PageRenderTime 147ms CodeModel.GetById 46ms app.highlight 95ms RepoModel.GetById 0ms app.codeStats 0ms

/fs/ubifs/journal.c

https://bitbucket.org/cyanogenmod/android_kernel_asus_tf300t
C | 1466 lines | 886 code | 161 blank | 419 comment | 156 complexity | 7e765ccfff9576019d981d4f7fefebe4 MD5 | raw file
Possible License(s): LGPL-2.0, AGPL-1.0, GPL-2.0
   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 */
  67static 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 */
  78static inline void zero_dent_node_unused(struct ubifs_dent_node *dent)
  79{
  80	dent->padding1 = 0;
  81	memset(dent->padding2, 0, 4);
  82}
  83
  84/**
  85 * zero_data_node_unused - zero out unused fields of an on-flash data node.
  86 * @data: the data node to zero out
  87 */
  88static inline void zero_data_node_unused(struct ubifs_data_node *data)
  89{
  90	memset(data->padding, 0, 2);
  91}
  92
  93/**
  94 * zero_trun_node_unused - zero out unused fields of an on-flash truncation
  95 *                         node.
  96 * @trun: the truncation node to zero out
  97 */
  98static inline void zero_trun_node_unused(struct ubifs_trun_node *trun)
  99{
 100	memset(trun->padding, 0, 12);
 101}
 102
 103/**
 104 * reserve_space - reserve space in the journal.
 105 * @c: UBIFS file-system description object
 106 * @jhead: journal head number
 107 * @len: node length
 108 *
 109 * This function reserves space in journal head @head. If the reservation
 110 * succeeded, the journal head stays locked and later has to be unlocked using
 111 * 'release_head()'. 'write_node()' and 'write_head()' functions also unlock
 112 * it. Returns zero in case of success, %-EAGAIN if commit has to be done, and
 113 * other negative error codes in case of other failures.
 114 */
 115static int reserve_space(struct ubifs_info *c, int jhead, int len)
 116{
 117	int err = 0, err1, retries = 0, avail, lnum, offs, squeeze;
 118	struct ubifs_wbuf *wbuf = &c->jheads[jhead].wbuf;
 119
 120	/*
 121	 * Typically, the base head has smaller nodes written to it, so it is
 122	 * better to try to allocate space at the ends of eraseblocks. This is
 123	 * what the squeeze parameter does.
 124	 */
 125	ubifs_assert(!c->ro_media && !c->ro_mount);
 126	squeeze = (jhead == BASEHD);
 127again:
 128	mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
 129
 130	if (c->ro_error) {
 131		err = -EROFS;
 132		goto out_unlock;
 133	}
 134
 135	avail = c->leb_size - wbuf->offs - wbuf->used;
 136	if (wbuf->lnum != -1 && avail >= len)
 137		return 0;
 138
 139	/*
 140	 * Write buffer wasn't seek'ed or there is no enough space - look for an
 141	 * LEB with some empty space.
 142	 */
 143	lnum = ubifs_find_free_space(c, len, &offs, squeeze);
 144	if (lnum >= 0)
 145		goto out;
 146
 147	err = lnum;
 148	if (err != -ENOSPC)
 149		goto out_unlock;
 150
 151	/*
 152	 * No free space, we have to run garbage collector to make
 153	 * some. But the write-buffer mutex has to be unlocked because
 154	 * GC also takes it.
 155	 */
 156	dbg_jnl("no free space in jhead %s, run GC", dbg_jhead(jhead));
 157	mutex_unlock(&wbuf->io_mutex);
 158
 159	lnum = ubifs_garbage_collect(c, 0);
 160	if (lnum < 0) {
 161		err = lnum;
 162		if (err != -ENOSPC)
 163			return err;
 164
 165		/*
 166		 * GC could not make a free LEB. But someone else may
 167		 * have allocated new bud for this journal head,
 168		 * because we dropped @wbuf->io_mutex, so try once
 169		 * again.
 170		 */
 171		dbg_jnl("GC couldn't make a free LEB for jhead %s",
 172			dbg_jhead(jhead));
 173		if (retries++ < 2) {
 174			dbg_jnl("retry (%d)", retries);
 175			goto again;
 176		}
 177
 178		dbg_jnl("return -ENOSPC");
 179		return err;
 180	}
 181
 182	mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
 183	dbg_jnl("got LEB %d for jhead %s", lnum, dbg_jhead(jhead));
 184	avail = c->leb_size - wbuf->offs - wbuf->used;
 185
 186	if (wbuf->lnum != -1 && avail >= len) {
 187		/*
 188		 * Someone else has switched the journal head and we have
 189		 * enough space now. This happens when more than one process is
 190		 * trying to write to the same journal head at the same time.
 191		 */
 192		dbg_jnl("return LEB %d back, already have LEB %d:%d",
 193			lnum, wbuf->lnum, wbuf->offs + wbuf->used);
 194		err = ubifs_return_leb(c, lnum);
 195		if (err)
 196			goto out_unlock;
 197		return 0;
 198	}
 199
 200	offs = 0;
 201
 202out:
 203	/*
 204	 * Make sure we synchronize the write-buffer before we add the new bud
 205	 * to the log. Otherwise we may have a power cut after the log
 206	 * reference node for the last bud (@lnum) is written but before the
 207	 * write-buffer data are written to the next-to-last bud
 208	 * (@wbuf->lnum). And the effect would be that the recovery would see
 209	 * that there is corruption in the next-to-last bud.
 210	 */
 211	err = ubifs_wbuf_sync_nolock(wbuf);
 212	if (err)
 213		goto out_return;
 214	err = ubifs_add_bud_to_log(c, jhead, lnum, offs);
 215	if (err)
 216		goto out_return;
 217	err = ubifs_wbuf_seek_nolock(wbuf, lnum, offs, wbuf->dtype);
 218	if (err)
 219		goto out_unlock;
 220
 221	return 0;
 222
 223out_unlock:
 224	mutex_unlock(&wbuf->io_mutex);
 225	return err;
 226
 227out_return:
 228	/* An error occurred and the LEB has to be returned to lprops */
 229	ubifs_assert(err < 0);
 230	err1 = ubifs_return_leb(c, lnum);
 231	if (err1 && err == -EAGAIN)
 232		/*
 233		 * Return original error code only if it is not %-EAGAIN,
 234		 * which is not really an error. Otherwise, return the error
 235		 * code of 'ubifs_return_leb()'.
 236		 */
 237		err = err1;
 238	mutex_unlock(&wbuf->io_mutex);
 239	return err;
 240}
 241
 242/**
 243 * write_node - write node to a journal head.
 244 * @c: UBIFS file-system description object
 245 * @jhead: journal head
 246 * @node: node to write
 247 * @len: node length
 248 * @lnum: LEB number written is returned here
 249 * @offs: offset written is returned here
 250 *
 251 * This function writes a node to reserved space of journal head @jhead.
 252 * Returns zero in case of success and a negative error code in case of
 253 * failure.
