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 (Битюцкий Артём)
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	if (err != -ENOENT) {
453		ubifs_err(c, "cannot find next direntry, error %d", err);
454		return err;
455	}
456
457	kfree(file->private_data);
458	file->private_data = NULL;
459	/* 2 is a special value indicating that there are no more direntries */
460	ctx->pos = 2;
461	return 0;
462}
463
464/* Free saved readdir() state when the directory is closed */
465static int ubifs_dir_release(struct inode *dir, struct file *file)
466{
467	kfree(file->private_data);
468	file->private_data = NULL;
469	return 0;
470}
471
472/**
473 * lock_2_inodes - a wrapper for locking two UBIFS inodes.
474 * @inode1: first inode
475 * @inode2: second inode
476 *
477 * We do not implement any tricks to guarantee strict lock ordering, because
478 * VFS has already done it for us on the @i_mutex. So this is just a simple
479 * wrapper function.
480 */
481static void lock_2_inodes(struct inode *inode1, struct inode *inode2)
482{
483	mutex_lock_nested(&ubifs_inode(inode1)->ui_mutex, WB_MUTEX_1);
484	mutex_lock_nested(&ubifs_inode(inode2)->ui_mutex, WB_MUTEX_2);
485}
486
487/**
488 * unlock_2_inodes - a wrapper for unlocking two UBIFS inodes.
489 * @inode1: first inode
490 * @inode2: second inode
491 */
492static void unlock_2_inodes(struct inode *inode1, struct inode *inode2)
493{
494	mutex_unlock(&ubifs_inode(inode2)->ui_mutex);
495	mutex_unlock(&ubifs_inode(inode1)->ui_mutex);
496}
497
498static int ubifs_link(struct dentry *old_dentry, struct inode *dir,
499		      struct dentry *dentry)
500{
501	struct ubifs_info *c = dir->i_sb->s_fs_info;
502	struct inode *inode = d_inode(old_dentry);
503	struct ubifs_inode *ui = ubifs_inode(inode);
504	struct ubifs_inode *dir_ui = ubifs_inode(dir);
505	int err, sz_change = CALC_DENT_SIZE(dentry->d_name.len);
506	struct ubifs_budget_req req = { .new_dent = 1, .dirtied_ino = 2,
507				.dirtied_ino_d = ALIGN(ui->data_len, 8) };
508
509	/*
510	 * Budget request settings: new direntry, changing the target inode,
511	 * changing the parent inode.
512	 */
513
514	dbg_gen("dent '%pd' to ino %lu (nlink %d) in dir ino %lu",
515		dentry, inode->i_ino,
516		inode->i_nlink, dir->i_ino);
517	ubifs_assert(mutex_is_locked(&dir->i_mutex));
518	ubifs_assert(mutex_is_locked(&inode->i_mutex));
519
520	err = dbg_check_synced_i_size(c, inode);
521	if (err)
522		return err;
523
524	err = ubifs_budget_space(c, &req);
525	if (err)
526		return err;
527
528	lock_2_inodes(dir, inode);
529	inc_nlink(inode);
530	ihold(inode);
531	inode->i_ctime = ubifs_current_time(inode);
532	dir->i_size += sz_change;
533	dir_ui->ui_size = dir->i_size;
534	dir->i_mtime = dir->i_ctime = inode->i_ctime;
535	err = ubifs_jnl_update(c, dir, &dentry->d_name, inode, 0, 0);
536	if (err)
537		goto out_cancel;
538	unlock_2_inodes(dir, inode);
539
540	ubifs_release_budget(c, &req);
541	d_instantiate(dentry, inode);
542	return 0;
543
544out_cancel:
545	dir->i_size -= sz_change;
546	dir_ui->ui_size = dir->i_size;
547	drop_nlink(inode);
548	unlock_2_inodes(dir, inode);
549	ubifs_release_budget(c, &req);
550	iput(inode);
551	return err;
552}
553
554static int ubifs_unlink(struct inode *dir, struct dentry *dentry)
555{
556	struct ubifs_info *c = dir->i_sb->s_fs_info;
557	struct inode *inode = d_inode(dentry);
558	struct ubifs_inode *dir_ui = ubifs_inode(dir);
559	int sz_change = CALC_DENT_SIZE(dentry->d_name.len);
560	int err, budgeted = 1;
561	struct ubifs_budget_req req = { .mod_dent = 1, .dirtied_ino = 2 };
562	unsigned int saved_nlink = inode->i_nlink;
563
564	/*
565	 * Budget request settings: deletion direntry, deletion inode (+1 for
566	 * @dirtied_ino), changing the parent directory inode. If budgeting
567	 * fails, go ahead anyway because we have extra space reserved for
568	 * deletions.
