fs/ubifs/journal.c

   1 /*
   2  * This file is part of UBIFS.
   3  *
   4  * Copyright (C) 2006-2008 Nokia Corporation.
   5  *
   6  * This program is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License version 2 as published by
   8  * the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13  * more details.
  14  *
  15  * You should have received a copy of the GNU General Public License along with
  16  * this program; if not, write to the Free Software Foundation, Inc., 51
  17  * Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
  18  *
  19  * Authors: Artem Bityutskiy (Битюцкий Артём)
  20  *          Adrian Hunter
  21  */
  22
  23 /*
  24  * This file implements UBIFS journal.
  25  *
  26  * The journal consists of 2 parts - the log and bud LEBs. The log has fixed
  27  * length and position, while a bud logical eraseblock is any LEB in the main
  28  * area. Buds contain file system data - data nodes, inode nodes, etc. The log
  29  * contains only references to buds and some other stuff like commit
  30  * start node. The idea is that when we commit the journal, we do
  31  * not copy the data, the buds just become indexed. Since after the commit the
  32  * nodes in bud eraseblocks become leaf nodes of the file system index tree, we
  33  * use term "bud". Analogy is obvious, bud eraseblocks contain nodes which will
  34  * become leafs in the future.
  35  *
  36  * The journal is multi-headed because we want to write data to the journal as
  37  * optimally as possible. It is nice to have nodes belonging to the same inode
  38  * in one LEB, so we may write data owned by different inodes to different
  39  * journal heads, although at present only one data head is used.
  40  *
  41  * For recovery reasons, the base head contains all inode nodes, all directory
  42  * entry nodes and all truncate nodes. This means that the other heads contain
  43  * only data nodes.
  44  *
  45  * Bud LEBs may be half-indexed. For example, if the bud was not full at the
  46  * time of commit, the bud is retained to continue to be used in the journal,
  47  * even though the "front" of the LEB is now indexed. In that case, the log
  48  * reference contains the offset where the bud starts for the purposes of the
  49  * journal.
  50  *
  51  * The journal size has to be limited, because the larger is the journal, the
  52  * longer it takes to mount UBIFS (scanning the journal) and the more memory it
  53  * takes (indexing in the TNC).
  54  *
  55  * All the journal write operations like 'ubifs_jnl_update()' here, which write
  56  * multiple UBIFS nodes to the journal at one go, are atomic with respect to
  57  * unclean reboots. Should the unclean reboot happen, the recovery code drops
  58  * all the nodes.
  59  */
  60
  61 #include "ubifs.h"
  62
  63 /**
  64  * zero_ino_node_unused - zero out unused fields of an on-flash inode node.
  65  * @ino: the inode to zero out
  66  */
  67 static inline void zero_ino_node_unused(struct ubifs_ino_node *ino)
  68 {
  69         memset(ino->padding1, 0, 4);
  70         memset(ino->padding2, 0, 26);
  71 }
  72
  73 /**
  74  * zero_dent_node_unused - zero out unused fields of an on-flash directory
  75  *                         entry node.
  76  * @dent: the directory entry to zero out
  77  */
  78 static inline void zero_dent_node_unused(struct ubifs_dent_node *dent)
  79 {
  80         dent->padding1 = 0;
  81         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  */
  88 static 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  */
  98 static 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  */
 115 static int reserve_space(struct ubifs_info *c, int jhead, int len)
 116 {
 117         int err = 0, err1, retries = 0, avail, lnum, offs, free, 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);
 126 again:
 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, &free, squeeze);
 143         if (lnum >= 0) {
 144                 /* Found an LEB, add it to the journal head */
 145                 offs = c->leb_size - free;
 146                 err = ubifs_add_bud_to_log(c, jhead, lnum, offs);
 147                 if (err)
 148                         goto out_return;
 149                 /* A new bud was successfully allocated and added to the log */
 150                 goto out;
 151         }
 152
 153         err = lnum;
 154         if (err != -ENOSPC)
 155                 goto out_unlock;
 156
 157         /*
 158          * No free space, we have to run garbage collector to make
 159          * some. But the write-buffer mutex has to be unlocked because
 160          * GC also takes it.
 161          */
 162         dbg_jnl("no free space  jhead %d, run GC", jhead);
 163         mutex_unlock(&wbuf->io_mutex);
 164
 165         lnum = ubifs_garbage_collect(c, 0);
 166         if (lnum < 0) {
 167                 err = lnum;
 168                 if (err != -ENOSPC)
 169                         return err;
 170
 171                 /*
 172                  * GC could not make a free LEB. But someone else may
 173                  * have allocated new bud for this journal head,
 174                  * because we dropped @wbuf->io_mutex, so try once
 175                  * again.
 176                  */
 177                 dbg_jnl("GC couldn't make a free LEB for jhead %d", 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 %d", lnum, 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
 210 out:
 211         err = ubifs_wbuf_seek_nolock(wbuf, lnum, offs, wbuf->dtype);
 212         if (err)
 213                 goto out_unlock;
 214
 215         return 0;
 216
 217 out_unlock:
 218         mutex_unlock(&wbuf->io_mutex);
 219         return err;
 220
 221 out_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  */
 249 static 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 %d, LEB %d:%d, len %d", jhead, *lnum, *offs, len);
 260         ubifs_prepare_node(c, node, len, 0);
 261
 262         return ubifs_wbuf_write_nolock(wbuf, node, len);
 263 }
 264
 265 /**
 266  * write_head - write data to a journal head.
 267  * @c: UBIFS file-system description object
 268  * @jhead: journal head
 269  * @buf: buffer to write
 270  * @len: length to write
 271  * @lnum: LEB number written is returned here
 272  * @offs: offset written is returned here
 273  * @sync: non-zero if the write-buffer has to by synchronized
 274  *
 275  * This function is the same as 'write_node()' but it does not assume the
 276  * buffer it is writing is a node, so it does not prepare it (which means
 277  * initializing common header and calculating CRC).
 278  */
 279 static int write_head(struct ubifs_info *c, int jhead, void *buf, int len,
 280                       int *lnum, int *offs, int sync)
 281 {
 282         int err;
 283         struct ubifs_wbuf *wbuf = &c->jheads[jhead].wbuf;
 284
 285         ubifs_assert(jhead != GCHD);
 286
 287         *lnum = c->jheads[jhead].wbuf.lnum;
 288         *offs = c->jheads[jhead].wbuf.offs + c->jheads[jhead].wbuf.used;
 289         dbg_jnl("jhead %d, LEB %d:%d, len %d", jhead, *lnum, *offs, len);
 290
 291         err = ubifs_wbuf_write_nolock(wbuf, buf, len);
 292         if (err)
 293                 return err;
 294         if (sync)
 295                 err = ubifs_wbuf_sync_nolock(wbuf);
 296         return err;
 297 }
 298
 299 /**
 300  * make_reservation - reserve journal space.