 254 */
 255static int write_node(struct ubifs_info *c, int jhead, void *node, int len,
 256		      int *lnum, int *offs)
 257{
 258	struct ubifs_wbuf *wbuf = &c->jheads[jhead].wbuf;
 259
 260	ubifs_assert(jhead != GCHD);
 261
 262	*lnum = c->jheads[jhead].wbuf.lnum;
 263	*offs = c->jheads[jhead].wbuf.offs + c->jheads[jhead].wbuf.used;
 264
 265	dbg_jnl("jhead %s, LEB %d:%d, len %d",
 266		dbg_jhead(jhead), *lnum, *offs, len);
 267	ubifs_prepare_node(c, node, len, 0);
 268
 269	return ubifs_wbuf_write_nolock(wbuf, node, len);
 270}
 271
 272/**
 273 * write_head - write data to a journal head.
 274 * @c: UBIFS file-system description object
 275 * @jhead: journal head
 276 * @buf: buffer to write
 277 * @len: length to write
 278 * @lnum: LEB number written is returned here
 279 * @offs: offset written is returned here
 280 * @sync: non-zero if the write-buffer has to by synchronized
 281 *
 282 * This function is the same as 'write_node()' but it does not assume the
 283 * buffer it is writing is a node, so it does not prepare it (which means
 284 * initializing common header and calculating CRC).
 285 */
 286static int write_head(struct ubifs_info *c, int jhead, void *buf, int len,
 287		      int *lnum, int *offs, int sync)
 288{
 289	int err;
 290	struct ubifs_wbuf *wbuf = &c->jheads[jhead].wbuf;
 291
 292	ubifs_assert(jhead != GCHD);
 293
 294	*lnum = c->jheads[jhead].wbuf.lnum;
 295	*offs = c->jheads[jhead].wbuf.offs + c->jheads[jhead].wbuf.used;
 296	dbg_jnl("jhead %s, LEB %d:%d, len %d",
 297		dbg_jhead(jhead), *lnum, *offs, len);
 298
 299	err = ubifs_wbuf_write_nolock(wbuf, buf, len);
 300	if (err)
 301		return err;
 302	if (sync)
 303		err = ubifs_wbuf_sync_nolock(wbuf);
 304	return err;
 305}
 306
 307/**
 308 * make_reservation - reserve journal space.
 309 * @c: UBIFS file-system description object
 310 * @jhead: journal head
 311 * @len: how many bytes to reserve
 312 *
 313 * This function makes space reservation in journal head @jhead. The function
 314 * takes the commit lock and locks the journal head, and the caller has to
 315 * unlock the head and finish the reservation with 'finish_reservation()'.
 316 * Returns zero in case of success and a negative error code in case of
 317 * failure.
 318 *
 319 * Note, the journal head may be unlocked as soon as the data is written, while
 320 * the commit lock has to be released after the data has been added to the
 321 * TNC.
 322 */
 323static int make_reservation(struct ubifs_info *c, int jhead, int len)
 324{
 325	int err, cmt_retries = 0, nospc_retries = 0;
 326
 327again:
 328	down_read(&c->commit_sem);
 329	err = reserve_space(c, jhead, len);
 330	if (!err)
 331		return 0;
 332	up_read(&c->commit_sem);
 333
 334	if (err == -ENOSPC) {
 335		/*
 336		 * GC could not make any progress. We should try to commit
 337		 * once because it could make some dirty space and GC would
 338		 * make progress, so make the error -EAGAIN so that the below
 339		 * will commit and re-try.
 340		 */
 341		if (nospc_retries++ < 2) {
 342			dbg_jnl("no space, retry");
 343			err = -EAGAIN;
 344		}
 345
 346		/*
 347		 * This means that the budgeting is incorrect. We always have
 348		 * to be able to write to the media, because all operations are
 349		 * budgeted. Deletions are not budgeted, though, but we reserve
 350		 * an extra LEB for them.
 351		 */
 352	}
 353
 354	if (err != -EAGAIN)
 355		goto out;
 356
 357	/*
 358	 * -EAGAIN means that the journal is full or too large, or the above
 359	 * code wants to do one commit. Do this and re-try.
 360	 */
 361	if (cmt_retries > 128) {
 362		/*
 363		 * This should not happen unless the journal size limitations
 364		 * are too tough.
 365		 */
 366		ubifs_err("stuck in space allocation");
 367		err = -ENOSPC;
 368		goto out;
 369	} else if (cmt_retries > 32)
 370		ubifs_warn("too many space allocation re-tries (%d)",
 371			   cmt_retries);
 372
 373	dbg_jnl("-EAGAIN, commit and retry (retried %d times)",
 374		cmt_retries);
 375	cmt_retries += 1;
 376
 377	err = ubifs_run_commit(c);
 378	if (err)
 379		return err;
 380	goto again;
 381
 382out:
 383	ubifs_err("cannot reserve %d bytes in jhead %d, error %d",
 384		  len, jhead, err);
 385	if (err == -ENOSPC) {
 386		/* This are some budgeting problems, print useful information */
 387		down_write(&c->commit_sem);
 388		dbg_dump_stack();
 389		dbg_dump_budg(c, &c->bi);
 390		dbg_dump_lprops(c);
 391		cmt_retries = dbg_check_lprops(c);
 392		up_write(&c->commit_sem);
 393	}
 394	return err;
 395}
 396
 397/**
 398 * release_head - release a journal head.
 399 * @c: UBIFS file-system description object
 400 * @jhead: journal head
 401 *
 402 * This function releases journal head @jhead which was locked by
 403 * the 'make_reservation()' function. It has to be called after each successful
 404 * 'make_reservation()' invocation.
 405 */
 406static inline void release_head(struct ubifs_info *c, int jhead)
 407{
 408	mutex_unlock(&c->jheads[jhead].wbuf.io_mutex);
 409}
 410
 411/**
 412 * finish_reservation - finish a reservation.
 413 * @c: UBIFS file-system description object
 414 *
 415 * This function finishes journal space reservation. It must be called after
 416 * 'make_reservation()'.
 417 */
 418static void finish_reservation(struct ubifs_info *c)
 419{
 420	up_read(&c->commit_sem);
 421}
 422
 423/**
 424 * get_dent_type - translate VFS inode mode to UBIFS directory entry type.
 425 * @mode: inode mode
 426 */
 427static int get_dent_type(int mode)
 428{
 429	switch (mode & S_IFMT) {
 430	case S_IFREG:
 431		return UBIFS_ITYPE_REG;
 432	case S_IFDIR:
 433		return UBIFS_ITYPE_DIR;
 434	case S_IFLNK:
 435		return UBIFS_ITYPE_LNK;
 436	case S_IFBLK:
 437		return UBIFS_ITYPE_BLK;
 438	case S_IFCHR:
 439		return UBIFS_ITYPE_CHR;
 440	case S_IFIFO:
 441		return UBIFS_ITYPE_FIFO;
 442	case S_IFSOCK:
 443		return UBIFS_ITYPE_SOCK;
 444	default:
 445		BUG();
 446	}
 447	return 0;
 448}
 449
 450/**
 451 * pack_inode - pack an inode node.
