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 *host_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(mutex_is_locked(&host_ui->ui_mutex));
 557
 558         dlen = UBIFS_DENT_NODE_SZ + nm->len + 1;
 559         ilen = UBIFS_INO_NODE_SZ;
 560
 561         /*
 562          * If the last reference to the inode is being deleted, then there is
 563          * no need to attach and write inode data, it is being deleted anyway.
 564          * And if the inode is being deleted, no need to synchronize
 565          * write-buffer even if the inode is synchronous.
 566          */
 567         if (!last_reference) {
 568                 ilen += ui->data_len;
 569                 sync |= IS_SYNC(inode);
 570         }
 571
 572         aligned_dlen = ALIGN(dlen, 8);
 573         aligned_ilen = ALIGN(ilen, 8);
 574
 575         len = aligned_dlen + aligned_ilen + UBIFS_INO_NODE_SZ;
 576         /* Make sure to also account for extended attributes */
 577         len += host_ui->data_len;
 578
 579         dent = kmalloc(len, GFP_NOFS);
 580         if (!dent)
 581                 return -ENOMEM;
 582
 583         /* Make reservation before allocating sequence numbers */
 584         err = make_reservation(c, BASEHD, len);
 585         if (err)
 586                 goto out_free;
 587
 588         if (!xent) {
 589                 dent->ch.node_type = UBIFS_DENT_NODE;
 590                 dent_key_init(c, &dent_key, dir->i_ino, nm);
 591         } else {
 592                 dent->ch.node_type = UBIFS_XENT_NODE;
 593                 xent_key_init(c, &dent_key, dir->i_ino, nm);
 594         }
 595
 596         key_write(c, &dent_key, dent->key);
 597         dent->inum = deletion ? 0 : cpu_to_le64(inode->i_ino);
 598         dent->type = get_dent_type(inode->i_mode);
 599         dent->nlen = cpu_to_le16(nm->len);
 600         memcpy(dent->name, nm->name, nm->len);
 601         dent->name[nm->len] = '\0';
 602         zero_dent_node_unused(dent);
 603         ubifs_prep_grp_node(c, dent, dlen, 0);
 604
 605         ino = (void *)dent + aligned_dlen;
 606         pack_inode(c, ino, inode, 0);
 607         ino = (void *)ino + aligned_ilen;
 608         pack_inode(c, ino, dir, 1);
 609
 610         if (last_reference) {
 611                 err = ubifs_add_orphan(c, inode->i_ino);
 612                 if (err) {
 613                         release_head(c, BASEHD);
 614                         goto out_finish;
 615                 }
 616                 ui->del_cmtno = c->cmt_no;
 617         }
 618
 619         err = write_head(c, BASEHD, dent, len, &lnum, &dent_offs, sync);
 620         if (err)
 621                 goto out_release;
 622         if (!sync) {
 623                 struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf;
 624
 625                 ubifs_wbuf_add_ino_nolock(wbuf, inode->i_ino);
 626                 ubifs_wbuf_add_ino_nolock(wbuf, dir->i_ino);
 627         }
 628         release_head(c, BASEHD);
 629         kfree(dent);
 630
 631         if (deletion) {
 632                 err = ubifs_tnc_remove_nm(c, &dent_key, nm);
 633                 if (err)
 634                         goto out_ro;
 635                 err = ubifs_add_dirt(c, lnum, dlen);
 636         } else
 637                 err = ubifs_tnc_add_nm(c, &dent_key, lnum, dent_offs, dlen, nm);
 638         if (err)
 639                 goto out_ro;
 640
 641         /*
 642          * Note, we do not remove the inode from TNC even if the last reference
 643          * to it has just been deleted, because the inode may still be opened.
 644          * Instead, the inode has been added to orphan lists and the orphan
 645          * subsystem will take further care about it.
