xref: /openbmc/linux/fs/ubifs/journal.c (revision 98ddec80)
1 /*
2  * This file is part of UBIFS.
3  *
4  * Copyright (C) 2006-2008 Nokia Corporation.
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  */
22 
23 /*
24  * This file implements UBIFS journal.
25  *
26  * The journal consists of 2 parts - the log and bud LEBs. The log has fixed
27  * length and position, while a bud logical eraseblock is any LEB in the main
28  * area. Buds contain file system data - data nodes, inode nodes, etc. The log
29  * contains only references to buds and some other stuff like commit
30  * start node. The idea is that when we commit the journal, we do
31  * not copy the data, the buds just become indexed. Since after the commit the
32  * nodes in bud eraseblocks become leaf nodes of the file system index tree, we
33  * use term "bud". Analogy is obvious, bud eraseblocks contain nodes which will
34  * become leafs in the future.
35  *
36  * The journal is multi-headed because we want to write data to the journal as
37  * optimally as possible. It is nice to have nodes belonging to the same inode
38  * in one LEB, so we may write data owned by different inodes to different
39  * journal heads, although at present only one data head is used.
40  *
41  * For recovery reasons, the base head contains all inode nodes, all directory
42  * entry nodes and all truncate nodes. This means that the other heads contain
43  * only data nodes.
44  *
45  * Bud LEBs may be half-indexed. For example, if the bud was not full at the
46  * time of commit, the bud is retained to continue to be used in the journal,
47  * even though the "front" of the LEB is now indexed. In that case, the log
48  * reference contains the offset where the bud starts for the purposes of the
49  * journal.
50  *
51  * The journal size has to be limited, because the larger is the journal, the
52  * longer it takes to mount UBIFS (scanning the journal) and the more memory it
53  * takes (indexing in the TNC).
54  *
55  * All the journal write operations like 'ubifs_jnl_update()' here, which write
56  * multiple UBIFS nodes to the journal at one go, are atomic with respect to
57  * unclean reboots. Should the unclean reboot happen, the recovery code drops
58  * all the nodes.
59  */
60 
61 #include "ubifs.h"
62 
63 /**
64  * zero_ino_node_unused - zero out unused fields of an on-flash inode node.
65  * @ino: the inode to zero out
66  */
67 static inline void zero_ino_node_unused(struct ubifs_ino_node *ino)
68 {
69 	memset(ino->padding1, 0, 4);
70 	memset(ino->padding2, 0, 26);
71 }
72 
73 /**
74  * zero_dent_node_unused - zero out unused fields of an on-flash directory
75  *                         entry node.
76  * @dent: the directory entry to zero out
77  */
78 static inline void zero_dent_node_unused(struct ubifs_dent_node *dent)
79 {
80 	dent->padding1 = 0;
81 }
82 
83 /**
84  * zero_trun_node_unused - zero out unused fields of an on-flash truncation
85  *                         node.
86  * @trun: the truncation node to zero out
87  */
88 static inline void zero_trun_node_unused(struct ubifs_trun_node *trun)
89 {
90 	memset(trun->padding, 0, 12);
91 }
92 
93 /**
94  * reserve_space - reserve space in the journal.
95  * @c: UBIFS file-system description object
96  * @jhead: journal head number
97  * @len: node length
98  *
99  * This function reserves space in journal head @head. If the reservation
100  * succeeded, the journal head stays locked and later has to be unlocked using
101  * 'release_head()'. Returns zero in case of success, %-EAGAIN if commit has to
102  * be done, and other negative error codes in case of other failures.
103  */
104 static int reserve_space(struct ubifs_info *c, int jhead, int len)
105 {
106 	int err = 0, err1, retries = 0, avail, lnum, offs, squeeze;
107 	struct ubifs_wbuf *wbuf = &c->jheads[jhead].wbuf;
108 
109 	/*
110 	 * Typically, the base head has smaller nodes written to it, so it is
111 	 * better to try to allocate space at the ends of eraseblocks. This is
112 	 * what the squeeze parameter does.
113 	 */
114 	ubifs_assert(!c->ro_media && !c->ro_mount);
115 	squeeze = (jhead == BASEHD);
116 again:
117 	mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
118 
119 	if (c->ro_error) {
120 		err = -EROFS;
121 		goto out_unlock;
122 	}
123 
124 	avail = c->leb_size - wbuf->offs - wbuf->used;
125 	if (wbuf->lnum != -1 && avail >= len)
126 		return 0;
127 
128 	/*
129 	 * Write buffer wasn't seek'ed or there is no enough space - look for an
130 	 * LEB with some empty space.
131 	 */
132 	lnum = ubifs_find_free_space(c, len, &offs, squeeze);
133 	if (lnum >= 0)
134 		goto out;
135 
136 	err = lnum;
137 	if (err != -ENOSPC)
138 		goto out_unlock;
139 
140 	/*
141 	 * No free space, we have to run garbage collector to make
142 	 * some. But the write-buffer mutex has to be unlocked because
143 	 * GC also takes it.
144 	 */
145 	dbg_jnl("no free space in jhead %s, run GC", dbg_jhead(jhead));
146 	mutex_unlock(&wbuf->io_mutex);
147 
148 	lnum = ubifs_garbage_collect(c, 0);
149 	if (lnum < 0) {
150 		err = lnum;
151 		if (err != -ENOSPC)
152 			return err;
153 
154 		/*
155 		 * GC could not make a free LEB. But someone else may
156 		 * have allocated new bud for this journal head,
157 		 * because we dropped @wbuf->io_mutex, so try once
158 		 * again.
159 		 */
160 		dbg_jnl("GC couldn't make a free LEB for jhead %s",
161 			dbg_jhead(jhead));
162 		if (retries++ < 2) {
163 			dbg_jnl("retry (%d)", retries);
164 			goto again;
165 		}
166 
167 		dbg_jnl("return -ENOSPC");
168 		return err;
169 	}
170 
171 	mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
172 	dbg_jnl("got LEB %d for jhead %s", lnum, dbg_jhead(jhead));
173 	avail = c->leb_size - wbuf->offs - wbuf->used;
174 
175 	if (wbuf->lnum != -1 && avail >= len) {
176 		/*
177 		 * Someone else has switched the journal head and we have
178 		 * enough space now. This happens when more than one process is
179 		 * trying to write to the same journal head at the same time.
180 		 */
181 		dbg_jnl("return LEB %d back, already have LEB %d:%d",
182 			lnum, wbuf->lnum, wbuf->offs + wbuf->used);
183 		err = ubifs_return_leb(c, lnum);
184 		if (err)
185 			goto out_unlock;
186 		return 0;
187 	}
188 
189 	offs = 0;
190 
191 out:
192 	/*
193 	 * Make sure we synchronize the write-buffer before we add the new bud
194 	 * to the log. Otherwise we may have a power cut after the log
195 	 * reference node for the last bud (@lnum) is written but before the
196 	 * write-buffer data are written to the next-to-last bud
197 	 * (@wbuf->lnum). And the effect would be that the recovery would see
198 	 * that there is corruption in the next-to-last bud.
199 	 */
200 	err = ubifs_wbuf_sync_nolock(wbuf);
201 	if (err)
202 		goto out_return;
203 	err = ubifs_add_bud_to_log(c, jhead, lnum, offs);
204 	if (err)
205 		goto out_return;
206 	err = ubifs_wbuf_seek_nolock(wbuf, lnum, offs);
207 	if (err)
208 		goto out_unlock;
209 
210 	return 0;
211 
212 out_unlock:
213 	mutex_unlock(&wbuf->io_mutex);
214 	return err;
215 
216 out_return:
217 	/* An error occurred and the LEB has to be returned to lprops */
218 	ubifs_assert(err < 0);
219 	err1 = ubifs_return_leb(c, lnum);
220 	if (err1 && err == -EAGAIN)
221 		/*
222 		 * Return original error code only if it is not %-EAGAIN,
223 		 * which is not really an error. Otherwise, return the error
224 		 * code of 'ubifs_return_leb()'.
225 		 */
226 		err = err1;
227 	mutex_unlock(&wbuf->io_mutex);
228 	return err;
229 }
230 
231 /**
232  * write_node - write node to a journal head.
233  * @c: UBIFS file-system description object
234  * @jhead: journal head
235  * @node: node to write
236  * @len: node length
237  * @lnum: LEB number written is returned here
238  * @offs: offset written is returned here
239  *
240  * This function writes a node to reserved space of journal head @jhead.
241  * Returns zero in case of success and a negative error code in case of
242  * failure.