 452 * @c: UBIFS file-system description object
 453 * @ino: buffer in which to pack inode node
 454 * @inode: inode to pack
 455 * @last: indicates the last node of the group
 456 */
 457static void pack_inode(struct ubifs_info *c, struct ubifs_ino_node *ino,
 458		       const struct inode *inode, int last)
 459{
 460	int data_len = 0, last_reference = !inode->i_nlink;
 461	struct ubifs_inode *ui = ubifs_inode(inode);
 462
 463	ino->ch.node_type = UBIFS_INO_NODE;
 464	ino_key_init_flash(c, &ino->key, inode->i_ino);
 465	ino->creat_sqnum = cpu_to_le64(ui->creat_sqnum);
 466	ino->atime_sec  = cpu_to_le64(inode->i_atime.tv_sec);
 467	ino->atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
 468	ino->ctime_sec  = cpu_to_le64(inode->i_ctime.tv_sec);
 469	ino->ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 470	ino->mtime_sec  = cpu_to_le64(inode->i_mtime.tv_sec);
 471	ino->mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 472	ino->uid   = cpu_to_le32(inode->i_uid);
 473	ino->gid   = cpu_to_le32(inode->i_gid);
 474	ino->mode  = cpu_to_le32(inode->i_mode);
 475	ino->flags = cpu_to_le32(ui->flags);
 476	ino->size  = cpu_to_le64(ui->ui_size);
 477	ino->nlink = cpu_to_le32(inode->i_nlink);
 478	ino->compr_type  = cpu_to_le16(ui->compr_type);
 479	ino->data_len    = cpu_to_le32(ui->data_len);
 480	ino->xattr_cnt   = cpu_to_le32(ui->xattr_cnt);
 481	ino->xattr_size  = cpu_to_le32(ui->xattr_size);
 482	ino->xattr_names = cpu_to_le32(ui->xattr_names);
 483	zero_ino_node_unused(ino);
 484
 485	/*
 486	 * Drop the attached data if this is a deletion inode, the data is not
 487	 * needed anymore.
 488	 */
 489	if (!last_reference) {
 490		memcpy(ino->data, ui->data, ui->data_len);
 491		data_len = ui->data_len;
 492	}
 493
 494	ubifs_prep_grp_node(c, ino, UBIFS_INO_NODE_SZ + data_len, last);
 495}
 496
 497/**
 498 * mark_inode_clean - mark UBIFS inode as clean.
 499 * @c: UBIFS file-system description object
 500 * @ui: UBIFS inode to mark as clean
 501 *
 502 * This helper function marks UBIFS inode @ui as clean by cleaning the
 503 * @ui->dirty flag and releasing its budget. Note, VFS may still treat the
 504 * inode as dirty and try to write it back, but 'ubifs_write_inode()' would
 505 * just do nothing.
 506 */
 507static void mark_inode_clean(struct ubifs_info *c, struct ubifs_inode *ui)
 508{
 509	if (ui->dirty)
 510		ubifs_release_dirty_inode_budget(c, ui);
 511	ui->dirty = 0;
 512}
 513
 514/**
 515 * ubifs_jnl_update - update inode.
 516 * @c: UBIFS file-system description object
 517 * @dir: parent inode or host inode in case of extended attributes
 518 * @nm: directory entry name
 519 * @inode: inode to update
 520 * @deletion: indicates a directory entry deletion i.e unlink or rmdir
 521 * @xent: non-zero if the directory entry is an extended attribute entry
 522 *
 523 * This function updates an inode by writing a directory entry (or extended
 524 * attribute entry), the inode itself, and the parent directory inode (or the
 525 * host inode) to the journal.
 526 *
 527 * The function writes the host inode @dir last, which is important in case of
 528 * extended attributes. Indeed, then we guarantee that if the host inode gets
 529 * synchronized (with 'fsync()'), and the write-buffer it sits in gets flushed,
 530 * the extended attribute inode gets flushed too. And this is exactly what the
 531 * user expects - synchronizing the host inode synchronizes its extended
 532 * attributes. Similarly, this guarantees that if @dir is synchronized, its
 533 * directory entry corresponding to @nm gets synchronized too.
 534 *
 535 * If the inode (@inode) or the parent directory (@dir) are synchronous, this
 536 * function synchronizes the write-buffer.
 537 *
 538 * This function marks the @dir and @inode inodes as clean and returns zero on
 539 * success. In case of failure, a negative error code is returned.
 540 */
 541int ubifs_jnl_update(struct ubifs_info *c, const struct inode *dir,
 542		     const struct qstr *nm, const struct inode *inode,
 543		     int deletion, int xent)
 544{
 545	int err, dlen, ilen, len, lnum, ino_offs, dent_offs;
 546	int aligned_dlen, aligned_ilen, sync = IS_DIRSYNC(dir);
 547	int last_reference = !!(deletion && inode->i_nlink == 0);
 548	struct ubifs_inode *ui = ubifs_inode(inode);
 549	struct ubifs_inode *dir_ui = ubifs_inode(dir);
 550	struct ubifs_dent_node *dent;
 551	struct ubifs_ino_node *ino;
 552	union ubifs_key dent_key, ino_key;
 553
 554	dbg_jnl("ino %lu, dent '%.*s', data len %d in dir ino %lu",
 555		inode->i_ino, nm->len, nm->name, ui->data_len, dir->i_ino);
 556	ubifs_assert(dir_ui->data_len == 0);
 557	ubifs_assert(mutex_is_locked(&dir_ui->ui_mutex));
 558
 559	dlen = UBIFS_DENT_NODE_SZ + nm->len + 1;
 560	ilen = UBIFS_INO_NODE_SZ;
 561
 562	/*
 563	 * If the last reference to the inode is being deleted, then there is
 564	 * no need to attach and write inode data, it is being deleted anyway.
 565	 * And if the inode is being deleted, no need to synchronize
 566	 * write-buffer even if the inode is synchronous.
 567	 */
 568	if (!last_reference) {
 569		ilen += ui->data_len;
 570		sync |= IS_SYNC(inode);
 571	}
 572
 573	aligned_dlen = ALIGN(dlen, 8);
 574	aligned_ilen = ALIGN(ilen, 8);
 575	len = aligned_dlen + aligned_ilen + UBIFS_INO_NODE_SZ;
 576	dent = kmalloc(len, GFP_NOFS);
 577	if (!dent)
 578		return -ENOMEM;
 579
 580	/* Make reservation before allocating sequence numbers */
 581	err = make_reservation(c, BASEHD, len);
 582	if (err)
 583		goto out_free;
 584
 585	if (!xent) {
 586		dent->ch.node_type = UBIFS_DENT_NODE;
 587		dent_key_init(c, &dent_key, dir->i_ino, nm);
 588	} else {
 589		dent->ch.node_type = UBIFS_XENT_NODE;
 590		xent_key_init(c, &dent_key, dir->i_ino, nm);
 591	}
 592
 593	key_write(c, &dent_key, dent->key);
 594	dent->inum = deletion ? 0 : cpu_to_le64(inode->i_ino);
 595	dent->type = get_dent_type(inode->i_mode);
 596	dent->nlen = cpu_to_le16(nm->len);
 597	memcpy(dent->name, nm->name, nm->len);
 598	dent->name[nm->len] = '\0';
 599	zero_dent_node_unused(dent);
 600	ubifs_prep_grp_node(c, dent, dlen, 0);
 601
 602	ino = (void *)dent + aligned_dlen;
 603	pack_inode(c, ino, inode, 0);
 604	ino = (void *)ino + aligned_ilen;
 605	pack_inode(c, ino, dir, 1);
 606
 607	if (last_reference) {
 608		err = ubifs_add_orphan(c, inode->i_ino);
 609		if (err) {
 610			release_head(c, BASEHD);
 611			goto out_finish;
 612		}
 613		ui->del_cmtno = c->cmt_no;
 614	}
 615
 616	err = write_head(c, BASEHD, dent, len, &lnum, &dent_offs, sync);
 617	if (err)
 618		goto out_release;
 619	if (!sync) {
 620		struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf;
 621
 622		ubifs_wbuf_add_ino_nolock(wbuf, inode->i_ino);
 623		ubifs_wbuf_add_ino_nolock(wbuf, dir->i_ino);
 624	}
 625	release_head(c, BASEHD);
 626	kfree(dent);
 627
 628	if (deletion) {
 629		err = ubifs_tnc_remove_nm(c, &dent_key, nm);
 630		if (err)
 631			goto out_ro;
 632		err = ubifs_add_dirt(c, lnum, dlen);
 633	} else
 634		err = ubifs_tnc_add_nm(c, &dent_key, lnum, dent_offs, dlen, nm);
 635	if (err)
 636		goto out_ro;
 637
 638	/*
 639	 * Note, we do not remove the inode from TNC even if the last reference
 640	 * to it has just been deleted, because the inode may still be opened.
