1/* * This file is part of UBIFS.
2 *
3 * Copyright (C) 2006-2008 Nokia Corporation.
4 * Copyright (C) 2006, 2007 University of Szeged, Hungary
5 *
6 * This program is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU General Public License version 2 as published by
8 * the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13 * more details.
14 *
15 * You should have received a copy of the GNU General Public License along with
16 * this program; if not, write to the Free Software Foundation, Inc., 51
17 * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 *
19 * Authors: Artem Bityutskiy (���������������� ����������)
20 *          Adrian Hunter
21 *          Zoltan Sogor
22 */
23
24/*
25 * This file implements directory operations.
26 *
27 * All FS operations in this file allocate budget before writing anything to the
28 * media. If they fail to allocate it, the error is returned. The only
29 * exceptions are 'ubifs_unlink()' and 'ubifs_rmdir()' which keep working even
30 * if they unable to allocate the budget, because deletion %-ENOSPC failure is
31 * not what users are usually ready to get. UBIFS budgeting subsystem has some
32 * space reserved for these purposes.
33 *
34 * All operations in this file write all inodes which they change straight
35 * away, instead of marking them dirty. For example, 'ubifs_link()' changes
36 * @i_size of the parent inode and writes the parent inode together with the
37 * target inode. This was done to simplify file-system recovery which would
38 * otherwise be very difficult to do. The only exception is rename which marks
39 * the re-named inode dirty (because its @i_ctime is updated) but does not
40 * write it, but just marks it as dirty.
41 */
42
43#include "ubifs.h"
44
45/**
46 * inherit_flags - inherit flags of the parent inode.
47 * @dir: parent inode
48 * @mode: new inode mode flags
49 *
50 * This is a helper function for 'ubifs_new_inode()' which inherits flag of the
51 * parent directory inode @dir. UBIFS inodes inherit the following flags:
52 * o %UBIFS_COMPR_FL, which is useful to switch compression on/of on
53 *   sub-directory basis;
54 * o %UBIFS_SYNC_FL - useful for the same reasons;
55 * o %UBIFS_DIRSYNC_FL - similar, but relevant only to directories.
56 *
57 * This function returns the inherited flags.
58 */
59static int inherit_flags(const struct inode *dir, umode_t mode)
60{
61	int flags;
62	const struct ubifs_inode *ui = ubifs_inode(dir);
63
64	if (!S_ISDIR(dir->i_mode))
65		/*
66		 * The parent is not a directory, which means that an extended
67		 * attribute inode is being created. No flags.
68		 */
69		return 0;
70
71	flags = ui->flags & (UBIFS_COMPR_FL | UBIFS_SYNC_FL | UBIFS_DIRSYNC_FL);
72	if (!S_ISDIR(mode))
73		/* The "DIRSYNC" flag only applies to directories */
74		flags &= ~UBIFS_DIRSYNC_FL;
75	return flags;
76}
77
78/**
79 * ubifs_new_inode - allocate new UBIFS inode object.
80 * @c: UBIFS file-system description object
81 * @dir: parent directory inode
82 * @mode: inode mode flags
83 *
84 * This function finds an unused inode number, allocates new inode and
85 * initializes it. Returns new inode in case of success and an error code in
86 * case of failure.
87 */
88struct inode *ubifs_new_inode(struct ubifs_info *c, const struct inode *dir,
89			      umode_t mode)
90{
91	struct inode *inode;
92	struct ubifs_inode *ui;
93
94	inode = new_inode(c->vfs_sb);
95	ui = ubifs_inode(inode);
96	if (!inode)
97		return ERR_PTR(-ENOMEM);
98
99	/*
100	 * Set 'S_NOCMTIME' to prevent VFS form updating [mc]time of inodes and
101	 * marking them dirty in file write path (see 'file_update_time()').
102	 * UBIFS has to fully control "clean <-> dirty" transitions of inodes
103	 * to make budgeting work.
104	 */
105	inode->i_flags |= S_NOCMTIME;
106
107	inode_init_owner(inode, dir, mode);
108	inode->i_mtime = inode->i_atime = inode->i_ctime =
109			 ubifs_current_time(inode);
110	inode->i_mapping->nrpages = 0;
111
112	switch (mode & S_IFMT) {
113	case S_IFREG:
114		inode->i_mapping->a_ops = &ubifs_file_address_operations;
115		inode->i_op = &ubifs_file_inode_operations;
116		inode->i_fop = &ubifs_file_operations;
117		break;
118	case S_IFDIR:
119		inode->i_op  = &ubifs_dir_inode_operations;
120		inode->i_fop = &ubifs_dir_operations;
121		inode->i_size = ui->ui_size = UBIFS_INO_NODE_SZ;
122		break;
123	case S_IFLNK:
124		inode->i_op = &ubifs_symlink_inode_operations;
125		break;
126	case S_IFSOCK:
127	case S_IFIFO:
128	case S_IFBLK:
129	case S_IFCHR:
130		inode->i_op  = &ubifs_file_inode_operations;
131		break;
132	default:
133		BUG();
134	}
135
136	ui->flags = inherit_flags(dir, mode);
137	ubifs_set_inode_flags(inode);
138	if (S_ISREG(mode))
139		ui->compr_type = c->default_compr;
140	else
141		ui->compr_type = UBIFS_COMPR_NONE;
142	ui->synced_i_size = 0;
143
144	spin_lock(&c->cnt_lock);
145	/* Inode number overflow is currently not supported */
146	if (c->highest_inum >= INUM_WARN_WATERMARK) {
147		if (c->highest_inum >= INUM_WATERMARK) {
148			spin_unlock(&c->cnt_lock);
149			ubifs_err(c, "out of inode numbers");
150			make_bad_inode(inode);
151			iput(inode);
152			return ERR_PTR(-EINVAL);
153		}
154		ubifs_warn(c, "running out of inode numbers (current %lu, max %u)",
155			   (unsigned long)c->highest_inum, INUM_WATERMARK);
156	}
157
158	inode->i_ino = ++c->highest_inum;
159	/*
160	 * The creation sequence number remains with this inode for its
161	 * lifetime. All nodes for this inode have a greater sequence number,
162	 * and so it is possible to distinguish obsolete nodes belonging to a
163	 * previous incarnation of the same inode number - for example, for the
164	 * purpose of rebuilding the index.