 646          */
 647         ino_key_init(c, &ino_key, inode->i_ino);
 648         ino_offs = dent_offs + aligned_dlen;
 649         err = ubifs_tnc_add(c, &ino_key, lnum, ino_offs, ilen);
 650         if (err)
 651                 goto out_ro;
 652
 653         ino_key_init(c, &ino_key, dir->i_ino);
 654         ino_offs += aligned_ilen;
 655         err = ubifs_tnc_add(c, &ino_key, lnum, ino_offs,
 656                             UBIFS_INO_NODE_SZ + host_ui->data_len);
 657         if (err)
 658                 goto out_ro;
 659
 660         finish_reservation(c);
 661         spin_lock(&ui->ui_lock);
 662         ui->synced_i_size = ui->ui_size;
 663         spin_unlock(&ui->ui_lock);
 664         mark_inode_clean(c, ui);
 665         mark_inode_clean(c, host_ui);
 666         return 0;
 667
 668 out_finish:
 669         finish_reservation(c);
 670 out_free:
 671         kfree(dent);
 672         return err;
 673
 674 out_release:
 675         release_head(c, BASEHD);
 676         kfree(dent);
 677 out_ro:
 678         ubifs_ro_mode(c, err);
 679         if (last_reference)
 680                 ubifs_delete_orphan(c, inode->i_ino);
 681         finish_reservation(c);
 682         return err;
 683 }
 684
 685 /**
 686  * ubifs_jnl_write_data - write a data node to the journal.
 687  * @c: UBIFS file-system description object
 688  * @inode: inode the data node belongs to
 689  * @key: node key
 690  * @buf: buffer to write
 691  * @len: data length (must not exceed %UBIFS_BLOCK_SIZE)
 692  *
 693  * This function writes a data node to the journal. Returns %0 if the data node
 694  * was successfully written, and a negative error code in case of failure.
 695  */
 696 int ubifs_jnl_write_data(struct ubifs_info *c, const struct inode *inode,
 697                          const union ubifs_key *key, const void *buf, int len)
 698 {
 699         struct ubifs_data_node *data;
 700         int err, lnum, offs, compr_type, out_len;
 701         int dlen = COMPRESSED_DATA_NODE_BUF_SZ, allocated = 1;
 702         struct ubifs_inode *ui = ubifs_inode(inode);
 703
 704         dbg_jnlk(key, "ino %lu, blk %u, len %d, key ",
 705                 (unsigned long)key_inum(c, key), key_block(c, key), len);
 706         ubifs_assert(len <= UBIFS_BLOCK_SIZE);
 707
 708         data = kmalloc(dlen, GFP_NOFS | __GFP_NOWARN);
 709         if (!data) {
 710                 /*
 711                  * Fall-back to the write reserve buffer. Note, we might be
 712                  * currently on the memory reclaim path, when the kernel is
 713                  * trying to free some memory by writing out dirty pages. The
 714                  * write reserve buffer helps us to guarantee that we are
 715                  * always able to write the data.
 716                  */
 717                 allocated = 0;
 718                 mutex_lock(&c->write_reserve_mutex);
 719                 data = c->write_reserve_buf;
 720         }
 721
 722         data->ch.node_type = UBIFS_DATA_NODE;
 723         key_write(c, key, &data->key);
 724         data->size = cpu_to_le32(len);
 725         zero_data_node_unused(data);
 726
 727         if (!(ui->flags & UBIFS_COMPR_FL))
 728                 /* Compression is disabled for this inode */
 729                 compr_type = UBIFS_COMPR_NONE;
 730         else
 731                 compr_type = ui->compr_type;
 732
 733         out_len = dlen - UBIFS_DATA_NODE_SZ;
 734         ubifs_compress(buf, len, &data->data, &out_len, &compr_type);
 735         ubifs_assert(out_len <= UBIFS_BLOCK_SIZE);
 736
 737         dlen = UBIFS_DATA_NODE_SZ + out_len;
 738         data->compr_type = cpu_to_le16(compr_type);
 739
 740         /* Make reservation before allocating sequence numbers */
 741         err = make_reservation(c, DATAHD, dlen);
 742         if (err)
 743                 goto out_free;
 744
 745         err = write_node(c, DATAHD, data, dlen, &lnum, &offs);
 746         if (err)
 747                 goto out_release;
 748         ubifs_wbuf_add_ino_nolock(&c->jheads[DATAHD].wbuf, key_inum(c, key));
 749         release_head(c, DATAHD);
 750
 751         err = ubifs_tnc_add(c, key, lnum, offs, dlen);
 752         if (err)
 753                 goto out_ro;
 754
 755         finish_reservation(c);
 756         if (!allocated)
 757                 mutex_unlock(&c->write_reserve_mutex);
 758         else
 759                 kfree(data);
 760         return 0;
 761
 762 out_release:
 763         release_head(c, DATAHD);
 764 out_ro:
 765         ubifs_ro_mode(c, err);
 766         finish_reservation(c);
 767 out_free:
 768         if (!allocated)
 769                 mutex_unlock(&c->write_reserve_mutex);
 770         else
 771                 kfree(data);
 772         return err;
 773 }
 774
 775 /**
 776  * ubifs_jnl_write_inode - flush inode to the journal.
