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