 301  * @c: UBIFS file-system description object
 302  * @jhead: journal head
 303  * @len: how many bytes to reserve
 304  *
 305  * This function makes space reservation in journal head @jhead. The function
 306  * takes the commit lock and locks the journal head, and the caller has to
 307  * unlock the head and finish the reservation with 'finish_reservation()'.
 308  * Returns zero in case of success and a negative error code in case of
 309  * failure.
 310  *
 311  * Note, the journal head may be unlocked as soon as the data is written, while
 312  * the commit lock has to be released after the data has been added to the
 313  * TNC.
 314  */
 315 static int make_reservation(struct ubifs_info *c, int jhead, int len)
 316 {
 317         int err, cmt_retries = 0, nospc_retries = 0;
 318
 319 again:
 320         down_read(&c->commit_sem);
 321         err = reserve_space(c, jhead, len);
 322         if (!err)
 323                 return 0;
 324         up_read(&c->commit_sem);
 325
 326         if (err == -ENOSPC) {
 327                 /*
 328                  * GC could not make any progress. We should try to commit
 329                  * once because it could make some dirty space and GC would
 330                  * make progress, so make the error -EAGAIN so that the below
 331                  * will commit and re-try.
 332                  */
 333                 if (nospc_retries++ < 2) {
 334                         dbg_jnl("no space, retry");
 335                         err = -EAGAIN;
 336                 }
 337
 338                 /*
 339                  * This means that the budgeting is incorrect. We always have
 340                  * to be able to write to the media, because all operations are
 341                  * budgeted. Deletions are not budgeted, though, but we reserve
 342                  * an extra LEB for them.
 343                  */
 344         }
 345
 346         if (err != -EAGAIN)
 347                 goto out;
 348
 349         /*
 350          * -EAGAIN means that the journal is full or too large, or the above
 351          * code wants to do one commit. Do this and re-try.
 352          */
 353         if (cmt_retries > 128) {
 354                 /*
 355                  * This should not happen unless the journal size limitations
 356                  * are too tough.
 357                  */
 358                 ubifs_err("stuck in space allocation");
 359                 err = -ENOSPC;
 360                 goto out;
 361         } else if (cmt_retries > 32)
 362                 ubifs_warn("too many space allocation re-tries (%d)",
 363                            cmt_retries);
 364
 365         dbg_jnl("-EAGAIN, commit and retry (retried %d times)",
 366                 cmt_retries);
 367         cmt_retries += 1;
 368
 369         err = ubifs_run_commit(c);
 370         if (err)
 371                 return err;
 372         goto again;
 373
 374 out:
 375         ubifs_err("cannot reserve %d bytes in jhead %d, error %d",
 376                   len, jhead, err);
 377         if (err == -ENOSPC) {
 378                 /* This are some budgeting problems, print useful information */
 379                 down_write(&c->commit_sem);
 380                 spin_lock(&c->space_lock);
 381                 dbg_dump_stack();
 382                 dbg_dump_budg(c);
 383                 spin_unlock(&c->space_lock);
 384                 dbg_dump_lprops(c);
 385                 cmt_retries = dbg_check_lprops(c);
 386                 up_write(&c->commit_sem);
 387         }
 388         return err;
 389 }
 390
 391 /**
 392  * release_head - release a journal head.
 393  * @c: UBIFS file-system description object
 394  * @jhead: journal head
 395  *
 396  * This function releases journal head @jhead which was locked by
 397  * the 'make_reservation()' function. It has to be called after each successful
 398  * 'make_reservation()' invocation.
 399  */
 400 static inline void release_head(struct ubifs_info *c, int jhead)
 401 {
 402         mutex_unlock(&c->jheads[jhead].wbuf.io_mutex);
 403 }
 404
 405 /**
 406  * finish_reservation - finish a reservation.
 407  * @c: UBIFS file-system description object
 408  *
 409  * This function finishes journal space reservation. It must be called after
 410  * 'make_reservation()'.
 411  */
 412 static void finish_reservation(struct ubifs_info *c)
 413 {
 414         up_read(&c->commit_sem);
 415 }
 416
 417 /**
 418  * get_dent_type - translate VFS inode mode to UBIFS directory entry type.
 419  * @mode: inode mode
 420  */
 421 static int get_dent_type(int mode)
 422 {
 423         switch (mode & S_IFMT) {
 424         case S_IFREG:
 425                 return UBIFS_ITYPE_REG;
 426         case S_IFDIR:
 427                 return UBIFS_ITYPE_DIR;
 428         case S_IFLNK:
 429                 return UBIFS_ITYPE_LNK;
 430         case S_IFBLK:
 431                 return UBIFS_ITYPE_BLK;
 432         case S_IFCHR:
 433                 return UBIFS_ITYPE_CHR;
 434         case S_IFIFO:
 435                 return UBIFS_ITYPE_FIFO;
 436         case S_IFSOCK:
 437                 return UBIFS_ITYPE_SOCK;
 438         default:
 439                 BUG();
 440         }
 441         return 0;
 442 }
 443
 444 /**
 445  * pack_inode - pack an inode node.
 446  * @c: UBIFS file-system description object
 447  * @ino: buffer in which to pack inode node
 448  * @inode: inode to pack
 449  * @last: indicates the last node of the group
 450  */
 451 static void pack_inode(struct ubifs_info *c, struct ubifs_ino_node *ino,
 452                        const struct inode *inode, int last)
 453 {
 454         int data_len = 0, last_reference = !inode->i_nlink;
 455         struct ubifs_inode *ui = ubifs_inode(inode);
 456
 457         ino->ch.node_type = UBIFS_INO_NODE;
 458         ino_key_init_flash(c, &ino->key, inode->i_ino);
 459         ino->creat_sqnum = cpu_to_le64(ui->creat_sqnum);
 460         ino->atime_sec  = cpu_to_le64(inode->i_atime.tv_sec);
 461         ino->atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
 462         ino->ctime_sec  = cpu_to_le64(inode->i_ctime.tv_sec);
 463         ino->ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 464         ino->mtime_sec  = cpu_to_le64(inode->i_mtime.tv_sec);
 465         ino->mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 466         ino->uid   = cpu_to_le32(inode->i_uid);
 467         ino->gid   = cpu_to_le32(inode->i_gid);
 468         ino->mode  = cpu_to_le32(inode->i_mode);
 469         ino->flags = cpu_to_le32(ui->flags);
 470         ino->size  = cpu_to_le64(ui->ui_size);
 471         ino->nlink = cpu_to_le32(inode->i_nlink);
 472         ino->compr_type  = cpu_to_le16(ui->compr_type);
 473         ino->data_len    = cpu_to_le32(ui->data_len);
 474         ino->xattr_cnt   = cpu_to_le32(ui->xattr_cnt);
 475         ino->xattr_size  = cpu_to_le32(ui->xattr_size);
 476         ino->xattr_names = cpu_to_le32(ui->xattr_names);
 477         zero_ino_node_unused(ino);
 478
 479         /*
 480          * Drop the attached data if this is a deletion inode, the data is not
 481          * needed anymore.
