journal.c

   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * This file is part of UBIFS.
   4 *
   5 * Copyright (C) 2006-2008 Nokia Corporation.
   6 *
   7 * Authors: Artem Bityutskiy (Битюцкий Артём)
   8 *          Adrian Hunter
   9 */
  10
  11/*
  12 * This file implements UBIFS journal.
  13 *
  14 * The journal consists of 2 parts - the log and bud LEBs. The log has fixed
  15 * length and position, while a bud logical eraseblock is any LEB in the main
  16 * area. Buds contain file system data - data nodes, inode nodes, etc. The log
  17 * contains only references to buds and some other stuff like commit
  18 * start node. The idea is that when we commit the journal, we do
  19 * not copy the data, the buds just become indexed. Since after the commit the
  20 * nodes in bud eraseblocks become leaf nodes of the file system index tree, we
  21 * use term "bud". Analogy is obvious, bud eraseblocks contain nodes which will
  22 * become leafs in the future.
  23 *
  24 * The journal is multi-headed because we want to write data to the journal as
  25 * optimally as possible. It is nice to have nodes belonging to the same inode
  26 * in one LEB, so we may write data owned by different inodes to different
  27 * journal heads, although at present only one data head is used.
  28 *
  29 * For recovery reasons, the base head contains all inode nodes, all directory
  30 * entry nodes and all truncate nodes. This means that the other heads contain
  31 * only data nodes.
  32 *
  33 * Bud LEBs may be half-indexed. For example, if the bud was not full at the
  34 * time of commit, the bud is retained to continue to be used in the journal,
  35 * even though the "front" of the LEB is now indexed. In that case, the log
  36 * reference contains the offset where the bud starts for the purposes of the
  37 * journal.
  38 *
  39 * The journal size has to be limited, because the larger is the journal, the
  40 * longer it takes to mount UBIFS (scanning the journal) and the more memory it
  41 * takes (indexing in the TNC).
  42 *
  43 * All the journal write operations like 'ubifs_jnl_update()' here, which write
  44 * multiple UBIFS nodes to the journal at one go, are atomic with respect to
  45 * unclean reboots. Should the unclean reboot happen, the recovery code drops
  46 * all the nodes.
  47 */
  48
  49#include "ubifs.h"
  50
  51/**
  52 * zero_ino_node_unused - zero out unused fields of an on-flash inode node.
  53 * @ino: the inode to zero out
  54 */
  55static inline void zero_ino_node_unused(struct ubifs_ino_node *ino)
  56{
  57	memset(ino->padding1, 0, 4);
  58	memset(ino->padding2, 0, 26);
  59}
  60
  61/**
  62 * zero_dent_node_unused - zero out unused fields of an on-flash directory
  63 *                         entry node.
  64 * @dent: the directory entry to zero out
  65 */
  66static inline void zero_dent_node_unused(struct ubifs_dent_node *dent)
  67{
  68	dent->padding1 = 0;
  69}
  70
  71/**
  72 * zero_trun_node_unused - zero out unused fields of an on-flash truncation
  73 *                         node.
  74 * @trun: the truncation node to zero out
  75 */
  76static inline void zero_trun_node_unused(struct ubifs_trun_node *trun)
  77{
  78	memset(trun->padding, 0, 12);
  79}
  80
  81static void ubifs_add_auth_dirt(struct ubifs_info *c, int lnum)
  82{
  83	if (ubifs_authenticated(c))
  84		ubifs_add_dirt(c, lnum, ubifs_auth_node_sz(c));
  85}
  86
  87/**
  88 * reserve_space - reserve space in the journal.
  89 * @c: UBIFS file-system description object
  90 * @jhead: journal head number
  91 * @len: node length
  92 *
  93 * This function reserves space in journal head @head. If the reservation
  94 * succeeded, the journal head stays locked and later has to be unlocked using
  95 * 'release_head()'. Returns zero in case of success, %-EAGAIN if commit has to
  96 * be done, and other negative error codes in case of other failures.
  97 */
  98static int reserve_space(struct ubifs_info *c, int jhead, int len)
  99{
 100	int err = 0, err1, retries = 0, avail, lnum, offs, squeeze;
 101	struct ubifs_wbuf *wbuf = &c->jheads[jhead].wbuf;
 102
 103	/*
 104	 * Typically, the base head has smaller nodes written to it, so it is
 105	 * better to try to allocate space at the ends of eraseblocks. This is
 106	 * what the squeeze parameter does.
 107	 */
 108	ubifs_assert(c, !c->ro_media && !c->ro_mount);
 109	squeeze = (jhead == BASEHD);
 110again:
 111	mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
 112
 113	if (c->ro_error) {
 114		err = -EROFS;
 115		goto out_unlock;
 116	}
 117
 118	avail = c->leb_size - wbuf->offs - wbuf->used;
 119	if (wbuf->lnum != -1 && avail >= len)
 120		return 0;
 121
 122	/*
 123	 * Write buffer wasn't seek'ed or there is no enough space - look for an
 124	 * LEB with some empty space.
 125	 */
 126	lnum = ubifs_find_free_space(c, len, &offs, squeeze);
 127	if (lnum >= 0)
 128		goto out;
 129
 130	err = lnum;
 131	if (err != -ENOSPC)
 132		goto out_unlock;
 133
 134	/*
 135	 * No free space, we have to run garbage collector to make
 136	 * some. But the write-buffer mutex has to be unlocked because
 137	 * GC also takes it.
 138	 */
 139	dbg_jnl("no free space in jhead %s, run GC", dbg_jhead(jhead));
 140	mutex_unlock(&wbuf->io_mutex);
 141
 142	lnum = ubifs_garbage_collect(c, 0);
 143	if (lnum < 0) {
 144		err = lnum;
 145		if (err != -ENOSPC)
 146			return err;
 147
 148		/*
 149		 * GC could not make a free LEB. But someone else may
 150		 * have allocated new bud for this journal head,
 151		 * because we dropped @wbuf->io_mutex, so try once
 152		 * again.
 153		 */
 154		dbg_jnl("GC couldn't make a free LEB for jhead %s",
 155			dbg_jhead(jhead));
 156		if (retries++ < 2) {
 157			dbg_jnl("retry (%d)", retries);
 158			goto again;
 159		}
 160
 161		dbg_jnl("return -ENOSPC");
 162		return err;
 163	}
 164
 165	mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
 166	dbg_jnl("got LEB %d for jhead %s", lnum, dbg_jhead(jhead));
 167	avail = c->leb_size - wbuf->offs - wbuf->used;
 168
 169	if (wbuf->lnum != -1 && avail >= len) {
 170		/*
 171		 * Someone else has switched the journal head and we have
 172		 * enough space now. This happens when more than one process is
 173		 * trying to write to the same journal head at the same time.
 174		 */
 175		dbg_jnl("return LEB %d back, already have LEB %d:%d",
 176			lnum, wbuf->lnum, wbuf->offs + wbuf->used);
 177		err = ubifs_return_leb(c, lnum);
 178		if (err)
 179			goto out_unlock;
 180		return 0;
 181	}
 182
 183	offs = 0;
 184
 185out:
 186	/*
 187	 * Make sure we synchronize the write-buffer before we add the new bud
 188	 * to the log. Otherwise we may have a power cut after the log
 189	 * reference node for the last bud (@lnum) is written but before the
 190	 * write-buffer data are written to the next-to-last bud
 191	 * (@wbuf->lnum). And the effect would be that the recovery would see
 192	 * that there is corruption in the next-to-last bud.
 193	 */
 194	err = ubifs_wbuf_sync_nolock(wbuf);
 195	if (err)
 196		goto out_return;
 197	err = ubifs_add_bud_to_log(c, jhead, lnum, offs);
 198	if (err)
 199		goto out_return;
 200	err = ubifs_wbuf_seek_nolock(wbuf, lnum, offs);
 201	if (err)
 202		goto out_unlock;
 203
 204	return 0;
 205
 206out_unlock:
 207	mutex_unlock(&wbuf->io_mutex);
 208	return err;
 209
 210out_return:
 211	/* An error occurred and the LEB has to be returned to lprops */
 212	ubifs_assert(c, err < 0);
 213	err1 = ubifs_return_leb(c, lnum);
 214	if (err1 && err == -EAGAIN)
 215		/*
 216		 * Return original error code only if it is not %-EAGAIN,
 217		 * which is not really an error. Otherwise, return the error
 218		 * code of 'ubifs_return_leb()'.
 219		 */
 220		err = err1;
 221	mutex_unlock(&wbuf->io_mutex);
 222	return err;
 223}
 224
 225static int ubifs_hash_nodes(struct ubifs_info *c, void *node,
 226			     int len, struct shash_desc *hash)
 227{
 228	int auth_node_size = ubifs_auth_node_sz(c);
 229	int err;
 230
 231	while (1) {
 232		const struct ubifs_ch *ch = node;
 233		int nodelen = le32_to_cpu(ch->len);
 234
 235		ubifs_assert(c, len >= auth_node_size);
 236
 237		if (len == auth_node_size)
 238			break;
 239
 240		ubifs_assert(c, len > nodelen);
 241		ubifs_assert(c, ch->magic == cpu_to_le32(UBIFS_NODE_MAGIC));
 242
 243		err = ubifs_shash_update(c, hash, (void *)node, nodelen);
 244		if (err)
 245			return err;
 246
 247		node += ALIGN(nodelen, 8);
 248		len -= ALIGN(nodelen, 8);
 249	}
 250
 251	return ubifs_prepare_auth_node(c, node, hash);
 252}
 253
 254/**
 255 * write_head - write data to a journal head.
 256 * @c: UBIFS file-system description object
 257 * @jhead: journal head
 258 * @buf: buffer to write
 259 * @len: length to write
 260 * @lnum: LEB number written is returned here
 261 * @offs: offset written is returned here
 262 * @sync: non-zero if the write-buffer has to by synchronized
 263 *
 264 * This function writes data to the reserved space of journal head @jhead.
 265 * Returns zero in case of success and a negative error code in case of
 266 * failure.