243  */
244 static int write_node(struct ubifs_info *c, int jhead, void *node, int len,
245 		      int *lnum, int *offs)
246 {
247 	struct ubifs_wbuf *wbuf = &c->jheads[jhead].wbuf;
248 
249 	ubifs_assert(jhead != GCHD);
250 
251 	*lnum = c->jheads[jhead].wbuf.lnum;
252 	*offs = c->jheads[jhead].wbuf.offs + c->jheads[jhead].wbuf.used;
253 
254 	dbg_jnl("jhead %s, LEB %d:%d, len %d",
255 		dbg_jhead(jhead), *lnum, *offs, len);
256 	ubifs_prepare_node(c, node, len, 0);
257 
258 	return ubifs_wbuf_write_nolock(wbuf, node, len);
259 }
260 
261 /**
262  * write_head - write data to a journal head.
263  * @c: UBIFS file-system description object
264  * @jhead: journal head
265  * @buf: buffer to write
266  * @len: length to write
267  * @lnum: LEB number written is returned here
268  * @offs: offset written is returned here
269  * @sync: non-zero if the write-buffer has to by synchronized
270  *
271  * This function is the same as 'write_node()' but it does not assume the
272  * buffer it is writing is a node, so it does not prepare it (which means
273  * initializing common header and calculating CRC).
274  */
275 static int write_head(struct ubifs_info *c, int jhead, void *buf, int len,
276 		      int *lnum, int *offs, int sync)
277 {
278 	int err;
279 	struct ubifs_wbuf *wbuf = &c->jheads[jhead].wbuf;
280 
281 	ubifs_assert(jhead != GCHD);
282 
283 	*lnum = c->jheads[jhead].wbuf.lnum;
284 	*offs = c->jheads[jhead].wbuf.offs + c->jheads[jhead].wbuf.used;
285 	dbg_jnl("jhead %s, LEB %d:%d, len %d",
286 		dbg_jhead(jhead), *lnum, *offs, len);
287 
288 	err = ubifs_wbuf_write_nolock(wbuf, buf, len);
289 	if (err)
290 		return err;
291 	if (sync)
292 		err = ubifs_wbuf_sync_nolock(wbuf);
293 	return err;
294 }
295 
296 /**
297  * make_reservation - reserve journal space.
298  * @c: UBIFS file-system description object
299  * @jhead: journal head
300  * @len: how many bytes to reserve
301  *
302  * This function makes space reservation in journal head @jhead. The function
303  * takes the commit lock and locks the journal head, and the caller has to
304  * unlock the head and finish the reservation with 'finish_reservation()'.
305  * Returns zero in case of success and a negative error code in case of
306  * failure.
307  *
308  * Note, the journal head may be unlocked as soon as the data is written, while
309  * the commit lock has to be released after the data has been added to the
310  * TNC.
311  */
312 static int make_reservation(struct ubifs_info *c, int jhead, int len)
313 {
314 	int err, cmt_retries = 0, nospc_retries = 0;
315 
316 again:
317 	down_read(&c->commit_sem);
318 	err = reserve_space(c, jhead, len);
319 	if (!err)
320 		return 0;
321 	up_read(&c->commit_sem);
322 
323 	if (err == -ENOSPC) {
324 		/*
325 		 * GC could not make any progress. We should try to commit
326 		 * once because it could make some dirty space and GC would
327 		 * make progress, so make the error -EAGAIN so that the below
328 		 * will commit and re-try.
329 		 */
330 		if (nospc_retries++ < 2) {
331 			dbg_jnl("no space, retry");
332 			err = -EAGAIN;
333 		}
334 
335 		/*
336 		 * This means that the budgeting is incorrect. We always have
337 		 * to be able to write to the media, because all operations are
338 		 * budgeted. Deletions are not budgeted, though, but we reserve
339 		 * an extra LEB for them.
340 		 */
341 	}
342 
343 	if (err != -EAGAIN)
344 		goto out;
345 
346 	/*
347 	 * -EAGAIN means that the journal is full or too large, or the above
348 	 * code wants to do one commit. Do this and re-try.
349 	 */
350 	if (cmt_retries > 128) {
351 		/*
352 		 * This should not happen unless the journal size limitations
353 		 * are too tough.
354 		 */
355 		ubifs_err(c, "stuck in space allocation");
356 		err = -ENOSPC;
357 		goto out;
358 	} else if (cmt_retries > 32)
359 		ubifs_warn(c, "too many space allocation re-tries (%d)",
360 			   cmt_retries);
361 
362 	dbg_jnl("-EAGAIN, commit and retry (retried %d times)",
363 		cmt_retries);
364 	cmt_retries += 1;
365 
366 	err = ubifs_run_commit(c);
367 	if (err)
368 		return err;
369 	goto again;
370 
371 out:
372 	ubifs_err(c, "cannot reserve %d bytes in jhead %d, error %d",
373 		  len, jhead, err);
374 	if (err == -ENOSPC) {
375 		/* This are some budgeting problems, print useful information */
376 		down_write(&c->commit_sem);
377 		dump_stack();
378 		ubifs_dump_budg(c, &c->bi);
379 		ubifs_dump_lprops(c);
380 		cmt_retries = dbg_check_lprops(c);
381 		up_write(&c->commit_sem);
382 	}
383 	return err;
384 }
385 
386 /**
387  * release_head - release a journal head.
388  * @c: UBIFS file-system description object
389  * @jhead: journal head
390  *
391  * This function releases journal head @jhead which was locked by
392  * the 'make_reservation()' function. It has to be called after each successful
393  * 'make_reservation()' invocation.
394  */
395 static inline void release_head(struct ubifs_info *c, int jhead)
396 {
397 	mutex_unlock(&c->jheads[jhead].wbuf.io_mutex);
398 }
399 
400 /**
401  * finish_reservation - finish a reservation.
402  * @c: UBIFS file-system description object
403  *
404  * This function finishes journal space reservation. It must be called after
405  * 'make_reservation()'.
406  */
407 static void finish_reservation(struct ubifs_info *c)
408 {
409 	up_read(&c->commit_sem);
410 }
411 
412 /**
413  * get_dent_type - translate VFS inode mode to UBIFS directory entry type.
414  * @mode: inode mode
415  */
416 static int get_dent_type(int mode)
417 {
418 	switch (mode & S_IFMT) {
419 	case S_IFREG:
420 		return UBIFS_ITYPE_REG;
421 	case S_IFDIR:
422 		return UBIFS_ITYPE_DIR;
423 	case S_IFLNK:
424 		return UBIFS_ITYPE_LNK;
425 	case S_IFBLK:
426 		return UBIFS_ITYPE_BLK;
427 	case S_IFCHR:
428 		return UBIFS_ITYPE_CHR;
429 	case S_IFIFO:
430 		return UBIFS_ITYPE_FIFO;
431 	case S_IFSOCK:
432 		return UBIFS_ITYPE_SOCK;
433 	default:
434 		BUG();
435 	}
436 	return 0;
437 }
438 
439 /**
440  * pack_inode - pack an inode node.
441  * @c: UBIFS file-system description object
442  * @ino: buffer in which to pack inode node
443  * @inode: inode to pack
444  * @last: indicates the last node of the group
445  */
446 static void pack_inode(struct ubifs_info *c, struct ubifs_ino_node *ino,
447 		       const struct inode *inode, int last)
448 {
449 	int data_len = 0, last_reference = !inode->i_nlink;
450 	struct ubifs_inode *ui = ubifs_inode(inode);
451 
452 	ino->ch.node_type = UBIFS_INO_NODE;
453 	ino_key_init_flash(c, &ino->key, inode->i_ino);
454 	ino->creat_sqnum = cpu_to_le64(ui->creat_sqnum);
455 	ino->atime_sec  = cpu_to_le64(inode->i_atime.tv_sec);
456 	ino->atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
457 	ino->ctime_sec  = cpu_to_le64(inode->i_ctime.tv_sec);
458 	ino->ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
459 	ino->mtime_sec  = cpu_to_le64(inode->i_mtime.tv_sec);
460 	ino->mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
461 	ino->uid   = cpu_to_le32(i_uid_read(inode));
462 	ino->gid   = cpu_to_le32(i_gid_read(inode));
463 	ino->mode  = cpu_to_le32(inode->i_mode);
464 	ino->flags = cpu_to_le32(ui->flags);
465 	ino->size  = cpu_to_le64(ui->ui_size);
466 	ino->nlink = cpu_to_le32(inode->i_nlink);
467 	ino->compr_type  = cpu_to_le16(ui->compr_type);
468 	ino->data_len    = cpu_to_le32(ui->data_len);
469 	ino->xattr_cnt   = cpu_to_le32(ui->xattr_cnt);
470 	ino->xattr_size  = cpu_to_le32(ui->xattr_size);
471 	ino->xattr_names = cpu_to_le32(ui->xattr_names);
472 	zero_ino_node_unused(ino);
473 
474 	/*
475 	 * Drop the attached data if this is a deletion inode, the data is not
476 	 * needed anymore.