569	 */
570
571	dbg_gen("dent '%pd' from ino %lu (nlink %d) in dir ino %lu",
572		dentry, inode->i_ino,
573		inode->i_nlink, dir->i_ino);
574	ubifs_assert(mutex_is_locked(&dir->i_mutex));
575	ubifs_assert(mutex_is_locked(&inode->i_mutex));
576	err = dbg_check_synced_i_size(c, inode);
577	if (err)
578		return err;
579
580	err = ubifs_budget_space(c, &req);
581	if (err) {
582		if (err != -ENOSPC)
583			return err;
584		budgeted = 0;
585	}
586
587	lock_2_inodes(dir, inode);
588	inode->i_ctime = ubifs_current_time(dir);
589	drop_nlink(inode);
590	dir->i_size -= sz_change;
591	dir_ui->ui_size = dir->i_size;
592	dir->i_mtime = dir->i_ctime = inode->i_ctime;
593	err = ubifs_jnl_update(c, dir, &dentry->d_name, inode, 1, 0);
594	if (err)
595		goto out_cancel;
596	unlock_2_inodes(dir, inode);
597
598	if (budgeted)
599		ubifs_release_budget(c, &req);
600	else {
601		/* We've deleted something - clean the "no space" flags */
602		c->bi.nospace = c->bi.nospace_rp = 0;
603		smp_wmb();
604	}
605	return 0;
606
607out_cancel:
608	dir->i_size += sz_change;
609	dir_ui->ui_size = dir->i_size;
610	set_nlink(inode, saved_nlink);
611	unlock_2_inodes(dir, inode);
612	if (budgeted)
613		ubifs_release_budget(c, &req);
614	return err;
615}
616
617/**
618 * check_dir_empty - check if a directory is empty or not.
619 * @c: UBIFS file-system description object
620 * @dir: VFS inode object of the directory to check
621 *
622 * This function checks if directory @dir is empty. Returns zero if the
623 * directory is empty, %-ENOTEMPTY if it is not, and other negative error codes
624 * in case of of errors.
625 */
626static int check_dir_empty(struct ubifs_info *c, struct inode *dir)
627{
628	struct qstr nm = { .name = NULL };
629	struct ubifs_dent_node *dent;
630	union ubifs_key key;
631	int err;
632
633	lowest_dent_key(c, &key, dir->i_ino);
634	dent = ubifs_tnc_next_ent(c, &key, &nm);
635	if (IS_ERR(dent)) {
636		err = PTR_ERR(dent);
637		if (err == -ENOENT)
638			err = 0;
639	} else {
640		kfree(dent);
641		err = -ENOTEMPTY;
642	}
643	return err;
644}
645
646static int ubifs_rmdir(struct inode *dir, struct dentry *dentry)
647{
648	struct ubifs_info *c = dir->i_sb->s_fs_info;
649	struct inode *inode = d_inode(dentry);
650	int sz_change = CALC_DENT_SIZE(dentry->d_name.len);
651	int err, budgeted = 1;
652	struct ubifs_inode *dir_ui = ubifs_inode(dir);
653	struct ubifs_budget_req req = { .mod_dent = 1, .dirtied_ino = 2 };
654
655	/*
656	 * Budget request settings: deletion direntry, deletion inode and
657	 * changing the parent inode. If budgeting fails, go ahead anyway
658	 * because we have extra space reserved for deletions.