 777  * @c: UBIFS file-system description object
 778  * @inode: inode to flush
 779  *
 780  * This function writes inode @inode to the journal. If the inode is
 781  * synchronous, it also synchronizes the write-buffer. Returns zero in case of
 782  * success and a negative error code in case of failure.
 783  */
 784 int ubifs_jnl_write_inode(struct ubifs_info *c, const struct inode *inode)
 785 {
 786         int err, lnum, offs;
 787         struct ubifs_ino_node *ino;
 788         struct ubifs_inode *ui = ubifs_inode(inode);
 789         int sync = 0, len = UBIFS_INO_NODE_SZ, last_reference = !inode->i_nlink;
 790
 791         dbg_jnl("ino %lu, nlink %u", inode->i_ino, inode->i_nlink);
 792
 793         /*
 794          * If the inode is being deleted, do not write the attached data. No
 795          * need to synchronize the write-buffer either.
 796          */
 797         if (!last_reference) {
 798                 len += ui->data_len;
 799                 sync = IS_SYNC(inode);
 800         }
 801         ino = kmalloc(len, GFP_NOFS);
 802         if (!ino)
 803                 return -ENOMEM;
 804
 805         /* Make reservation before allocating sequence numbers */
 806         err = make_reservation(c, BASEHD, len);
 807         if (err)
 808                 goto out_free;
 809
 810         pack_inode(c, ino, inode, 1);
 811         err = write_head(c, BASEHD, ino, len, &lnum, &offs, sync);
 812         if (err)
 813                 goto out_release;
 814         if (!sync)
 815                 ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf,
 816                                           inode->i_ino);
 817         release_head(c, BASEHD);
 818
 819         if (last_reference) {
 820                 err = ubifs_tnc_remove_ino(c, inode->i_ino);
 821                 if (err)
 822                         goto out_ro;
 823                 ubifs_delete_orphan(c, inode->i_ino);
 824                 err = ubifs_add_dirt(c, lnum, len);
 825         } else {
 826                 union ubifs_key key;
 827
 828                 ino_key_init(c, &key, inode->i_ino);
 829                 err = ubifs_tnc_add(c, &key, lnum, offs, len);
 830         }
 831         if (err)
 832                 goto out_ro;
 833
 834         finish_reservation(c);
 835         spin_lock(&ui->ui_lock);
 836         ui->synced_i_size = ui->ui_size;
 837         spin_unlock(&ui->ui_lock);
 838         kfree(ino);
 839         return 0;
 840
 841 out_release:
 842         release_head(c, BASEHD);
 843 out_ro:
 844         ubifs_ro_mode(c, err);
 845         finish_reservation(c);
 846 out_free:
 847         kfree(ino);
 848         return err;
 849 }
 850
 851 /**
 852  * ubifs_jnl_delete_inode - delete an inode.
 853  * @c: UBIFS file-system description object
 854  * @inode: inode to delete
 855  *
 856  * This function deletes inode @inode which includes removing it from orphans,
 857  * deleting it from TNC and, in some cases, writing a deletion inode to the
 858  * journal.
 859  *
 860  * When regular file inodes are unlinked or a directory inode is removed, the
 861  * 'ubifs_jnl_update()' function writes a corresponding deletion inode and
 862  * direntry to the media, and adds the inode to orphans. After this, when the
 863  * last reference to this inode has been dropped, this function is called. In
 864  * general, it has to write one more deletion inode to the media, because if
 865  * a commit happened between 'ubifs_jnl_update()' and
 866  * 'ubifs_jnl_delete_inode()', the deletion inode is not in the journal
 867  * anymore, and in fact it might not be on the flash anymore, because it might
 868  * have been garbage-collected already. And for optimization reasons UBIFS does
 869  * not read the orphan area if it has been unmounted cleanly, so it would have
 870  * no indication in the journal that there is a deleted inode which has to be
 871  * removed from TNC.