477 	 */
478 	if (!last_reference) {
479 		memcpy(ino->data, ui->data, ui->data_len);
480 		data_len = ui->data_len;
481 	}
482 
483 	ubifs_prep_grp_node(c, ino, UBIFS_INO_NODE_SZ + data_len, last);
484 }
485 
486 /**
487  * mark_inode_clean - mark UBIFS inode as clean.
488  * @c: UBIFS file-system description object
489  * @ui: UBIFS inode to mark as clean
490  *
491  * This helper function marks UBIFS inode @ui as clean by cleaning the
492  * @ui->dirty flag and releasing its budget. Note, VFS may still treat the
493  * inode as dirty and try to write it back, but 'ubifs_write_inode()' would
494  * just do nothing.
495  */
496 static void mark_inode_clean(struct ubifs_info *c, struct ubifs_inode *ui)
497 {
498 	if (ui->dirty)
499 		ubifs_release_dirty_inode_budget(c, ui);
500 	ui->dirty = 0;
501 }
502 
503 static void set_dent_cookie(struct ubifs_info *c, struct ubifs_dent_node *dent)
504 {
505 	if (c->double_hash)
506 		dent->cookie = prandom_u32();
507 	else
508 		dent->cookie = 0;
509 }
510 
511 /**
512  * ubifs_jnl_update - update inode.
513  * @c: UBIFS file-system description object
514  * @dir: parent inode or host inode in case of extended attributes
515  * @nm: directory entry name
516  * @inode: inode to update
517  * @deletion: indicates a directory entry deletion i.e unlink or rmdir
518  * @xent: non-zero if the directory entry is an extended attribute entry
519  *
520  * This function updates an inode by writing a directory entry (or extended
521  * attribute entry), the inode itself, and the parent directory inode (or the
522  * host inode) to the journal.
523  *
524  * The function writes the host inode @dir last, which is important in case of
525  * extended attributes. Indeed, then we guarantee that if the host inode gets
526  * synchronized (with 'fsync()'), and the write-buffer it sits in gets flushed,
527  * the extended attribute inode gets flushed too. And this is exactly what the
528  * user expects - synchronizing the host inode synchronizes its extended
529  * attributes. Similarly, this guarantees that if @dir is synchronized, its
530  * directory entry corresponding to @nm gets synchronized too.
531  *
532  * If the inode (@inode) or the parent directory (@dir) are synchronous, this
533  * function synchronizes the write-buffer.
534  *
535  * This function marks the @dir and @inode inodes as clean and returns zero on
536  * success. In case of failure, a negative error code is returned.
537  */
538 int ubifs_jnl_update(struct ubifs_info *c, const struct inode *dir,
539 		     const struct fscrypt_name *nm, const struct inode *inode,
540 		     int deletion, int xent)
541 {
542 	int err, dlen, ilen, len, lnum, ino_offs, dent_offs;
543 	int aligned_dlen, aligned_ilen, sync = IS_DIRSYNC(dir);
544 	int last_reference = !!(deletion && inode->i_nlink == 0);
545 	struct ubifs_inode *ui = ubifs_inode(inode);
546 	struct ubifs_inode *host_ui = ubifs_inode(dir);
547 	struct ubifs_dent_node *dent;
548 	struct ubifs_ino_node *ino;
549 	union ubifs_key dent_key, ino_key;
550 
551 	ubifs_assert(mutex_is_locked(&host_ui->ui_mutex));
552 
553 	dlen = UBIFS_DENT_NODE_SZ + fname_len(nm) + 1;
554 	ilen = UBIFS_INO_NODE_SZ;
555 
556 	/*
557 	 * If the last reference to the inode is being deleted, then there is
558 	 * no need to attach and write inode data, it is being deleted anyway.
559 	 * And if the inode is being deleted, no need to synchronize
560 	 * write-buffer even if the inode is synchronous.
561 	 */
562 	if (!last_reference) {
563 		ilen += ui->data_len;
564 		sync |= IS_SYNC(inode);
565 	}
566 
567 	aligned_dlen = ALIGN(dlen, 8);
568 	aligned_ilen = ALIGN(ilen, 8);
569 
570 	len = aligned_dlen + aligned_ilen + UBIFS_INO_NODE_SZ;
571 	/* Make sure to also account for extended attributes */
572 	len += host_ui->data_len;
573 
574 	dent = kzalloc(len, GFP_NOFS);
575 	if (!dent)
576 		return -ENOMEM;
577 
578 	/* Make reservation before allocating sequence numbers */
579 	err = make_reservation(c, BASEHD, len);
580 	if (err)
581 		goto out_free;
582 
583 	if (!xent) {
584 		dent->ch.node_type = UBIFS_DENT_NODE;
585 		if (nm->hash)
586 			dent_key_init_hash(c, &dent_key, dir->i_ino, nm->hash);
587 		else
588 			dent_key_init(c, &dent_key, dir->i_ino, nm);
589 	} else {
590 		dent->ch.node_type = UBIFS_XENT_NODE;
591 		xent_key_init(c, &dent_key, dir->i_ino, nm);
592 	}
593 
594 	key_write(c, &dent_key, dent->key);
595 	dent->inum = deletion ? 0 : cpu_to_le64(inode->i_ino);
596 	dent->type = get_dent_type(inode->i_mode);
597 	dent->nlen = cpu_to_le16(fname_len(nm));
598 	memcpy(dent->name, fname_name(nm), fname_len(nm));
599 	dent->name[fname_len(nm)] = '\0';
600 	set_dent_cookie(c, dent);
601 
602 	zero_dent_node_unused(dent);
603 	ubifs_prep_grp_node(c, dent, dlen, 0);
604 
605 	ino = (void *)dent + aligned_dlen;
606 	pack_inode(c, ino, inode, 0);
607 	ino = (void *)ino + aligned_ilen;
608 	pack_inode(c, ino, dir, 1);
609 
610 	if (last_reference) {
611 		err = ubifs_add_orphan(c, inode->i_ino);
612 		if (err) {
613 			release_head(c, BASEHD);
614 			goto out_finish;
615 		}
616 		ui->del_cmtno = c->cmt_no;
617 	}
618 
619 	err = write_head(c, BASEHD, dent, len, &lnum, &dent_offs, sync);
620 	if (err)
621 		goto out_release;
622 	if (!sync) {
623 		struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf;
624 
625 		ubifs_wbuf_add_ino_nolock(wbuf, inode->i_ino);
626 		ubifs_wbuf_add_ino_nolock(wbuf, dir->i_ino);
627 	}
628 	release_head(c, BASEHD);
629 	kfree(dent);
630 
631 	if (deletion) {
632 		if (nm->hash)
633 			err = ubifs_tnc_remove_dh(c, &dent_key, nm->minor_hash);
634 		else
635 			err = ubifs_tnc_remove_nm(c, &dent_key, nm);
636 		if (err)
637 			goto out_ro;
638 		err = ubifs_add_dirt(c, lnum, dlen);
639 	} else
640 		err = ubifs_tnc_add_nm(c, &dent_key, lnum, dent_offs, dlen, nm);
641 	if (err)
642 		goto out_ro;
643 
644 	/*
645 	 * Note, we do not remove the inode from TNC even if the last reference
646 	 * to it has just been deleted, because the inode may still be opened.
647 	 * Instead, the inode has been added to orphan lists and the orphan
648 	 * subsystem will take further care about it.
649 	 */
650 	ino_key_init(c, &ino_key, inode->i_ino);
651 	ino_offs = dent_offs + aligned_dlen;
652 	err = ubifs_tnc_add(c, &ino_key, lnum, ino_offs, ilen);
653 	if (err)
654 		goto out_ro;
655 
656 	ino_key_init(c, &ino_key, dir->i_ino);
657 	ino_offs += aligned_ilen;
658 	err = ubifs_tnc_add(c, &ino_key, lnum, ino_offs,
659 			    UBIFS_INO_NODE_SZ + host_ui->data_len);
660 	if (err)
661 		goto out_ro;
662 
663 	finish_reservation(c);
664 	spin_lock(&ui->ui_lock);
665 	ui->synced_i_size = ui->ui_size;
666 	spin_unlock(&ui->ui_lock);
667 	mark_inode_clean(c, ui);
668 	mark_inode_clean(c, host_ui);
669 	return 0;
670 
671 out_finish:
672 	finish_reservation(c);
673 out_free:
674 	kfree(dent);
675 	return err;
676 
677 out_release:
678 	release_head(c, BASEHD);
679 	kfree(dent);
680 out_ro:
681 	ubifs_ro_mode(c, err);
682 	if (last_reference)
683 		ubifs_delete_orphan(c, inode->i_ino);
684 	finish_reservation(c);
685 	return err;
686 }
687 
688 /**
689  * ubifs_jnl_write_data - write a data node to the journal.