659	 */
660
661	dbg_gen("directory '%pd', ino %lu in dir ino %lu", dentry,
662		inode->i_ino, dir->i_ino);
663	ubifs_assert(mutex_is_locked(&dir->i_mutex));
664	ubifs_assert(mutex_is_locked(&inode->i_mutex));
665	err = check_dir_empty(c, d_inode(dentry));
666	if (err)
667		return err;
668
669	err = ubifs_budget_space(c, &req);
670	if (err) {
671		if (err != -ENOSPC)
672			return err;
673		budgeted = 0;
674	}
675
676	lock_2_inodes(dir, inode);
677	inode->i_ctime = ubifs_current_time(dir);
678	clear_nlink(inode);
679	drop_nlink(dir);
680	dir->i_size -= sz_change;
681	dir_ui->ui_size = dir->i_size;
682	dir->i_mtime = dir->i_ctime = inode->i_ctime;
683	err = ubifs_jnl_update(c, dir, &dentry->d_name, inode, 1, 0);
684	if (err)
685		goto out_cancel;
686	unlock_2_inodes(dir, inode);
687
688	if (budgeted)
689		ubifs_release_budget(c, &req);
690	else {
691		/* We've deleted something - clean the "no space" flags */
692		c->bi.nospace = c->bi.nospace_rp = 0;
693		smp_wmb();
694	}
695	return 0;
696
697out_cancel:
698	dir->i_size += sz_change;
699	dir_ui->ui_size = dir->i_size;
700	inc_nlink(dir);
701	set_nlink(inode, 2);
702	unlock_2_inodes(dir, inode);
703	if (budgeted)
704		ubifs_release_budget(c, &req);
705	return err;
706}
707
708static int ubifs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
709{
710	struct inode *inode;
711	struct ubifs_inode *dir_ui = ubifs_inode(dir);
712	struct ubifs_info *c = dir->i_sb->s_fs_info;
713	int err, sz_change = CALC_DENT_SIZE(dentry->d_name.len);
714	struct ubifs_budget_req req = { .new_ino = 1, .new_dent = 1 };
715
716	/*
717	 * Budget request settings: new inode, new direntry and changing parent
718	 * directory inode.
719	 */
720
721	dbg_gen("dent '%pd', mode %#hx in dir ino %lu",
722		dentry, mode, dir->i_ino);
723
724	err = ubifs_budget_space(c, &req);
725	if (err)
726		return err;
727
728	inode = ubifs_new_inode(c, dir, S_IFDIR | mode);
729	if (IS_ERR(inode)) {
730		err = PTR_ERR(inode);
731		goto out_budg;
732	}
733
734	err = ubifs_init_security(dir, inode, &dentry->d_name);
735	if (err)
736		goto out_inode;
737
738	mutex_lock(&dir_ui->ui_mutex);
739	insert_inode_hash(inode);
740	inc_nlink(inode);
741	inc_nlink(dir);
742	dir->i_size += sz_change;
743	dir_ui->ui_size = dir->i_size;
744	dir->i_mtime = dir->i_ctime = inode->i_ctime;
745	err = ubifs_jnl_update(c, dir, &dentry->d_name, inode, 0, 0);
746	if (err) {
747		ubifs_err(c, "cannot create directory, error %d", err);
748		goto out_cancel;
749	}
750	mutex_unlock(&dir_ui->ui_mutex);
751
752	ubifs_release_budget(c, &req);
753	d_instantiate(dentry, inode);
754	return 0;
755
756out_cancel:
757	dir->i_size -= sz_change;
758	dir_ui->ui_size = dir->i_size;
759	drop_nlink(dir);
760	mutex_unlock(&dir_ui->ui_mutex);
761out_inode:
762	make_bad_inode(inode);
763	iput(inode);
764out_budg:
765	ubifs_release_budget(c, &req);
766	return err;
767}
768
769static int ubifs_mknod(struct inode *dir, struct dentry *dentry,
770		       umode_t mode, dev_t rdev)
771{
772	struct inode *inode;
773	struct ubifs_inode *ui;
774	struct ubifs_inode *dir_ui = ubifs_inode(dir);
775	struct ubifs_info *c = dir->i_sb->s_fs_info;
776	union ubifs_dev_desc *dev = NULL;
777	int sz_change = CALC_DENT_SIZE(dentry->d_name.len);
778	int err, devlen = 0;
779	struct ubifs_budget_req req = { .new_ino = 1, .new_dent = 1,
780					.new_ino_d = ALIGN(devlen, 8),
781					.dirtied_ino = 1 };
782
783	/*
784	 * Budget request settings: new inode, new direntry and changing parent
785	 * directory inode.