 482          */
 483         if (!last_reference) {
 484                 memcpy(ino->data, ui->data, ui->data_len);
 485                 data_len = ui->data_len;
 486         }
 487
 488         ubifs_prep_grp_node(c, ino, UBIFS_INO_NODE_SZ + data_len, last);
 489 }
 490
 491 /**
 492  * mark_inode_clean - mark UBIFS inode as clean.
 493  * @c: UBIFS file-system description object
 494  * @ui: UBIFS inode to mark as clean
 495  *
 496  * This helper function marks UBIFS inode @ui as clean by cleaning the
 497  * @ui->dirty flag and releasing its budget. Note, VFS may still treat the
 498  * inode as dirty and try to write it back, but 'ubifs_write_inode()' would
 499  * just do nothing.
 500  */
 501 static void mark_inode_clean(struct ubifs_info *c, struct ubifs_inode *ui)
 502 {
 503         if (ui->dirty)
 504                 ubifs_release_dirty_inode_budget(c, ui);
 505         ui->dirty = 0;
 506 }
 507
 508 /**
 509  * ubifs_jnl_update - update inode.
 510  * @c: UBIFS file-system description object
 511  * @dir: parent inode or host inode in case of extended attributes
 512  * @nm: directory entry name
 513  * @inode: inode to update
 514  * @deletion: indicates a directory entry deletion i.e unlink or rmdir
 515  * @xent: non-zero if the directory entry is an extended attribute entry
 516  *
 517  * This function updates an inode by writing a directory entry (or extended
 518  * attribute entry), the inode itself, and the parent directory inode (or the
 519  * host inode) to the journal.
 520  *
 521  * The function writes the host inode @dir last, which is important in case of
 522  * extended attributes. Indeed, then we guarantee that if the host inode gets
 523  * synchronized (with 'fsync()'), and the write-buffer it sits in gets flushed,
 524  * the extended attribute inode gets flushed too. And this is exactly what the
 525  * user expects - synchronizing the host inode synchronizes its extended
 526  * attributes. Similarly, this guarantees that if @dir is synchronized, its
 527  * directory entry corresponding to @nm gets synchronized too.
 528  *
 529  * If the inode (@inode) or the parent directory (@dir) are synchronous, this
 530  * function synchronizes the write-buffer.
 531  *
 532  * This function marks the @dir and @inode inodes as clean and returns zero on
 533  * success. In case of failure, a negative error code is returned.
 534  */
 535 int ubifs_jnl_update(struct ubifs_info *c, const struct inode *dir,
 536                      const struct qstr *nm, const struct inode *inode,
 537                      int deletion, int xent)
 538 {
 539         int err, dlen, ilen, len, lnum, ino_offs, dent_offs;
 540         int aligned_dlen, aligned_ilen, sync = IS_DIRSYNC(dir);
 541         int last_reference = !!(deletion && inode->i_nlink == 0);
 542         struct ubifs_inode *ui = ubifs_inode(inode);
 543         struct ubifs_inode *dir_ui = ubifs_inode(dir);
 544         struct ubifs_dent_node *dent;
 545         struct ubifs_ino_node *ino;
 546         union ubifs_key dent_key, ino_key;
 547
 548         dbg_jnl("ino %lu, dent '%.*s', data len %d in dir ino %lu",
 549                 inode->i_ino, nm->len, nm->name, ui->data_len, dir->i_ino);
 550         ubifs_assert(dir_ui->data_len == 0);
 551         ubifs_assert(mutex_is_locked(&dir_ui->ui_mutex));
 552
 553         dlen = UBIFS_DENT_NODE_SZ + nm->len + 1;
 554         ilen = UBIFS_INO_NODE_SZ;
 555
 556         /*
 557          * If the last reference to the inode is being deleted, then there is
 558          * no need to attach and write inode data, it is being deleted anyway.
 559          * And if the inode is being deleted, no need to synchronize
 560          * write-buffer even if the inode is synchronous.
 561          */
 562         if (!last_reference) {
 563                 ilen += ui->data_len;
 564                 sync |= IS_SYNC(inode);
 565         }
 566
 567         aligned_dlen = ALIGN(dlen, 8);
 568         aligned_ilen = ALIGN(ilen, 8);
 569         len = aligned_dlen + aligned_ilen + UBIFS_INO_NODE_SZ;
 570         dent = kmalloc(len, GFP_NOFS);
 571         if (!dent)
 572                 return -ENOMEM;
 573
 574         /* Make reservation before allocating sequence numbers */
 575         err = make_reservation(c, BASEHD, len);
 576         if (err)
 577                 goto out_free;
 578
 579         if (!xent) {
 580                 dent->ch.node_type = UBIFS_DENT_NODE;
 581                 dent_key_init(c, &dent_key, dir->i_ino, nm);
 582         } else {
 583                 dent->ch.node_type = UBIFS_XENT_NODE;
 584                 xent_key_init(c, &dent_key, dir->i_ino, nm);
 585         }
 586
 587         key_write(c, &dent_key, dent->key);
 588         dent->inum = deletion ? 0 : cpu_to_le64(inode->i_ino);
 589         dent->type = get_dent_type(inode->i_mode);
 590         dent->nlen = cpu_to_le16(nm->len);
 591         memcpy(dent->name, nm->name, nm->len);
 592         dent->name[nm->len] = '\0';
 593         zero_dent_node_unused(dent);
 594         ubifs_prep_grp_node(c, dent, dlen, 0);
 595
 596         ino = (void *)dent + aligned_dlen;
 597         pack_inode(c, ino, inode, 0);
 598         ino = (void *)ino + aligned_ilen;
 599         pack_inode(c, ino, dir, 1);
 600
 601         if (last_reference) {
 602                 err = ubifs_add_orphan(c, inode->i_ino);
 603                 if (err) {
 604                         release_head(c, BASEHD);
 605                         goto out_finish;
 606                 }
 607                 ui->del_cmtno = c->cmt_no;
 608         }
 609
 610         err = write_head(c, BASEHD, dent, len, &lnum, &dent_offs, sync);
 611         if (err)
 612                 goto out_release;
 613         if (!sync) {
 614                 struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf;
 615
 616                 ubifs_wbuf_add_ino_nolock(wbuf, inode->i_ino);
 617                 ubifs_wbuf_add_ino_nolock(wbuf, dir->i_ino);
 618         }
 619         release_head(c, BASEHD);
 620         kfree(dent);
 621
 622         if (deletion) {
 623                 err = ubifs_tnc_remove_nm(c, &dent_key, nm);
 624                 if (err)
 625                         goto out_ro;
 626                 err = ubifs_add_dirt(c, lnum, dlen);
 627         } else
 628                 err = ubifs_tnc_add_nm(c, &dent_key, lnum, dent_offs, dlen, nm);
 629         if (err)
 630                 goto out_ro;
 631
 632         /*
 633          * Note, we do not remove the inode from TNC even if the last reference
 634          * to it has just been deleted, because the inode may still be opened.
