2_journal_dirty(handle, orphan_dir_bh);
1939 if (status < 0) {
1940 mlog_errno(status);
1941 goto leave;
1942 }
1943
1944 status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
1945 OCFS2_ORPHAN_NAMELEN, inode,
1946 OCFS2_I(inode)->ip_blkno,
1947 orphan_dir_bh, lookup);
1948 if (status < 0) {
1949 mlog_errno(status);
1950 goto leave;
1951 }
1952
1953 le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
1954
1955 /* Record which orphan dir our inode now resides
1956 * in. delete_inode will use this to determine which orphan
1957 * dir to lock. */
1958 fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
1959
1960 mlog(0, "Inode %llu orphaned in slot %d\n",
1961 (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
1962
1963 leave:
1964 brelse(orphan_dir_bh);
1965
1966 mlog_exit(status);
1967 return status;
1968 }
1969
1970 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
1971 int ocfs2_orphan_del(struct ocfs2_super *osb,
1972 handle_t *handle,
1973 struct inode *orphan_dir_inode,
1974 struct inode *inode,
1975 struct buffer_head *orphan_dir_bh)
1976 {
1977 char name[OCFS2_ORPHAN_NAMELEN + 1];
1978 struct ocfs2_dinode *orphan_fe;
1979 int status = 0;
1980 struct ocfs2_dir_lookup_result lookup = { NULL, };
1981
1982 mlog_entry_void();
1983
1984 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1985 if (status < 0) {
1986 mlog_errno(status);
1987 goto leave;
1988 }
1989
1990 mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
1991 name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
1992 OCFS2_ORPHAN_NAMELEN);
1993
1994 /* find it's spot in the orphan directory */
1995 status = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN, orphan_dir_inode,
1996 &lookup);
1997 if (status) {
1998 mlog_errno(status);
1999 goto leave;
2000 }
2001
2002 /* remove it from the orphan directory */
2003 status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2004 if (status < 0) {
2005 mlog_errno(status);
2006 goto leave;
2007 }
2008
2009 status = ocfs2_journal_access_di(handle,orphan_dir_inode, orphan_dir_bh,
2010 OCFS2_JOURNAL_ACCESS_WRITE);
2011 if (status < 0) {
2012 mlog_errno(status);
2013 goto leave;
2014 }
2015
2016 /* do the i_nlink dance! :) */
2017 orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2018 if (S_ISDIR(inode->i_mode))
2019 ocfs2_add_links_count(orphan_fe, -1);
2020 orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
2021
2022 status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2023 if (status < 0) {
2024 mlog_errno(status);
2025 goto leave;
2026 }
2027
2028 leave:
2029 ocfs2_free_dir_lookup_result(&lookup);
2030
2031 mlog_exit(status);
2032 return status;
2033 }
2034
2035 const struct inode_operations ocfs2_dir_iops = {
2036 .create = ocfs2_create,
2037 .lookup = ocfs2_lookup,
2038 .link = ocfs2_link,
2039 .unlink = ocfs2_unlink,
2040 .rmdir = ocfs2_unlink,
2041 .symlink = ocfs2_symlink,
2042 .mkdir = ocfs2_mkdir,
2043 .mknod = ocfs2_mknod,
2044 .rename = ocfs2_rename,
2045 .setattr = ocfs2_setattr,
2046 .getattr = ocfs2_getattr,
2047 .permission = ocfs2_permission,
2048 .setxattr = generic_setxattr,
2049 .getxattr = generic_getxattr,
2050 .listxattr = ocfs2_listxattr,
2051 .removexattr = generic_removexattr,
2052 };
ubuntu-zesty-kernel mirror