786	 */
787
788	dbg_gen("dent '%pd' in dir ino %lu", dentry, dir->i_ino);
789
790	if (!new_valid_dev(rdev))
791		return -EINVAL;
792
793	if (S_ISBLK(mode) || S_ISCHR(mode)) {
794		dev = kmalloc(sizeof(union ubifs_dev_desc), GFP_NOFS);
795		if (!dev)
796			return -ENOMEM;
797		devlen = ubifs_encode_dev(dev, rdev);
798	}
799
800	err = ubifs_budget_space(c, &req);
801	if (err) {
802		kfree(dev);
803		return err;
804	}
805
806	inode = ubifs_new_inode(c, dir, mode);
807	if (IS_ERR(inode)) {
808		kfree(dev);
809		err = PTR_ERR(inode);
810		goto out_budg;
811	}
812
813	init_special_inode(inode, inode->i_mode, rdev);
814	inode->i_size = ubifs_inode(inode)->ui_size = devlen;
815	ui = ubifs_inode(inode);
816	ui->data = dev;
817	ui->data_len = devlen;
818
819	err = ubifs_init_security(dir, inode, &dentry->d_name);
820	if (err)
821		goto out_inode;
822
823	mutex_lock(&dir_ui->ui_mutex);
824	dir->i_size += sz_change;
825	dir_ui->ui_size = dir->i_size;
826	dir->i_mtime = dir->i_ctime = inode->i_ctime;
827	err = ubifs_jnl_update(c, dir, &dentry->d_name, inode, 0, 0);
828	if (err)
829		goto out_cancel;
830	mutex_unlock(&dir_ui->ui_mutex);
831
832	ubifs_release_budget(c, &req);
833	insert_inode_hash(inode);
834	d_instantiate(dentry, inode);
835	return 0;
836
837out_cancel:
838	dir->i_size -= sz_change;
839	dir_ui->ui_size = dir->i_size;
840	mutex_unlock(&dir_ui->ui_mutex);
841out_inode:
842	make_bad_inode(inode);
843	iput(inode);
844out_budg:
845	ubifs_release_budget(c, &req);
846	return err;
847}
848
849static int ubifs_symlink(struct inode *dir, struct dentry *dentry,
850			 const char *symname)
851{
852	struct inode *inode;
853	struct ubifs_inode *ui;
854	struct ubifs_inode *dir_ui = ubifs_inode(dir);
855	struct ubifs_info *c = dir->i_sb->s_fs_info;
856	int err, len = strlen(symname);
857	int sz_change = CALC_DENT_SIZE(dentry->d_name.len);
858	struct ubifs_budget_req req = { .new_ino = 1, .new_dent = 1,
859					.new_ino_d = ALIGN(len, 8),
860					.dirtied_ino = 1 };
861
862	/*
863	 * Budget request settings: new inode, new direntry and changing parent
864	 * directory inode.
865	 */
866
867	dbg_gen("dent '%pd', target '%s' in dir ino %lu", dentry,
868		symname, dir->i_ino);
869
870	if (len > UBIFS_MAX_INO_DATA)
871		return -ENAMETOOLONG;
872
873	err = ubifs_budget_space(c, &req);
874	if (err)
875		return err;
876
877	inode = ubifs_new_inode(c, dir, S_IFLNK | S_IRWXUGO);
878	if (IS_ERR(inode)) {
879		err = PTR_ERR(inode);
880		goto out_budg;
881	}
882
883	ui = ubifs_inode(inode);
884	ui->data = kmalloc(len + 1, GFP_NOFS);
885	if (!ui->data) {
886		err = -ENOMEM;
887		goto out_inode;
888	}
889
890	memcpy(ui->data, symname, len);
891	((char *)ui->data)[len] = '\0';
892	/*
893	 * The terminating zero byte is not written to the flash media and it
894	 * is put just to make later in-memory string processing simpler. Thus,
895	 * data length is @len, not @len + %1.
