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