 267 */
 268static int write_head(struct ubifs_info *c, int jhead, void *buf, int len,
 269		      int *lnum, int *offs, int sync)
 270{
 271	int err;
 272	struct ubifs_wbuf *wbuf = &c->jheads[jhead].wbuf;
 273
 274	ubifs_assert(c, jhead != GCHD);
 275
 276	*lnum = c->jheads[jhead].wbuf.lnum;
 277	*offs = c->jheads[jhead].wbuf.offs + c->jheads[jhead].wbuf.used;
 278	dbg_jnl("jhead %s, LEB %d:%d, len %d",
 279		dbg_jhead(jhead), *lnum, *offs, len);
 280
 281	if (ubifs_authenticated(c)) {
 282		err = ubifs_hash_nodes(c, buf, len, c->jheads[jhead].log_hash);
 283		if (err)
 284			return err;
 285	}
 286
 287	err = ubifs_wbuf_write_nolock(wbuf, buf, len);
 288	if (err)
 289		return err;
 290	if (sync)
 291		err = ubifs_wbuf_sync_nolock(wbuf);
 292	return err;
 293}
 294
 295/**
 296 * __queue_and_wait - queue a task and wait until the task is waked up.
 297 * @c: UBIFS file-system description object
 298 *
 299 * This function adds current task in queue and waits until the task is waked
 300 * up. This function should be called with @c->reserve_space_wq locked.
 301 */
 302static void __queue_and_wait(struct ubifs_info *c)
 303{
 304	DEFINE_WAIT(wait);
 305
 306	__add_wait_queue_entry_tail_exclusive(&c->reserve_space_wq, &wait);
 307	set_current_state(TASK_UNINTERRUPTIBLE);
 308	spin_unlock(&c->reserve_space_wq.lock);
 309
 310	schedule();
 311	finish_wait(&c->reserve_space_wq, &wait);
 312}
 313
 314/**
 315 * wait_for_reservation - try queuing current task to wait until waked up.
 316 * @c: UBIFS file-system description object
 317 *
 318 * This function queues current task to wait until waked up, if queuing is
 319 * started(@c->need_wait_space is not %0). Returns %true if current task is
 320 * added in queue, otherwise %false is returned.
 321 */
 322static bool wait_for_reservation(struct ubifs_info *c)
 323{
 324	if (likely(atomic_read(&c->need_wait_space) == 0))
 325		/* Quick path to check whether queuing is started. */
 326		return false;
 327
 328	spin_lock(&c->reserve_space_wq.lock);
 329	if (atomic_read(&c->need_wait_space) == 0) {
 330		/* Queuing is not started, don't queue current task. */
 331		spin_unlock(&c->reserve_space_wq.lock);
 332		return false;
 333	}
 334
 335	__queue_and_wait(c);
 336	return true;
 337}
 338
 339/**
 340 * wake_up_reservation - wake up first task in queue or stop queuing.
 341 * @c: UBIFS file-system description object
 342 *
 343 * This function wakes up the first task in queue if it exists, or stops
 344 * queuing if no tasks in queue.
 345 */
 346static void wake_up_reservation(struct ubifs_info *c)
 347{
 348	spin_lock(&c->reserve_space_wq.lock);
 349	if (waitqueue_active(&c->reserve_space_wq))
 350		wake_up_locked(&c->reserve_space_wq);
 351	else
 352		/*
 353		 * Compared with wait_for_reservation(), set @c->need_wait_space
 354		 * under the protection of wait queue lock, which can avoid that
 355		 * @c->need_wait_space is set to 0 after new task queued.
 356		 */
 357		atomic_set(&c->need_wait_space, 0);
 358	spin_unlock(&c->reserve_space_wq.lock);
 359}
 360
 361/**
 362 * add_or_start_queue - add current task in queue or start queuing.
 363 * @c: UBIFS file-system description object
 364 *
 365 * This function starts queuing if queuing is not started, otherwise adds
 366 * current task in queue.
 367 */
 368static void add_or_start_queue(struct ubifs_info *c)
 369{
 370	spin_lock(&c->reserve_space_wq.lock);
 371	if (atomic_cmpxchg(&c->need_wait_space, 0, 1) == 0) {
 372		/* Starts queuing, task can go on directly. */
 373		spin_unlock(&c->reserve_space_wq.lock);
 374		return;
 375	}
 376
 377	/*
 378	 * There are at least two tasks have retried more than 32 times
 379	 * at certain point, first task has started queuing, just queue
 380	 * the left tasks.
 381	 */
 382	__queue_and_wait(c);
 383}
 384
 385/**
 386 * make_reservation - reserve journal space.
 387 * @c: UBIFS file-system description object
 388 * @jhead: journal head
 389 * @len: how many bytes to reserve
 390 *
 391 * This function makes space reservation in journal head @jhead. The function
 392 * takes the commit lock and locks the journal head, and the caller has to
 393 * unlock the head and finish the reservation with 'finish_reservation()'.
 394 * Returns zero in case of success and a negative error code in case of
 395 * failure.
 396 *
 397 * Note, the journal head may be unlocked as soon as the data is written, while
 398 * the commit lock has to be released after the data has been added to the
 399 * TNC.
 400 */
 401static int make_reservation(struct ubifs_info *c, int jhead, int len)
 402{
 403	int err, cmt_retries = 0, nospc_retries = 0;
 404	bool blocked = wait_for_reservation(c);
 405
 406again:
 407	down_read(&c->commit_sem);
 408	err = reserve_space(c, jhead, len);
 409	if (!err) {
 410		/* c->commit_sem will get released via finish_reservation(). */
 411		goto out_wake_up;
 412	}
 413	up_read(&c->commit_sem);
 414
 415	if (err == -ENOSPC) {
 416		/*
 417		 * GC could not make any progress. We should try to commit
 418		 * because it could make some dirty space and GC would make
 419		 * progress, so make the error -EAGAIN so that the below
 420		 * will commit and re-try.
 421		 */
 422		nospc_retries++;
 423		dbg_jnl("no space, retry");
 424		err = -EAGAIN;
 425	}
 426
 427	if (err != -EAGAIN)
 428		goto out;
 429
 430	/*
 431	 * -EAGAIN means that the journal is full or too large, or the above
 432	 * code wants to do one commit. Do this and re-try.
 433	 */
 434	if (cmt_retries > 128) {
 435		/*
 436		 * This should not happen unless:
 437		 * 1. The journal size limitations are too tough.
 438		 * 2. The budgeting is incorrect. We always have to be able to
 439		 *    write to the media, because all operations are budgeted.
 440		 *    Deletions are not budgeted, though, but we reserve an
 441		 *    extra LEB for them.
 442		 */
 443		ubifs_err(c, "stuck in space allocation, nospc_retries %d",
 444			  nospc_retries);
 445		err = -ENOSPC;
 446		goto out;
 447	} else if (cmt_retries > 32) {
 448		/*
 449		 * It's almost impossible to happen, unless there are many tasks
 450		 * making reservation concurrently and someone task has retried
 451		 * gc + commit for many times, generated available space during
 452		 * this period are grabbed by other tasks.
 453		 * But if it happens, start queuing up all tasks that will make
 454		 * space reservation, then there is only one task making space
 455		 * reservation at any time, and it can always make success under
 456		 * the premise of correct budgeting.
 457		 */
 458		ubifs_warn(c, "too many space allocation cmt_retries (%d) "
 459			   "nospc_retries (%d), start queuing tasks",
 460			   cmt_retries, nospc_retries);
 461
 462		if (!blocked) {
 463			blocked = true;
 464			add_or_start_queue(c);
 465		}
 466	}
 467
 468	dbg_jnl("-EAGAIN, commit and retry (retried %d times)",
 469		cmt_retries);
 470	cmt_retries += 1;
 471
 472	err = ubifs_run_commit(c);
 473	if (err)
 474		goto out_wake_up;
 475	goto again;
 476
 477out:
 478	ubifs_err(c, "cannot reserve %d bytes in jhead %d, error %d",
 479		  len, jhead, err);
 480	if (err == -ENOSPC) {
 481		/* This are some budgeting problems, print useful information */
 482		down_write(&c->commit_sem);
 483		dump_stack();
 484		ubifs_dump_budg(c, &c->bi);
 485		ubifs_dump_lprops(c);
 486		cmt_retries = dbg_check_lprops(c);
 487		up_write(&c->commit_sem);
 488	}
 489out_wake_up:
 490	if (blocked) {
 491		/*
 492		 * Only tasks that have ever started queuing or ever been queued
 493		 * can wake up other queued tasks, which can make sure that
 494		 * there is only one task waked up to make space reservation.
 495		 * For example:
 496		 *      task A          task B           task C
 497		 *                 make_reservation  make_reservation
 498		 * reserve_space // 0
 499		 * wake_up_reservation
 500		 *                  atomic_cmpxchg // 0, start queuing
 501		 *                  reserve_space
 502		 *                                    wait_for_reservation
 503		 *                                     __queue_and_wait
 504		 *                                      add_wait_queue
 505		 *  if (blocked) // false
 506		 *  // So that task C won't be waked up to race with task B
 507		 */
 508		wake_up_reservation(c);
 509	}
 510	return err;
 511}
 512
 513/**
 514 * release_head - release a journal head.
 515 * @c: UBIFS file-system description object
 516 * @jhead: journal head
 517 *
 518 * This function releases journal head @jhead which was locked by
 519 * the 'make_reservation()' function. It has to be called after each successful
 520 * 'make_reservation()' invocation.
 521 */
 522static inline void release_head(struct ubifs_info *c, int jhead)
 523{
 524	mutex_unlock(&c->jheads[jhead].wbuf.io_mutex);
 525}
 526
 527/**
 528 * finish_reservation - finish a reservation.
 529 * @c: UBIFS file-system description object
 530 *
 531 * This function finishes journal space reservation. It must be called after
 532 * 'make_reservation()'.
 533 */
 534static void finish_reservation(struct ubifs_info *c)
 535{
 536	up_read(&c->commit_sem);
 537}
 538
 539/**
 540 * get_dent_type - translate VFS inode mode to UBIFS directory entry type.
 541 * @mode: inode mode
 542 */
 543static int get_dent_type(int mode)
 544{
 545	switch (mode & S_IFMT) {
 546	case S_IFREG:
 547		return UBIFS_ITYPE_REG;
 548	case S_IFDIR:
 549		return UBIFS_ITYPE_DIR;
 550	case S_IFLNK:
 551		return UBIFS_ITYPE_LNK;
 552	case S_IFBLK:
 553		return UBIFS_ITYPE_BLK;
 554	case S_IFCHR:
 555		return UBIFS_ITYPE_CHR;
 556	case S_IFIFO:
 557		return UBIFS_ITYPE_FIFO;
 558	case S_IFSOCK:
 559		return UBIFS_ITYPE_SOCK;
 560	default:
 561		BUG();
 562	}
 563	return 0;
 564}
 565
 566/**
 567 * pack_inode - pack an inode node.