896	 */
897	ui->data_len = len;
898	inode->i_size = ubifs_inode(inode)->ui_size = len;
899
900	err = ubifs_init_security(dir, inode, &dentry->d_name);
901	if (err)
902		goto out_inode;
903
904	mutex_lock(&dir_ui->ui_mutex);
905	dir->i_size += sz_change;
906	dir_ui->ui_size = dir->i_size;
907	dir->i_mtime = dir->i_ctime = inode->i_ctime;
908	err = ubifs_jnl_update(c, dir, &dentry->d_name, inode, 0, 0);
909	if (err)
910		goto out_cancel;
911	mutex_unlock(&dir_ui->ui_mutex);
912
913	ubifs_release_budget(c, &req);
914	insert_inode_hash(inode);
915	d_instantiate(dentry, inode);
916	return 0;
917
918out_cancel:
919	dir->i_size -= sz_change;
920	dir_ui->ui_size = dir->i_size;
921	mutex_unlock(&dir_ui->ui_mutex);
922out_inode:
923	make_bad_inode(inode);
924	iput(inode);
925out_budg:
926	ubifs_release_budget(c, &req);
927	return err;
928}
929
930/**
931 * lock_3_inodes - a wrapper for locking three UBIFS inodes.
932 * @inode1: first inode
933 * @inode2: second inode
934 * @inode3: third inode
935 *
936 * This function is used for 'ubifs_rename()' and @inode1 may be the same as
937 * @inode2 whereas @inode3 may be %NULL.
938 *
939 * We do not implement any tricks to guarantee strict lock ordering, because
940 * VFS has already done it for us on the @i_mutex. So this is just a simple
941 * wrapper function.
942 */
943static void lock_3_inodes(struct inode *inode1, struct inode *inode2,
944			  struct inode *inode3)
945{
946	mutex_lock_nested(&ubifs_inode(inode1)->ui_mutex, WB_MUTEX_1);
947	if (inode2 != inode1)
948		mutex_lock_nested(&ubifs_inode(inode2)->ui_mutex, WB_MUTEX_2);
949	if (inode3)
950		mutex_lock_nested(&ubifs_inode(inode3)->ui_mutex, WB_MUTEX_3);
951}
952
953/**
954 * unlock_3_inodes - a wrapper for unlocking three UBIFS inodes for rename.
955 * @inode1: first inode
956 * @inode2: second inode
957 * @inode3: third inode
958 */
959static void unlock_3_inodes(struct inode *inode1, struct inode *inode2,
960			    struct inode *inode3)
961{
962	if (inode3)
963		mutex_unlock(&ubifs_inode(inode3)->ui_mutex);
964	if (inode1 != inode2)
965		mutex_unlock(&ubifs_inode(inode2)->ui_mutex);
966	mutex_unlock(&ubifs_inode(inode1)->ui_mutex);
967}
968
969static int ubifs_rename(struct inode *old_dir, struct dentry *old_dentry,
970			struct inode *new_dir, struct dentry *new_dentry)
971{
972	struct ubifs_info *c = old_dir->i_sb->s_fs_info;
973	struct inode *old_inode = d_inode(old_dentry);
974	struct inode *new_inode = d_inode(new_dentry);
975	struct ubifs_inode *old_inode_ui = ubifs_inode(old_inode);
976	int err, release, sync = 0, move = (new_dir != old_dir);
977	int is_dir = S_ISDIR(old_inode->i_mode);
978	int unlink = !!new_inode;
979	int new_sz = CALC_DENT_SIZE(new_dentry->d_name.len);
980	int old_sz = CALC_DENT_SIZE(old_dentry->d_name.len);
981	struct ubifs_budget_req req = { .new_dent = 1, .mod_dent = 1,
982					.dirtied_ino = 3 };
983	struct ubifs_budget_req ino_req = { .dirtied_ino = 1,
984			.dirtied_ino_d = ALIGN(old_inode_ui->data_len, 8) };
985	struct timespec time;
986	unsigned int uninitialized_var(saved_nlink);
987
988	/*
989	 * Budget request settings: deletion direntry, new direntry, removing
990	 * the old inode, and changing old and new parent directory inodes.