 635          * Instead, the inode has been added to orphan lists and the orphan
 636          * subsystem will take further care about it.
 637          */
 638         ino_key_init(c, &ino_key, inode->i_ino);
 639         ino_offs = dent_offs + aligned_dlen;
 640         err = ubifs_tnc_add(c, &ino_key, lnum, ino_offs, ilen);
 641         if (err)
 642                 goto out_ro;
 643
 644         ino_key_init(c, &ino_key, dir->i_ino);
 645         ino_offs += aligned_ilen;
 646         err = ubifs_tnc_add(c, &ino_key, lnum, ino_offs, UBIFS_INO_NODE_SZ);
 647         if (err)
 648                 goto out_ro;
 649
 650         finish_reservation(c);
 651         spin_lock(&ui->ui_lock);
 652         ui->synced_i_size = ui->ui_size;
 653         spin_unlock(&ui->ui_lock);
 654         mark_inode_clean(c, ui);
 655         mark_inode_clean(c, dir_ui);
 656         return 0;
 657
 658 out_finish:
 659         finish_reservation(c);
 660 out_free:
 661         kfree(dent);
 662         return err;
 663
 664 out_release:
 665         release_head(c, BASEHD);
 666 out_ro:
 667         ubifs_ro_mode(c, err);
 668         if (last_reference)
 669                 ubifs_delete_orphan(c, inode->i_ino);
 670         finish_reservation(c);
 671         return err;
 672 }
 673
 674 /**
 675  * ubifs_jnl_write_data - write a data node to the journal.
 676  * @c: UBIFS file-system description object
 677  * @inode: inode the data node belongs to
 678  * @key: node key
 679  * @buf: buffer to write
 680  * @len: data length (must not exceed %UBIFS_BLOCK_SIZE)
 681  *
 682  * This function writes a data node to the journal. Returns %0 if the data node
 683  * was successfully written, and a negative error code in case of failure.
 684  */
 685 int ubifs_jnl_write_data(struct ubifs_info *c, const struct inode *inode,
 686                          const union ubifs_key *key, const void *buf, int len)
 687 {
 688         struct ubifs_data_node *data;
 689         int err, lnum, offs, compr_type, out_len;
 690         int dlen = UBIFS_DATA_NODE_SZ + UBIFS_BLOCK_SIZE * WORST_COMPR_FACTOR;
 691         struct ubifs_inode *ui = ubifs_inode(inode);
 692
 693         dbg_jnl("ino %lu, blk %u, len %d, key %s",
 694                 (unsigned long)key_inum(c, key), key_block(c, key), len,
 695                 DBGKEY(key));
 696         ubifs_assert(len <= UBIFS_BLOCK_SIZE);
 697
 698         data = kmalloc(dlen, GFP_NOFS);
 699         if (!data)
 700                 return -ENOMEM;
 701
 702         data->ch.node_type = UBIFS_DATA_NODE;
 703         key_write(c, key, &data->key);
 704         data->size = cpu_to_le32(len);
 705         zero_data_node_unused(data);
 706
 707         if (!(ui->flags & UBIFS_COMPR_FL))
 708                 /* Compression is disabled for this inode */
 709                 compr_type = UBIFS_COMPR_NONE;
 710         else
 711                 compr_type = ui->compr_type;
 712
 713         out_len = dlen - UBIFS_DATA_NODE_SZ;
 714         ubifs_compress(buf, len, &data->data, &out_len, &compr_type);
 715         ubifs_assert(out_len <= UBIFS_BLOCK_SIZE);
 716
 717         dlen = UBIFS_DATA_NODE_SZ + out_len;
 718         data->compr_type = cpu_to_le16(compr_type);
 719
 720         /* Make reservation before allocating sequence numbers */
 721         err = make_reservation(c, DATAHD, dlen);
 722         if (err)
 723                 goto out_free;
 724
 725         err = write_node(c, DATAHD, data, dlen, &lnum, &offs);
 726         if (err)
 727                 goto out_release;
 728         ubifs_wbuf_add_ino_nolock(&c->jheads[DATAHD].wbuf, key_inum(c, key));
 729         release_head(c, DATAHD);
 730
 731         err = ubifs_tnc_add(c, key, lnum, offs, dlen);
 732         if (err)
 733                 goto out_ro;
 734
 735         finish_reservation(c);
 736         kfree(data);
 737         return 0;
 738
 739 out_release:
 740         release_head(c, DATAHD);
 741 out_ro:
 742         ubifs_ro_mode(c, err);
 743         finish_reservation(c);
 744 out_free:
 745         kfree(data);
 746         return err;
 747 }
 748
 749 /**
 750  * ubifs_jnl_write_inode - flush inode to the journal.
 751  * @c: UBIFS file-system description object
 752  * @inode: inode to flush
 753  *
 754  * This function writes inode @inode to the journal. If the inode is
 755  * synchronous, it also synchronizes the write-buffer. Returns zero in case of
 756  * success and a negative error code in case of failure.
 757  */
 758 int ubifs_jnl_write_inode(struct ubifs_info *c, const struct inode *inode)
 759 {
 760         int err, lnum, offs;
 761         struct ubifs_ino_node *ino;
 762         struct ubifs_inode *ui = ubifs_inode(inode);
 763         int sync = 0, len = UBIFS_INO_NODE_SZ, last_reference = !inode->i_nlink;
 764
 765         dbg_jnl("ino %lu, nlink %u", inode->i_ino, inode->i_nlink);
 766
 767         /*
 768          * If the inode is being deleted, do not write the attached data. No
 769          * need to synchronize the write-buffer either.
 770          */
 771         if (!last_reference) {
 772                 len += ui->data_len;
 773                 sync = IS_SYNC(inode);
 774         }
 775         ino = kmalloc(len, GFP_NOFS);
 776         if (!ino)
 777                 return -ENOMEM;
 778
 779         /* Make reservation before allocating sequence numbers */
 780         err = make_reservation(c, BASEHD, len);
 781         if (err)
 782                 goto out_free;
 783
 784         pack_inode(c, ino, inode, 1);
 785         err = write_head(c, BASEHD, ino, len, &lnum, &offs, sync);
 786         if (err)
 787                 goto out_release;
 788         if (!sync)
 789                 ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf,
 790                                           inode->i_ino);
 791         release_head(c, BASEHD);
 792
 793         if (last_reference) {
 794                 err = ubifs_tnc_remove_ino(c, inode->i_ino);
 795                 if (err)
 796                         goto out_ro;
 797                 ubifs_delete_orphan(c, inode->i_ino);
 798                 err = ubifs_add_dirt(c, lnum, len);
 799         } else {
 800                 union ubifs_key key;
 801
 802                 ino_key_init(c, &key, inode->i_ino);
 803                 err = ubifs_tnc_add(c, &key, lnum, offs, len);
 804         }
 805         if (err)
 806                 goto out_ro;
 807
 808         finish_reservation(c);
 809         spin_lock(&ui->ui_lock);
 810         ui->synced_i_size = ui->ui_size;
 811         spin_unlock(&ui->ui_lock);
 812         kfree(ino);
 813         return 0;
 814
 815 out_release:
 816         release_head(c, BASEHD);
 817 out_ro:
 818         ubifs_ro_mode(c, err);
 819         finish_reservation(c);
 820 out_free:
 821         kfree(ino);
 822         return err;
 823 }
 824
 825 /**
 826  * ubifs_jnl_delete_inode - delete an inode.