165	 */
166	ui->creat_sqnum = ++c->max_sqnum;
167	spin_unlock(&c->cnt_lock);
168	return inode;
169}
170
171static int dbg_check_name(const struct ubifs_info *c,
172			  const struct ubifs_dent_node *dent,
173			  const struct qstr *nm)
174{
175	if (!dbg_is_chk_gen(c))
176		return 0;
177	if (le16_to_cpu(dent->nlen) != nm->len)
178		return -EINVAL;
179	if (memcmp(dent->name, nm->name, nm->len))
180		return -EINVAL;
181	return 0;
182}
183
184static struct dentry *ubifs_lookup(struct inode *dir, struct dentry *dentry,
185				   unsigned int flags)
186{
187	int err;
188	union ubifs_key key;
189	struct inode *inode = NULL;
190	struct ubifs_dent_node *dent;
191	struct ubifs_info *c = dir->i_sb->s_fs_info;
192
193	dbg_gen("'%pd' in dir ino %lu", dentry, dir->i_ino);
194
195	if (dentry->d_name.len > UBIFS_MAX_NLEN)
196		return ERR_PTR(-ENAMETOOLONG);
197
198	dent = kmalloc(UBIFS_MAX_DENT_NODE_SZ, GFP_NOFS);
199	if (!dent)
200		return ERR_PTR(-ENOMEM);
201
202	dent_key_init(c, &key, dir->i_ino, &dentry->d_name);
203
204	err = ubifs_tnc_lookup_nm(c, &key, dent, &dentry->d_name);
205	if (err) {
206		if (err == -ENOENT) {
207			dbg_gen("not found");
208			goto done;
209		}
210		goto out;
211	}
212
213	if (dbg_check_name(c, dent, &dentry->d_name)) {
214		err = -EINVAL;
215		goto out;
216	}
217
218	inode = ubifs_iget(dir->i_sb, le64_to_cpu(dent->inum));
219	if (IS_ERR(inode)) {
220		/*
221		 * This should not happen. Probably the file-system needs
222		 * checking.
223		 */
224		err = PTR_ERR(inode);
225		ubifs_err(c, "dead directory entry '%pd', error %d",
226			  dentry, err);
227		ubifs_ro_mode(c, err);
228		goto out;
229	}
230
231done:
232	kfree(dent);
233	/*
234	 * Note, d_splice_alias() would be required instead if we supported
235	 * NFS.
236	 */
237	d_add(dentry, inode);
238	return NULL;
239
240out:
241	kfree(dent);
242	return ERR_PTR(err);
243}
244
245static int ubifs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
246			bool excl)
247{
248	struct inode *inode;
249	struct ubifs_info *c = dir->i_sb->s_fs_info;
250	int err, sz_change = CALC_DENT_SIZE(dentry->d_name.len);
251	struct ubifs_budget_req req = { .new_ino = 1, .new_dent = 1,
252					.dirtied_ino = 1 };
253	struct ubifs_inode *dir_ui = ubifs_inode(dir);
254
255	/*
256	 * Budget request settings: new inode, new direntry, changing the
257	 * parent directory inode.
258	 */
259
260	dbg_gen("dent '%pd', mode %#hx in dir ino %lu",
261		dentry, mode, dir->i_ino);
262
263	err = ubifs_budget_space(c, &req);
264	if (err)
265		return err;
266
267	inode = ubifs_new_inode(c, dir, mode);
268	if (IS_ERR(inode)) {
269		err = PTR_ERR(inode);
270		goto out_budg;
271	}
272
273	err = ubifs_init_security(dir, inode, &dentry->d_name);
274	if (err)
275		goto out_inode;
276
277	mutex_lock(&dir_ui->ui_mutex);
278	dir->i_size += sz_change;
279	dir_ui->ui_size = dir->i_size;
280	dir->i_mtime = dir->i_ctime = inode->i_ctime;
281	err = ubifs_jnl_update(c, dir, &dentry->d_name, inode, 0, 0);
282	if (err)
283		goto out_cancel;
284	mutex_unlock(&dir_ui->ui_mutex);
285
286	ubifs_release_budget(c, &req);
287	insert_inode_hash(inode);
288	d_instantiate(dentry, inode);
289	return 0;
290
291out_cancel:
292	dir->i_size -= sz_change;
293	dir_ui->ui_size = dir->i_size;
294	mutex_unlock(&dir_ui->ui_mutex);
295out_inode:
296	make_bad_inode(inode);
297	iput(inode);
298out_budg:
299	ubifs_release_budget(c, &req);
300	ubifs_err(c, "cannot create regular file, error %d", err);
301	return err;
302}
303
304/**
305 * vfs_dent_type - get VFS directory entry type.
306 * @type: UBIFS directory entry type
307 *
308 * This function converts UBIFS directory entry type into VFS directory entry
309 * type.