 641	 * Instead, the inode has been added to orphan lists and the orphan
 642	 * subsystem will take further care about it.
 643	 */
 644	ino_key_init(c, &ino_key, inode->i_ino);
 645	ino_offs = dent_offs + aligned_dlen;
 646	err = ubifs_tnc_add(c, &ino_key, lnum, ino_offs, ilen);
 647	if (err)
 648		goto out_ro;
 649
 650	ino_key_init(c, &ino_key, dir->i_ino);
 651	ino_offs += aligned_ilen;
 652	err = ubifs_tnc_add(c, &ino_key, lnum, ino_offs, UBIFS_INO_NODE_SZ);
 653	if (err)
 654		goto out_ro;
 655
 656	finish_reservation(c);
 657	spin_lock(&ui->ui_lock);
 658	ui->synced_i_size = ui->ui_size;
 659	spin_unlock(&ui->ui_lock);
 660	mark_inode_clean(c, ui);
 661	mark_inode_clean(c, dir_ui);
 662	return 0;
 663
 664out_finish:
 665	finish_reservation(c);
 666out_free:
 667	kfree(dent);
 668	return err;
 669
 670out_release:
 671	release_head(c, BASEHD);
 672	kfree(dent);
 673out_ro:
 674	ubifs_ro_mode(c, err);
 675	if (last_reference)
 676		ubifs_delete_orphan(c, inode->i_ino);
 677	finish_reservation(c);
 678	return err;
 679}
 680
 681/**
 682 * ubifs_jnl_write_data - write a data node to the journal.
 683 * @c: UBIFS file-system description object
 684 * @inode: inode the data node belongs to
 685 * @key: node key
 686 * @buf: buffer to write
 687 * @len: data length (must not exceed %UBIFS_BLOCK_SIZE)
 688 *
 689 * This function writes a data node to the journal. Returns %0 if the data node
 690 * was successfully written, and a negative error code in case of failure.
 691 */
 692int ubifs_jnl_write_data(struct ubifs_info *c, const struct inode *inode,
 693			 const union ubifs_key *key, const void *buf, int len)
 694{
 695	struct ubifs_data_node *data;
 696	int err, lnum, offs, compr_type, out_len;
 697	int dlen = COMPRESSED_DATA_NODE_BUF_SZ, allocated = 1;
 698	struct ubifs_inode *ui = ubifs_inode(inode);
 699
 700	dbg_jnl("ino %lu, blk %u, len %d, key %s",
 701		(unsigned long)key_inum(c, key), key_block(c, key), len,
 702		DBGKEY(key));
 703	ubifs_assert(len <= UBIFS_BLOCK_SIZE);
 704
 705	data = kmalloc(dlen, GFP_NOFS | __GFP_NOWARN);
 706	if (!data) {
 707		/*
 708		 * Fall-back to the write reserve buffer. Note, we might be
 709		 * currently on the memory reclaim path, when the kernel is
 710		 * trying to free some memory by writing out dirty pages. The
 711		 * write reserve buffer helps us to guarantee that we are
 712		 * always able to write the data.
 713		 */
 714		allocated = 0;
 715		mutex_lock(&c->write_reserve_mutex);
 716		data = c->write_reserve_buf;
 717	}
 718
 719	data->ch.node_type = UBIFS_DATA_NODE;
 720	key_write(c, key, &data->key);
 721	data->size = cpu_to_le32(len);
 722	zero_data_node_unused(data);
 723
 724	if (!(ui->flags & UBIFS_COMPR_FL))
 725		/* Compression is disabled for this inode */
 726		compr_type = UBIFS_COMPR_NONE;
 727	else
 728		compr_type = ui->compr_type;
 729
 730	out_len = dlen - UBIFS_DATA_NODE_SZ;
 731	ubifs_compress(buf, len, &data->data, &out_len, &compr_type);
 732	ubifs_assert(out_len <= UBIFS_BLOCK_SIZE);
 733
 734	dlen = UBIFS_DATA_NODE_SZ + out_len;
 735	data->compr_type = cpu_to_le16(compr_type);
 736
 737	/* Make reservation before allocating sequence numbers */
 738	err = make_reservation(c, DATAHD, dlen);
 739	if (err)
 740		goto out_free;
 741
 742	err = write_node(c, DATAHD, data, dlen, &lnum, &offs);
 743	if (err)
 744		goto out_release;
 745	ubifs_wbuf_add_ino_nolock(&c->jheads[DATAHD].wbuf, key_inum(c, key));
 746	release_head(c, DATAHD);
 747
 748	err = ubifs_tnc_add(c, key, lnum, offs, dlen);
 749	if (err)
 750		goto out_ro;
 751
 752	finish_reservation(c);
 753	if (!allocated)
 754		mutex_unlock(&c->write_reserve_mutex);
 755	else
 756		kfree(data);
 757	return 0;
 758
 759out_release:
 760	release_head(c, DATAHD);
 761out_ro:
 762	ubifs_ro_mode(c, err);
 763	finish_reservation(c);
 764out_free:
 765	if (!allocated)
 766		mutex_unlock(&c->write_reserve_mutex);
 767	else
 768		kfree(data);
 769	return err;
 770}
 771
 772/**
 773 * ubifs_jnl_write_inode - flush inode to the journal.
 774 * @c: UBIFS file-system description object
 775 * @inode: inode to flush
 776 *
 777 * This function writes inode @inode to the journal. If the inode is
 778 * synchronous, it also synchronizes the write-buffer. Returns zero in case of
 779 * success and a negative error code in case of failure.
 780 */
 781int ubifs_jnl_write_inode(struct ubifs_info *c, const struct inode *inode)
 782{
 783	int err, lnum, offs;
 784	struct ubifs_ino_node *ino;
 785	struct ubifs_inode *ui = ubifs_inode(inode);
 786	int sync = 0, len = UBIFS_INO_NODE_SZ, last_reference = !inode->i_nlink;
 787
 788	dbg_jnl("ino %lu, nlink %u", inode->i_ino, inode->i_nlink);
 789
 790	/*
 791	 * If the inode is being deleted, do not write the attached data. No
 792	 * need to synchronize the write-buffer either.