991	 *
992	 * However, this operation also marks the target inode as dirty and
993	 * does not write it, so we allocate budget for the target inode
994	 * separately.
995	 */
996
997	dbg_gen("dent '%pd' ino %lu in dir ino %lu to dent '%pd' in dir ino %lu",
998		old_dentry, old_inode->i_ino, old_dir->i_ino,
999		new_dentry, new_dir->i_ino);
1000	ubifs_assert(mutex_is_locked(&old_dir->i_mutex));
1001	ubifs_assert(mutex_is_locked(&new_dir->i_mutex));
1002	if (unlink)
1003		ubifs_assert(mutex_is_locked(&new_inode->i_mutex));
1004
1005
1006	if (unlink && is_dir) {
1007		err = check_dir_empty(c, new_inode);
1008		if (err)
1009			return err;
1010	}
1011
1012	err = ubifs_budget_space(c, &req);
1013	if (err)
1014		return err;
1015	err = ubifs_budget_space(c, &ino_req);
1016	if (err) {
1017		ubifs_release_budget(c, &req);
1018		return err;
1019	}
1020
1021	lock_3_inodes(old_dir, new_dir, new_inode);
1022
1023	/*
1024	 * Like most other Unix systems, set the @i_ctime for inodes on a
1025	 * rename.
1026	 */
1027	time = ubifs_current_time(old_dir);
1028	old_inode->i_ctime = time;
1029
1030	/* We must adjust parent link count when renaming directories */
1031	if (is_dir) {
1032		if (move) {
1033			/*
1034			 * @old_dir loses a link because we are moving
1035			 * @old_inode to a different directory.
1036			 */
1037			drop_nlink(old_dir);
1038			/*
1039			 * @new_dir only gains a link if we are not also
1040			 * overwriting an existing directory.
1041			 */
1042			if (!unlink)
1043				inc_nlink(new_dir);
1044		} else {
1045			/*
1046			 * @old_inode is not moving to a different directory,
1047			 * but @old_dir still loses a link if we are
1048			 * overwriting an existing directory.
1049			 */
1050			if (unlink)
1051				drop_nlink(old_dir);
1052		}
1053	}
1054
1055	old_dir->i_size -= old_sz;
1056	ubifs_inode(old_dir)->ui_size = old_dir->i_size;
1057	old_dir->i_mtime = old_dir->i_ctime = time;
1058	new_dir->i_mtime = new_dir->i_ctime = time;
1059
1060	/*
1061	 * And finally, if we unlinked a direntry which happened to have the
1062	 * same name as the moved direntry, we have to decrement @i_nlink of
1063	 * the unlinked inode and change its ctime.
1064	 */
1065	if (unlink) {
1066		/*
1067		 * Directories cannot have hard-links, so if this is a
1068		 * directory, just clear @i_nlink.
1069		 */
1070		saved_nlink = new_inode->i_nlink;
1071		if (is_dir)
1072			clear_nlink(new_inode);
1073		else
1074			drop_nlink(new_inode);
1075		new_inode->i_ctime = time;
1076	} else {
1077		new_dir->i_size += new_sz;
1078		ubifs_inode(new_dir)->ui_size = new_dir->i_size;
1079	}
1080
1081	/*
1082	 * Do not ask 'ubifs_jnl_rename()' to flush write-buffer if @old_inode
1083	 * is dirty, because this will be done later on at the end of
1084	 * 'ubifs_rename()'.
