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