 793	 */
 794	if (!last_reference) {
 795		len += ui->data_len;
 796		sync = IS_SYNC(inode);
 797	}
 798	ino = kmalloc(len, GFP_NOFS);
 799	if (!ino)
 800		return -ENOMEM;
 801
 802	/* Make reservation before allocating sequence numbers */
 803	err = make_reservation(c, BASEHD, len);
 804	if (err)
 805		goto out_free;
 806
 807	pack_inode(c, ino, inode, 1);
 808	err = write_head(c, BASEHD, ino, len, &lnum, &offs, sync);
 809	if (err)
 810		goto out_release;
 811	if (!sync)
 812		ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf,
 813					  inode->i_ino);
 814	release_head(c, BASEHD);
 815
 816	if (last_reference) {
 817		err = ubifs_tnc_remove_ino(c, inode->i_ino);
 818		if (err)
 819			goto out_ro;
 820		ubifs_delete_orphan(c, inode->i_ino);
 821		err = ubifs_add_dirt(c, lnum, len);
 822	} else {
 823		union ubifs_key key;
 824
 825		ino_key_init(c, &key, inode->i_ino);
 826		err = ubifs_tnc_add(c, &key, lnum, offs, len);
 827	}
 828	if (err)
 829		goto out_ro;
 830
 831	finish_reservation(c);
 832	spin_lock(&ui->ui_lock);
 833	ui->synced_i_size = ui->ui_size;
 834	spin_unlock(&ui->ui_lock);
 835	kfree(ino);
 836	return 0;
 837
 838out_release:
 839	release_head(c, BASEHD);
 840out_ro:
 841	ubifs_ro_mode(c, err);
 842	finish_reservation(c);
 843out_free:
 844	kfree(ino);
 845	return err;
 846}
 847
 848/**
 849 * ubifs_jnl_delete_inode - delete an inode.
 850 * @c: UBIFS file-system description object
 851 * @inode: inode to delete
 852 *
 853 * This function deletes inode @inode which includes removing it from orphans,
 854 * deleting it from TNC and, in some cases, writing a deletion inode to the
 855 * journal.
 856 *
 857 * When regular file inodes are unlinked or a directory inode is removed, the
 858 * 'ubifs_jnl_update()' function writes a corresponding deletion inode and
 859 * direntry to the media, and adds the inode to orphans. After this, when the
 860 * last reference to this inode has been dropped, this function is called. In
 861 * general, it has to write one more deletion inode to the media, because if
 862 * a commit happened between 'ubifs_jnl_update()' and
 863 * 'ubifs_jnl_delete_inode()', the deletion inode is not in the journal
 864 * anymore, and in fact it might not be on the flash anymore, because it might
 865 * have been garbage-collected already. And for optimization reasons UBIFS does
 866 * not read the orphan area if it has been unmounted cleanly, so it would have
 867 * no indication in the journal that there is a deleted inode which has to be
 868 * removed from TNC.
 869 *
 870 * However, if there was no commit between 'ubifs_jnl_update()' and
 871 * 'ubifs_jnl_delete_inode()', then there is no need to write the deletion
 872 * inode to the media for the second time. And this is quite a typical case.
 873 *
 874 * This function returns zero in case of success and a negative error code in
 875 * case of failure.
 876 */
 877int ubifs_jnl_delete_inode(struct ubifs_info *c, const struct inode *inode)
 878{
 879	int err;
 880	struct ubifs_inode *ui = ubifs_inode(inode);
 881
 882	ubifs_assert(inode->i_nlink == 0);
 883
 884	if (ui->del_cmtno != c->cmt_no)
 885		/* A commit happened for sure */
 886		return ubifs_jnl_write_inode(c, inode);
 887
 888	down_read(&c->commit_sem);
 889	/*
 890	 * Check commit number again, because the first test has been done
 891	 * without @c->commit_sem, so a commit might have happened.
 892	 */
 893	if (ui->del_cmtno != c->cmt_no) {
 894		up_read(&c->commit_sem);
 895		return ubifs_jnl_write_inode(c, inode);
 896	}
 897
 898	err = ubifs_tnc_remove_ino(c, inode->i_ino);
 899	if (err)
 900		ubifs_ro_mode(c, err);
 901	else
 902		ubifs_delete_orphan(c, inode->i_ino);
 903	up_read(&c->commit_sem);
 904	return err;
 905}
 906
 907/**
 908 * ubifs_jnl_rename - rename a directory entry.
 909 * @c: UBIFS file-system description object
 910 * @old_dir: parent inode of directory entry to rename
 911 * @old_dentry: directory entry to rename
 912 * @new_dir: parent inode of directory entry to rename
 913 * @new_dentry: new directory entry (or directory entry to replace)
 914 * @sync: non-zero if the write-buffer has to be synchronized
 915 *
 916 * This function implements the re-name operation which may involve writing up
 917 * to 3 inodes and 2 directory entries. It marks the written inodes as clean
 918 * and returns zero on success. In case of failure, a negative error code is
 919 * returned.