690  * @c: UBIFS file-system description object
691  * @inode: inode the data node belongs to
692  * @key: node key
693  * @buf: buffer to write
694  * @len: data length (must not exceed %UBIFS_BLOCK_SIZE)
695  *
696  * This function writes a data node to the journal. Returns %0 if the data node
697  * was successfully written, and a negative error code in case of failure.
698  */
699 int ubifs_jnl_write_data(struct ubifs_info *c, const struct inode *inode,
700 			 const union ubifs_key *key, const void *buf, int len)
701 {
702 	struct ubifs_data_node *data;
703 	int err, lnum, offs, compr_type, out_len, compr_len;
704 	int dlen = COMPRESSED_DATA_NODE_BUF_SZ, allocated = 1;
705 	struct ubifs_inode *ui = ubifs_inode(inode);
706 	bool encrypted = ubifs_crypt_is_encrypted(inode);
707 
708 	dbg_jnlk(key, "ino %lu, blk %u, len %d, key ",
709 		(unsigned long)key_inum(c, key), key_block(c, key), len);
710 	ubifs_assert(len <= UBIFS_BLOCK_SIZE);
711 
712 	if (encrypted)
713 		dlen += UBIFS_CIPHER_BLOCK_SIZE;
714 
715 	data = kmalloc(dlen, GFP_NOFS | __GFP_NOWARN);
716 	if (!data) {
717 		/*
718 		 * Fall-back to the write reserve buffer. Note, we might be
719 		 * currently on the memory reclaim path, when the kernel is
720 		 * trying to free some memory by writing out dirty pages. The
721 		 * write reserve buffer helps us to guarantee that we are
722 		 * always able to write the data.
723 		 */
724 		allocated = 0;
725 		mutex_lock(&c->write_reserve_mutex);
726 		data = c->write_reserve_buf;
727 	}
728 
729 	data->ch.node_type = UBIFS_DATA_NODE;
730 	key_write(c, key, &data->key);
731 	data->size = cpu_to_le32(len);
732 
733 	if (!(ui->flags & UBIFS_COMPR_FL))
734 		/* Compression is disabled for this inode */
735 		compr_type = UBIFS_COMPR_NONE;
736 	else
737 		compr_type = ui->compr_type;
738 
739 	out_len = compr_len = dlen - UBIFS_DATA_NODE_SZ;
740 	ubifs_compress(c, buf, len, &data->data, &compr_len, &compr_type);
741 	ubifs_assert(compr_len <= UBIFS_BLOCK_SIZE);
742 
743 	if (encrypted) {
744 		err = ubifs_encrypt(inode, data, compr_len, &out_len, key_block(c, key));
745 		if (err)
746 			goto out_free;
747 
748 	} else {
749 		data->compr_size = 0;
750 		out_len = compr_len;
751 	}
752 
753 	dlen = UBIFS_DATA_NODE_SZ + out_len;
754 	data->compr_type = cpu_to_le16(compr_type);
755 
756 	/* Make reservation before allocating sequence numbers */
757 	err = make_reservation(c, DATAHD, dlen);
758 	if (err)
759 		goto out_free;
760 
761 	err = write_node(c, DATAHD, data, dlen, &lnum, &offs);
762 	if (err)
763 		goto out_release;
764 	ubifs_wbuf_add_ino_nolock(&c->jheads[DATAHD].wbuf, key_inum(c, key));
765 	release_head(c, DATAHD);
766 
767 	err = ubifs_tnc_add(c, key, lnum, offs, dlen);
768 	if (err)
769 		goto out_ro;
770 
771 	finish_reservation(c);
772 	if (!allocated)
773 		mutex_unlock(&c->write_reserve_mutex);
774 	else
775 		kfree(data);
776 	return 0;
777 
778 out_release:
779 	release_head(c, DATAHD);
780 out_ro:
781 	ubifs_ro_mode(c, err);
782 	finish_reservation(c);
783 out_free:
784 	if (!allocated)
785 		mutex_unlock(&c->write_reserve_mutex);
786 	else
787 		kfree(data);
788 	return err;
789 }
790 
791 /**
792  * ubifs_jnl_write_inode - flush inode to the journal.
793  * @c: UBIFS file-system description object
794  * @inode: inode to flush
795  *
796  * This function writes inode @inode to the journal. If the inode is
797  * synchronous, it also synchronizes the write-buffer. Returns zero in case of
798  * success and a negative error code in case of failure.
799  */
800 int ubifs_jnl_write_inode(struct ubifs_info *c, const struct inode *inode)
801 {
802 	int err, lnum, offs;
803 	struct ubifs_ino_node *ino;
804 	struct ubifs_inode *ui = ubifs_inode(inode);
805 	int sync = 0, len = UBIFS_INO_NODE_SZ, last_reference = !inode->i_nlink;
806 
807 	dbg_jnl("ino %lu, nlink %u", inode->i_ino, inode->i_nlink);
808 
809 	/*
810 	 * If the inode is being deleted, do not write the attached data. No
811 	 * need to synchronize the write-buffer either.
812 	 */
813 	if (!last_reference) {
814 		len += ui->data_len;
815 		sync = IS_SYNC(inode);
816 	}
817 	ino = kmalloc(len, GFP_NOFS);
818 	if (!ino)
819 		return -ENOMEM;
820 
821 	/* Make reservation before allocating sequence numbers */
822 	err = make_reservation(c, BASEHD, len);
823 	if (err)
824 		goto out_free;
825 
826 	pack_inode(c, ino, inode, 1);
827 	err = write_head(c, BASEHD, ino, len, &lnum, &offs, sync);
828 	if (err)
829 		goto out_release;
830 	if (!sync)
831 		ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf,
832 					  inode->i_ino);
833 	release_head(c, BASEHD);
834 
835 	if (last_reference) {
836 		err = ubifs_tnc_remove_ino(c, inode->i_ino);
837 		if (err)
838 			goto out_ro;
839 		ubifs_delete_orphan(c, inode->i_ino);
840 		err = ubifs_add_dirt(c, lnum, len);
841 	} else {
842 		union ubifs_key key;
843 
844 		ino_key_init(c, &key, inode->i_ino);
845 		err = ubifs_tnc_add(c, &key, lnum, offs, len);
846 	}
847 	if (err)
848 		goto out_ro;
849 
850 	finish_reservation(c);
851 	spin_lock(&ui->ui_lock);
852 	ui->synced_i_size = ui->ui_size;
853 	spin_unlock(&ui->ui_lock);
854 	kfree(ino);
855 	return 0;
856 
857 out_release:
858 	release_head(c, BASEHD);
859 out_ro:
860 	ubifs_ro_mode(c, err);
861 	finish_reservation(c);
862 out_free:
863 	kfree(ino);
864 	return err;
865 }
866 
867 /**
868  * ubifs_jnl_delete_inode - delete an inode.
869  * @c: UBIFS file-system description object
870  * @inode: inode to delete
871  *
872  * This function deletes inode @inode which includes removing it from orphans,
873  * deleting it from TNC and, in some cases, writing a deletion inode to the
874  * journal.
875  *
876  * When regular file inodes are unlinked or a directory inode is removed, the
877  * 'ubifs_jnl_update()' function writes a corresponding deletion inode and
878  * direntry to the media, and adds the inode to orphans. After this, when the
879  * last reference to this inode has been dropped, this function is called. In
880  * general, it has to write one more deletion inode to the media, because if
881  * a commit happened between 'ubifs_jnl_update()' and
882  * 'ubifs_jnl_delete_inode()', the deletion inode is not in the journal
883  * anymore, and in fact it might not be on the flash anymore, because it might
884  * have been garbage-collected already. And for optimization reasons UBIFS does
885  * not read the orphan area if it has been unmounted cleanly, so it would have
886  * no indication in the journal that there is a deleted inode which has to be
887  * removed from TNC.
888  *
889  * However, if there was no commit between 'ubifs_jnl_update()' and
890  * 'ubifs_jnl_delete_inode()', then there is no need to write the deletion
891  * inode to the media for the second time. And this is quite a typical case.