 827  * @c: UBIFS file-system description object
 828  * @inode: inode to delete
 829  *
 830  * This function deletes inode @inode which includes removing it from orphans,
 831  * deleting it from TNC and, in some cases, writing a deletion inode to the
 832  * journal.
 833  *
 834  * When regular file inodes are unlinked or a directory inode is removed, the
 835  * 'ubifs_jnl_update()' function writes a corresponding deletion inode and
 836  * direntry to the media, and adds the inode to orphans. After this, when the
 837  * last reference to this inode has been dropped, this function is called. In
 838  * general, it has to write one more deletion inode to the media, because if
 839  * a commit happened between 'ubifs_jnl_update()' and
 840  * 'ubifs_jnl_delete_inode()', the deletion inode is not in the journal
 841  * anymore, and in fact it might not be on the flash anymore, because it might
 842  * have been garbage-collected already. And for optimization reasons UBIFS does
 843  * not read the orphan area if it has been unmounted cleanly, so it would have
 844  * no indication in the journal that there is a deleted inode which has to be
 845  * removed from TNC.
 846  *
 847  * However, if there was no commit between 'ubifs_jnl_update()' and
 848  * 'ubifs_jnl_delete_inode()', then there is no need to write the deletion
 849  * inode to the media for the second time. And this is quite a typical case.
 850  *
 851  * This function returns zero in case of success and a negative error code in
 852  * case of failure.
 853  */
 854 int ubifs_jnl_delete_inode(struct ubifs_info *c, const struct inode *inode)
 855 {
 856         int err;
 857         struct ubifs_inode *ui = ubifs_inode(inode);
 858
 859         ubifs_assert(inode->i_nlink == 0);
 860
 861         if (ui->del_cmtno != c->cmt_no)
 862                 /* A commit happened for sure */
 863                 return ubifs_jnl_write_inode(c, inode);
 864
 865         down_read(&c->commit_sem);
 866         /*
 867          * Check commit number again, because the first test has been done
 868          * without @c->commit_sem, so a commit might have happened.
 869          */
 870         if (ui->del_cmtno != c->cmt_no) {
 871                 up_read(&c->commit_sem);
 872                 return ubifs_jnl_write_inode(c, inode);
 873         }
 874
 875         err = ubifs_tnc_remove_ino(c, inode->i_ino);
 876         if (err)
 877                 ubifs_ro_mode(c, err);
 878         else
 879                 ubifs_delete_orphan(c, inode->i_ino);
 880         up_read(&c->commit_sem);
 881         return err;
 882 }
 883
 884 /**
 885  * ubifs_jnl_rename - rename a directory entry.
 886  * @c: UBIFS file-system description object
 887  * @old_dir: parent inode of directory entry to rename
 888  * @old_dentry: directory entry to rename
 889  * @new_dir: parent inode of directory entry to rename
 890  * @new_dentry: new directory entry (or directory entry to replace)
 891  * @sync: non-zero if the write-buffer has to be synchronized
 892  *
 893  * This function implements the re-name operation which may involve writing up
 894  * to 3 inodes and 2 directory entries. It marks the written inodes as clean
 895  * and returns zero on success. In case of failure, a negative error code is
 896  * returned.
 897  */
 898 int ubifs_jnl_rename(struct ubifs_info *c, const struct inode *old_dir,
 899                      const struct dentry *old_dentry,
 900                      const struct inode *new_dir,
 901                      const struct dentry *new_dentry, int sync)
 902 {
 903         void *p;
 904         union ubifs_key key;
 905         struct ubifs_dent_node *dent, *dent2;
 906         int err, dlen1, dlen2, ilen, lnum, offs, len;
 907         const struct inode *old_inode = old_dentry->d_inode;
 908         const struct inode *new_inode = new_dentry->d_inode;
 909         int aligned_dlen1, aligned_dlen2, plen = UBIFS_INO_NODE_SZ;
 910         int last_reference = !!(new_inode && new_inode->i_nlink == 0);
 911         int move = (old_dir != new_dir);
 912         struct ubifs_inode *uninitialized_var(new_ui);
 913
 914         dbg_jnl("dent '%.*s' in dir ino %lu to dent '%.*s' in dir ino %lu",
 915                 old_dentry->d_name.len, old_dentry->d_name.name,
 916                 old_dir->i_ino, new_dentry->d_name.len,
 917                 new_dentry->d_name.name, new_dir->i_ino);
 918         ubifs_assert(ubifs_inode(old_dir)->data_len == 0);
 919         ubifs_assert(ubifs_inode(new_dir)->data_len == 0);
 920         ubifs_assert(mutex_is_locked(&ubifs_inode(old_dir)->ui_mutex));
 921         ubifs_assert(mutex_is_locked(&ubifs_inode(new_dir)->ui_mutex));
 922
 923         dlen1 = UBIFS_DENT_NODE_SZ + new_dentry->d_name.len + 1;
 924         dlen2 = UBIFS_DENT_NODE_SZ + old_dentry->d_name.len + 1;
 925         if (new_inode) {
 926                 new_ui = ubifs_inode(new_inode);
 927                 ubifs_assert(mutex_is_locked(&new_ui->ui_mutex));
 928                 ilen = UBIFS_INO_NODE_SZ;
 929                 if (!last_reference)
 930                         ilen += new_ui->data_len;
 931         } else
 932                 ilen = 0;
 933
 934         aligned_dlen1 = ALIGN(dlen1, 8);
 935         aligned_dlen2 = ALIGN(dlen2, 8);
 936         len = aligned_dlen1 + aligned_dlen2 + ALIGN(ilen, 8) + ALIGN(plen, 8);
 937         if (old_dir != new_dir)
 938                 len += plen;
 939         dent = kmalloc(len, GFP_NOFS);
 940         if (!dent)
 941                 return -ENOMEM;
 942
 943         /* Make reservation before allocating sequence numbers */
 944         err = make_reservation(c, BASEHD, len);
 945         if (err)
 946                 goto out_free;
 947
 948         /* Make new dent */
 949         dent->ch.node_type = UBIFS_DENT_NODE;
 950         dent_key_init_flash(c, &dent->key, new_dir->i_ino, &new_dentry->d_name);
 951         dent->inum = cpu_to_le64(old_inode->i_ino);
 952         dent->type = get_dent_type(old_inode->i_mode);
 953         dent->nlen = cpu_to_le16(new_dentry->d_name.len);
 954         memcpy(dent->name, new_dentry->d_name.name, new_dentry->d_name.len);
 955         dent->name[new_dentry->d_name.len] = '\0';