 920 */
 921int ubifs_jnl_rename(struct ubifs_info *c, const struct inode *old_dir,
 922		     const struct dentry *old_dentry,
 923		     const struct inode *new_dir,
 924		     const struct dentry *new_dentry, int sync)
 925{
 926	void *p;
 927	union ubifs_key key;
 928	struct ubifs_dent_node *dent, *dent2;
 929	int err, dlen1, dlen2, ilen, lnum, offs, len;
 930	const struct inode *old_inode = old_dentry->d_inode;
 931	const struct inode *new_inode = new_dentry->d_inode;
 932	int aligned_dlen1, aligned_dlen2, plen = UBIFS_INO_NODE_SZ;
 933	int last_reference = !!(new_inode && new_inode->i_nlink == 0);
 934	int move = (old_dir != new_dir);
 935	struct ubifs_inode *uninitialized_var(new_ui);
 936
 937	dbg_jnl("dent '%.*s' in dir ino %lu to dent '%.*s' in dir ino %lu",
 938		old_dentry->d_name.len, old_dentry->d_name.name,
 939		old_dir->i_ino, new_dentry->d_name.len,
 940		new_dentry->d_name.name, new_dir->i_ino);
 941	ubifs_assert(ubifs_inode(old_dir)->data_len == 0);
 942	ubifs_assert(ubifs_inode(new_dir)->data_len == 0);
 943	ubifs_assert(mutex_is_locked(&ubifs_inode(old_dir)->ui_mutex));
 944	ubifs_assert(mutex_is_locked(&ubifs_inode(new_dir)->ui_mutex));
 945
 946	dlen1 = UBIFS_DENT_NODE_SZ + new_dentry->d_name.len + 1;
 947	dlen2 = UBIFS_DENT_NODE_SZ + old_dentry->d_name.len + 1;
 948	if (new_inode) {
 949		new_ui = ubifs_inode(new_inode);
 950		ubifs_assert(mutex_is_locked(&new_ui->ui_mutex));
 951		ilen = UBIFS_INO_NODE_SZ;
 952		if (!last_reference)
 953			ilen += new_ui->data_len;
 954	} else
 955		ilen = 0;
 956
 957	aligned_dlen1 = ALIGN(dlen1, 8);
 958	aligned_dlen2 = ALIGN(dlen2, 8);
 959	len = aligned_dlen1 + aligned_dlen2 + ALIGN(ilen, 8) + ALIGN(plen, 8);
 960	if (old_dir != new_dir)
 961		len += plen;
 962	dent = kmalloc(len, GFP_NOFS);
 963	if (!dent)
 964		return -ENOMEM;
 965
 966	/* Make reservation before allocating sequence numbers */
 967	err = make_reservation(c, BASEHD, len);
 968	if (err)
 969		goto out_free;
 970
 971	/* Make new dent */
 972	dent->ch.node_type = UBIFS_DENT_NODE;
 973	dent_key_init_flash(c, &dent->key, new_dir->i_ino, &new_dentry->d_name);
 974	dent->inum = cpu_to_le64(old_inode->i_ino);
 975	dent->type = get_dent_type(old_inode->i_mode);
 976	dent->nlen = cpu_to_le16(new_dentry->d_name.len);
 977	memcpy(dent->name, new_dentry->d_name.name, new_dentry->d_name.len);
 978	dent->name[new_dentry->d_name.len] = '\0';
 979	zero_dent_node_unused(dent);
 980	ubifs_prep_grp_node(c, dent, dlen1, 0);
 981
 982	/* Make deletion dent */
 983	dent2 = (void *)dent + aligned_dlen1;
 984	dent2->ch.node_type = UBIFS_DENT_NODE;
 985	dent_key_init_flash(c, &dent2->key, old_dir->i_ino,
 986			    &old_dentry->d_name);
 987	dent2->inum = 0;
 988	dent2->type = DT_UNKNOWN;
 989	dent2->nlen = cpu_to_le16(old_dentry->d_name.len);
 990	memcpy(dent2->name, old_dentry->d_name.name, old_dentry->d_name.len);
 991	dent2->name[old_dentry->d_name.len] = '\0';
 992	zero_dent_node_unused(dent2);
 993	ubifs_prep_grp_node(c, dent2, dlen2, 0);
 994
 995	p = (void *)dent2 + aligned_dlen2;
 996	if (new_inode) {
 997		pack_inode(c, p, new_inode, 0);
 998		p += ALIGN(ilen, 8);
 999	}
1000
1001	if (!move)
1002		pack_inode(c, p, old_dir, 1);
1003	else {
1004		pack_inode(c, p, old_dir, 0);
1005		p += ALIGN(plen, 8);
1006		pack_inode(c, p, new_dir, 1);
1007	}
1008
1009	if (last_reference) {
1010		err = ubifs_add_orphan(c, new_inode->i_ino);
1011		if (err) {
1012			release_head(c, BASEHD);
1013			goto out_finish;
1014		}
1015		new_ui->del_cmtno = c->cmt_no;
1016	}
1017
1018	err = write_head(c, BASEHD, dent, len, &lnum, &offs, sync);
1019	if (err)
1020		goto out_release;
1021	if (!sync) {
1022		struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf;
1023
1024		ubifs_wbuf_add_ino_nolock(wbuf, new_dir->i_ino);
1025		ubifs_wbuf_add_ino_nolock(wbuf, old_dir->i_ino);
1026		if (new_inode)
1027			ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf,
1028						  new_inode->i_ino);
1029	}
1030	release_head(c, BASEHD);
1031
1032	dent_key_init(c, &key, new_dir->i_ino, &new_dentry->d_name);
1033	err = ubifs_tnc_add_nm(c, &key, lnum, offs, dlen1, &new_dentry->d_name);
1034	if (err)
1035		goto out_ro;
1036
1037	err = ubifs_add_dirt(c, lnum, dlen2);
1038	if (err)
1039		goto out_ro;
1040
1041	dent_key_init(c, &key, old_dir->i_ino, &old_dentry->d_name);
1042	err = ubifs_tnc_remove_nm(c, &key, &old_dentry->d_name);
1043	if (err)
1044		goto out_ro;
1045
1046	offs += aligned_dlen1 + aligned_dlen2;
1047	if (new_inode) {
1048		ino_key_init(c, &key, new_inode->i_ino);
1049		err = ubifs_tnc_add(c, &key, lnum, offs, ilen);
1050		if (err)
1051			goto out_ro;
1052		offs += ALIGN(ilen, 8);
1053	}
1054
1055	ino_key_init(c, &key, old_dir->i_ino);
1056	err = ubifs_tnc_add(c, &key, lnum, offs, plen);
1057	if (err)
1058		goto out_ro;
1059
1060	if (old_dir != new_dir) {
1061		offs += ALIGN(plen, 8);
1062		ino_key_init(c, &key, new_dir->i_ino);
1063		err = ubifs_tnc_add(c, &key, lnum, offs, plen);
1064		if (err)
1065			goto out_ro;
1066	}
1067
1068	finish_reservation(c);
1069	if (new_inode) {
1070		mark_inode_clean(c, new_ui);
1071		spin_lock(&new_ui->ui_lock);
1072		new_ui->synced_i_size = new_ui->ui_size;
1073		spin_unlock(&new_ui->ui_lock);
1074	}
1075	mark_inode_clean(c, ubifs_inode(old_dir));
1076	if (move)
1077		mark_inode_clean(c, ubifs_inode(new_dir));
1078	kfree(dent);
1079	return 0;
1080
1081out_release:
1082	release_head(c, BASEHD);
1083out_ro:
1084	ubifs_ro_mode(c, err);
1085	if (last_reference)
1086		ubifs_delete_orphan(c, new_inode->i_ino);
1087out_finish:
1088	finish_reservation(c);
1089out_free:
1090	kfree(dent);
1091	return err;
1092}
1093
1094/**
1095 * recomp_data_node - re-compress a truncated data node.
1096 * @dn: data node to re-compress
1097 * @new_len: new length
1098 *
1099 * This function is used when an inode is truncated and the last data node of
1100 * the inode has to be re-compressed and re-written.
1101 */
1102static int recomp_data_node(struct ubifs_data_node *dn, int *new_len)
1103{
1104	void *buf;
1105	int err, len, compr_type, out_len;
1106
1107	out_len = le32_to_cpu(dn->size);
1108	buf = kmalloc(out_len * WORST_COMPR_FACTOR, GFP_NOFS);
1109	if (!buf)
1110		return -ENOMEM;
1111
1112	len = le32_to_cpu(dn->ch.len) - UBIFS_DATA_NODE_SZ;
1113	compr_type = le16_to_cpu(dn->compr_type);
1114	err = ubifs_decompress(&dn->data, len, buf, &out_len, compr_type);
1115	if (err)
1116		goto out;
1117
1118	ubifs_compress(buf, *new_len, &dn->data, &out_len, &compr_type);
1119	ubifs_assert(out_len <= UBIFS_BLOCK_SIZE);
1120	dn->compr_type = cpu_to_le16(compr_type);
1121	dn->size = cpu_to_le32(*new_len);
1122	*new_len = UBIFS_DATA_NODE_SZ + out_len;
1123out:
1124	kfree(buf);
1125	return err;
1126}
1127
1128/**
1129 * ubifs_jnl_truncate - update the journal for a truncation.