 568 * @c: UBIFS file-system description object
 569 * @ino: buffer in which to pack inode node
 570 * @inode: inode to pack
 571 * @last: indicates the last node of the group
 572 */
 573static void pack_inode(struct ubifs_info *c, struct ubifs_ino_node *ino,
 574		       const struct inode *inode, int last)
 575{
 576	int data_len = 0, last_reference = !inode->i_nlink;
 577	struct ubifs_inode *ui = ubifs_inode(inode);
 578
 579	ino->ch.node_type = UBIFS_INO_NODE;
 580	ino_key_init_flash(c, &ino->key, inode->i_ino);
 581	ino->creat_sqnum = cpu_to_le64(ui->creat_sqnum);
 582	ino->atime_sec  = cpu_to_le64(inode_get_atime_sec(inode));
 583	ino->atime_nsec = cpu_to_le32(inode_get_atime_nsec(inode));
 584	ino->ctime_sec  = cpu_to_le64(inode_get_ctime_sec(inode));
 585	ino->ctime_nsec = cpu_to_le32(inode_get_ctime_nsec(inode));
 586	ino->mtime_sec  = cpu_to_le64(inode_get_mtime_sec(inode));
 587	ino->mtime_nsec = cpu_to_le32(inode_get_mtime_nsec(inode));
 588	ino->uid   = cpu_to_le32(i_uid_read(inode));
 589	ino->gid   = cpu_to_le32(i_gid_read(inode));
 590	ino->mode  = cpu_to_le32(inode->i_mode);
 591	ino->flags = cpu_to_le32(ui->flags);
 592	ino->size  = cpu_to_le64(ui->ui_size);
 593	ino->nlink = cpu_to_le32(inode->i_nlink);
 594	ino->compr_type  = cpu_to_le16(ui->compr_type);
 595	ino->data_len    = cpu_to_le32(ui->data_len);
 596	ino->xattr_cnt   = cpu_to_le32(ui->xattr_cnt);
 597	ino->xattr_size  = cpu_to_le32(ui->xattr_size);
 598	ino->xattr_names = cpu_to_le32(ui->xattr_names);
 599	zero_ino_node_unused(ino);
 600
 601	/*
 602	 * Drop the attached data if this is a deletion inode, the data is not
 603	 * needed anymore.
 604	 */
 605	if (!last_reference) {
 606		memcpy(ino->data, ui->data, ui->data_len);
 607		data_len = ui->data_len;
 608	}
 609
 610	ubifs_prep_grp_node(c, ino, UBIFS_INO_NODE_SZ + data_len, last);
 611}
 612
 613/**
 614 * mark_inode_clean - mark UBIFS inode as clean.
 615 * @c: UBIFS file-system description object
 616 * @ui: UBIFS inode to mark as clean
 617 *
 618 * This helper function marks UBIFS inode @ui as clean by cleaning the
 619 * @ui->dirty flag and releasing its budget. Note, VFS may still treat the
 620 * inode as dirty and try to write it back, but 'ubifs_write_inode()' would
 621 * just do nothing.
 622 */
 623static void mark_inode_clean(struct ubifs_info *c, struct ubifs_inode *ui)
 624{
 625	if (ui->dirty)
 626		ubifs_release_dirty_inode_budget(c, ui);
 627	ui->dirty = 0;
 628}
 629
 630static void set_dent_cookie(struct ubifs_info *c, struct ubifs_dent_node *dent)
 631{
 632	if (c->double_hash)
 633		dent->cookie = (__force __le32) get_random_u32();
 634	else
 635		dent->cookie = 0;
 636}
 637
 638/**
 639 * ubifs_jnl_update - update inode.
 640 * @c: UBIFS file-system description object
 641 * @dir: parent inode or host inode in case of extended attributes
 642 * @nm: directory entry name
 643 * @inode: inode to update
 644 * @deletion: indicates a directory entry deletion i.e unlink or rmdir
 645 * @xent: non-zero if the directory entry is an extended attribute entry
 646 * @in_orphan: indicates whether the @inode is in orphan list
 647 *
 648 * This function updates an inode by writing a directory entry (or extended
 649 * attribute entry), the inode itself, and the parent directory inode (or the
 650 * host inode) to the journal.
 651 *
 652 * The function writes the host inode @dir last, which is important in case of
 653 * extended attributes. Indeed, then we guarantee that if the host inode gets
 654 * synchronized (with 'fsync()'), and the write-buffer it sits in gets flushed,
 655 * the extended attribute inode gets flushed too. And this is exactly what the
 656 * user expects - synchronizing the host inode synchronizes its extended
 657 * attributes. Similarly, this guarantees that if @dir is synchronized, its
 658 * directory entry corresponding to @nm gets synchronized too.
 659 *
 660 * If the inode (@inode) or the parent directory (@dir) are synchronous, this
 661 * function synchronizes the write-buffer.
 662 *
 663 * This function marks the @dir and @inode inodes as clean and returns zero on
 664 * success. In case of failure, a negative error code is returned.
 665 */
 666int ubifs_jnl_update(struct ubifs_info *c, const struct inode *dir,
 667		     const struct fscrypt_name *nm, const struct inode *inode,
 668		     int deletion, int xent, int in_orphan)
 669{
 670	int err, dlen, ilen, len, lnum, ino_offs, dent_offs, orphan_added = 0;
 671	int aligned_dlen, aligned_ilen, sync = IS_DIRSYNC(dir);
 672	int last_reference = !!(deletion && inode->i_nlink == 0);
 673	struct ubifs_inode *ui = ubifs_inode(inode);
 674	struct ubifs_inode *host_ui = ubifs_inode(dir);
 675	struct ubifs_dent_node *dent;
 676	struct ubifs_ino_node *ino;
 677	union ubifs_key dent_key, ino_key;
 678	u8 hash_dent[UBIFS_HASH_ARR_SZ];
 679	u8 hash_ino[UBIFS_HASH_ARR_SZ];
 680	u8 hash_ino_host[UBIFS_HASH_ARR_SZ];
 681
 682	ubifs_assert(c, mutex_is_locked(&host_ui->ui_mutex));
 683
 684	dlen = UBIFS_DENT_NODE_SZ + fname_len(nm) + 1;
 685	ilen = UBIFS_INO_NODE_SZ;
 686
 687	/*
 688	 * If the last reference to the inode is being deleted, then there is
 689	 * no need to attach and write inode data, it is being deleted anyway.
 690	 * And if the inode is being deleted, no need to synchronize
 691	 * write-buffer even if the inode is synchronous.
 692	 */
 693	if (!last_reference) {
 694		ilen += ui->data_len;
 695		sync |= IS_SYNC(inode);
 696	}
 697
 698	aligned_dlen = ALIGN(dlen, 8);
 699	aligned_ilen = ALIGN(ilen, 8);
 700
 701	len = aligned_dlen + aligned_ilen + UBIFS_INO_NODE_SZ;
 702	/* Make sure to also account for extended attributes */
 703	if (ubifs_authenticated(c))
 704		len += ALIGN(host_ui->data_len, 8) + ubifs_auth_node_sz(c);
 705	else
 706		len += host_ui->data_len;
 707
 708	dent = kzalloc(len, GFP_NOFS);
 709	if (!dent)
 710		return -ENOMEM;
 711
 712	/* Make reservation before allocating sequence numbers */
 713	err = make_reservation(c, BASEHD, len);
 714	if (err)
 715		goto out_free;
 716
 717	if (!xent) {
 718		dent->ch.node_type = UBIFS_DENT_NODE;
 719		if (fname_name(nm) == NULL)
 720			dent_key_init_hash(c, &dent_key, dir->i_ino, nm->hash);
 721		else
 722			dent_key_init(c, &dent_key, dir->i_ino, nm);
 723	} else {
 724		dent->ch.node_type = UBIFS_XENT_NODE;
 725		xent_key_init(c, &dent_key, dir->i_ino, nm);
 726	}
 727
 728	key_write(c, &dent_key, dent->key);
 729	dent->inum = deletion ? 0 : cpu_to_le64(inode->i_ino);
 730	dent->type = get_dent_type(inode->i_mode);
 731	dent->nlen = cpu_to_le16(fname_len(nm));
 732	memcpy(dent->name, fname_name(nm), fname_len(nm));
 733	dent->name[fname_len(nm)] = '\0';
 734	set_dent_cookie(c, dent);
 735
 736	zero_dent_node_unused(dent);
 737	ubifs_prep_grp_node(c, dent, dlen, 0);
 738	err = ubifs_node_calc_hash(c, dent, hash_dent);
 739	if (err)
 740		goto out_release;
 741
 742	ino = (void *)dent + aligned_dlen;
 743	pack_inode(c, ino, inode, 0);
 744	err = ubifs_node_calc_hash(c, ino, hash_ino);
 745	if (err)
 746		goto out_release;
 747
 748	ino = (void *)ino + aligned_ilen;
 749	pack_inode(c, ino, dir, 1);
 750	err = ubifs_node_calc_hash(c, ino, hash_ino_host);
 751	if (err)
 752		goto out_release;
 753
 754	if (last_reference && !in_orphan) {
 755		err = ubifs_add_orphan(c, inode->i_ino);
 756		if (err) {
 757			release_head(c, BASEHD);
 758			goto out_finish;
 759		}
 760		ui->del_cmtno = c->cmt_no;
 761		orphan_added = 1;
 762	}
 763
 764	err = write_head(c, BASEHD, dent, len, &lnum, &dent_offs, sync);
 765	if (err)
 766		goto out_release;
 767	if (!sync) {
 768		struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf;
 769
 770		ubifs_wbuf_add_ino_nolock(wbuf, inode->i_ino);
 771		ubifs_wbuf_add_ino_nolock(wbuf, dir->i_ino);
 772	}
 773	release_head(c, BASEHD);
 774	kfree(dent);
 775	ubifs_add_auth_dirt(c, lnum);
 776
 777	if (deletion) {
 778		if (fname_name(nm) == NULL)
 779			err = ubifs_tnc_remove_dh(c, &dent_key, nm->minor_hash);
 780		else
 781			err = ubifs_tnc_remove_nm(c, &dent_key, nm);
 782		if (err)
 783			goto out_ro;
 784		err = ubifs_add_dirt(c, lnum, dlen);
 785	} else
 786		err = ubifs_tnc_add_nm(c, &dent_key, lnum, dent_offs, dlen,
 787				       hash_dent, nm);
 788	if (err)
 789		goto out_ro;
 790
 791	/*
 792	 * Note, we do not remove the inode from TNC even if the last reference
 793	 * to it has just been deleted, because the inode may still be opened.
 794	 * Instead, the inode has been added to orphan lists and the orphan
 795	 * subsystem will take further care about it.