892  *
893  * This function returns zero in case of success and a negative error code in
894  * case of failure.
895  */
896 int ubifs_jnl_delete_inode(struct ubifs_info *c, const struct inode *inode)
897 {
898 	int err;
899 	struct ubifs_inode *ui = ubifs_inode(inode);
900 
901 	ubifs_assert(inode->i_nlink == 0);
902 
903 	if (ui->del_cmtno != c->cmt_no)
904 		/* A commit happened for sure */
905 		return ubifs_jnl_write_inode(c, inode);
906 
907 	down_read(&c->commit_sem);
908 	/*
909 	 * Check commit number again, because the first test has been done
910 	 * without @c->commit_sem, so a commit might have happened.
911 	 */
912 	if (ui->del_cmtno != c->cmt_no) {
913 		up_read(&c->commit_sem);
914 		return ubifs_jnl_write_inode(c, inode);
915 	}
916 
917 	err = ubifs_tnc_remove_ino(c, inode->i_ino);
918 	if (err)
919 		ubifs_ro_mode(c, err);
920 	else
921 		ubifs_delete_orphan(c, inode->i_ino);
922 	up_read(&c->commit_sem);
923 	return err;
924 }
925 
926 /**
927  * ubifs_jnl_xrename - cross rename two directory entries.
928  * @c: UBIFS file-system description object
929  * @fst_dir: parent inode of 1st directory entry to exchange
930  * @fst_inode: 1st inode to exchange
931  * @fst_nm: name of 1st inode to exchange
932  * @snd_dir: parent inode of 2nd directory entry to exchange
933  * @snd_inode: 2nd inode to exchange
934  * @snd_nm: name of 2nd inode to exchange
935  * @sync: non-zero if the write-buffer has to be synchronized
936  *
937  * This function implements the cross rename operation which may involve
938  * writing 2 inodes and 2 directory entries. It marks the written inodes as clean
939  * and returns zero on success. In case of failure, a negative error code is
940  * returned.
941  */
942 int ubifs_jnl_xrename(struct ubifs_info *c, const struct inode *fst_dir,
943 		      const struct inode *fst_inode,
944 		      const struct fscrypt_name *fst_nm,
945 		      const struct inode *snd_dir,
946 		      const struct inode *snd_inode,
947 		      const struct fscrypt_name *snd_nm, int sync)
948 {
949 	union ubifs_key key;
950 	struct ubifs_dent_node *dent1, *dent2;
951 	int err, dlen1, dlen2, lnum, offs, len, plen = UBIFS_INO_NODE_SZ;
952 	int aligned_dlen1, aligned_dlen2;
953 	int twoparents = (fst_dir != snd_dir);
954 	void *p;
955 
956 	ubifs_assert(ubifs_inode(fst_dir)->data_len == 0);
957 	ubifs_assert(ubifs_inode(snd_dir)->data_len == 0);
958 	ubifs_assert(mutex_is_locked(&ubifs_inode(fst_dir)->ui_mutex));
959 	ubifs_assert(mutex_is_locked(&ubifs_inode(snd_dir)->ui_mutex));
960 
961 	dlen1 = UBIFS_DENT_NODE_SZ + fname_len(snd_nm) + 1;
962 	dlen2 = UBIFS_DENT_NODE_SZ + fname_len(fst_nm) + 1;
963 	aligned_dlen1 = ALIGN(dlen1, 8);
964 	aligned_dlen2 = ALIGN(dlen2, 8);
965 
966 	len = aligned_dlen1 + aligned_dlen2 + ALIGN(plen, 8);
967 	if (twoparents)
968 		len += plen;
969 
970 	dent1 = kzalloc(len, GFP_NOFS);
971 	if (!dent1)
972 		return -ENOMEM;
973 
974 	/* Make reservation before allocating sequence numbers */
975 	err = make_reservation(c, BASEHD, len);
976 	if (err)
977 		goto out_free;
978 
979 	/* Make new dent for 1st entry */
980 	dent1->ch.node_type = UBIFS_DENT_NODE;
981 	dent_key_init_flash(c, &dent1->key, snd_dir->i_ino, snd_nm);
982 	dent1->inum = cpu_to_le64(fst_inode->i_ino);
983 	dent1->type = get_dent_type(fst_inode->i_mode);
984 	dent1->nlen = cpu_to_le16(fname_len(snd_nm));
985 	memcpy(dent1->name, fname_name(snd_nm), fname_len(snd_nm));
986 	dent1->name[fname_len(snd_nm)] = '\0';
987 	set_dent_cookie(c, dent1);
988 	zero_dent_node_unused(dent1);
989 	ubifs_prep_grp_node(c, dent1, dlen1, 0);
990 
991 	/* Make new dent for 2nd entry */
992 	dent2 = (void *)dent1 + aligned_dlen1;
993 	dent2->ch.node_type = UBIFS_DENT_NODE;
994 	dent_key_init_flash(c, &dent2->key, fst_dir->i_ino, fst_nm);
995 	dent2->inum = cpu_to_le64(snd_inode->i_ino);
996 	dent2->type = get_dent_type(snd_inode->i_mode);
997 	dent2->nlen = cpu_to_le16(fname_len(fst_nm));
998 	memcpy(dent2->name, fname_name(fst_nm), fname_len(fst_nm));
999 	dent2->name[fname_len(fst_nm)] = '\0';
1000 	set_dent_cookie(c, dent2);
1001 	zero_dent_node_unused(dent2);
1002 	ubifs_prep_grp_node(c, dent2, dlen2, 0);
1003 
1004 	p = (void *)dent2 + aligned_dlen2;
1005 	if (!twoparents)
1006 		pack_inode(c, p, fst_dir, 1);
1007 	else {
1008 		pack_inode(c, p, fst_dir, 0);
1009 		p += ALIGN(plen, 8);
1010 		pack_inode(c, p, snd_dir, 1);
1011 	}
1012 
1013 	err = write_head(c, BASEHD, dent1, len, &lnum, &offs, sync);
1014 	if (err)
1015 		goto out_release;
1016 	if (!sync) {
1017 		struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf;
1018 
1019 		ubifs_wbuf_add_ino_nolock(wbuf, fst_dir->i_ino);
1020 		ubifs_wbuf_add_ino_nolock(wbuf, snd_dir->i_ino);
1021 	}
1022 	release_head(c, BASEHD);
1023 
1024 	dent_key_init(c, &key, snd_dir->i_ino, snd_nm);
1025 	err = ubifs_tnc_add_nm(c, &key, lnum, offs, dlen1, snd_nm);
1026 	if (err)
1027 		goto out_ro;
1028 
1029 	offs += aligned_dlen1;
1030 	dent_key_init(c, &key, fst_dir->i_ino, fst_nm);
1031 	err = ubifs_tnc_add_nm(c, &key, lnum, offs, dlen2, fst_nm);
1032 	if (err)
1033 		goto out_ro;
1034 
1035 	offs += aligned_dlen2;
1036 
1037 	ino_key_init(c, &key, fst_dir->i_ino);
1038 	err = ubifs_tnc_add(c, &key, lnum, offs, plen);
1039 	if (err)
1040 		goto out_ro;
1041 
1042 	if (twoparents) {
1043 		offs += ALIGN(plen, 8);
1044 		ino_key_init(c, &key, snd_dir->i_ino);
1045 		err = ubifs_tnc_add(c, &key, lnum, offs, plen);
1046 		if (err)
1047 			goto out_ro;
1048 	}
1049 
1050 	finish_reservation(c);
1051 
1052 	mark_inode_clean(c, ubifs_inode(fst_dir));
1053 	if (twoparents)
1054 		mark_inode_clean(c, ubifs_inode(snd_dir));
1055 	kfree(dent1);
1056 	return 0;
1057 
1058 out_release:
1059 	release_head(c, BASEHD);
1060 out_ro:
1061 	ubifs_ro_mode(c, err);
1062 	finish_reservation(c);
1063 out_free:
1064 	kfree(dent1);
1065 	return err;
1066 }
1067 
1068 /**
1069  * ubifs_jnl_rename - rename a directory entry.
1070  * @c: UBIFS file-system description object
1071  * @old_dir: parent inode of directory entry to rename
1072  * @old_dentry: directory entry to rename
1073  * @new_dir: parent inode of directory entry to rename
1074  * @new_dentry: new directory entry (or directory entry to replace)
1075  * @sync: non-zero if the write-buffer has to be synchronized
1076  *
1077  * This function implements the re-name operation which may involve writing up
1078  * to 4 inodes and 2 directory entries. It marks the written inodes as clean
1079  * and returns zero on success. In case of failure, a negative error code is
1080  * returned.
