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