310 */
311static unsigned int vfs_dent_type(uint8_t type)
312{
313	switch (type) {
314	case UBIFS_ITYPE_REG:
315		return DT_REG;
316	case UBIFS_ITYPE_DIR:
317		return DT_DIR;
318	case UBIFS_ITYPE_LNK:
319		return DT_LNK;
320	case UBIFS_ITYPE_BLK:
321		return DT_BLK;
322	case UBIFS_ITYPE_CHR:
323		return DT_CHR;
324	case UBIFS_ITYPE_FIFO:
325		return DT_FIFO;
326	case UBIFS_ITYPE_SOCK:
327		return DT_SOCK;
328	default:
329		BUG();
330	}
331	return 0;
332}
333
334/*
335 * The classical Unix view for directory is that it is a linear array of
336 * (name, inode number) entries. Linux/VFS assumes this model as well.
337 * Particularly, 'readdir()' call wants us to return a directory entry offset
338 * which later may be used to continue 'readdir()'ing the directory or to
339 * 'seek()' to that specific direntry. Obviously UBIFS does not really fit this
340 * model because directory entries are identified by keys, which may collide.
341 *
342 * UBIFS uses directory entry hash value for directory offsets, so
343 * 'seekdir()'/'telldir()' may not always work because of possible key
344 * collisions. But UBIFS guarantees that consecutive 'readdir()' calls work
345 * properly by means of saving full directory entry name in the private field
346 * of the file description object.
347 *
348 * This means that UBIFS cannot support NFS which requires full
349 * 'seekdir()'/'telldir()' support.
350 */
351static int ubifs_readdir(struct file *file, struct dir_context *ctx)
352{
353	int err;
354	struct qstr nm;
355	union ubifs_key key;
356	struct ubifs_dent_node *dent;
357	struct inode *dir = file_inode(file);
358	struct ubifs_info *c = dir->i_sb->s_fs_info;
359
360	dbg_gen("dir ino %lu, f_pos %#llx", dir->i_ino, ctx->pos);
361
362	if (ctx->pos > UBIFS_S_KEY_HASH_MASK || ctx->pos == 2)
363		/*
364		 * The directory was seek'ed to a senseless position or there
365		 * are no more entries.
366		 */
367		return 0;
368
369	if (file->f_version == 0) {
370		/*
371		 * The file was seek'ed, which means that @file->private_data
372		 * is now invalid. This may also be just the first
373		 * 'ubifs_readdir()' invocation, in which case
374		 * @file->private_data is NULL, and the below code is
375		 * basically a no-op.
376		 */
377		kfree(file->private_data);
378		file->private_data = NULL;
379	}
380
381	/*
382	 * 'generic_file_llseek()' unconditionally sets @file->f_version to
383	 * zero, and we use this for detecting whether the file was seek'ed.
384	 */
385	file->f_version = 1;
386
387	/* File positions 0 and 1 correspond to "." and ".." */
388	if (ctx->pos < 2) {
389		ubifs_assert(!file->private_data);
390		if (!dir_emit_dots(file, ctx))
391			return 0;
392
393		/* Find the first entry in TNC and save it */
394		lowest_dent_key(c, &key, dir->i_ino);
395		nm.name = NULL;
396		dent = ubifs_tnc_next_ent(c, &key, &nm);
397		if (IS_ERR(dent)) {
398			err = PTR_ERR(dent);
399			goto out;
400		}
401
402		ctx->pos = key_hash_flash(c, &dent->key);
403		file->private_data = dent;
404	}
405
406	dent = file->private_data;
407	if (!dent) {
408		/*
409		 * The directory was seek'ed to and is now readdir'ed.
410		 * Find the entry corresponding to @ctx->pos or the closest one.
411		 */
412		dent_key_init_hash(c, &key, dir->i_ino, ctx->pos);
413		nm.name = NULL;
414		dent = ubifs_tnc_next_ent(c, &key, &nm);
415		if (IS_ERR(dent)) {
416			err = PTR_ERR(dent);
417			goto out;
418		}
419		ctx->pos = key_hash_flash(c, &dent->key);
420		file->private_data = dent;
421	}
422
423	while (1) {
424		dbg_gen("feed '%s', ino %llu, new f_pos %#x",
425			dent->name, (unsigned long long)le64_to_cpu(dent->inum),
426			key_hash_flash(c, &dent->key));
427		ubifs_assert(le64_to_cpu(dent->ch.sqnum) >
428			     ubifs_inode(dir)->creat_sqnum);
429
430		nm.len = le16_to_cpu(dent->nlen);
431		if (!dir_emit(ctx, dent->name, nm.len,
432			       le64_to_cpu(dent->inum),
433			       vfs_dent_type(dent->type)))
434			return 0;
435
436		/* Switch to the next entry */
437		key_read(c, &dent->key, &key);
438		nm.name = dent->name;
439		dent = ubifs_tnc_next_ent(c, &key, &nm);
440		if (IS_ERR(dent)) {
441			err = PTR_ERR(dent);
442			goto out;
443		}
444
445		kfree(file->private_data);
446		ctx->pos = key_hash_flash(c, &dent->key);
447		file->private_data = dent;
448		cond_resched();
449	}
450
451out:
452	kfree(file->private_data);
453	file->private_data = NULL;
454
455	if (err != -ENOENT) {
456		ubifs_err(c, "cannot find next direntry, error %d", err);
457		return err;
458	}
459
460	/* 2 is a special value indicating that there are no more direntries */
461	ctx->pos = 2;
462	return 0;
463}
464
465/* Free saved readdir() state when the directory is closed */
466static int ubifs_dir_release(struct inode *dir, struct file *file)
467{
468	kfree(file->private_data);
469	file->private_data = NULL;
470	return 0;
471}
472
473/**
474 * lock_2_inodes - a wrapper for locking two UBIFS inodes.