 956         zero_dent_node_unused(dent);
 957         ubifs_prep_grp_node(c, dent, dlen1, 0);
 958
 959         /* Make deletion dent */
 960         dent2 = (void *)dent + aligned_dlen1;
 961         dent2->ch.node_type = UBIFS_DENT_NODE;
 962         dent_key_init_flash(c, &dent2->key, old_dir->i_ino,
 963                             &old_dentry->d_name);
 964         dent2->inum = 0;
 965         dent2->type = DT_UNKNOWN;
 966         dent2->nlen = cpu_to_le16(old_dentry->d_name.len);
 967         memcpy(dent2->name, old_dentry->d_name.name, old_dentry->d_name.len);
 968         dent2->name[old_dentry->d_name.len] = '\0';
 969         zero_dent_node_unused(dent2);
 970         ubifs_prep_grp_node(c, dent2, dlen2, 0);
 971
 972         p = (void *)dent2 + aligned_dlen2;
 973         if (new_inode) {
 974                 pack_inode(c, p, new_inode, 0);
 975                 p += ALIGN(ilen, 8);
 976         }
 977
 978         if (!move)
 979                 pack_inode(c, p, old_dir, 1);
 980         else {
 981                 pack_inode(c, p, old_dir, 0);
 982                 p += ALIGN(plen, 8);
 983                 pack_inode(c, p, new_dir, 1);
 984         }
 985
 986         if (last_reference) {
 987                 err = ubifs_add_orphan(c, new_inode->i_ino);
 988                 if (err) {
 989                         release_head(c, BASEHD);
 990                         goto out_finish;
 991                 }
 992                 new_ui->del_cmtno = c->cmt_no;
 993         }
 994
 995         err = write_head(c, BASEHD, dent, len, &lnum, &offs, sync);
 996         if (err)
 997                 goto out_release;
 998         if (!sync) {
 999                 struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf;
1000
1001                 ubifs_wbuf_add_ino_nolock(wbuf, new_dir->i_ino);
1002                 ubifs_wbuf_add_ino_nolock(wbuf, old_dir->i_ino);
1003                 if (new_inode)
1004                         ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf,
1005                                                   new_inode->i_ino);
1006         }
1007         release_head(c, BASEHD);
1008
1009         dent_key_init(c, &key, new_dir->i_ino, &new_dentry->d_name);
1010         err = ubifs_tnc_add_nm(c, &key, lnum, offs, dlen1, &new_dentry->d_name);
1011         if (err)
1012                 goto out_ro;
1013
1014         err = ubifs_add_dirt(c, lnum, dlen2);
1015         if (err)
1016                 goto out_ro;
1017
1018         dent_key_init(c, &key, old_dir->i_ino, &old_dentry->d_name);
1019         err = ubifs_tnc_remove_nm(c, &key, &old_dentry->d_name);
1020         if (err)
1021                 goto out_ro;
1022
1023         offs += aligned_dlen1 + aligned_dlen2;
1024         if (new_inode) {
1025                 ino_key_init(c, &key, new_inode->i_ino);
1026                 err = ubifs_tnc_add(c, &key, lnum, offs, ilen);
1027                 if (err)
1028                         goto out_ro;
1029                 offs += ALIGN(ilen, 8);
1030         }
1031
1032         ino_key_init(c, &key, old_dir->i_ino);
1033         err = ubifs_tnc_add(c, &key, lnum, offs, plen);
1034         if (err)
1035                 goto out_ro;
1036
1037         if (old_dir != new_dir) {
1038                 offs += ALIGN(plen, 8);
1039                 ino_key_init(c, &key, new_dir->i_ino);
1040                 err = ubifs_tnc_add(c, &key, lnum, offs, plen);
1041                 if (err)
1042                         goto out_ro;
1043         }
1044
1045         finish_reservation(c);
1046         if (new_inode) {
1047                 mark_inode_clean(c, new_ui);
1048                 spin_lock(&new_ui->ui_lock);
1049                 new_ui->synced_i_size = new_ui->ui_size;
1050                 spin_unlock(&new_ui->ui_lock);
1051         }
1052         mark_inode_clean(c, ubifs_inode(old_dir));
1053         if (move)
1054                 mark_inode_clean(c, ubifs_inode(new_dir));
1055         kfree(dent);
1056         return 0;
1057
1058 out_release:
1059         release_head(c, BASEHD);
1060 out_ro:
1061         ubifs_ro_mode(c, err);
1062         if (last_reference)
1063                 ubifs_delete_orphan(c, new_inode->i_ino);
1064 out_finish:
1065         finish_reservation(c);
1066 out_free:
1067         kfree(dent);
1068         return err;
1069 }
1070
1071 /**
1072  * recomp_data_node - re-compress a truncated data node.
1073  * @dn: data node to re-compress
1074  * @new_len: new length
1075  *
1076  * This function is used when an inode is truncated and the last data node of
1077  * the inode has to be re-compressed and re-written.
1078  */
1079 static int recomp_data_node(struct ubifs_data_node *dn, int *new_len)
1080 {
1081         void *buf;
1082         int err, len, compr_type, out_len;
1083
1084         out_len = le32_to_cpu(dn->size);
1085         buf = kmalloc(out_len * WORST_COMPR_FACTOR, GFP_NOFS);
1086         if (!buf)
1087                 return -ENOMEM;
1088
1089         len = le32_to_cpu(dn->ch.len) - UBIFS_DATA_NODE_SZ;
1090         compr_type = le16_to_cpu(dn->compr_type);
1091         err = ubifs_decompress(&dn->data, len, buf, &out_len, compr_type);
1092         if (err)
1093                 goto out;
1094
1095         ubifs_compress(buf, *new_len, &dn->data, &out_len, &compr_type);
1096         ubifs_assert(out_len <= UBIFS_BLOCK_SIZE);
1097         dn->compr_type = cpu_to_le16(compr_type);
1098         dn->size = cpu_to_le32(*new_len);
1099         *new_len = UBIFS_DATA_NODE_SZ + out_len;
1100 out:
1101         kfree(buf);
1102         return err;
1103 }
1104
1105 /**
1106  * ubifs_jnl_truncate - update the journal for a truncation.
1107  * @c: UBIFS file-system description object
1108  * @inode: inode to truncate
1109  * @old_size: old size
1110  * @new_size: new size
1111  *
1112  * When the size of a file decreases due to truncation, a truncation node is
1113  * written, the journal tree is updated, and the last data block is re-written
1114  * if it has been affected. The inode is also updated in order to synchronize
1115  * the new inode size.
1116  *
1117  * This function marks the inode as clean and returns zero on success. In case
1118  * of failure, a negative error code is returned.