 872  *
 873  * However, if there was no commit between 'ubifs_jnl_update()' and
 874  * 'ubifs_jnl_delete_inode()', then there is no need to write the deletion
 875  * inode to the media for the second time. And this is quite a typical case.
 876  *
 877  * This function returns zero in case of success and a negative error code in
 878  * case of failure.
 879  */
 880 int ubifs_jnl_delete_inode(struct ubifs_info *c, const struct inode *inode)
 881 {
 882         int err;
 883         struct ubifs_inode *ui = ubifs_inode(inode);
 884
 885         ubifs_assert(inode->i_nlink == 0);
 886
 887         if (ui->del_cmtno != c->cmt_no)
 888                 /* A commit happened for sure */
 889                 return ubifs_jnl_write_inode(c, inode);
 890
 891         down_read(&c->commit_sem);
 892         /*
 893          * Check commit number again, because the first test has been done
 894          * without @c->commit_sem, so a commit might have happened.
 895          */
 896         if (ui->del_cmtno != c->cmt_no) {
 897                 up_read(&c->commit_sem);
 898                 return ubifs_jnl_write_inode(c, inode);
 899         }
 900
 901         err = ubifs_tnc_remove_ino(c, inode->i_ino);
 902         if (err)
 903                 ubifs_ro_mode(c, err);
 904         else
 905                 ubifs_delete_orphan(c, inode->i_ino);
 906         up_read(&c->commit_sem);
 907         return err;
 908 }
 909
 910 /**
 911  * ubifs_jnl_rename - rename a directory entry.
 912  * @c: UBIFS file-system description object
 913  * @old_dir: parent inode of directory entry to rename
 914  * @old_dentry: directory entry to rename
 915  * @new_dir: parent inode of directory entry to rename
 916  * @new_dentry: new directory entry (or directory entry to replace)
 917  * @sync: non-zero if the write-buffer has to be synchronized
 918  *
 919  * This function implements the re-name operation which may involve writing up
 920  * to 3 inodes and 2 directory entries. It marks the written inodes as clean
 921  * and returns zero on success. In case of failure, a negative error code is
 922  * returned.
 923  */
 924 int ubifs_jnl_rename(struct ubifs_info *c, const struct inode *old_dir,
 925                      const struct dentry *old_dentry,
 926                      const struct inode *new_dir,
 927                      const struct dentry *new_dentry, int sync)
 928 {
 929         void *p;
 930         union ubifs_key key;
 931         struct ubifs_dent_node *dent, *dent2;
 932         int err, dlen1, dlen2, ilen, lnum, offs, len;
 933         const struct inode *old_inode = old_dentry->d_inode;
 934         const struct inode *new_inode = new_dentry->d_inode;
 935         int aligned_dlen1, aligned_dlen2, plen = UBIFS_INO_NODE_SZ;
 936         int last_reference = !!(new_inode && new_inode->i_nlink == 0);
 937         int move = (old_dir != new_dir);
 938         struct ubifs_inode *uninitialized_var(new_ui);
 939
 940         dbg_jnl("dent '%pd' in dir ino %lu to dent '%pd' in dir ino %lu",
 941                 old_dentry, old_dir->i_ino, new_dentry, new_dir->i_ino);
 942         ubifs_assert(ubifs_inode(old_dir)->data_len == 0);
 943         ubifs_assert(ubifs_inode(new_dir)->data_len == 0);
 944         ubifs_assert(mutex_is_locked(&ubifs_inode(old_dir)->ui_mutex));
 945         ubifs_assert(mutex_is_locked(&ubifs_inode(new_dir)->ui_mutex));
 946
 947         dlen1 = UBIFS_DENT_NODE_SZ + new_dentry->d_name.len + 1;
 948         dlen2 = UBIFS_DENT_NODE_SZ + old_dentry->d_name.len + 1;
 949         if (new_inode) {
 950                 new_ui = ubifs_inode(new_inode);
 951                 ubifs_assert(mutex_is_locked(&new_ui->ui_mutex));
 952                 ilen = UBIFS_INO_NODE_SZ;