475 * @inode1: first inode
476 * @inode2: second inode
477 *
478 * We do not implement any tricks to guarantee strict lock ordering, because
479 * VFS has already done it for us on the @i_mutex. So this is just a simple
480 * wrapper function.
481 */
482static void lock_2_inodes(struct inode *inode1, struct inode *inode2)
483{
484	mutex_lock_nested(&ubifs_inode(inode1)->ui_mutex, WB_MUTEX_1);
485	mutex_lock_nested(&ubifs_inode(inode2)->ui_mutex, WB_MUTEX_2);
486}
487
488/**
489 * unlock_2_inodes - a wrapper for unlocking two UBIFS inodes.
490 * @inode1: first inode
491 * @inode2: second inode
492 */
493static void unlock_2_inodes(struct inode *inode1, struct inode *inode2)
494{
495	mutex_unlock(&ubifs_inode(inode2)->ui_mutex);
496	mutex_unlock(&ubifs_inode(inode1)->ui_mutex);
497}
498
499static int ubifs_link(struct dentry *old_dentry, struct inode *dir,
500		      struct dentry *dentry)
501{
502	struct ubifs_info *c = dir->i_sb->s_fs_info;
503	struct inode *inode = d_inode(old_dentry);
504	struct ubifs_inode *ui = ubifs_inode(inode);
505	struct ubifs_inode *dir_ui = ubifs_inode(dir);
506	int err, sz_change = CALC_DENT_SIZE(dentry->d_name.len);
507	struct ubifs_budget_req req = { .new_dent = 1, .dirtied_ino = 2,
508				.dirtied_ino_d = ALIGN(ui->data_len, 8) };
509
510	/*
511	 * Budget request settings: new direntry, changing the target inode,
512	 * changing the parent inode.
513	 */
514
515	dbg_gen("dent '%pd' to ino %lu (nlink %d) in dir ino %lu",
516		dentry, inode->i_ino,
517		inode->i_nlink, dir->i_ino);
518	ubifs_assert(mutex_is_locked(&dir->i_mutex));
519	ubifs_assert(mutex_is_locked(&inode->i_mutex));
520
521	err = dbg_check_synced_i_size(c, inode);
522	if (err)
523		return err;
524
525	err = ubifs_budget_space(c, &req);
526	if (err)
527		return err;
528
529	lock_2_inodes(dir, inode);
530	inc_nlink(inode);
531	ihold(inode);
532	inode->i_ctime = ubifs_current_time(inode);
533	dir->i_size += sz_change;
534	dir_ui->ui_size = dir->i_size;
535	dir->i_mtime = dir->i_ctime = inode->i_ctime;
536	err = ubifs_jnl_update(c, dir, &dentry->d_name, inode, 0, 0);
537	if (err)
538		goto out_cancel;
539	unlock_2_inodes(dir, inode);
540
541	ubifs_release_budget(c, &req);
542	d_instantiate(dentry, inode);
543	return 0;
544
545out_cancel:
546	dir->i_size -= sz_change;
547	dir_ui->ui_size = dir->i_size;
548	drop_nlink(inode);
549	unlock_2_inodes(dir, inode);
550	ubifs_release_budget(c, &req);
551	iput(inode);
552	return err;
553}
554
555static int ubifs_unlink(struct inode *dir, struct dentry *dentry)
556{
557	struct ubifs_info *c = dir->i_sb->s_fs_info;
558	struct inode *inode = d_inode(dentry);
559	struct ubifs_inode *dir_ui = ubifs_inode(dir);
560	int sz_change = CALC_DENT_SIZE(dentry->d_name.len);
561	int err, budgeted = 1;
562	struct ubifs_budget_req req = { .mod_dent = 1, .dirtied_ino = 2 };
563	unsigned int saved_nlink = inode->i_nlink;
564
565	/*
566	 * Budget request settings: deletion direntry, deletion inode (+1 for
567	 * @dirtied_ino), changing the parent directory inode. If budgeting
568	 * fails, go ahead anyway because we have extra space reserved for
569	 * deletions.
570	 */
571
572	dbg_gen("dent '%pd' from ino %lu (nlink %d) in dir ino %lu",
573		dentry, inode->i_ino,
574		inode->i_nlink, dir->i_ino);
575	ubifs_assert(mutex_is_locked(&dir->i_mutex));
576	ubifs_assert(mutex_is_locked(&inode->i_mutex));
577	err = dbg_check_synced_i_size(c, inode);
578	if (err)
579		return err;
580
581	err = ubifs_budget_space(c, &req);
582	if (err) {
583		if (err != -ENOSPC)
584			return err;
585		budgeted = 0;
586	}
587
588	lock_2_inodes(dir, inode);
589	inode->i_ctime = ubifs_current_time(dir);
590	drop_nlink(inode);
591	dir->i_size -= sz_change;
592	dir_ui->ui_size = dir->i_size;
593	dir->i_mtime = dir->i_ctime = inode->i_ctime;
594	err = ubifs_jnl_update(c, dir, &dentry->d_name, inode, 1, 0);
595	if (err)
596		goto out_cancel;
597	unlock_2_inodes(dir, inode);
598
599	if (budgeted)
600		ubifs_release_budget(c, &req);
601	else {
602		/* We've deleted something - clean the "no space" flags */
603		c->bi.nospace = c->bi.nospace_rp = 0;
604		smp_wmb();
605	}
606	return 0;
607
608out_cancel:
609	dir->i_size += sz_change;
610	dir_ui->ui_size = dir->i_size;
611	set_nlink(inode, saved_nlink);
612	unlock_2_inodes(dir, inode);
613	if (budgeted)
614		ubifs_release_budget(c, &req);
615	return err;
616}
617
618/**
619 * check_dir_empty - check if a directory is empty or not.