1081  */
1082 int ubifs_jnl_rename(struct ubifs_info *c, const struct inode *old_dir,
1083 		     const struct inode *old_inode,
1084 		     const struct fscrypt_name *old_nm,
1085 		     const struct inode *new_dir,
1086 		     const struct inode *new_inode,
1087 		     const struct fscrypt_name *new_nm,
1088 		     const struct inode *whiteout, int sync)
1089 {
1090 	void *p;
1091 	union ubifs_key key;
1092 	struct ubifs_dent_node *dent, *dent2;
1093 	int err, dlen1, dlen2, ilen, lnum, offs, len;
1094 	int aligned_dlen1, aligned_dlen2, plen = UBIFS_INO_NODE_SZ;
1095 	int last_reference = !!(new_inode && new_inode->i_nlink == 0);
1096 	int move = (old_dir != new_dir);
1097 	struct ubifs_inode *uninitialized_var(new_ui);
1098 
1099 	ubifs_assert(ubifs_inode(old_dir)->data_len == 0);
1100 	ubifs_assert(ubifs_inode(new_dir)->data_len == 0);
1101 	ubifs_assert(mutex_is_locked(&ubifs_inode(old_dir)->ui_mutex));
1102 	ubifs_assert(mutex_is_locked(&ubifs_inode(new_dir)->ui_mutex));
1103 
1104 	dlen1 = UBIFS_DENT_NODE_SZ + fname_len(new_nm) + 1;
1105 	dlen2 = UBIFS_DENT_NODE_SZ + fname_len(old_nm) + 1;
1106 	if (new_inode) {
1107 		new_ui = ubifs_inode(new_inode);
1108 		ubifs_assert(mutex_is_locked(&new_ui->ui_mutex));
1109 		ilen = UBIFS_INO_NODE_SZ;
1110 		if (!last_reference)
1111 			ilen += new_ui->data_len;
1112 	} else
1113 		ilen = 0;
1114 
1115 	aligned_dlen1 = ALIGN(dlen1, 8);
1116 	aligned_dlen2 = ALIGN(dlen2, 8);
1117 	len = aligned_dlen1 + aligned_dlen2 + ALIGN(ilen, 8) + ALIGN(plen, 8);
1118 	if (move)
1119 		len += plen;
1120 	dent = kzalloc(len, GFP_NOFS);
1121 	if (!dent)
1122 		return -ENOMEM;
1123 
1124 	/* Make reservation before allocating sequence numbers */
1125 	err = make_reservation(c, BASEHD, len);
1126 	if (err)
1127 		goto out_free;
1128 
1129 	/* Make new dent */
1130 	dent->ch.node_type = UBIFS_DENT_NODE;
1131 	dent_key_init_flash(c, &dent->key, new_dir->i_ino, new_nm);
1132 	dent->inum = cpu_to_le64(old_inode->i_ino);
1133 	dent->type = get_dent_type(old_inode->i_mode);
1134 	dent->nlen = cpu_to_le16(fname_len(new_nm));
1135 	memcpy(dent->name, fname_name(new_nm), fname_len(new_nm));
1136 	dent->name[fname_len(new_nm)] = '\0';
1137 	set_dent_cookie(c, dent);
1138 	zero_dent_node_unused(dent);
1139 	ubifs_prep_grp_node(c, dent, dlen1, 0);
1140 
1141 	dent2 = (void *)dent + aligned_dlen1;
1142 	dent2->ch.node_type = UBIFS_DENT_NODE;
1143 	dent_key_init_flash(c, &dent2->key, old_dir->i_ino, old_nm);
1144 
1145 	if (whiteout) {
1146 		dent2->inum = cpu_to_le64(whiteout->i_ino);
1147 		dent2->type = get_dent_type(whiteout->i_mode);
1148 	} else {
1149 		/* Make deletion dent */
1150 		dent2->inum = 0;
1151 		dent2->type = DT_UNKNOWN;
1152 	}
1153 	dent2->nlen = cpu_to_le16(fname_len(old_nm));
1154 	memcpy(dent2->name, fname_name(old_nm), fname_len(old_nm));
1155 	dent2->name[fname_len(old_nm)] = '\0';
1156 	set_dent_cookie(c, dent2);
1157 	zero_dent_node_unused(dent2);
1158 	ubifs_prep_grp_node(c, dent2, dlen2, 0);
1159 
1160 	p = (void *)dent2 + aligned_dlen2;
1161 	if (new_inode) {
1162 		pack_inode(c, p, new_inode, 0);
1163 		p += ALIGN(ilen, 8);
1164 	}
1165 
1166 	if (!move)
1167 		pack_inode(c, p, old_dir, 1);
1168 	else {
1169 		pack_inode(c, p, old_dir, 0);
1170 		p += ALIGN(plen, 8);
1171 		pack_inode(c, p, new_dir, 1);
1172 	}
1173 
1174 	if (last_reference) {
1175 		err = ubifs_add_orphan(c, new_inode->i_ino);
1176 		if (err) {
1177 			release_head(c, BASEHD);
1178 			goto out_finish;
1179 		}
1180 		new_ui->del_cmtno = c->cmt_no;
1181 	}
1182 
1183 	err = write_head(c, BASEHD, dent, len, &lnum, &offs, sync);
1184 	if (err)
1185 		goto out_release;
1186 	if (!sync) {
1187 		struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf;
1188 
1189 		ubifs_wbuf_add_ino_nolock(wbuf, new_dir->i_ino);
1190 		ubifs_wbuf_add_ino_nolock(wbuf, old_dir->i_ino);
1191 		if (new_inode)
1192 			ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf,
1193 						  new_inode->i_ino);
1194 	}
1195 	release_head(c, BASEHD);
1196 
1197 	dent_key_init(c, &key, new_dir->i_ino, new_nm);
1198 	err = ubifs_tnc_add_nm(c, &key, lnum, offs, dlen1, new_nm);
1199 	if (err)
1200 		goto out_ro;
1201 
1202 	offs += aligned_dlen1;
1203 	if (whiteout) {
1204 		dent_key_init(c, &key, old_dir->i_ino, old_nm);
1205 		err = ubifs_tnc_add_nm(c, &key, lnum, offs, dlen2, old_nm);
1206 		if (err)
1207 			goto out_ro;
1208 
1209 		ubifs_delete_orphan(c, whiteout->i_ino);
1210 	} else {
1211 		err = ubifs_add_dirt(c, lnum, dlen2);
1212 		if (err)
1213 			goto out_ro;
1214 
1215 		dent_key_init(c, &key, old_dir->i_ino, old_nm);
1216 		err = ubifs_tnc_remove_nm(c, &key, old_nm);
1217 		if (err)
1218 			goto out_ro;
1219 	}
1220 
1221 	offs += aligned_dlen2;
1222 	if (new_inode) {
1223 		ino_key_init(c, &key, new_inode->i_ino);
1224 		err = ubifs_tnc_add(c, &key, lnum, offs, ilen);
1225 		if (err)
1226 			goto out_ro;
1227 		offs += ALIGN(ilen, 8);
1228 	}
1229 
1230 	ino_key_init(c, &key, old_dir->i_ino);
1231 	err = ubifs_tnc_add(c, &key, lnum, offs, plen);
1232 	if (err)
1233 		goto out_ro;
1234 
1235 	if (move) {
1236 		offs += ALIGN(plen, 8);
1237 		ino_key_init(c, &key, new_dir->i_ino);
1238 		err = ubifs_tnc_add(c, &key, lnum, offs, plen);
1239 		if (err)
1240 			goto out_ro;
1241 	}
1242 
1243 	finish_reservation(c);
1244 	if (new_inode) {
1245 		mark_inode_clean(c, new_ui);
1246 		spin_lock(&new_ui->ui_lock);
1247 		new_ui->synced_i_size = new_ui->ui_size;
1248 		spin_unlock(&new_ui->ui_lock);
1249 	}
1250 	mark_inode_clean(c, ubifs_inode(old_dir));
1251 	if (move)
1252 		mark_inode_clean(c, ubifs_inode(new_dir));
1253 	kfree(dent);
1254 	return 0;
1255 
1256 out_release:
1257 	release_head(c, BASEHD);
1258 out_ro:
1259 	ubifs_ro_mode(c, err);
1260 	if (last_reference)
1261 		ubifs_delete_orphan(c, new_inode->i_ino);
1262 out_finish:
1263 	finish_reservation(c);
1264 out_free:
1265 	kfree(dent);
1266 	return err;
1267 }
1268 
1269 /**
1270  * truncate_data_node - re-compress/encrypt a truncated data node.