 953                 if (!last_reference)
 954                         ilen += new_ui->data_len;
 955         } else
 956                 ilen = 0;
 957
 958         aligned_dlen1 = ALIGN(dlen1, 8);
 959         aligned_dlen2 = ALIGN(dlen2, 8);
 960         len = aligned_dlen1 + aligned_dlen2 + ALIGN(ilen, 8) + ALIGN(plen, 8);
 961         if (old_dir != new_dir)
 962                 len += plen;
 963         dent = kmalloc(len, GFP_NOFS);
 964         if (!dent)
 965                 return -ENOMEM;
 966
 967         /* Make reservation before allocating sequence numbers */
 968         err = make_reservation(c, BASEHD, len);
 969         if (err)
 970                 goto out_free;
 971
 972         /* Make new dent */
 973         dent->ch.node_type = UBIFS_DENT_NODE;
 974         dent_key_init_flash(c, &dent->key, new_dir->i_ino, &new_dentry->d_name);
 975         dent->inum = cpu_to_le64(old_inode->i_ino);
 976         dent->type = get_dent_type(old_inode->i_mode);
 977         dent->nlen = cpu_to_le16(new_dentry->d_name.len);
 978         memcpy(dent->name, new_dentry->d_name.name, new_dentry->d_name.len);
 979         dent->name[new_dentry->d_name.len] = '\0';
 980         zero_dent_node_unused(dent);
 981         ubifs_prep_grp_node(c, dent, dlen1, 0);
 982
 983         /* Make deletion dent */
 984         dent2 = (void *)dent + aligned_dlen1;
 985         dent2->ch.node_type = UBIFS_DENT_NODE;
 986         dent_key_init_flash(c, &dent2->key, old_dir->i_ino,
 987                             &old_dentry->d_name);
 988         dent2->inum = 0;
 989         dent2->type = DT_UNKNOWN;
 990         dent2->nlen = cpu_to_le16(old_dentry->d_name.len);
 991         memcpy(dent2->name, old_dentry->d_name.name, old_dentry->d_name.len);
 992         dent2->name[old_dentry->d_name.len] = '\0';
 993         zero_dent_node_unused(dent2);
 994         ubifs_prep_grp_node(c, dent2, dlen2, 0);
 995
 996         p = (void *)dent2 + aligned_dlen2;
 997         if (new_inode) {
 998                 pack_inode(c, p, new_inode, 0);
 999                 p += ALIGN(ilen, 8);
1000         }
1001
1002         if (!move)
1003                 pack_inode(c, p, old_dir, 1);
1004         else {
1005                 pack_inode(c, p, old_dir, 0);
1006                 p += ALIGN(plen, 8);
1007                 pack_inode(c, p, new_dir, 1);
1008         }
1009
1010         if (last_reference) {
1011                 err = ubifs_add_orphan(c, new_inode->i_ino);
1012                 if (err) {
1013                         release_head(c, BASEHD);
1014                         goto out_finish;
1015                 }
1016                 new_ui->del_cmtno = c->cmt_no;
1017         }
1018
1019         err = write_head(c, BASEHD, dent, len, &lnum, &offs, sync);
1020         if (err)
1021                 goto out_release;
1022         if (!sync) {
1023                 struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf;
1024
1025                 ubifs_wbuf_add_ino_nolock(wbuf, new_dir->i_ino);
1026                 ubifs_wbuf_add_ino_nolock(wbuf, old_dir->i_ino);
1027                 if (new_inode)
1028                         ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf,
1029                                                   new_inode->i_ino);
1030         }
1031         release_head(c, BASEHD);
1032
1033         dent_key_init(c, &key, new_dir->i_ino, &new_dentry->d_name);
1034         err = ubifs_tnc_add_nm(c, &key, lnum, offs, dlen1, &new_dentry->d_name);
1035         if (err)
1036                 goto out_ro;
1037
1038         err = ubifs_add_dirt(c, lnum, dlen2);
1039         if (err)
1040                 goto out_ro;
1041
1042         dent_key_init(c, &key, old_dir->i_ino, &old_dentry->d_name);
1043         err = ubifs_tnc_remove_nm(c, &key, &old_dentry->d_name);