620 * @c: UBIFS file-system description object
621 * @dir: VFS inode object of the directory to check
622 *
623 * This function checks if directory @dir is empty. Returns zero if the
624 * directory is empty, %-ENOTEMPTY if it is not, and other negative error codes
625 * in case of of errors.
626 */
627static int check_dir_empty(struct ubifs_info *c, struct inode *dir)
628{
629	struct qstr nm = { .name = NULL };
630	struct ubifs_dent_node *dent;
631	union ubifs_key key;
632	int err;
633
634	lowest_dent_key(c, &key, dir->i_ino);
635	dent = ubifs_tnc_next_ent(c, &key, &nm);
636	if (IS_ERR(dent)) {
637		err = PTR_ERR(dent);
638		if (err == -ENOENT)
639			err = 0;
640	} else {
641		kfree(dent);
642		err = -ENOTEMPTY;
643	}
644	return err;
645}
646
647static int ubifs_rmdir(struct inode *dir, struct dentry *dentry)
648{
649	struct ubifs_info *c = dir->i_sb->s_fs_info;
650	struct inode *inode = d_inode(dentry);
651	int sz_change = CALC_DENT_SIZE(dentry->d_name.len);
652	int err, budgeted = 1;
653	struct ubifs_inode *dir_ui = ubifs_inode(dir);
654	struct ubifs_budget_req req = { .mod_dent = 1, .dirtied_ino = 2 };
655
656	/*
657	 * Budget request settings: deletion direntry, deletion inode and
658	 * changing the parent inode. If budgeting fails, go ahead anyway
659	 * because we have extra space reserved for deletions.
660	 */
661
662	dbg_gen("directory '%pd', ino %lu in dir ino %lu", dentry,
663		inode->i_ino, dir->i_ino);
664	ubifs_assert(mutex_is_locked(&dir->i_mutex));
665	ubifs_assert(mutex_is_locked(&inode->i_mutex));
666	err = check_dir_empty(c, d_inode(dentry));
667	if (err)
668		return err;
669
670	err = ubifs_budget_space(c, &req);
671	if (err) {
672		if (err != -ENOSPC)
673			return err;
674		budgeted = 0;
675	}
676
677	lock_2_inodes(dir, inode);
678	inode->i_ctime = ubifs_current_time(dir);
679	clear_nlink(inode);
680	drop_nlink(dir);
681	dir->i_size -= sz_change;
682	dir_ui->ui_size = dir->i_size;
683	dir->i_mtime = dir->i_ctime = inode->i_ctime;
684	err = ubifs_jnl_update(c, dir, &dentry->d_name, inode, 1, 0);
685	if (err)
686		goto out_cancel;
687	unlock_2_inodes(dir, inode);
688
689	if (budgeted)
690		ubifs_release_budget(c, &req);
691	else {
692		/* We've deleted something - clean the "no space" flags */
693		c->bi.nospace = c->bi.nospace_rp = 0;
694		smp_wmb();
695	}
696	return 0;
697
698out_cancel:
699	dir->i_size += sz_change;
700	dir_ui->ui_size = dir->i_size;
701	inc_nlink(dir);
702	set_nlink(inode, 2);
703	unlock_2_inodes(dir, inode);
704	if (budgeted)
705		ubifs_release_budget(c, &req);
706	return err;
707}
708
709static int ubifs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
710{
711	struct inode *inode;
712	struct ubifs_inode *dir_ui = ubifs_inode(dir);
713	struct ubifs_info *c = dir->i_sb->s_fs_info;
714	int err, sz_change = CALC_DENT_SIZE(dentry->d_name.len);
715	struct ubifs_budget_req req = { .new_ino = 1, .new_dent = 1 };
716
717	/*
718	 * Budget request settings: new inode, new direntry and changing parent
719	 * directory inode.
720	 */
721
722	dbg_gen("dent '%pd', mode %#hx in dir ino %lu",
723		dentry, mode, dir->i_ino);
724
725	err = ubifs_budget_space(c, &req);
726	if (err)
727		return err;
728
729	inode = ubifs_new_inode(c, dir, S_IFDIR | mode);
730	if (IS_ERR(inode)) {
731		err = PTR_ERR(inode);
732		goto out_budg;
733	}
734
735	err = ubifs_init_security(dir, inode, &dentry->d_name);
736	if (err)
737		goto out_inode;
738
739	mutex_lock(&dir_ui->ui_mutex);
740	insert_inode_hash(inode);
741	inc_nlink(inode);
742	inc_nlink(dir);
743	dir->i_size += sz_change;
744	dir_ui->ui_size = dir->i_size;
745	dir->i_mtime = dir->i_ctime = inode->i_ctime;
746	err = ubifs_jnl_update(c, dir, &dentry->d_name, inode, 0, 0);
747	if (err) {
748		ubifs_err(c, "cannot create directory, error %d", err);
749		goto out_cancel;
750	}
751	mutex_unlock(&dir_ui->ui_mutex);
752
753	ubifs_release_budget(c, &req);
754	d_instantiate(dentry, inode);
755	return 0;
756
757out_cancel:
758	dir->i_size -= sz_change;
759	dir_ui->ui_size = dir->i_size;
760	drop_nlink(dir);
761	mutex_unlock(&dir_ui->ui_mutex);
762out_inode:
763	make_bad_inode(inode);
764	iput(inode);
765out_budg:
766	ubifs_release_budget(c, &req);
767	return err;
768}
769
770static int ubifs_mknod(struct inode *dir, struct dentry *dentry,
771		       umode_t mode, dev_t rdev)
772{
773	struct inode *inode;
774	struct ubifs_inode *ui;
775	struct ubifs_inode *dir_ui = ubifs_inode(dir);
776	struct ubifs_info *c = dir->i_sb->s_fs_info;
777	union ubifs_dev_desc *dev = NULL;
778	int sz_change = CALC_DENT_SIZE(dentry->d_name.len);
779	int err, devlen = 0;
780	struct ubifs_budget_req req = { .new_ino = 1, .new_dent = 1,
781					.new_ino_d = ALIGN(devlen, 8),
782					.dirtied_ino = 1 };
783
784	/*
785	 * Budget request settings: new inode, new direntry and changing parent
786	 * directory inode.