1130 * @c: UBIFS file-system description object
1131 * @inode: inode to truncate
1132 * @old_size: old size
1133 * @new_size: new size
1134 *
1135 * When the size of a file decreases due to truncation, a truncation node is
1136 * written, the journal tree is updated, and the last data block is re-written
1137 * if it has been affected. The inode is also updated in order to synchronize
1138 * the new inode size.
1139 *
1140 * This function marks the inode as clean and returns zero on success. In case
1141 * of failure, a negative error code is returned.
1142 */
1143int ubifs_jnl_truncate(struct ubifs_info *c, const struct inode *inode,
1144		       loff_t old_size, loff_t new_size)
1145{
1146	union ubifs_key key, to_key;
1147	struct ubifs_ino_node *ino;
1148	struct ubifs_trun_node *trun;
1149	struct ubifs_data_node *uninitialized_var(dn);
1150	int err, dlen, len, lnum, offs, bit, sz, sync = IS_SYNC(inode);
1151	struct ubifs_inode *ui = ubifs_inode(inode);
1152	ino_t inum = inode->i_ino;
1153	unsigned int blk;
1154
1155	dbg_jnl("ino %lu, size %lld -> %lld",
1156		(unsigned long)inum, old_size, new_size);
1157	ubifs_assert(!ui->data_len);
1158	ubifs_assert(S_ISREG(inode->i_mode));
1159	ubifs_assert(mutex_is_locked(&ui->ui_mutex));
1160
1161	sz = UBIFS_TRUN_NODE_SZ + UBIFS_INO_NODE_SZ +
1162	     UBIFS_MAX_DATA_NODE_SZ * WORST_COMPR_FACTOR;
1163	ino = kmalloc(sz, GFP_NOFS);
1164	if (!ino)
1165		return -ENOMEM;
1166
1167	trun = (void *)ino + UBIFS_INO_NODE_SZ;
1168	trun->ch.node_type = UBIFS_TRUN_NODE;
1169	trun->inum = cpu_to_le32(inum);
1170	trun->old_size = cpu_to_le64(old_size);
1171	trun->new_size = cpu_to_le64(new_size);
1172	zero_trun_node_unused(trun);
1173
1174	dlen = new_size & (UBIFS_BLOCK_SIZE - 1);
1175	if (dlen) {
1176		/* Get last data block so it can be truncated */
1177		dn = (void *)trun + UBIFS_TRUN_NODE_SZ;
1178		blk = new_size >> UBIFS_BLOCK_SHIFT;
1179		data_key_init(c, &key, inum, blk);
1180		dbg_jnl("last block key %s", DBGKEY(&key));
1181		err = ubifs_tnc_lookup(c, &key, dn);
1182		if (err == -ENOENT)
1183			dlen = 0; /* Not found (so it is a hole) */
1184		else if (err)
1185			goto out_free;
1186		else {
1187			if (le32_to_cpu(dn->size) <= dlen)
1188				dlen = 0; /* Nothing to do */
1189			else {
1190				int compr_type = le16_to_cpu(dn->compr_type);
1191
1192				if (compr_type != UBIFS_COMPR_NONE) {
1193					err = recomp_data_node(dn, &dlen);
1194					if (err)
1195						goto out_free;
1196				} else {
1197					dn->size = cpu_to_le32(dlen);
1198					dlen += UBIFS_DATA_NODE_SZ;
1199				}
1200				zero_data_node_unused(dn);
1201			}
1202		}
1203	}
1204
1205	/* Must make reservation before allocating sequence numbers */
1206	len = UBIFS_TRUN_NODE_SZ + UBIFS_INO_NODE_SZ;
1207	if (dlen)
1208		len += dlen;
1209	err = make_reservation(c, BASEHD, len);
1210	if (err)
1211		goto out_free;
1212
1213	pack_inode(c, ino, inode, 0);
1214	ubifs_prep_grp_node(c, trun, UBIFS_TRUN_NODE_SZ, dlen ? 0 : 1);
1215	if (dlen)
1216		ubifs_prep_grp_node(c, dn, dlen, 1);
1217
1218	err = write_head(c, BASEHD, ino, len, &lnum, &offs, sync);
1219	if (err)
1220		goto out_release;
1221	if (!sync)
1222		ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf, inum);
1223	release_head(c, BASEHD);
1224
1225	if (dlen) {
1226		sz = offs + UBIFS_INO_NODE_SZ + UBIFS_TRUN_NODE_SZ;
1227		err = ubifs_tnc_add(c, &key, lnum, sz, dlen);
1228		if (err)
1229			goto out_ro;
1230	}
1231
1232	ino_key_init(c, &key, inum);
1233	err = ubifs_tnc_add(c, &key, lnum, offs, UBIFS_INO_NODE_SZ);
1234	if (err)
1235		goto out_ro;
1236
1237	err = ubifs_add_dirt(c, lnum, UBIFS_TRUN_NODE_SZ);
1238	if (err)
1239		goto out_ro;
1240
1241	bit = new_size & (UBIFS_BLOCK_SIZE - 1);
1242	blk = (new_size >> UBIFS_BLOCK_SHIFT) + (bit ? 1 : 0);
1243	data_key_init(c, &key, inum, blk);
1244
1245	bit = old_size & (UBIFS_BLOCK_SIZE - 1);
1246	blk = (old_size >> UBIFS_BLOCK_SHIFT) - (bit ? 0 : 1);
1247	data_key_init(c, &to_key, inum, blk);
1248
1249	err = ubifs_tnc_remove_range(c, &key, &to_key);
1250	if (err)
1251		goto out_ro;
1252
1253	finish_reservation(c);
1254	spin_lock(&ui->ui_lock);
1255	ui->synced_i_size = ui->ui_size;
1256	spin_unlock(&ui->ui_lock);
1257	mark_inode_clean(c, ui);
1258	kfree(ino);
1259	return 0;
1260
1261out_release:
1262	release_head(c, BASEHD);
1263out_ro:
1264	ubifs_ro_mode(c, err);
1265	finish_reservation(c);
1266out_free:
1267	kfree(ino);
1268	return err;
1269}
1270
1271#ifdef CONFIG_UBIFS_FS_XATTR
1272
1273/**
1274 * ubifs_jnl_delete_xattr - delete an extended attribute.
1275 * @c: UBIFS file-system description object
1276 * @host: host inode
1277 * @inode: extended attribute inode
1278 * @nm: extended attribute entry name
1279 *
1280 * This function delete an extended attribute which is very similar to
1281 * un-linking regular files - it writes a deletion xentry, a deletion inode and
1282 * updates the target inode. Returns zero in case of success and a negative
1283 * error code in case of failure.
1284 */
1285int ubifs_jnl_delete_xattr(struct ubifs_info *c, const struct inode *host,
1286			   const struct inode *inode, const struct qstr *nm)
1287{
1288	int err, xlen, hlen, len, lnum, xent_offs, aligned_xlen;
1289	struct ubifs_dent_node *xent;
1290	struct ubifs_ino_node *ino;
1291	union ubifs_key xent_key, key1, key2;
1292	int sync = IS_DIRSYNC(host);
1293	struct ubifs_inode *host_ui = ubifs_inode(host);
1294
1295	dbg_jnl("host %lu, xattr ino %lu, name '%s', data len %d",
1296		host->i_ino, inode->i_ino, nm->name,
1297		ubifs_inode(inode)->data_len);
1298	ubifs_assert(inode->i_nlink == 0);
1299	ubifs_assert(mutex_is_locked(&host_ui->ui_mutex));
1300
1301	/*
1302	 * Since we are deleting the inode, we do not bother to attach any data
1303	 * to it and assume its length is %UBIFS_INO_NODE_SZ.