1271  * @c: UBIFS file-system description object
1272  * @inode: inode which referes to the data node
1273  * @block: data block number
1274  * @dn: data node to re-compress
1275  * @new_len: new length
1276  *
1277  * This function is used when an inode is truncated and the last data node of
1278  * the inode has to be re-compressed/encrypted and re-written.
1279  */
1280 static int truncate_data_node(const struct ubifs_info *c, const struct inode *inode,
1281 			      unsigned int block, struct ubifs_data_node *dn,
1282 			      int *new_len)
1283 {
1284 	void *buf;
1285 	int err, compr_type;
1286 	u32 dlen, out_len, old_dlen;
1287 
1288 	out_len = le32_to_cpu(dn->size);
1289 	buf = kmalloc_array(out_len, WORST_COMPR_FACTOR, GFP_NOFS);
1290 	if (!buf)
1291 		return -ENOMEM;
1292 
1293 	dlen = old_dlen = le32_to_cpu(dn->ch.len) - UBIFS_DATA_NODE_SZ;
1294 	compr_type = le16_to_cpu(dn->compr_type);
1295 
1296 	if (ubifs_crypt_is_encrypted(inode)) {
1297 		err = ubifs_decrypt(inode, dn, &dlen, block);
1298 		if (err)
1299 			goto out;
1300 	}
1301 
1302 	if (compr_type == UBIFS_COMPR_NONE) {
1303 		out_len = *new_len;
1304 	} else {
1305 		err = ubifs_decompress(c, &dn->data, dlen, buf, &out_len, compr_type);
1306 		if (err)
1307 			goto out;
1308 
1309 		ubifs_compress(c, buf, *new_len, &dn->data, &out_len, &compr_type);
1310 	}
1311 
1312 	if (ubifs_crypt_is_encrypted(inode)) {
1313 		err = ubifs_encrypt(inode, dn, out_len, &old_dlen, block);
1314 		if (err)
1315 			goto out;
1316 
1317 		out_len = old_dlen;
1318 	} else {
1319 		dn->compr_size = 0;
1320 	}
1321 
1322 	ubifs_assert(out_len <= UBIFS_BLOCK_SIZE);
1323 	dn->compr_type = cpu_to_le16(compr_type);
1324 	dn->size = cpu_to_le32(*new_len);
1325 	*new_len = UBIFS_DATA_NODE_SZ + out_len;
1326 	err = 0;
1327 out:
1328 	kfree(buf);
1329 	return err;
1330 }
1331 
1332 /**
1333  * ubifs_jnl_truncate - update the journal for a truncation.
1334  * @c: UBIFS file-system description object
1335  * @inode: inode to truncate
1336  * @old_size: old size
1337  * @new_size: new size
1338  *
1339  * When the size of a file decreases due to truncation, a truncation node is
1340  * written, the journal tree is updated, and the last data block is re-written
1341  * if it has been affected. The inode is also updated in order to synchronize
1342  * the new inode size.
1343  *
1344  * This function marks the inode as clean and returns zero on success. In case
1345  * of failure, a negative error code is returned.
1346  */
1347 int ubifs_jnl_truncate(struct ubifs_info *c, const struct inode *inode,
1348 		       loff_t old_size, loff_t new_size)
1349 {
1350 	union ubifs_key key, to_key;
1351 	struct ubifs_ino_node *ino;
1352 	struct ubifs_trun_node *trun;
1353 	struct ubifs_data_node *uninitialized_var(dn);
1354 	int err, dlen, len, lnum, offs, bit, sz, sync = IS_SYNC(inode);
1355 	struct ubifs_inode *ui = ubifs_inode(inode);
1356 	ino_t inum = inode->i_ino;
1357 	unsigned int blk;
1358 
1359 	dbg_jnl("ino %lu, size %lld -> %lld",
1360 		(unsigned long)inum, old_size, new_size);
1361 	ubifs_assert(!ui->data_len);
1362 	ubifs_assert(S_ISREG(inode->i_mode));
1363 	ubifs_assert(mutex_is_locked(&ui->ui_mutex));
1364 
1365 	sz = UBIFS_TRUN_NODE_SZ + UBIFS_INO_NODE_SZ +
1366 	     UBIFS_MAX_DATA_NODE_SZ * WORST_COMPR_FACTOR;
1367 	ino = kmalloc(sz, GFP_NOFS);
1368 	if (!ino)
1369 		return -ENOMEM;
1370 
1371 	trun = (void *)ino + UBIFS_INO_NODE_SZ;
1372 	trun->ch.node_type = UBIFS_TRUN_NODE;
1373 	trun->inum = cpu_to_le32(inum);
1374 	trun->old_size = cpu_to_le64(old_size);
1375 	trun->new_size = cpu_to_le64(new_size);
1376 	zero_trun_node_unused(trun);
1377 
1378 	dlen = new_size & (UBIFS_BLOCK_SIZE - 1);
1379 	if (dlen) {
1380 		/* Get last data block so it can be truncated */
1381 		dn = (void *)trun + UBIFS_TRUN_NODE_SZ;
1382 		blk = new_size >> UBIFS_BLOCK_SHIFT;
1383 		data_key_init(c, &key, inum, blk);
1384 		dbg_jnlk(&key, "last block key ");
1385 		err = ubifs_tnc_lookup(c, &key, dn);
1386 		if (err == -ENOENT)
1387 			dlen = 0; /* Not found (so it is a hole) */
1388 		else if (err)
1389 			goto out_free;
1390 		else {
1391 			if (le32_to_cpu(dn->size) <= dlen)
1392 				dlen = 0; /* Nothing to do */
1393 			else {
1394 				err = truncate_data_node(c, inode, blk, dn, &dlen);
1395 				if (err)
1396 					goto out_free;
1397 			}
1398 		}
1399 	}
1400 
1401 	/* Must make reservation before allocating sequence numbers */
1402 	len = UBIFS_TRUN_NODE_SZ + UBIFS_INO_NODE_SZ;
1403 	if (dlen)
1404 		len += dlen;
1405 	err = make_reservation(c, BASEHD, len);
1406 	if (err)
1407 		goto out_free;
1408 
1409 	pack_inode(c, ino, inode, 0);
1410 	ubifs_prep_grp_node(c, trun, UBIFS_TRUN_NODE_SZ, dlen ? 0 : 1);
1411 	if (dlen)
1412 		ubifs_prep_grp_node(c, dn, dlen, 1);
1413 
1414 	err = write_head(c, BASEHD, ino, len, &lnum, &offs, sync);
1415 	if (err)
1416 		goto out_release;
1417 	if (!sync)
1418 		ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf, inum);
1419 	release_head(c, BASEHD);
1420 
1421 	if (dlen) {
1422 		sz = offs + UBIFS_INO_NODE_SZ + UBIFS_TRUN_NODE_SZ;
1423 		err = ubifs_tnc_add(c, &key, lnum, sz, dlen);
1424 		if (err)
1425 			goto out_ro;
1426 	}
1427 
1428 	ino_key_init(c, &key, inum);
1429 	err = ubifs_tnc_add(c, &key, lnum, offs, UBIFS_INO_NODE_SZ);
1430 	if (err)
1431 		goto out_ro;
1432 
1433 	err = ubifs_add_dirt(c, lnum, UBIFS_TRUN_NODE_SZ);
1434 	if (err)
1435 		goto out_ro;
1436 
1437 	bit = new_size & (UBIFS_BLOCK_SIZE - 1);
1438 	blk = (new_size >> UBIFS_BLOCK_SHIFT) + (bit ? 1 : 0);
1439 	data_key_init(c, &key, inum, blk);
1440 
1441 	bit = old_size & (UBIFS_BLOCK_SIZE - 1);
1442 	blk = (old_size >> UBIFS_BLOCK_SHIFT) - (bit ? 0 : 1);
1443 	data_key_init(c, &to_key, inum, blk);
1444 
1445 	err = ubifs_tnc_remove_range(c, &key, &to_key);
1446 	if (err)
1447 		goto out_ro;
1448 
1449 	finish_reservation(c);
1450 	spin_lock(&ui->ui_lock);
1451 	ui->synced_i_size = ui->ui_size;
1452 	spin_unlock(&ui->ui_lock);
1453 	mark_inode_clean(c, ui);
1454 	kfree(ino);
1455 	return 0;
1456 
1457 out_release:
1458 	release_head(c, BASEHD);
1459 out_ro:
1460 	ubifs_ro_mode(c, err);
1461 	finish_reservation(c);
1462 out_free:
1463 	kfree(ino);
1464 	return err;
1465 }
1466 
1467 
1468 /**
1469  * ubifs_jnl_delete_xattr - delete an extended attribute.