787	 */
788
789	dbg_gen("dent '%pd' in dir ino %lu", dentry, dir->i_ino);
790
791	if (S_ISBLK(mode) || S_ISCHR(mode)) {
792		dev = kmalloc(sizeof(union ubifs_dev_desc), GFP_NOFS);
793		if (!dev)
794			return -ENOMEM;
795		devlen = ubifs_encode_dev(dev, rdev);
796	}
797
798	err = ubifs_budget_space(c, &req);
799	if (err) {
800		kfree(dev);
801		return err;
802	}
803
804	inode = ubifs_new_inode(c, dir, mode);
805	if (IS_ERR(inode)) {
806		kfree(dev);
807		err = PTR_ERR(inode);
808		goto out_budg;
809	}
810
811	init_special_inode(inode, inode->i_mode, rdev);
812	inode->i_size = ubifs_inode(inode)->ui_size = devlen;
813	ui = ubifs_inode(inode);
814	ui->data = dev;
815	ui->data_len = devlen;
816
817	err = ubifs_init_security(dir, inode, &dentry->d_name);
818	if (err)
819		goto out_inode;
820
821	mutex_lock(&dir_ui->ui_mutex);
822	dir->i_size += sz_change;
823	dir_ui->ui_size = dir->i_size;
824	dir->i_mtime = dir->i_ctime = inode->i_ctime;
825	err = ubifs_jnl_update(c, dir, &dentry->d_name, inode, 0, 0);
826	if (err)
827		goto out_cancel;
828	mutex_unlock(&dir_ui->ui_mutex);
829
830	ubifs_release_budget(c, &req);
831	insert_inode_hash(inode);
832	d_instantiate(dentry, inode);
833	return 0;
834
835out_cancel:
836	dir->i_size -= sz_change;
837	dir_ui->ui_size = dir->i_size;
838	mutex_unlock(&dir_ui->ui_mutex);
839out_inode:
840	make_bad_inode(inode);
841	iput(inode);
842out_budg:
843	ubifs_release_budget(c, &req);
844	return err;
845}
846
847static int ubifs_symlink(struct inode *dir, struct dentry *dentry,
848			 const char *symname)
849{
850	struct inode *inode;
851	struct ubifs_inode *ui;
852	struct ubifs_inode *dir_ui = ubifs_inode(dir);
853	struct ubifs_info *c = dir->i_sb->s_fs_info;
854	int err, len = strlen(symname);
855	int sz_change = CALC_DENT_SIZE(dentry->d_name.len);
856	struct ubifs_budget_req req = { .new_ino = 1, .new_dent = 1,
857					.new_ino_d = ALIGN(len, 8),
858					.dirtied_ino = 1 };
859
860	/*
861	 * Budget request settings: new inode, new direntry and changing parent
862	 * directory inode.
863	 */
864
865	dbg_gen("dent '%pd', target '%s' in dir ino %lu", dentry,
866		symname, dir->i_ino);
867
868	if (len > UBIFS_MAX_INO_DATA)
869		return -ENAMETOOLONG;
870
871	err = ubifs_budget_space(c, &req);
872	if (err)
873		return err;
874
875	inode = ubifs_new_inode(c, dir, S_IFLNK | S_IRWXUGO);
876	if (IS_ERR(inode)) {
877		err = PTR_ERR(inode);
878		goto out_budg;
879	}
880
881	ui = ubifs_inode(inode);
882	ui->data = kmalloc(len + 1, GFP_NOFS);
883	if (!ui->data) {
884		err = -ENOMEM;
885		goto out_inode;
886	}
887
888	memcpy(ui->data, symname, len);
889	((char *)ui->data)[len] = '\0';
890	inode->i_link = ui->data;
891	/*
892	 * The terminating zero byte is not written to the flash media and it
893	 * is put just to make later in-memory string processing simpler. Thus,
894	 * data length is @len, not @len + %1.
895	 */
896	ui->data_len = len;
897	inode->i_size = ubifs_inode(inode)->ui_size = len;
898
899	err = ubifs_init_security(dir, inode, &dentry->d_name);
900	if (err)
901		goto out_inode;
902
903	mutex_lock(&dir_ui->ui_mutex);
904	dir->i_size += sz_change;
905	dir_ui->ui_size = dir->i_size;
906	dir->i_mtime = dir->i_ctime = inode->i_ctime;
907	err = ubifs_jnl_update(c, dir, &dentry->d_name, inode, 0, 0);
908	if (err)
909		goto out_cancel;
910	mutex_unlock(&dir_ui->ui_mutex);
911
912	ubifs_release_budget(c, &req);
913	insert_inode_hash(inode);
914	d_instantiate(dentry, inode);
915	return 0;
916
917out_cancel:
918	dir->i_size -= sz_change;
919	dir_ui->ui_size = dir->i_size;
920	mutex_unlock(&dir_ui->ui_mutex);
921out_inode:
922	make_bad_inode(inode);
923	iput(inode);
924out_budg:
925	ubifs_release_budget(c, &req);
926	return err;
927}
928
929/**
930 * lock_3_inodes - a wrapper for locking three UBIFS inodes.