 796	 */
 797	ino_key_init(c, &ino_key, inode->i_ino);
 798	ino_offs = dent_offs + aligned_dlen;
 799	err = ubifs_tnc_add(c, &ino_key, lnum, ino_offs, ilen, hash_ino);
 800	if (err)
 801		goto out_ro;
 802
 803	ino_key_init(c, &ino_key, dir->i_ino);
 804	ino_offs += aligned_ilen;
 805	err = ubifs_tnc_add(c, &ino_key, lnum, ino_offs,
 806			    UBIFS_INO_NODE_SZ + host_ui->data_len, hash_ino_host);
 807	if (err)
 808		goto out_ro;
 809
 810	if (in_orphan && inode->i_nlink)
 811		ubifs_delete_orphan(c, inode->i_ino);
 812
 813	finish_reservation(c);
 814	spin_lock(&ui->ui_lock);
 815	ui->synced_i_size = ui->ui_size;
 816	spin_unlock(&ui->ui_lock);
 817	if (xent) {
 818		spin_lock(&host_ui->ui_lock);
 819		host_ui->synced_i_size = host_ui->ui_size;
 820		spin_unlock(&host_ui->ui_lock);
 821	}
 822	mark_inode_clean(c, ui);
 823	mark_inode_clean(c, host_ui);
 824	return 0;
 825
 826out_finish:
 827	finish_reservation(c);
 828out_free:
 829	kfree(dent);
 830	return err;
 831
 832out_release:
 833	release_head(c, BASEHD);
 834	kfree(dent);
 835out_ro:
 836	ubifs_ro_mode(c, err);
 837	if (orphan_added)
 838		ubifs_delete_orphan(c, inode->i_ino);
 839	finish_reservation(c);
 840	return err;
 841}
 842
 843/**
 844 * ubifs_jnl_write_data - write a data node to the journal.
 845 * @c: UBIFS file-system description object
 846 * @inode: inode the data node belongs to
 847 * @key: node key
 848 * @folio: buffer to write
 849 * @offset: offset to write at
 850 * @len: data length (must not exceed %UBIFS_BLOCK_SIZE)
 851 *
 852 * This function writes a data node to the journal. Returns %0 if the data node
 853 * was successfully written, and a negative error code in case of failure.
 854 */
 855int ubifs_jnl_write_data(struct ubifs_info *c, const struct inode *inode,
 856			 const union ubifs_key *key, struct folio *folio,
 857			 size_t offset, int len)
 858{
 859	struct ubifs_data_node *data;
 860	int err, lnum, offs, compr_type, out_len, compr_len, auth_len;
 861	int dlen = COMPRESSED_DATA_NODE_BUF_SZ, allocated = 1;
 862	int write_len;
 863	struct ubifs_inode *ui = ubifs_inode(inode);
 864	bool encrypted = IS_ENCRYPTED(inode);
 865	u8 hash[UBIFS_HASH_ARR_SZ];
 866
 867	dbg_jnlk(key, "ino %lu, blk %u, len %d, key ",
 868		(unsigned long)key_inum(c, key), key_block(c, key), len);
 869	ubifs_assert(c, len <= UBIFS_BLOCK_SIZE);
 870
 871	if (encrypted)
 872		dlen += UBIFS_CIPHER_BLOCK_SIZE;
 873
 874	auth_len = ubifs_auth_node_sz(c);
 875
 876	data = kmalloc(dlen + auth_len, GFP_NOFS | __GFP_NOWARN);
 877	if (!data) {
 878		/*
 879		 * Fall-back to the write reserve buffer. Note, we might be
 880		 * currently on the memory reclaim path, when the kernel is
 881		 * trying to free some memory by writing out dirty pages. The
 882		 * write reserve buffer helps us to guarantee that we are
 883		 * always able to write the data.
 884		 */
 885		allocated = 0;
 886		mutex_lock(&c->write_reserve_mutex);
 887		data = c->write_reserve_buf;
 888	}
 889
 890	data->ch.node_type = UBIFS_DATA_NODE;
 891	key_write(c, key, &data->key);
 892	data->size = cpu_to_le32(len);
 893
 894	if (!(ui->flags & UBIFS_COMPR_FL))
 895		/* Compression is disabled for this inode */
 896		compr_type = UBIFS_COMPR_NONE;
 897	else
 898		compr_type = ui->compr_type;
 899
 900	out_len = compr_len = dlen - UBIFS_DATA_NODE_SZ;
 901	ubifs_compress_folio(c, folio, offset, len, &data->data, &compr_len,
 902			     &compr_type);
 903	ubifs_assert(c, compr_len <= UBIFS_BLOCK_SIZE);
 904
 905	if (encrypted) {
 906		err = ubifs_encrypt(inode, data, compr_len, &out_len, key_block(c, key));
 907		if (err)
 908			goto out_free;
 909
 910	} else {
 911		data->compr_size = 0;
 912		out_len = compr_len;
 913	}
 914
 915	dlen = UBIFS_DATA_NODE_SZ + out_len;
 916	if (ubifs_authenticated(c))
 917		write_len = ALIGN(dlen, 8) + auth_len;
 918	else
 919		write_len = dlen;
 920
 921	data->compr_type = cpu_to_le16(compr_type);
 922
 923	/* Make reservation before allocating sequence numbers */
 924	err = make_reservation(c, DATAHD, write_len);
 925	if (err)
 926		goto out_free;
 927
 928	ubifs_prepare_node(c, data, dlen, 0);
 929	err = write_head(c, DATAHD, data, write_len, &lnum, &offs, 0);
 930	if (err)
 931		goto out_release;
 932
 933	err = ubifs_node_calc_hash(c, data, hash);
 934	if (err)
 935		goto out_release;
 936
 937	ubifs_wbuf_add_ino_nolock(&c->jheads[DATAHD].wbuf, key_inum(c, key));
 938	release_head(c, DATAHD);
 939
 940	ubifs_add_auth_dirt(c, lnum);
 941
 942	err = ubifs_tnc_add(c, key, lnum, offs, dlen, hash);
 943	if (err)
 944		goto out_ro;
 945
 946	finish_reservation(c);
 947	if (!allocated)
 948		mutex_unlock(&c->write_reserve_mutex);
 949	else
 950		kfree(data);
 951	return 0;
 952
 953out_release:
 954	release_head(c, DATAHD);
 955out_ro:
 956	ubifs_ro_mode(c, err);
 957	finish_reservation(c);
 958out_free:
 959	if (!allocated)
 960		mutex_unlock(&c->write_reserve_mutex);
 961	else
 962		kfree(data);
 963	return err;
 964}
 965
 966/**
 967 * ubifs_jnl_write_inode - flush inode to the journal.
 968 * @c: UBIFS file-system description object
 969 * @inode: inode to flush
 970 *
 971 * This function writes inode @inode to the journal. If the inode is
 972 * synchronous, it also synchronizes the write-buffer. Returns zero in case of
 973 * success and a negative error code in case of failure.
 974 */
 975int ubifs_jnl_write_inode(struct ubifs_info *c, const struct inode *inode)
 976{
 977	int err, lnum, offs;
 978	struct ubifs_ino_node *ino, *ino_start;
 979	struct ubifs_inode *ui = ubifs_inode(inode);
 980	int sync = 0, write_len = 0, ilen = UBIFS_INO_NODE_SZ;
 981	int last_reference = !inode->i_nlink;
 982	int kill_xattrs = ui->xattr_cnt && last_reference;
 983	u8 hash[UBIFS_HASH_ARR_SZ];
 984
 985	dbg_jnl("ino %lu, nlink %u", inode->i_ino, inode->i_nlink);
 986
 987	if (kill_xattrs && ui->xattr_cnt > ubifs_xattr_max_cnt(c)) {
 988		ubifs_err(c, "Cannot delete inode, it has too many xattrs!");
 989		err = -EPERM;
 990		ubifs_ro_mode(c, err);
 991		return err;
 992	}
 993
 994	/*
 995	 * If the inode is being deleted, do not write the attached data. No
 996	 * need to synchronize the write-buffer either.
 997	 */
 998	if (!last_reference) {
 999		ilen += ui->data_len;
1000		sync = IS_SYNC(inode);
1001	} else if (kill_xattrs) {
1002		write_len += UBIFS_INO_NODE_SZ * ui->xattr_cnt;
1003	}
1004
1005	if (ubifs_authenticated(c))
1006		write_len += ALIGN(ilen, 8) + ubifs_auth_node_sz(c);
1007	else
1008		write_len += ilen;
1009
1010	ino_start = ino = kmalloc(write_len, GFP_NOFS);
1011	if (!ino)
1012		return -ENOMEM;
1013
1014	/* Make reservation before allocating sequence numbers */
1015	err = make_reservation(c, BASEHD, write_len);
1016	if (err)
1017		goto out_free;
1018
1019	if (kill_xattrs) {
1020		union ubifs_key key;
1021		struct fscrypt_name nm = {0};
1022		struct inode *xino;
1023		struct ubifs_dent_node *xent, *pxent = NULL;
1024
1025		lowest_xent_key(c, &key, inode->i_ino);
1026		while (1) {
1027			xent = ubifs_tnc_next_ent(c, &key, &nm);
1028			if (IS_ERR(xent)) {
1029				err = PTR_ERR(xent);
1030				if (err == -ENOENT)
1031					break;
1032
1033				kfree(pxent);
1034				goto out_release;
1035			}
1036
1037			fname_name(&nm) = xent->name;
1038			fname_len(&nm) = le16_to_cpu(xent->nlen);
1039
1040			xino = ubifs_iget(c->vfs_sb, le64_to_cpu(xent->inum));
1041			if (IS_ERR(xino)) {
1042				err = PTR_ERR(xino);
1043				ubifs_err(c, "dead directory entry '%s', error %d",
1044					  xent->name, err);
1045				ubifs_ro_mode(c, err);
1046				kfree(pxent);
1047				kfree(xent);
1048				goto out_release;
1049			}
1050			ubifs_assert(c, ubifs_inode(xino)->xattr);
1051
1052			clear_nlink(xino);
1053			pack_inode(c, ino, xino, 0);
1054			ino = (void *)ino + UBIFS_INO_NODE_SZ;
1055			iput(xino);
1056
1057			kfree(pxent);
1058			pxent = xent;
1059			key_read(c, &xent->key, &key);
1060		}
1061		kfree(pxent);
1062	}
1063
1064	pack_inode(c, ino, inode, 1);
1065	err = ubifs_node_calc_hash(c, ino, hash);
1066	if (err)
1067		goto out_release;
1068
1069	err = write_head(c, BASEHD, ino_start, write_len, &lnum, &offs, sync);
1070	if (err)
1071		goto out_release;
1072	if (!sync)
1073		ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf,
1074					  inode->i_ino);
1075	release_head(c, BASEHD);
1076
1077	if (last_reference) {
1078		err = ubifs_tnc_remove_ino(c, inode->i_ino);
1079		if (err)
1080			goto out_ro;
1081		ubifs_delete_orphan(c, inode->i_ino);
1082		err = ubifs_add_dirt(c, lnum, write_len);
1083	} else {
1084		union ubifs_key key;
1085
1086		ubifs_add_auth_dirt(c, lnum);
1087
1088		ino_key_init(c, &key, inode->i_ino);
1089		err = ubifs_tnc_add(c, &key, lnum, offs, ilen, hash);
1090	}
1091	if (err)
1092		goto out_ro;
1093
1094	finish_reservation(c);
1095	spin_lock(&ui->ui_lock);
1096	ui->synced_i_size = ui->ui_size;
1097	spin_unlock(&ui->ui_lock);
1098	kfree(ino_start);
1099	return 0;
1100
1101out_release:
1102	release_head(c, BASEHD);
1103out_ro:
1104	ubifs_ro_mode(c, err);
1105	finish_reservation(c);
1106out_free:
1107	kfree(ino_start);
1108	return err;
1109}
1110
1111/**
1112 * ubifs_jnl_delete_inode - delete an inode.