1085	 */
1086	if (IS_SYNC(old_inode)) {
1087		sync = IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir);
1088		if (unlink && IS_SYNC(new_inode))
1089			sync = 1;
1090	}
1091	err = ubifs_jnl_rename(c, old_dir, old_dentry, new_dir, new_dentry,
1092			       sync);
1093	if (err)
1094		goto out_cancel;
1095
1096	unlock_3_inodes(old_dir, new_dir, new_inode);
1097	ubifs_release_budget(c, &req);
1098
1099	mutex_lock(&old_inode_ui->ui_mutex);
1100	release = old_inode_ui->dirty;
1101	mark_inode_dirty_sync(old_inode);
1102	mutex_unlock(&old_inode_ui->ui_mutex);
1103
1104	if (release)
1105		ubifs_release_budget(c, &ino_req);
1106	if (IS_SYNC(old_inode))
1107		err = old_inode->i_sb->s_op->write_inode(old_inode, NULL);
1108	return err;
1109
1110out_cancel:
1111	if (unlink) {
1112		set_nlink(new_inode, saved_nlink);
1113	} else {
1114		new_dir->i_size -= new_sz;
1115		ubifs_inode(new_dir)->ui_size = new_dir->i_size;
1116	}
1117	old_dir->i_size += old_sz;
1118	ubifs_inode(old_dir)->ui_size = old_dir->i_size;
1119	if (is_dir) {
1120		if (move) {
1121			inc_nlink(old_dir);
1122			if (!unlink)
1123				drop_nlink(new_dir);
1124		} else {
1125			if (unlink)
1126				inc_nlink(old_dir);
1127		}
1128	}
1129	unlock_3_inodes(old_dir, new_dir, new_inode);
1130	ubifs_release_budget(c, &ino_req);
1131	ubifs_release_budget(c, &req);
1132	return err;
1133}
1134
1135int ubifs_getattr(struct vfsmount *mnt, struct dentry *dentry,
1136		  struct kstat *stat)
1137{
1138	loff_t size;
1139	struct inode *inode = d_inode(dentry);
1140	struct ubifs_inode *ui = ubifs_inode(inode);
1141
1142	mutex_lock(&ui->ui_mutex);
1143	generic_fillattr(inode, stat);
1144	stat->blksize = UBIFS_BLOCK_SIZE;
1145	stat->size = ui->ui_size;
1146
1147	/*
1148	 * Unfortunately, the 'stat()' system call was designed for block
1149	 * device based file systems, and it is not appropriate for UBIFS,
1150	 * because UBIFS does not have notion of "block". For example, it is
1151	 * difficult to tell how many block a directory takes - it actually
1152	 * takes less than 300 bytes, but we have to round it to block size,
1153	 * which introduces large mistake. This makes utilities like 'du' to
1154	 * report completely senseless numbers. This is the reason why UBIFS
1155	 * goes the same way as JFFS2 - it reports zero blocks for everything
1156	 * but regular files, which makes more sense than reporting completely
1157	 * wrong sizes.
1158	 */
1159	if (S_ISREG(inode->i_mode)) {
1160		size = ui->xattr_size;
1161		size += stat->size;
1162		size = ALIGN(size, UBIFS_BLOCK_SIZE);
1163		/*
1164		 * Note, user-space expects 512-byte blocks count irrespectively
1165		 * of what was reported in @stat->size.
1166		 */
1167		stat->blocks = size >> 9;
1168	} else
1169		stat->blocks = 0;
1170	mutex_unlock(&ui->ui_mutex);
1171	return 0;
1172}
1173
1174const struct inode_operations ubifs_dir_inode_operations = {
1175	.lookup      = ubifs_lookup,
1176	.create      = ubifs_create,
1177	.link        = ubifs_link,
1178	.symlink     = ubifs_symlink,
1179	.unlink      = ubifs_unlink,
1180	.mkdir       = ubifs_mkdir,
1181	.rmdir       = ubifs_rmdir,
1182	.mknod       = ubifs_mknod,
1183	.rename      = ubifs_rename,
1184	.setattr     = ubifs_setattr,
1185	.getattr     = ubifs_getattr,
1186	.setxattr    = ubifs_setxattr,
1187	.getxattr    = ubifs_getxattr,
1188	.listxattr   = ubifs_listxattr,
1189	.removexattr = ubifs_removexattr,
1190};
1191
1192const struct file_operations ubifs_dir_operations = {
1193	.llseek         = generic_file_llseek,
1194	.release        = ubifs_dir_release,
1195	.read           = generic_read_dir,
1196	.iterate        = ubifs_readdir,
1197	.fsync          = ubifs_fsync,
1198	.unlocked_ioctl = ubifs_ioctl,
1199#ifdef CONFIG_COMPAT
1200	.compat_ioctl   = ubifs_compat_ioctl,
1201#endif
1202};
1203