931 * @inode1: first inode
932 * @inode2: second inode
933 * @inode3: third inode
934 *
935 * This function is used for 'ubifs_rename()' and @inode1 may be the same as
936 * @inode2 whereas @inode3 may be %NULL.
937 *
938 * We do not implement any tricks to guarantee strict lock ordering, because
939 * VFS has already done it for us on the @i_mutex. So this is just a simple
940 * wrapper function.
941 */
942static void lock_3_inodes(struct inode *inode1, struct inode *inode2,
943			  struct inode *inode3)
944{
945	mutex_lock_nested(&ubifs_inode(inode1)->ui_mutex, WB_MUTEX_1);
946	if (inode2 != inode1)
947		mutex_lock_nested(&ubifs_inode(inode2)->ui_mutex, WB_MUTEX_2);
948	if (inode3)
949		mutex_lock_nested(&ubifs_inode(inode3)->ui_mutex, WB_MUTEX_3);
950}
951
952/**
953 * unlock_3_inodes - a wrapper for unlocking three UBIFS inodes for rename.
954 * @inode1: first inode
955 * @inode2: second inode
956 * @inode3: third inode
957 */
958static void unlock_3_inodes(struct inode *inode1, struct inode *inode2,
959			    struct inode *inode3)
960{
961	if (inode3)
962		mutex_unlock(&ubifs_inode(inode3)->ui_mutex);
963	if (inode1 != inode2)
964		mutex_unlock(&ubifs_inode(inode2)->ui_mutex);
965	mutex_unlock(&ubifs_inode(inode1)->ui_mutex);
966}
967
968static int ubifs_rename(struct inode *old_dir, struct dentry *old_dentry,
969			struct inode *new_dir, struct dentry *new_dentry)
970{
971	struct ubifs_info *c = old_dir->i_sb->s_fs_info;
972	struct inode *old_inode = d_inode(old_dentry);
973	struct inode *new_inode = d_inode(new_dentry);
974	struct ubifs_inode *old_inode_ui = ubifs_inode(old_inode);
975	int err, release, sync = 0, move = (new_dir != old_dir);
976	int is_dir = S_ISDIR(old_inode->i_mode);
977	int unlink = !!new_inode;
978	int new_sz = CALC_DENT_SIZE(new_dentry->d_name.len);
979	int old_sz = CALC_DENT_SIZE(old_dentry->d_name.len);
980	struct ubifs_budget_req req = { .new_dent = 1, .mod_dent = 1,
981					.dirtied_ino = 3 };
982	struct ubifs_budget_req ino_req = { .dirtied_ino = 1,
983			.dirtied_ino_d = ALIGN(old_inode_ui->data_len, 8) };
984	struct timespec time;
985	unsigned int uninitialized_var(saved_nlink);
986
987	/*
988	 * Budget request settings: deletion direntry, new direntry, removing
989	 * the old inode, and changing old and new parent directory inodes.
990	 *
991	 * However, this operation also marks the target inode as dirty and
992	 * does not write it, so we allocate budget for the target inode
993	 * separately.
994	 */
995
996	dbg_gen("dent '%pd' ino %lu in dir ino %lu to dent '%pd' in dir ino %lu",
997		old_dentry, old_inode->i_ino, old_dir->i_ino,
998		new_dentry, new_dir->i_ino);
999	ubifs_assert(mutex_is_locked(&old_dir->i_mutex));
1000	ubifs_assert(mutex_is_locked(&new_dir->i_mutex));
1001	if (unlink)
1002		ubifs_assert(mutex_is_locked(&new_inode->i_mutex));
1003
1004
1005	if (unlink && is_dir) {
1006		err = check_dir_empty(c, new_inode);
1007		if (err)
1008			return err;
1009	}
1010
1011	err = ubifs_budget_space(c, &req);
1012	if (err)
1013		return err;
1014	err = ubifs_budget_space(c, &ino_req);
1015	if (err) {
1016		ubifs_release_budget(c, &req);
1017		return err;
1018	}
1019
1020	lock_3_inodes(old_dir, new_dir, new_inode);
1021
1022	/*
1023	 * Like most other Unix systems, set the @i_ctime for inodes on a
1024	 * rename.
1025	 */
1026	time = ubifs_current_time(old_dir);
1027	old_inode->i_ctime = time;
1028
1029	/* We must adjust parent link count when renaming directories */
1030	if (is_dir) {
1031		if (move) {
1032			/*
1033			 * @old_dir loses a link because we are moving
1034			 * @old_inode to a different directory.
1035			 */
1036			drop_nlink(old_dir);
1037			/*
1038			 * @new_dir only gains a link if we are not also
1039			 * overwriting an existing directory.
1040			 */
1041			if (!unlink)
1042				inc_nlink(new_dir);
1043		} else {
1044			/*
1045			 * @old_inode is not moving to a different directory,
1046			 * but @old_dir still loses a link if we are
1047			 * overwriting an existing directory.
1048			 */
1049			if (unlink)
1050				drop_nlink(old_dir);
1051		}
1052	}
1053
1054	old_dir->i_size -= old_sz;
1055	ubifs_inode(old_dir)->ui_size = old_dir->i_size;
1056	old_dir->i_mtime = old_dir->i_ctime = time;
1057	new_dir->i_mtime = new_dir->i_ctime = time;
1058
1059	/*
1060	 * And finally, if we unlinked a direntry which happened to have the
1061	 * same name as the moved direntry, we have to decrement @i_nlink of
1062	 * the unlinked inode and change its ctime.