1113 * @c: UBIFS file-system description object
1114 * @inode: inode to delete
1115 *
1116 * This function deletes inode @inode which includes removing it from orphans,
1117 * deleting it from TNC and, in some cases, writing a deletion inode to the
1118 * journal.
1119 *
1120 * When regular file inodes are unlinked or a directory inode is removed, the
1121 * 'ubifs_jnl_update()' function writes a corresponding deletion inode and
1122 * direntry to the media, and adds the inode to orphans. After this, when the
1123 * last reference to this inode has been dropped, this function is called. In
1124 * general, it has to write one more deletion inode to the media, because if
1125 * a commit happened between 'ubifs_jnl_update()' and
1126 * 'ubifs_jnl_delete_inode()', the deletion inode is not in the journal
1127 * anymore, and in fact it might not be on the flash anymore, because it might
1128 * have been garbage-collected already. And for optimization reasons UBIFS does
1129 * not read the orphan area if it has been unmounted cleanly, so it would have
1130 * no indication in the journal that there is a deleted inode which has to be
1131 * removed from TNC.
1132 *
1133 * However, if there was no commit between 'ubifs_jnl_update()' and
1134 * 'ubifs_jnl_delete_inode()', then there is no need to write the deletion
1135 * inode to the media for the second time. And this is quite a typical case.
1136 *
1137 * This function returns zero in case of success and a negative error code in
1138 * case of failure.
1139 */
1140int ubifs_jnl_delete_inode(struct ubifs_info *c, const struct inode *inode)
1141{
1142	int err;
1143	struct ubifs_inode *ui = ubifs_inode(inode);
1144
1145	ubifs_assert(c, inode->i_nlink == 0);
1146
1147	if (ui->xattr_cnt || ui->del_cmtno != c->cmt_no)
1148		/* A commit happened for sure or inode hosts xattrs */
1149		return ubifs_jnl_write_inode(c, inode);
1150
1151	down_read(&c->commit_sem);
1152	/*
1153	 * Check commit number again, because the first test has been done
1154	 * without @c->commit_sem, so a commit might have happened.
1155	 */
1156	if (ui->del_cmtno != c->cmt_no) {
1157		up_read(&c->commit_sem);
1158		return ubifs_jnl_write_inode(c, inode);
1159	}
1160
1161	err = ubifs_tnc_remove_ino(c, inode->i_ino);
1162	if (err)
1163		ubifs_ro_mode(c, err);
1164	else
1165		ubifs_delete_orphan(c, inode->i_ino);
1166	up_read(&c->commit_sem);
1167	return err;
1168}
1169
1170/**
1171 * ubifs_jnl_xrename - cross rename two directory entries.
1172 * @c: UBIFS file-system description object
1173 * @fst_dir: parent inode of 1st directory entry to exchange
1174 * @fst_inode: 1st inode to exchange
1175 * @fst_nm: name of 1st inode to exchange
1176 * @snd_dir: parent inode of 2nd directory entry to exchange
1177 * @snd_inode: 2nd inode to exchange
1178 * @snd_nm: name of 2nd inode to exchange
1179 * @sync: non-zero if the write-buffer has to be synchronized
1180 *
1181 * This function implements the cross rename operation which may involve
1182 * writing 2 inodes and 2 directory entries. It marks the written inodes as clean
1183 * and returns zero on success. In case of failure, a negative error code is
1184 * returned.
1185 */
1186int ubifs_jnl_xrename(struct ubifs_info *c, const struct inode *fst_dir,
1187		      const struct inode *fst_inode,
1188		      const struct fscrypt_name *fst_nm,
1189		      const struct inode *snd_dir,
1190		      const struct inode *snd_inode,
1191		      const struct fscrypt_name *snd_nm, int sync)
1192{
1193	union ubifs_key key;
1194	struct ubifs_dent_node *dent1, *dent2;
1195	int err, dlen1, dlen2, lnum, offs, len, plen = UBIFS_INO_NODE_SZ;
1196	int aligned_dlen1, aligned_dlen2;
1197	int twoparents = (fst_dir != snd_dir);
1198	void *p;
1199	u8 hash_dent1[UBIFS_HASH_ARR_SZ];
1200	u8 hash_dent2[UBIFS_HASH_ARR_SZ];
1201	u8 hash_p1[UBIFS_HASH_ARR_SZ];
1202	u8 hash_p2[UBIFS_HASH_ARR_SZ];
1203
1204	ubifs_assert(c, ubifs_inode(fst_dir)->data_len == 0);
1205	ubifs_assert(c, ubifs_inode(snd_dir)->data_len == 0);
1206	ubifs_assert(c, mutex_is_locked(&ubifs_inode(fst_dir)->ui_mutex));
1207	ubifs_assert(c, mutex_is_locked(&ubifs_inode(snd_dir)->ui_mutex));
1208
1209	dlen1 = UBIFS_DENT_NODE_SZ + fname_len(snd_nm) + 1;
1210	dlen2 = UBIFS_DENT_NODE_SZ + fname_len(fst_nm) + 1;
1211	aligned_dlen1 = ALIGN(dlen1, 8);
1212	aligned_dlen2 = ALIGN(dlen2, 8);
1213
1214	len = aligned_dlen1 + aligned_dlen2 + ALIGN(plen, 8);
1215	if (twoparents)
1216		len += plen;
1217
1218	len += ubifs_auth_node_sz(c);
1219
1220	dent1 = kzalloc(len, GFP_NOFS);
1221	if (!dent1)
1222		return -ENOMEM;
1223
1224	/* Make reservation before allocating sequence numbers */
1225	err = make_reservation(c, BASEHD, len);
1226	if (err)
1227		goto out_free;
1228
1229	/* Make new dent for 1st entry */
1230	dent1->ch.node_type = UBIFS_DENT_NODE;
1231	dent_key_init_flash(c, &dent1->key, snd_dir->i_ino, snd_nm);
1232	dent1->inum = cpu_to_le64(fst_inode->i_ino);
1233	dent1->type = get_dent_type(fst_inode->i_mode);
1234	dent1->nlen = cpu_to_le16(fname_len(snd_nm));
1235	memcpy(dent1->name, fname_name(snd_nm), fname_len(snd_nm));
1236	dent1->name[fname_len(snd_nm)] = '\0';
1237	set_dent_cookie(c, dent1);
1238	zero_dent_node_unused(dent1);
1239	ubifs_prep_grp_node(c, dent1, dlen1, 0);
1240	err = ubifs_node_calc_hash(c, dent1, hash_dent1);
1241	if (err)
1242		goto out_release;
1243
1244	/* Make new dent for 2nd entry */
1245	dent2 = (void *)dent1 + aligned_dlen1;
1246	dent2->ch.node_type = UBIFS_DENT_NODE;
1247	dent_key_init_flash(c, &dent2->key, fst_dir->i_ino, fst_nm);
1248	dent2->inum = cpu_to_le64(snd_inode->i_ino);
1249	dent2->type = get_dent_type(snd_inode->i_mode);
1250	dent2->nlen = cpu_to_le16(fname_len(fst_nm));
1251	memcpy(dent2->name, fname_name(fst_nm), fname_len(fst_nm));
1252	dent2->name[fname_len(fst_nm)] = '\0';
1253	set_dent_cookie(c, dent2);
1254	zero_dent_node_unused(dent2);
1255	ubifs_prep_grp_node(c, dent2, dlen2, 0);
1256	err = ubifs_node_calc_hash(c, dent2, hash_dent2);
1257	if (err)
1258		goto out_release;
1259
1260	p = (void *)dent2 + aligned_dlen2;
1261	if (!twoparents) {
1262		pack_inode(c, p, fst_dir, 1);
1263		err = ubifs_node_calc_hash(c, p, hash_p1);
1264		if (err)
1265			goto out_release;
1266	} else {
1267		pack_inode(c, p, fst_dir, 0);
1268		err = ubifs_node_calc_hash(c, p, hash_p1);
1269		if (err)
1270			goto out_release;
1271		p += ALIGN(plen, 8);
1272		pack_inode(c, p, snd_dir, 1);
1273		err = ubifs_node_calc_hash(c, p, hash_p2);
1274		if (err)
1275			goto out_release;
1276	}
1277
1278	err = write_head(c, BASEHD, dent1, len, &lnum, &offs, sync);
1279	if (err)
1280		goto out_release;
1281	if (!sync) {
1282		struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf;
1283
1284		ubifs_wbuf_add_ino_nolock(wbuf, fst_dir->i_ino);
1285		ubifs_wbuf_add_ino_nolock(wbuf, snd_dir->i_ino);
1286	}
1287	release_head(c, BASEHD);
1288
1289	ubifs_add_auth_dirt(c, lnum);
1290
1291	dent_key_init(c, &key, snd_dir->i_ino, snd_nm);
1292	err = ubifs_tnc_add_nm(c, &key, lnum, offs, dlen1, hash_dent1, snd_nm);
1293	if (err)
1294		goto out_ro;
1295
1296	offs += aligned_dlen1;
1297	dent_key_init(c, &key, fst_dir->i_ino, fst_nm);
1298	err = ubifs_tnc_add_nm(c, &key, lnum, offs, dlen2, hash_dent2, fst_nm);
1299	if (err)
1300		goto out_ro;
1301
1302	offs += aligned_dlen2;
1303
1304	ino_key_init(c, &key, fst_dir->i_ino);
1305	err = ubifs_tnc_add(c, &key, lnum, offs, plen, hash_p1);
1306	if (err)
1307		goto out_ro;
1308
1309	if (twoparents) {
1310		offs += ALIGN(plen, 8);
1311		ino_key_init(c, &key, snd_dir->i_ino);
1312		err = ubifs_tnc_add(c, &key, lnum, offs, plen, hash_p2);
1313		if (err)
1314			goto out_ro;
1315	}
1316
1317	finish_reservation(c);
1318
1319	mark_inode_clean(c, ubifs_inode(fst_dir));
1320	if (twoparents)
1321		mark_inode_clean(c, ubifs_inode(snd_dir));
1322	kfree(dent1);
1323	return 0;
1324
1325out_release:
1326	release_head(c, BASEHD);
1327out_ro:
1328	ubifs_ro_mode(c, err);
1329	finish_reservation(c);
1330out_free:
1331	kfree(dent1);
1332	return err;
1333}
1334
1335/**
1336 * ubifs_jnl_rename - rename a directory entry.