1304	 */
1305	xlen = UBIFS_DENT_NODE_SZ + nm->len + 1;
1306	aligned_xlen = ALIGN(xlen, 8);
1307	hlen = host_ui->data_len + UBIFS_INO_NODE_SZ;
1308	len = aligned_xlen + UBIFS_INO_NODE_SZ + ALIGN(hlen, 8);
1309
1310	xent = kmalloc(len, GFP_NOFS);
1311	if (!xent)
1312		return -ENOMEM;
1313
1314	/* Make reservation before allocating sequence numbers */
1315	err = make_reservation(c, BASEHD, len);
1316	if (err) {
1317		kfree(xent);
1318		return err;
1319	}
1320
1321	xent->ch.node_type = UBIFS_XENT_NODE;
1322	xent_key_init(c, &xent_key, host->i_ino, nm);
1323	key_write(c, &xent_key, xent->key);
1324	xent->inum = 0;
1325	xent->type = get_dent_type(inode->i_mode);
1326	xent->nlen = cpu_to_le16(nm->len);
1327	memcpy(xent->name, nm->name, nm->len);
1328	xent->name[nm->len] = '\0';
1329	zero_dent_node_unused(xent);
1330	ubifs_prep_grp_node(c, xent, xlen, 0);
1331
1332	ino = (void *)xent + aligned_xlen;
1333	pack_inode(c, ino, inode, 0);
1334	ino = (void *)ino + UBIFS_INO_NODE_SZ;
1335	pack_inode(c, ino, host, 1);
1336
1337	err = write_head(c, BASEHD, xent, len, &lnum, &xent_offs, sync);
1338	if (!sync && !err)
1339		ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf, host->i_ino);
1340	release_head(c, BASEHD);
1341	kfree(xent);
1342	if (err)
1343		goto out_ro;
1344
1345	/* Remove the extended attribute entry from TNC */
1346	err = ubifs_tnc_remove_nm(c, &xent_key, nm);
1347	if (err)
1348		goto out_ro;
1349	err = ubifs_add_dirt(c, lnum, xlen);
1350	if (err)
1351		goto out_ro;
1352
1353	/*
1354	 * Remove all nodes belonging to the extended attribute inode from TNC.
1355	 * Well, there actually must be only one node - the inode itself.
1356	 */
1357	lowest_ino_key(c, &key1, inode->i_ino);
1358	highest_ino_key(c, &key2, inode->i_ino);
1359	err = ubifs_tnc_remove_range(c, &key1, &key2);
1360	if (err)
1361		goto out_ro;
1362	err = ubifs_add_dirt(c, lnum, UBIFS_INO_NODE_SZ);
1363	if (err)
1364		goto out_ro;
1365
1366	/* And update TNC with the new host inode position */
1367	ino_key_init(c, &key1, host->i_ino);
1368	err = ubifs_tnc_add(c, &key1, lnum, xent_offs + len - hlen, hlen);
1369	if (err)
1370		goto out_ro;
1371
1372	finish_reservation(c);
1373	spin_lock(&host_ui->ui_lock);
1374	host_ui->synced_i_size = host_ui->ui_size;
1375	spin_unlock(&host_ui->ui_lock);
1376	mark_inode_clean(c, host_ui);
1377	return 0;
1378
1379out_ro:
1380	ubifs_ro_mode(c, err);
1381	finish_reservation(c);
1382	return err;
1383}
1384
1385/**
1386 * ubifs_jnl_change_xattr - change an extended attribute.
1387 * @c: UBIFS file-system description object
1388 * @inode: extended attribute inode
1389 * @host: host inode
1390 *
1391 * This function writes the updated version of an extended attribute inode and
1392 * the host inode to the journal (to the base head). The host inode is written
1393 * after the extended attribute inode in order to guarantee that the extended
1394 * attribute will be flushed when the inode is synchronized by 'fsync()' and
1395 * consequently, the write-buffer is synchronized. This function returns zero
1396 * in case of success and a negative error code in case of failure.
1397 */
1398int ubifs_jnl_change_xattr(struct ubifs_info *c, const struct inode *inode,
1399			   const struct inode *host)
1400{
1401	int err, len1, len2, aligned_len, aligned_len1, lnum, offs;
1402	struct ubifs_inode *host_ui = ubifs_inode(host);
1403	struct ubifs_ino_node *ino;
1404	union ubifs_key key;
1405	int sync = IS_DIRSYNC(host);
1406
1407	dbg_jnl("ino %lu, ino %lu", host->i_ino, inode->i_ino);
1408	ubifs_assert(host->i_nlink > 0);
1409	ubifs_assert(inode->i_nlink > 0);
1410	ubifs_assert(mutex_is_locked(&host_ui->ui_mutex));
1411
1412	len1 = UBIFS_INO_NODE_SZ + host_ui->data_len;
1413	len2 = UBIFS_INO_NODE_SZ + ubifs_inode(inode)->data_len;
1414	aligned_len1 = ALIGN(len1, 8);
1415	aligned_len = aligned_len1 + ALIGN(len2, 8);
1416
1417	ino = kmalloc(aligned_len, GFP_NOFS);
1418	if (!ino)
1419		return -ENOMEM;
1420
1421	/* Make reservation before allocating sequence numbers */
1422	err = make_reservation(c, BASEHD, aligned_len);
1423	if (err)
1424		goto out_free;
1425
1426	pack_inode(c, ino, host, 0);
1427	pack_inode(c, (void *)ino + aligned_len1, inode, 1);
1428
1429	err = write_head(c, BASEHD, ino, aligned_len, &lnum, &offs, 0);
1430	if (!sync && !err) {
1431		struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf;
1432
1433		ubifs_wbuf_add_ino_nolock(wbuf, host->i_ino);
1434		ubifs_wbuf_add_ino_nolock(wbuf, inode->i_ino);
1435	}
1436	release_head(c, BASEHD);
1437	if (err)
1438		goto out_ro;
1439
1440	ino_key_init(c, &key, host->i_ino);
1441	err = ubifs_tnc_add(c, &key, lnum, offs, len1);
1442	if (err)
1443		goto out_ro;
1444
1445	ino_key_init(c, &key, inode->i_ino);
1446	err = ubifs_tnc_add(c, &key, lnum, offs + aligned_len1, len2);
1447	if (err)
1448		goto out_ro;
1449
1450	finish_reservation(c);
1451	spin_lock(&host_ui->ui_lock);
1452	host_ui->synced_i_size = host_ui->ui_size;
1453	spin_unlock(&host_ui->ui_lock);
1454	mark_inode_clean(c, host_ui);
1455	kfree(ino);
1456	return 0;
1457
1458out_ro:
1459	ubifs_ro_mode(c, err);
1460	finish_reservation(c);
1461out_free:
1462	kfree(ino);
1463	return err;
1464}
1465
1466#endif /* CONFIG_UBIFS_FS_XATTR */