1063	 */
1064	if (unlink) {
1065		/*
1066		 * Directories cannot have hard-links, so if this is a
1067		 * directory, just clear @i_nlink.
1068		 */
1069		saved_nlink = new_inode->i_nlink;
1070		if (is_dir)
1071			clear_nlink(new_inode);
1072		else
1073			drop_nlink(new_inode);
1074		new_inode->i_ctime = time;
1075	} else {
1076		new_dir->i_size += new_sz;
1077		ubifs_inode(new_dir)->ui_size = new_dir->i_size;
1078	}
1079
1080	/*
1081	 * Do not ask 'ubifs_jnl_rename()' to flush write-buffer if @old_inode
1082	 * is dirty, because this will be done later on at the end of
1083	 * 'ubifs_rename()'.
1084	 */
1085	if (IS_SYNC(old_inode)) {
1086		sync = IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir);
1087		if (unlink && IS_SYNC(new_inode))
1088			sync = 1;
1089	}
1090	err = ubifs_jnl_rename(c, old_dir, old_dentry, new_dir, new_dentry,
1091			       sync);
1092	if (err)
1093		goto out_cancel;
1094
1095	unlock_3_inodes(old_dir, new_dir, new_inode);
1096	ubifs_release_budget(c, &req);
1097
1098	mutex_lock(&old_inode_ui->ui_mutex);
1099	release = old_inode_ui->dirty;
1100	mark_inode_dirty_sync(old_inode);
1101	mutex_unlock(&old_inode_ui->ui_mutex);
1102
1103	if (release)
1104		ubifs_release_budget(c, &ino_req);
1105	if (IS_SYNC(old_inode))
1106		err = old_inode->i_sb->s_op->write_inode(old_inode, NULL);
1107	return err;
1108
1109out_cancel:
1110	if (unlink) {
1111		set_nlink(new_inode, saved_nlink);
1112	} else {
1113		new_dir->i_size -= new_sz;
1114		ubifs_inode(new_dir)->ui_size = new_dir->i_size;
1115	}
1116	old_dir->i_size += old_sz;
1117	ubifs_inode(old_dir)->ui_size = old_dir->i_size;
1118	if (is_dir) {
1119		if (move) {
1120			inc_nlink(old_dir);
1121			if (!unlink)
1122				drop_nlink(new_dir);
1123		} else {
1124			if (unlink)
1125				inc_nlink(old_dir);
1126		}
1127	}
1128	unlock_3_inodes(old_dir, new_dir, new_inode);
1129	ubifs_release_budget(c, &ino_req);
1130	ubifs_release_budget(c, &req);
1131	return err;
1132}
1133
1134int ubifs_getattr(struct vfsmount *mnt, struct dentry *dentry,
1135		  struct kstat *stat)
1136{
1137	loff_t size;
1138	struct inode *inode = d_inode(dentry);
1139	struct ubifs_inode *ui = ubifs_inode(inode);
1140
1141	mutex_lock(&ui->ui_mutex);
1142	generic_fillattr(inode, stat);
1143	stat->blksize = UBIFS_BLOCK_SIZE;
1144	stat->size = ui->ui_size;
1145
1146	/*
1147	 * Unfortunately, the 'stat()' system call was designed for block
1148	 * device based file systems, and it is not appropriate for UBIFS,
1149	 * because UBIFS does not have notion of "block". For example, it is
1150	 * difficult to tell how many block a directory takes - it actually
1151	 * takes less than 300 bytes, but we have to round it to block size,
1152	 * which introduces large mistake. This makes utilities like 'du' to
1153	 * report completely senseless numbers. This is the reason why UBIFS
1154	 * goes the same way as JFFS2 - it reports zero blocks for everything
1155	 * but regular files, which makes more sense than reporting completely
1156	 * wrong sizes.
1157	 */
1158	if (S_ISREG(inode->i_mode)) {
1159		size = ui->xattr_size;
1160		size += stat->size;
1161		size = ALIGN(size, UBIFS_BLOCK_SIZE);
1162		/*
1163		 * Note, user-space expects 512-byte blocks count irrespectively
1164		 * of what was reported in @stat->size.
1165		 */
1166		stat->blocks = size >> 9;
1167	} else
1168		stat->blocks = 0;
1169	mutex_unlock(&ui->ui_mutex);
1170	return 0;
1171}
1172
1173const struct inode_operations ubifs_dir_inode_operations = {
1174	.lookup      = ubifs_lookup,
1175	.create      = ubifs_create,
1176	.link        = ubifs_link,
1177	.symlink     = ubifs_symlink,
1178	.unlink      = ubifs_unlink,
1179	.mkdir       = ubifs_mkdir,
1180	.rmdir       = ubifs_rmdir,
1181	.mknod       = ubifs_mknod,
1182	.rename      = ubifs_rename,
1183	.setattr     = ubifs_setattr,
1184	.getattr     = ubifs_getattr,
1185	.setxattr    = ubifs_setxattr,
1186	.getxattr    = ubifs_getxattr,
1187	.listxattr   = ubifs_listxattr,
1188	.removexattr = ubifs_removexattr,
1189#ifdef CONFIG_UBIFS_ATIME_SUPPORT
1190	.update_time = ubifs_update_time,
1191#endif
1192};
1193
1194const struct file_operations ubifs_dir_operations = {
1195	.llseek         = generic_file_llseek,
1196	.release        = ubifs_dir_release,
1197	.read           = generic_read_dir,
1198	.iterate        = ubifs_readdir,
1199	.fsync          = ubifs_fsync,
1200	.unlocked_ioctl = ubifs_ioctl,
1201#ifdef CONFIG_COMPAT
1202	.compat_ioctl   = ubifs_compat_ioctl,
1203#endif
1204};
1205