1119  */
1120 int ubifs_jnl_truncate(struct ubifs_info *c, const struct inode *inode,
1121                        loff_t old_size, loff_t new_size)
1122 {
1123         union ubifs_key key, to_key;
1124         struct ubifs_ino_node *ino;
1125         struct ubifs_trun_node *trun;
1126         struct ubifs_data_node *uninitialized_var(dn);
1127         int err, dlen, len, lnum, offs, bit, sz, sync = IS_SYNC(inode);
1128         struct ubifs_inode *ui = ubifs_inode(inode);
1129         ino_t inum = inode->i_ino;
1130         unsigned int blk;
1131
1132         dbg_jnl("ino %lu, size %lld -> %lld",
1133                 (unsigned long)inum, old_size, new_size);
1134         ubifs_assert(!ui->data_len);
1135         ubifs_assert(S_ISREG(inode->i_mode));
1136         ubifs_assert(mutex_is_locked(&ui->ui_mutex));
1137
1138         sz = UBIFS_TRUN_NODE_SZ + UBIFS_INO_NODE_SZ +
1139              UBIFS_MAX_DATA_NODE_SZ * WORST_COMPR_FACTOR;
1140         ino = kmalloc(sz, GFP_NOFS);
1141         if (!ino)
1142                 return -ENOMEM;
1143
1144         trun = (void *)ino + UBIFS_INO_NODE_SZ;
1145         trun->ch.node_type = UBIFS_TRUN_NODE;
1146         trun->inum = cpu_to_le32(inum);
1147         trun->old_size = cpu_to_le64(old_size);
1148         trun->new_size = cpu_to_le64(new_size);
1149         zero_trun_node_unused(trun);
1150
1151         dlen = new_size & (UBIFS_BLOCK_SIZE - 1);
1152         if (dlen) {
1153                 /* Get last data block so it can be truncated */
1154                 dn = (void *)trun + UBIFS_TRUN_NODE_SZ;
1155                 blk = new_size >> UBIFS_BLOCK_SHIFT;
1156                 data_key_init(c, &key, inum, blk);
1157                 dbg_jnl("last block key %s", DBGKEY(&key));
1158                 err = ubifs_tnc_lookup(c, &key, dn);
1159                 if (err == -ENOENT)
1160                         dlen = 0; /* Not found (so it is a hole) */
1161                 else if (err)
1162                         goto out_free;
1163                 else {
1164                         if (le32_to_cpu(dn->size) <= dlen)
1165                                 dlen = 0; /* Nothing to do */
1166                         else {
1167                                 int compr_type = le16_to_cpu(dn->compr_type);
1168
1169                                 if (compr_type != UBIFS_COMPR_NONE) {
1170                                         err = recomp_data_node(dn, &dlen);
1171                                         if (err)
1172                                                 goto out_free;
1173                                 } else {
1174                                         dn->size = cpu_to_le32(dlen);
1175                                         dlen += UBIFS_DATA_NODE_SZ;
1176                                 }
1177                                 zero_data_node_unused(dn);
1178                         }
1179                 }
1180         }
1181
1182         /* Must make reservation before allocating sequence numbers */
1183         len = UBIFS_TRUN_NODE_SZ + UBIFS_INO_NODE_SZ;
1184         if (dlen)
1185                 len += dlen;
1186         err = make_reservation(c, BASEHD, len);
1187         if (err)
1188                 goto out_free;
1189
1190         pack_inode(c, ino, inode, 0);
1191         ubifs_prep_grp_node(c, trun, UBIFS_TRUN_NODE_SZ, dlen ? 0 : 1);
1192         if (dlen)
1193                 ubifs_prep_grp_node(c, dn, dlen, 1);
1194
1195         err = write_head(c, BASEHD, ino, len, &lnum, &offs, sync);
1196         if (err)
1197                 goto out_release;
1198         if (!sync)
1199                 ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf, inum);
1200         release_head(c, BASEHD);
1201
1202         if (dlen) {
1203                 sz = offs + UBIFS_INO_NODE_SZ + UBIFS_TRUN_NODE_SZ;
1204                 err = ubifs_tnc_add(c, &key, lnum, sz, dlen);
1205                 if (err)
1206                         goto out_ro;
1207         }
1208
1209         ino_key_init(c, &key, inum);
1210         err = ubifs_tnc_add(c, &key, lnum, offs, UBIFS_INO_NODE_SZ);
1211         if (err)
1212                 goto out_ro;
1213
1214         err = ubifs_add_dirt(c, lnum, UBIFS_TRUN_NODE_SZ);
1215         if (err)
1216                 goto out_ro;
1217
1218         bit = new_size & (UBIFS_BLOCK_SIZE - 1);
1219         blk = (new_size >> UBIFS_BLOCK_SHIFT) + (bit ? 1 : 0);
1220         data_key_init(c, &key, inum, blk);
1221
1222         bit = old_size & (UBIFS_BLOCK_SIZE - 1);
1223         blk = (old_size >> UBIFS_BLOCK_SHIFT) - (bit ? 0 : 1);
1224         data_key_init(c, &to_key, inum, blk);
1225
1226         err = ubifs_tnc_remove_range(c, &key, &to_key);
1227         if (err)
1228                 goto out_ro;
1229
1230         finish_reservation(c);
1231         spin_lock(&ui->ui_lock);
1232         ui->synced_i_size = ui->ui_size;
1233         spin_unlock(&ui->ui_lock);
1234         mark_inode_clean(c, ui);
1235         kfree(ino);
1236         return 0;
1237
1238 out_release:
1239         release_head(c, BASEHD);
1240 out_ro:
1241         ubifs_ro_mode(c, err);
1242         finish_reservation(c);
1243 out_free:
1244         kfree(ino);
1245         return err;
1246 }
1247
1248 #ifdef CONFIG_UBIFS_FS_XATTR
1249
1250 /**
1251  * ubifs_jnl_delete_xattr - delete an extended attribute.
1252  * @c: UBIFS file-system description object
1253  * @host: host inode
1254  * @inode: extended attribute inode
1255  * @nm: extended attribute entry name
1256  *
1257  * This function delete an extended attribute which is very similar to
1258  * un-linking regular files - it writes a deletion xentry, a deletion inode and
1259  * updates the target inode. Returns zero in case of success and a negative
1260  * error code in case of failure.
1261  */
1262 int ubifs_jnl_delete_xattr(struct ubifs_info *c, const struct inode *host,
1263                            const struct inode *inode, const struct qstr *nm)
1264 {
1265         int err, xlen, hlen, len, lnum, xent_offs, aligned_xlen;
1266         struct ubifs_dent_node *xent;
1267         struct ubifs_ino_node *ino;
1268         union ubifs_key xent_key, key1, key2;
1269         int sync = IS_DIRSYNC(host);
1270         struct ubifs_inode *host_ui = ubifs_inode(host);
1271
1272         dbg_jnl("host %lu, xattr ino %lu, name '%s', data len %d",
1273                 host->i_ino, inode->i_ino, nm->name,
1274                 ubifs_inode(inode)->data_len);
1275         ubifs_assert(inode->i_nlink == 0);
1276         ubifs_assert(mutex_is_locked(&host_ui->ui_mutex));
1277
1278         /*
1279          * Since we are deleting the inode, we do not bother to attach any data
1280          * to it and assume its length is %UBIFS_INO_NODE_SZ.