1337 * @c: UBIFS file-system description object
1338 * @old_dir: parent inode of directory entry to rename
1339 * @old_inode: directory entry's inode to rename
1340 * @old_nm: name of the old directory entry to rename
1341 * @new_dir: parent inode of directory entry to rename
1342 * @new_inode: new directory entry's inode (or directory entry's inode to
1343 *		replace)
1344 * @new_nm: new name of the new directory entry
1345 * @whiteout: whiteout inode
1346 * @sync: non-zero if the write-buffer has to be synchronized
1347 * @delete_orphan: indicates an orphan entry deletion for @whiteout
1348 *
1349 * This function implements the re-name operation which may involve writing up
1350 * to 4 inodes(new inode, whiteout inode, old and new parent directory inodes)
1351 * and 2 directory entries. It marks the written inodes as clean and returns
1352 * zero on success. In case of failure, a negative error code is returned.
1353 */
1354int ubifs_jnl_rename(struct ubifs_info *c, const struct inode *old_dir,
1355		     const struct inode *old_inode,
1356		     const struct fscrypt_name *old_nm,
1357		     const struct inode *new_dir,
1358		     const struct inode *new_inode,
1359		     const struct fscrypt_name *new_nm,
1360		     const struct inode *whiteout, int sync, int delete_orphan)
1361{
1362	void *p;
1363	union ubifs_key key;
1364	struct ubifs_dent_node *dent, *dent2;
1365	int err, dlen1, dlen2, ilen, wlen, lnum, offs, len, orphan_added = 0;
1366	int aligned_dlen1, aligned_dlen2, plen = UBIFS_INO_NODE_SZ;
1367	int last_reference = !!(new_inode && new_inode->i_nlink == 0);
1368	int move = (old_dir != new_dir);
1369	struct ubifs_inode *new_ui, *whiteout_ui;
1370	u8 hash_old_dir[UBIFS_HASH_ARR_SZ];
1371	u8 hash_new_dir[UBIFS_HASH_ARR_SZ];
1372	u8 hash_new_inode[UBIFS_HASH_ARR_SZ];
1373	u8 hash_whiteout_inode[UBIFS_HASH_ARR_SZ];
1374	u8 hash_dent1[UBIFS_HASH_ARR_SZ];
1375	u8 hash_dent2[UBIFS_HASH_ARR_SZ];
1376
1377	ubifs_assert(c, ubifs_inode(old_dir)->data_len == 0);
1378	ubifs_assert(c, ubifs_inode(new_dir)->data_len == 0);
1379	ubifs_assert(c, mutex_is_locked(&ubifs_inode(old_dir)->ui_mutex));
1380	ubifs_assert(c, mutex_is_locked(&ubifs_inode(new_dir)->ui_mutex));
1381
1382	dlen1 = UBIFS_DENT_NODE_SZ + fname_len(new_nm) + 1;
1383	dlen2 = UBIFS_DENT_NODE_SZ + fname_len(old_nm) + 1;
1384	if (new_inode) {
1385		new_ui = ubifs_inode(new_inode);
1386		ubifs_assert(c, mutex_is_locked(&new_ui->ui_mutex));
1387		ilen = UBIFS_INO_NODE_SZ;
1388		if (!last_reference)
1389			ilen += new_ui->data_len;
1390	} else
1391		ilen = 0;
1392
1393	if (whiteout) {
1394		whiteout_ui = ubifs_inode(whiteout);
1395		ubifs_assert(c, mutex_is_locked(&whiteout_ui->ui_mutex));
1396		ubifs_assert(c, whiteout->i_nlink == 1);
1397		ubifs_assert(c, !whiteout_ui->dirty);
1398		wlen = UBIFS_INO_NODE_SZ;
1399		wlen += whiteout_ui->data_len;
1400	} else
1401		wlen = 0;
1402
1403	aligned_dlen1 = ALIGN(dlen1, 8);
1404	aligned_dlen2 = ALIGN(dlen2, 8);
1405	len = aligned_dlen1 + aligned_dlen2 + ALIGN(ilen, 8) +
1406	      ALIGN(wlen, 8) + ALIGN(plen, 8);
1407	if (move)
1408		len += plen;
1409
1410	len += ubifs_auth_node_sz(c);
1411
1412	dent = kzalloc(len, GFP_NOFS);
1413	if (!dent)
1414		return -ENOMEM;
1415
1416	/* Make reservation before allocating sequence numbers */
1417	err = make_reservation(c, BASEHD, len);
1418	if (err)
1419		goto out_free;
1420
1421	/* Make new dent */
1422	dent->ch.node_type = UBIFS_DENT_NODE;
1423	dent_key_init_flash(c, &dent->key, new_dir->i_ino, new_nm);
1424	dent->inum = cpu_to_le64(old_inode->i_ino);
1425	dent->type = get_dent_type(old_inode->i_mode);
1426	dent->nlen = cpu_to_le16(fname_len(new_nm));
1427	memcpy(dent->name, fname_name(new_nm), fname_len(new_nm));
1428	dent->name[fname_len(new_nm)] = '\0';
1429	set_dent_cookie(c, dent);
1430	zero_dent_node_unused(dent);
1431	ubifs_prep_grp_node(c, dent, dlen1, 0);
1432	err = ubifs_node_calc_hash(c, dent, hash_dent1);
1433	if (err)
1434		goto out_release;
1435
1436	dent2 = (void *)dent + aligned_dlen1;
1437	dent2->ch.node_type = UBIFS_DENT_NODE;
1438	dent_key_init_flash(c, &dent2->key, old_dir->i_ino, old_nm);
1439
1440	if (whiteout) {
1441		dent2->inum = cpu_to_le64(whiteout->i_ino);
1442		dent2->type = get_dent_type(whiteout->i_mode);
1443	} else {
1444		/* Make deletion dent */
1445		dent2->inum = 0;
1446		dent2->type = DT_UNKNOWN;
1447	}
1448	dent2->nlen = cpu_to_le16(fname_len(old_nm));
1449	memcpy(dent2->name, fname_name(old_nm), fname_len(old_nm));
1450	dent2->name[fname_len(old_nm)] = '\0';
1451	set_dent_cookie(c, dent2);
1452	zero_dent_node_unused(dent2);
1453	ubifs_prep_grp_node(c, dent2, dlen2, 0);
1454	err = ubifs_node_calc_hash(c, dent2, hash_dent2);
1455	if (err)
1456		goto out_release;
1457
1458	p = (void *)dent2 + aligned_dlen2;
1459	if (new_inode) {
1460		pack_inode(c, p, new_inode, 0);
1461		err = ubifs_node_calc_hash(c, p, hash_new_inode);
1462		if (err)
1463			goto out_release;
1464
1465		p += ALIGN(ilen, 8);
1466	}
1467
1468	if (whiteout) {
1469		pack_inode(c, p, whiteout, 0);
1470		err = ubifs_node_calc_hash(c, p, hash_whiteout_inode);
1471		if (err)
1472			goto out_release;
1473
1474		p += ALIGN(wlen, 8);
1475	}
1476
1477	if (!move) {
1478		pack_inode(c, p, old_dir, 1);
1479		err = ubifs_node_calc_hash(c, p, hash_old_dir);
1480		if (err)
1481			goto out_release;
1482	} else {
1483		pack_inode(c, p, old_dir, 0);
1484		err = ubifs_node_calc_hash(c, p, hash_old_dir);
1485		if (err)
1486			goto out_release;
1487
1488		p += ALIGN(plen, 8);
1489		pack_inode(c, p, new_dir, 1);
1490		err = ubifs_node_calc_hash(c, p, hash_new_dir);
1491		if (err)
1492			goto out_release;
1493	}
1494
1495	if (last_reference) {
1496		err = ubifs_add_orphan(c, new_inode->i_ino);
1497		if (err) {
1498			release_head(c, BASEHD);
1499			goto out_finish;
1500		}
1501		new_ui->del_cmtno = c->cmt_no;
1502		orphan_added = 1;
1503	}
1504
1505	err = write_head(c, BASEHD, dent, len, &lnum, &offs, sync);
1506	if (err)
1507		goto out_release;
1508	if (!sync) {
1509		struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf;
1510
1511		ubifs_wbuf_add_ino_nolock(wbuf, new_dir->i_ino);
1512		ubifs_wbuf_add_ino_nolock(wbuf, old_dir->i_ino);
1513		if (new_inode)
1514			ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf,
1515						  new_inode->i_ino);
1516		if (whiteout)
1517			ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf,
1518						  whiteout->i_ino);
1519	}
1520	release_head(c, BASEHD);
1521
1522	ubifs_add_auth_dirt(c, lnum);
1523
1524	dent_key_init(c, &key, new_dir->i_ino, new_nm);
1525	err = ubifs_tnc_add_nm(c, &key, lnum, offs, dlen1, hash_dent1, new_nm);
1526	if (err)
1527		goto out_ro;
1528
1529	offs += aligned_dlen1;
1530	if (whiteout) {
1531		dent_key_init(c, &key, old_dir->i_ino, old_nm);
1532		err = ubifs_tnc_add_nm(c, &key, lnum, offs, dlen2, hash_dent2, old_nm);
1533		if (err)
1534			goto out_ro;
1535	} else {
1536		err = ubifs_add_dirt(c, lnum, dlen2);
1537		if (err)
1538			goto out_ro;
1539
1540		dent_key_init(c, &key, old_dir->i_ino, old_nm);
1541		err = ubifs_tnc_remove_nm(c, &key, old_nm);
1542		if (err)
1543			goto out_ro;
1544	}
1545
1546	offs += aligned_dlen2;
1547	if (new_inode) {
1548		ino_key_init(c, &key, new_inode->i_ino);
1549		err = ubifs_tnc_add(c, &key, lnum, offs, ilen, hash_new_inode);
1550		if (err)
1551			goto out_ro;
1552		offs += ALIGN(ilen, 8);
1553	}
1554
1555	if (whiteout) {
1556		ino_key_init(c, &key, whiteout->i_ino);
1557		err = ubifs_tnc_add(c, &key, lnum, offs, wlen,
1558				    hash_whiteout_inode);
1559		if (err)
1560			goto out_ro;
1561		offs += ALIGN(wlen, 8);
1562	}
1563
1564	ino_key_init(c, &key, old_dir->i_ino);
1565	err = ubifs_tnc_add(c, &key, lnum, offs, plen, hash_old_dir);
1566	if (err)
1567		goto out_ro;
1568
1569	if (move) {
1570		offs += ALIGN(plen, 8);
1571		ino_key_init(c, &key, new_dir->i_ino);
1572		err = ubifs_tnc_add(c, &key, lnum, offs, plen, hash_new_dir);
1573		if (err)
1574			goto out_ro;
1575	}
1576
1577	if (delete_orphan)
1578		ubifs_delete_orphan(c, whiteout->i_ino);
1579
1580	finish_reservation(c);
1581	if (new_inode) {
1582		mark_inode_clean(c, new_ui);
1583		spin_lock(&new_ui->ui_lock);
1584		new_ui->synced_i_size = new_ui->ui_size;
1585		spin_unlock(&new_ui->ui_lock);
1586	}
1587	/*
1588	 * No need to mark whiteout inode clean.