1470  * @c: UBIFS file-system description object
1471  * @host: host inode
1472  * @inode: extended attribute inode
1473  * @nm: extended attribute entry name
1474  *
1475  * This function delete an extended attribute which is very similar to
1476  * un-linking regular files - it writes a deletion xentry, a deletion inode and
1477  * updates the target inode. Returns zero in case of success and a negative
1478  * error code in case of failure.
1479  */
1480 int ubifs_jnl_delete_xattr(struct ubifs_info *c, const struct inode *host,
1481 			   const struct inode *inode,
1482 			   const struct fscrypt_name *nm)
1483 {
1484 	int err, xlen, hlen, len, lnum, xent_offs, aligned_xlen;
1485 	struct ubifs_dent_node *xent;
1486 	struct ubifs_ino_node *ino;
1487 	union ubifs_key xent_key, key1, key2;
1488 	int sync = IS_DIRSYNC(host);
1489 	struct ubifs_inode *host_ui = ubifs_inode(host);
1490 
1491 	ubifs_assert(inode->i_nlink == 0);
1492 	ubifs_assert(mutex_is_locked(&host_ui->ui_mutex));
1493 
1494 	/*
1495 	 * Since we are deleting the inode, we do not bother to attach any data
1496 	 * to it and assume its length is %UBIFS_INO_NODE_SZ.
1497 	 */
1498 	xlen = UBIFS_DENT_NODE_SZ + fname_len(nm) + 1;
1499 	aligned_xlen = ALIGN(xlen, 8);
1500 	hlen = host_ui->data_len + UBIFS_INO_NODE_SZ;
1501 	len = aligned_xlen + UBIFS_INO_NODE_SZ + ALIGN(hlen, 8);
1502 
1503 	xent = kzalloc(len, GFP_NOFS);
1504 	if (!xent)
1505 		return -ENOMEM;
1506 
1507 	/* Make reservation before allocating sequence numbers */
1508 	err = make_reservation(c, BASEHD, len);
1509 	if (err) {
1510 		kfree(xent);
1511 		return err;
1512 	}
1513 
1514 	xent->ch.node_type = UBIFS_XENT_NODE;
1515 	xent_key_init(c, &xent_key, host->i_ino, nm);
1516 	key_write(c, &xent_key, xent->key);
1517 	xent->inum = 0;
1518 	xent->type = get_dent_type(inode->i_mode);
1519 	xent->nlen = cpu_to_le16(fname_len(nm));
1520 	memcpy(xent->name, fname_name(nm), fname_len(nm));
1521 	xent->name[fname_len(nm)] = '\0';
1522 	zero_dent_node_unused(xent);
1523 	ubifs_prep_grp_node(c, xent, xlen, 0);
1524 
1525 	ino = (void *)xent + aligned_xlen;
1526 	pack_inode(c, ino, inode, 0);
1527 	ino = (void *)ino + UBIFS_INO_NODE_SZ;
1528 	pack_inode(c, ino, host, 1);
1529 
1530 	err = write_head(c, BASEHD, xent, len, &lnum, &xent_offs, sync);
1531 	if (!sync && !err)
1532 		ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf, host->i_ino);
1533 	release_head(c, BASEHD);
1534 	kfree(xent);
1535 	if (err)
1536 		goto out_ro;
1537 
1538 	/* Remove the extended attribute entry from TNC */
1539 	err = ubifs_tnc_remove_nm(c, &xent_key, nm);
1540 	if (err)
1541 		goto out_ro;
1542 	err = ubifs_add_dirt(c, lnum, xlen);
1543 	if (err)
1544 		goto out_ro;
1545 
1546 	/*
1547 	 * Remove all nodes belonging to the extended attribute inode from TNC.
1548 	 * Well, there actually must be only one node - the inode itself.
1549 	 */
1550 	lowest_ino_key(c, &key1, inode->i_ino);
1551 	highest_ino_key(c, &key2, inode->i_ino);
1552 	err = ubifs_tnc_remove_range(c, &key1, &key2);
1553 	if (err)
1554 		goto out_ro;
1555 	err = ubifs_add_dirt(c, lnum, UBIFS_INO_NODE_SZ);
1556 	if (err)
1557 		goto out_ro;
1558 
1559 	/* And update TNC with the new host inode position */
1560 	ino_key_init(c, &key1, host->i_ino);
1561 	err = ubifs_tnc_add(c, &key1, lnum, xent_offs + len - hlen, hlen);
1562 	if (err)
1563 		goto out_ro;
1564 
1565 	finish_reservation(c);
1566 	spin_lock(&host_ui->ui_lock);
1567 	host_ui->synced_i_size = host_ui->ui_size;
1568 	spin_unlock(&host_ui->ui_lock);
1569 	mark_inode_clean(c, host_ui);
1570 	return 0;
1571 
1572 out_ro:
1573 	ubifs_ro_mode(c, err);
1574 	finish_reservation(c);
1575 	return err;
1576 }
1577 
1578 /**
1579  * ubifs_jnl_change_xattr - change an extended attribute.
1580  * @c: UBIFS file-system description object
1581  * @inode: extended attribute inode
1582  * @host: host inode
1583  *
1584  * This function writes the updated version of an extended attribute inode and
1585  * the host inode to the journal (to the base head). The host inode is written
1586  * after the extended attribute inode in order to guarantee that the extended
1587  * attribute will be flushed when the inode is synchronized by 'fsync()' and
1588  * consequently, the write-buffer is synchronized. This function returns zero
1589  * in case of success and a negative error code in case of failure.
1590  */
1591 int ubifs_jnl_change_xattr(struct ubifs_info *c, const struct inode *inode,
1592 			   const struct inode *host)
1593 {
1594 	int err, len1, len2, aligned_len, aligned_len1, lnum, offs;
1595 	struct ubifs_inode *host_ui = ubifs_inode(host);
1596 	struct ubifs_ino_node *ino;
1597 	union ubifs_key key;
1598 	int sync = IS_DIRSYNC(host);
1599 
1600 	dbg_jnl("ino %lu, ino %lu", host->i_ino, inode->i_ino);
1601 	ubifs_assert(host->i_nlink > 0);
1602 	ubifs_assert(inode->i_nlink > 0);
1603 	ubifs_assert(mutex_is_locked(&host_ui->ui_mutex));
1604 
1605 	len1 = UBIFS_INO_NODE_SZ + host_ui->data_len;
1606 	len2 = UBIFS_INO_NODE_SZ + ubifs_inode(inode)->data_len;
1607 	aligned_len1 = ALIGN(len1, 8);
1608 	aligned_len = aligned_len1 + ALIGN(len2, 8);
1609 
1610 	ino = kzalloc(aligned_len, GFP_NOFS);
1611 	if (!ino)
1612 		return -ENOMEM;
1613 
1614 	/* Make reservation before allocating sequence numbers */
1615 	err = make_reservation(c, BASEHD, aligned_len);
1616 	if (err)
1617 		goto out_free;
1618 
1619 	pack_inode(c, ino, host, 0);
1620 	pack_inode(c, (void *)ino + aligned_len1, inode, 1);
1621 
1622 	err = write_head(c, BASEHD, ino, aligned_len, &lnum, &offs, 0);
1623 	if (!sync && !err) {
1624 		struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf;
1625 
1626 		ubifs_wbuf_add_ino_nolock(wbuf, host->i_ino);
1627 		ubifs_wbuf_add_ino_nolock(wbuf, inode->i_ino);
1628 	}
1629 	release_head(c, BASEHD);
1630 	if (err)
1631 		goto out_ro;
1632 
1633 	ino_key_init(c, &key, host->i_ino);
1634 	err = ubifs_tnc_add(c, &key, lnum, offs, len1);
1635 	if (err)
1636 		goto out_ro;
1637 
1638 	ino_key_init(c, &key, inode->i_ino);
1639 	err = ubifs_tnc_add(c, &key, lnum, offs + aligned_len1, len2);
1640 	if (err)
1641 		goto out_ro;
1642 
1643 	finish_reservation(c);
1644 	spin_lock(&host_ui->ui_lock);
1645 	host_ui->synced_i_size = host_ui->ui_size;
1646 	spin_unlock(&host_ui->ui_lock);
1647 	mark_inode_clean(c, host_ui);
1648 	kfree(ino);
1649 	return 0;
1650 
1651 out_ro:
1652 	ubifs_ro_mode(c, err);
1653 	finish_reservation(c);
1654 out_free:
1655 	kfree(ino);
1656 	return err;
1657 }
1658 
1659