1281          */
1282         xlen = UBIFS_DENT_NODE_SZ + nm->len + 1;
1283         aligned_xlen = ALIGN(xlen, 8);
1284         hlen = host_ui->data_len + UBIFS_INO_NODE_SZ;
1285         len = aligned_xlen + UBIFS_INO_NODE_SZ + ALIGN(hlen, 8);
1286
1287         xent = kmalloc(len, GFP_NOFS);
1288         if (!xent)
1289                 return -ENOMEM;
1290
1291         /* Make reservation before allocating sequence numbers */
1292         err = make_reservation(c, BASEHD, len);
1293         if (err) {
1294                 kfree(xent);
1295                 return err;
1296         }
1297
1298         xent->ch.node_type = UBIFS_XENT_NODE;
1299         xent_key_init(c, &xent_key, host->i_ino, nm);
1300         key_write(c, &xent_key, xent->key);
1301         xent->inum = 0;
1302         xent->type = get_dent_type(inode->i_mode);
1303         xent->nlen = cpu_to_le16(nm->len);
1304         memcpy(xent->name, nm->name, nm->len);
1305         xent->name[nm->len] = '\0';
1306         zero_dent_node_unused(xent);
1307         ubifs_prep_grp_node(c, xent, xlen, 0);
1308
1309         ino = (void *)xent + aligned_xlen;
1310         pack_inode(c, ino, inode, 0);
1311         ino = (void *)ino + UBIFS_INO_NODE_SZ;
1312         pack_inode(c, ino, host, 1);
1313
1314         err = write_head(c, BASEHD, xent, len, &lnum, &xent_offs, sync);
1315         if (!sync && !err)
1316                 ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf, host->i_ino);
1317         release_head(c, BASEHD);
1318         kfree(xent);
1319         if (err)
1320                 goto out_ro;
1321
1322         /* Remove the extended attribute entry from TNC */
1323         err = ubifs_tnc_remove_nm(c, &xent_key, nm);
1324         if (err)
1325                 goto out_ro;
1326         err = ubifs_add_dirt(c, lnum, xlen);
1327         if (err)
1328                 goto out_ro;
1329
1330         /*
1331          * Remove all nodes belonging to the extended attribute inode from TNC.
1332          * Well, there actually must be only one node - the inode itself.
1333          */
1334         lowest_ino_key(c, &key1, inode->i_ino);
1335         highest_ino_key(c, &key2, inode->i_ino);
1336         err = ubifs_tnc_remove_range(c, &key1, &key2);
1337         if (err)
1338                 goto out_ro;
1339         err = ubifs_add_dirt(c, lnum, UBIFS_INO_NODE_SZ);
1340         if (err)
1341                 goto out_ro;
1342
1343         /* And update TNC with the new host inode position */
1344         ino_key_init(c, &key1, host->i_ino);
1345         err = ubifs_tnc_add(c, &key1, lnum, xent_offs + len - hlen, hlen);
1346         if (err)
1347                 goto out_ro;
1348
1349         finish_reservation(c);
1350         spin_lock(&host_ui->ui_lock);
1351         host_ui->synced_i_size = host_ui->ui_size;
1352         spin_unlock(&host_ui->ui_lock);
1353         mark_inode_clean(c, host_ui);
1354         return 0;
1355
1356 out_ro:
1357         ubifs_ro_mode(c, err);
1358         finish_reservation(c);
1359         return err;
1360 }
1361
1362 /**
1363  * ubifs_jnl_change_xattr - change an extended attribute.
1364  * @c: UBIFS file-system description object
1365  * @inode: extended attribute inode
1366  * @host: host inode
1367  *
1368  * This function writes the updated version of an extended attribute inode and
1369  * the host inode tho the journal (to the base head). The host inode is written
1370  * after the extended attribute inode in order to guarantee that the extended
1371  * attribute will be flushed when the inode is synchronized by 'fsync()' and
1372  * consequently, the write-buffer is synchronized. This function returns zero
1373  * in case of success and a negative error code in case of failure.
1374  */
1375 int ubifs_jnl_change_xattr(struct ubifs_info *c, const struct inode *inode,
1376                            const struct inode *host)
1377 {
1378         int err, len1, len2, aligned_len, aligned_len1, lnum, offs;
1379         struct ubifs_inode *host_ui = ubifs_inode(host);
1380         struct ubifs_ino_node *ino;
1381         union ubifs_key key;
1382         int sync = IS_DIRSYNC(host);
1383
1384         dbg_jnl("ino %lu, ino %lu", host->i_ino, inode->i_ino);
1385         ubifs_assert(host->i_nlink > 0);
1386         ubifs_assert(inode->i_nlink > 0);
1387         ubifs_assert(mutex_is_locked(&host_ui->ui_mutex));
1388
1389         len1 = UBIFS_INO_NODE_SZ + host_ui->data_len;
1390         len2 = UBIFS_INO_NODE_SZ + ubifs_inode(inode)->data_len;
1391         aligned_len1 = ALIGN(len1, 8);
1392         aligned_len = aligned_len1 + ALIGN(len2, 8);
1393
1394         ino = kmalloc(aligned_len, GFP_NOFS);
1395         if (!ino)
1396                 return -ENOMEM;
1397
1398         /* Make reservation before allocating sequence numbers */
1399         err = make_reservation(c, BASEHD, aligned_len);
1400         if (err)
1401                 goto out_free;
1402
1403         pack_inode(c, ino, host, 0);
1404         pack_inode(c, (void *)ino + aligned_len1, inode, 1);
1405
1406         err = write_head(c, BASEHD, ino, aligned_len, &lnum, &offs, 0);
1407         if (!sync && !err) {
1408                 struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf;
1409
1410                 ubifs_wbuf_add_ino_nolock(wbuf, host->i_ino);
1411                 ubifs_wbuf_add_ino_nolock(wbuf, inode->i_ino);
1412         }
1413         release_head(c, BASEHD);
1414         if (err)
1415                 goto out_ro;
1416
1417         ino_key_init(c, &key, host->i_ino);
1418         err = ubifs_tnc_add(c, &key, lnum, offs, len1);
1419         if (err)
1420                 goto out_ro;
1421
1422         ino_key_init(c, &key, inode->i_ino);
1423         err = ubifs_tnc_add(c, &key, lnum, offs + aligned_len1, len2);
1424         if (err)
1425                 goto out_ro;
1426
1427         finish_reservation(c);
1428         spin_lock(&host_ui->ui_lock);
1429         host_ui->synced_i_size = host_ui->ui_size;
1430         spin_unlock(&host_ui->ui_lock);
1431         mark_inode_clean(c, host_ui);
1432         kfree(ino);
1433         return 0;
1434
1435 out_ro:
1436         ubifs_ro_mode(c, err);
1437         finish_reservation(c);
1438 out_free:
1439         kfree(ino);
1440         return err;
1441 }
1442
1443 #endif /* CONFIG_UBIFS_FS_XATTR */