1589	 * Whiteout doesn't have non-zero size, no need to update
1590	 * synced_i_size for whiteout_ui.
1591	 */
1592	mark_inode_clean(c, ubifs_inode(old_dir));
1593	if (move)
1594		mark_inode_clean(c, ubifs_inode(new_dir));
1595	kfree(dent);
1596	return 0;
1597
1598out_release:
1599	release_head(c, BASEHD);
1600out_ro:
1601	ubifs_ro_mode(c, err);
1602	if (orphan_added)
1603		ubifs_delete_orphan(c, new_inode->i_ino);
1604out_finish:
1605	finish_reservation(c);
1606out_free:
1607	kfree(dent);
1608	return err;
1609}
1610
1611/**
1612 * truncate_data_node - re-compress/encrypt a truncated data node.
1613 * @c: UBIFS file-system description object
1614 * @inode: inode which refers to the data node
1615 * @block: data block number
1616 * @dn: data node to re-compress
1617 * @new_len: new length
1618 * @dn_size: size of the data node @dn in memory
1619 *
1620 * This function is used when an inode is truncated and the last data node of
1621 * the inode has to be re-compressed/encrypted and re-written.
1622 */
1623static int truncate_data_node(const struct ubifs_info *c, const struct inode *inode,
1624			      unsigned int block, struct ubifs_data_node *dn,
1625			      int *new_len, int dn_size)
1626{
1627	void *buf;
1628	int err, dlen, compr_type, out_len, data_size;
1629
1630	out_len = le32_to_cpu(dn->size);
1631	buf = kmalloc(out_len, GFP_NOFS);
1632	if (!buf)
1633		return -ENOMEM;
1634
1635	dlen = le32_to_cpu(dn->ch.len) - UBIFS_DATA_NODE_SZ;
1636	data_size = dn_size - UBIFS_DATA_NODE_SZ;
1637	compr_type = le16_to_cpu(dn->compr_type);
1638
1639	if (IS_ENCRYPTED(inode)) {
1640		err = ubifs_decrypt(inode, dn, &dlen, block);
1641		if (err)
1642			goto out;
1643	}
1644
1645	if (compr_type == UBIFS_COMPR_NONE) {
1646		out_len = *new_len;
1647	} else {
1648		err = ubifs_decompress(c, &dn->data, dlen, buf, &out_len, compr_type);
1649		if (err)
1650			goto out;
1651
1652		ubifs_compress(c, buf, *new_len, &dn->data, &out_len, &compr_type);
1653	}
1654
1655	if (IS_ENCRYPTED(inode)) {
1656		err = ubifs_encrypt(inode, dn, out_len, &data_size, block);
1657		if (err)
1658			goto out;
1659
1660		out_len = data_size;
1661	} else {
1662		dn->compr_size = 0;
1663	}
1664
1665	ubifs_assert(c, out_len <= UBIFS_BLOCK_SIZE);
1666	dn->compr_type = cpu_to_le16(compr_type);
1667	dn->size = cpu_to_le32(*new_len);
1668	*new_len = UBIFS_DATA_NODE_SZ + out_len;
1669	err = 0;
1670out:
1671	kfree(buf);
1672	return err;
1673}
1674
1675/**
1676 * ubifs_jnl_truncate - update the journal for a truncation.
1677 * @c: UBIFS file-system description object
1678 * @inode: inode to truncate
1679 * @old_size: old size
1680 * @new_size: new size
1681 *
1682 * When the size of a file decreases due to truncation, a truncation node is
1683 * written, the journal tree is updated, and the last data block is re-written
1684 * if it has been affected. The inode is also updated in order to synchronize
1685 * the new inode size.
1686 *
1687 * This function marks the inode as clean and returns zero on success. In case
1688 * of failure, a negative error code is returned.
1689 */
1690int ubifs_jnl_truncate(struct ubifs_info *c, const struct inode *inode,
1691		       loff_t old_size, loff_t new_size)
1692{
1693	union ubifs_key key, to_key;
1694	struct ubifs_ino_node *ino;
1695	struct ubifs_trun_node *trun;
1696	struct ubifs_data_node *dn;
1697	int err, dlen, len, lnum, offs, bit, sz, sync = IS_SYNC(inode);
1698	int dn_size;
1699	struct ubifs_inode *ui = ubifs_inode(inode);
1700	ino_t inum = inode->i_ino;
1701	unsigned int blk;
1702	u8 hash_ino[UBIFS_HASH_ARR_SZ];
1703	u8 hash_dn[UBIFS_HASH_ARR_SZ];
1704
1705	dbg_jnl("ino %lu, size %lld -> %lld",
1706		(unsigned long)inum, old_size, new_size);
1707	ubifs_assert(c, !ui->data_len);
1708	ubifs_assert(c, S_ISREG(inode->i_mode));
1709	ubifs_assert(c, mutex_is_locked(&ui->ui_mutex));
1710
1711	dn_size = COMPRESSED_DATA_NODE_BUF_SZ;
1712
1713	if (IS_ENCRYPTED(inode))
1714		dn_size += UBIFS_CIPHER_BLOCK_SIZE;
1715
1716	sz =  UBIFS_TRUN_NODE_SZ + UBIFS_INO_NODE_SZ +
1717		dn_size + ubifs_auth_node_sz(c);
1718
1719	ino = kmalloc(sz, GFP_NOFS);
1720	if (!ino)
1721		return -ENOMEM;
1722
1723	trun = (void *)ino + UBIFS_INO_NODE_SZ;
1724	trun->ch.node_type = UBIFS_TRUN_NODE;
1725	trun->inum = cpu_to_le32(inum);
1726	trun->old_size = cpu_to_le64(old_size);
1727	trun->new_size = cpu_to_le64(new_size);
1728	zero_trun_node_unused(trun);
1729
1730	dlen = new_size & (UBIFS_BLOCK_SIZE - 1);
1731	if (dlen) {
1732		/* Get last data block so it can be truncated */
1733		dn = (void *)trun + UBIFS_TRUN_NODE_SZ;
1734		blk = new_size >> UBIFS_BLOCK_SHIFT;
1735		data_key_init(c, &key, inum, blk);
1736		dbg_jnlk(&key, "last block key ");
1737		err = ubifs_tnc_lookup(c, &key, dn);
1738		if (err == -ENOENT)
1739			dlen = 0; /* Not found (so it is a hole) */
1740		else if (err)
1741			goto out_free;
1742		else {
1743			int dn_len = le32_to_cpu(dn->size);
1744
1745			if (dn_len <= 0 || dn_len > UBIFS_BLOCK_SIZE) {
1746				ubifs_err(c, "bad data node (block %u, inode %lu)",
1747					  blk, inode->i_ino);
1748				ubifs_dump_node(c, dn, dn_size);
1749				err = -EUCLEAN;
1750				goto out_free;
1751			}
1752
1753			if (dn_len <= dlen)
1754				dlen = 0; /* Nothing to do */
1755			else {
1756				err = truncate_data_node(c, inode, blk, dn,
1757						&dlen, dn_size);
1758				if (err)
1759					goto out_free;
1760			}
1761		}
1762	}
1763
1764	/* Must make reservation before allocating sequence numbers */
1765	len = UBIFS_TRUN_NODE_SZ + UBIFS_INO_NODE_SZ;
1766
1767	if (ubifs_authenticated(c))
1768		len += ALIGN(dlen, 8) + ubifs_auth_node_sz(c);
1769	else
1770		len += dlen;
1771
1772	err = make_reservation(c, BASEHD, len);
1773	if (err)
1774		goto out_free;
1775
1776	pack_inode(c, ino, inode, 0);
1777	err = ubifs_node_calc_hash(c, ino, hash_ino);
1778	if (err)
1779		goto out_release;
1780
1781	ubifs_prep_grp_node(c, trun, UBIFS_TRUN_NODE_SZ, dlen ? 0 : 1);
1782	if (dlen) {
1783		ubifs_prep_grp_node(c, dn, dlen, 1);
1784		err = ubifs_node_calc_hash(c, dn, hash_dn);
1785		if (err)
1786			goto out_release;
1787	}
1788
1789	err = write_head(c, BASEHD, ino, len, &lnum, &offs, sync);
1790	if (err)
1791		goto out_release;
1792	if (!sync)
1793		ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf, inum);
1794	release_head(c, BASEHD);
1795
1796	ubifs_add_auth_dirt(c, lnum);
1797
1798	if (dlen) {
1799		sz = offs + UBIFS_INO_NODE_SZ + UBIFS_TRUN_NODE_SZ;
1800		err = ubifs_tnc_add(c, &key, lnum, sz, dlen, hash_dn);
1801		if (err)
1802			goto out_ro;
1803	}
1804
1805	ino_key_init(c, &key, inum);
1806	err = ubifs_tnc_add(c, &key, lnum, offs, UBIFS_INO_NODE_SZ, hash_ino);
1807	if (err)
1808		goto out_ro;
1809
1810	err = ubifs_add_dirt(c, lnum, UBIFS_TRUN_NODE_SZ);
1811	if (err)
1812		goto out_ro;
1813
1814	bit = new_size & (UBIFS_BLOCK_SIZE - 1);
1815	blk = (new_size >> UBIFS_BLOCK_SHIFT) + (bit ? 1 : 0);
1816	data_key_init(c, &key, inum, blk);
1817
1818	bit = old_size & (UBIFS_BLOCK_SIZE - 1);
1819	blk = (old_size >> UBIFS_BLOCK_SHIFT) - (bit ? 0 : 1);
1820	data_key_init(c, &to_key, inum, blk);
1821
1822	err = ubifs_tnc_remove_range(c, &key, &to_key);
1823	if (err)
1824		goto out_ro;
1825
1826	finish_reservation(c);
1827	spin_lock(&ui->ui_lock);
1828	ui->synced_i_size = ui->ui_size;
1829	spin_unlock(&ui->ui_lock);
1830	mark_inode_clean(c, ui);
1831	kfree(ino);
1832	return 0;
1833
1834out_release:
1835	release_head(c, BASEHD);
1836out_ro:
1837	ubifs_ro_mode(c, err);
1838	finish_reservation(c);
1839out_free:
1840	kfree(ino);
1841	return err;
1842}
1843
1844
1845/**
1846 * ubifs_jnl_delete_xattr - delete an extended attribute.
1847 * @c: UBIFS file-system description object
1848 * @host: host inode
1849 * @inode: extended attribute inode
1850 * @nm: extended attribute entry name
1851 *
1852 * This function delete an extended attribute which is very similar to
1853 * un-linking regular files - it writes a deletion xentry, a deletion inode and
1854 * updates the target inode. Returns zero in case of success and a negative
1855 * error code in case of failure.
1856 */
1857int ubifs_jnl_delete_xattr(struct ubifs_info *c, const struct inode *host,
1858			   const struct inode *inode,
1859			   const struct fscrypt_name *nm)
1860{
1861	int err, xlen, hlen, len, lnum, xent_offs, aligned_xlen, write_len;
1862	struct ubifs_dent_node *xent;
1863	struct ubifs_ino_node *ino;
1864	union ubifs_key xent_key, key1, key2;
1865	int sync = IS_DIRSYNC(host);
1866	struct ubifs_inode *host_ui = ubifs_inode(host);
1867	u8 hash[UBIFS_HASH_ARR_SZ];
1868
1869	ubifs_assert(c, inode->i_nlink == 0);
1870	ubifs_assert(c, mutex_is_locked(&host_ui->ui_mutex));
1871
1872	/*
1873	 * Since we are deleting the inode, we do not bother to attach any data
1874	 * to it and assume its length is %UBIFS_INO_NODE_SZ.
1875	 */
1876	xlen = UBIFS_DENT_NODE_SZ + fname_len(nm) + 1;
1877	aligned_xlen = ALIGN(xlen, 8);
1878	hlen = host_ui->data_len + UBIFS_INO_NODE_SZ;
1879	len = aligned_xlen + UBIFS_INO_NODE_SZ + ALIGN(hlen, 8);
1880
1881	write_len = len + ubifs_auth_node_sz(c);
1882
1883	xent = kzalloc(write_len, GFP_NOFS);
1884	if (!xent)
1885		return -ENOMEM;
1886
1887	/* Make reservation before allocating sequence numbers */
1888	err = make_reservation(c, BASEHD, write_len);
1889	if (err) {
1890		kfree(xent);
1891		return err;
1892	}
1893
1894	xent->ch.node_type = UBIFS_XENT_NODE;
1895	xent_key_init(c, &xent_key, host->i_ino, nm);
1896	key_write(c, &xent_key, xent->key);
1897	xent->inum = 0;
1898	xent->type = get_dent_type(inode->i_mode);
1899	xent->nlen = cpu_to_le16(fname_len(nm));
1900	memcpy(xent->name, fname_name(nm), fname_len(nm));
1901	xent->name[fname_len(nm)] = '\0';
1902	zero_dent_node_unused(xent);
1903	ubifs_prep_grp_node(c, xent, xlen, 0);
1904
1905	ino = (void *)xent + aligned_xlen;
1906	pack_inode(c, ino, inode, 0);
1907	ino = (void *)ino + UBIFS_INO_NODE_SZ;
1908	pack_inode(c, ino, host, 1);
1909	err = ubifs_node_calc_hash(c, ino, hash);
1910	if (err)
1911		goto out_release;
1912
1913	err = write_head(c, BASEHD, xent, write_len, &lnum, &xent_offs, sync);
1914	if (!sync && !err)
1915		ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf, host->i_ino);
1916	release_head(c, BASEHD);
1917
1918	ubifs_add_auth_dirt(c, lnum);
1919	kfree(xent);
1920	if (err)
1921		goto out_ro;
1922
1923	/* Remove the extended attribute entry from TNC */
1924	err = ubifs_tnc_remove_nm(c, &xent_key, nm);
1925	if (err)
1926		goto out_ro;
1927	err = ubifs_add_dirt(c, lnum, xlen);
1928	if (err)
1929		goto out_ro;
1930
1931	/*
1932	 * Remove all nodes belonging to the extended attribute inode from TNC.
1933	 * Well, there actually must be only one node - the inode itself.
1934	 */
1935	lowest_ino_key(c, &key1, inode->i_ino);
1936	highest_ino_key(c, &key2, inode->i_ino);
1937	err = ubifs_tnc_remove_range(c, &key1, &key2);
1938	if (err)
1939		goto out_ro;
1940	err = ubifs_add_dirt(c, lnum, UBIFS_INO_NODE_SZ);
1941	if (err)
1942		goto out_ro;
1943
1944	/* And update TNC with the new host inode position */
1945	ino_key_init(c, &key1, host->i_ino);
1946	err = ubifs_tnc_add(c, &key1, lnum, xent_offs + len - hlen, hlen, hash);
1947	if (err)
1948		goto out_ro;
1949
1950	finish_reservation(c);
1951	spin_lock(&host_ui->ui_lock);
1952	host_ui->synced_i_size = host_ui->ui_size;
1953	spin_unlock(&host_ui->ui_lock);
1954	mark_inode_clean(c, host_ui);
1955	return 0;
1956
1957out_release:
1958	kfree(xent);
1959	release_head(c, BASEHD);
1960out_ro:
1961	ubifs_ro_mode(c, err);
1962	finish_reservation(c);
1963	return err;
1964}
1965
1966/**
1967 * ubifs_jnl_change_xattr - change an extended attribute.
1968 * @c: UBIFS file-system description object
1969 * @inode: extended attribute inode
1970 * @host: host inode
1971 *
1972 * This function writes the updated version of an extended attribute inode and
1973 * the host inode to the journal (to the base head). The host inode is written
1974 * after the extended attribute inode in order to guarantee that the extended
1975 * attribute will be flushed when the inode is synchronized by 'fsync()' and
1976 * consequently, the write-buffer is synchronized. This function returns zero
1977 * in case of success and a negative error code in case of failure.
1978 */
1979int ubifs_jnl_change_xattr(struct ubifs_info *c, const struct inode *inode,
1980			   const struct inode *host)
1981{
1982	int err, len1, len2, aligned_len, aligned_len1, lnum, offs;
1983	struct ubifs_inode *host_ui = ubifs_inode(host);
1984	struct ubifs_ino_node *ino;
1985	union ubifs_key key;
1986	int sync = IS_DIRSYNC(host);
1987	u8 hash_host[UBIFS_HASH_ARR_SZ];
1988	u8 hash[UBIFS_HASH_ARR_SZ];
1989
1990	dbg_jnl("ino %lu, ino %lu", host->i_ino, inode->i_ino);
1991	ubifs_assert(c, inode->i_nlink > 0);
1992	ubifs_assert(c, mutex_is_locked(&host_ui->ui_mutex));
1993
1994	len1 = UBIFS_INO_NODE_SZ + host_ui->data_len;
1995	len2 = UBIFS_INO_NODE_SZ + ubifs_inode(inode)->data_len;
1996	aligned_len1 = ALIGN(len1, 8);
1997	aligned_len = aligned_len1 + ALIGN(len2, 8);
1998
1999	aligned_len += ubifs_auth_node_sz(c);
2000
2001	ino = kzalloc(aligned_len, GFP_NOFS);
2002	if (!ino)
2003		return -ENOMEM;
2004
2005	/* Make reservation before allocating sequence numbers */
2006	err = make_reservation(c, BASEHD, aligned_len);
2007	if (err)
2008		goto out_free;
2009
2010	pack_inode(c, ino, host, 0);
2011	err = ubifs_node_calc_hash(c, ino, hash_host);
2012	if (err)
2013		goto out_release;
2014	pack_inode(c, (void *)ino + aligned_len1, inode, 1);
2015	err = ubifs_node_calc_hash(c, (void *)ino + aligned_len1, hash);
2016	if (err)
2017		goto out_release;
2018
2019	err = write_head(c, BASEHD, ino, aligned_len, &lnum, &offs, 0);
2020	if (!sync && !err) {
2021		struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf;
2022
2023		ubifs_wbuf_add_ino_nolock(wbuf, host->i_ino);
2024		ubifs_wbuf_add_ino_nolock(wbuf, inode->i_ino);
2025	}
2026	release_head(c, BASEHD);
2027	if (err)
2028		goto out_ro;
2029
2030	ubifs_add_auth_dirt(c, lnum);
2031
2032	ino_key_init(c, &key, host->i_ino);
2033	err = ubifs_tnc_add(c, &key, lnum, offs, len1, hash_host);
2034	if (err)
2035		goto out_ro;
2036
2037	ino_key_init(c, &key, inode->i_ino);
2038	err = ubifs_tnc_add(c, &key, lnum, offs + aligned_len1, len2, hash);
2039	if (err)
2040		goto out_ro;
2041
2042	finish_reservation(c);
2043	spin_lock(&host_ui->ui_lock);
2044	host_ui->synced_i_size = host_ui->ui_size;
2045	spin_unlock(&host_ui->ui_lock);
2046	mark_inode_clean(c, host_ui);
2047	kfree(ino);
2048	return 0;
2049
2050out_release:
2051	release_head(c, BASEHD);
2052out_ro:
2053	ubifs_ro_mode(c, err);
2054	finish_reservation(c);
2055out_free:
2056	kfree(ino);
2057	return err;
2058}
2059