1044         if (err)
1045                 goto out_ro;
1046
1047         offs += aligned_dlen1 + aligned_dlen2;
1048         if (new_inode) {
1049                 ino_key_init(c, &key, new_inode->i_ino);
1050                 err = ubifs_tnc_add(c, &key, lnum, offs, ilen);
1051                 if (err)
1052                         goto out_ro;
1053                 offs += ALIGN(ilen, 8);
1054         }
1055
1056         ino_key_init(c, &key, old_dir->i_ino);
1057         err = ubifs_tnc_add(c, &key, lnum, offs, plen);
1058         if (err)
1059                 goto out_ro;
1060
1061         if (old_dir != new_dir) {
1062                 offs += ALIGN(plen, 8);
1063                 ino_key_init(c, &key, new_dir->i_ino);
1064                 err = ubifs_tnc_add(c, &key, lnum, offs, plen);
1065                 if (err)
1066                         goto out_ro;
1067         }
1068
1069         finish_reservation(c);
1070         if (new_inode) {
1071                 mark_inode_clean(c, new_ui);
1072                 spin_lock(&new_ui->ui_lock);
1073                 new_ui->synced_i_size = new_ui->ui_size;
1074                 spin_unlock(&new_ui->ui_lock);
1075         }
1076         mark_inode_clean(c, ubifs_inode(old_dir));
1077         if (move)
1078                 mark_inode_clean(c, ubifs_inode(new_dir));
1079         kfree(dent);
1080         return 0;
1081
1082 out_release:
1083         release_head(c, BASEHD);
1084 out_ro:
1085         ubifs_ro_mode(c, err);
1086         if (last_reference)
1087                 ubifs_delete_orphan(c, new_inode->i_ino);
1088 out_finish:
1089         finish_reservation(c);
1090 out_free:
1091         kfree(dent);
1092         return err;
1093 }
1094
1095 /**
1096  * recomp_data_node - re-compress a truncated data node.
1097  * @dn: data node to re-compress
1098  * @new_len: new length
1099  *
1100  * This function is used when an inode is truncated and the last data node of
1101  * the inode has to be re-compressed and re-written.
1102  */
1103 static int recomp_data_node(struct ubifs_data_node *dn, int *new_len)
1104 {
1105         void *buf;
1106         int err, len, compr_type, out_len;
1107
1108         out_len = le32_to_cpu(dn->size);
1109         buf = kmalloc(out_len * WORST_COMPR_FACTOR, GFP_NOFS);
1110         if (!buf)
1111                 return -ENOMEM;
1112
1113         len = le32_to_cpu(dn->ch.len) - UBIFS_DATA_NODE_SZ;
1114         compr_type = le16_to_cpu(dn->compr_type);
1115         err = ubifs_decompress(&dn->data, len, buf, &out_len, compr_type);
1116         if (err)
1117                 goto out;
1118
1119         ubifs_compress(buf, *new_len, &dn->data, &out_len, &compr_type);
1120         ubifs_assert(out_len <= UBIFS_BLOCK_SIZE);
1121         dn->compr_type = cpu_to_le16(compr_type);
1122         dn->size = cpu_to_le32(*new_len);
1123         *new_len = UBIFS_DATA_NODE_SZ + out_len;
1124 out:
1125         kfree(buf);
1126         return err;
1127 }
1128
1129 /**
1130  * ubifs_jnl_truncate - update the journal for a truncation.
1131  * @c: UBIFS file-system description object
1132  * @inode: inode to truncate
1133  * @old_size: old size
1134  * @new_size: new size
1135  *
1136  * When the size of a file decreases due to truncation, a truncation node is
1137  * written, the journal tree is updated, and the last data block is re-written
1138  * if it has been affected. The inode is also updated in order to synchronize
1139  * the new inode size.
1140  *
1141  * This function marks the inode as clean and returns zero on success. In case
1142  * of failure, a negative error code is returned.
1143  */
1144 int ubifs_jnl_truncate(struct ubifs_info *c, const struct inode *inode,
1145                        loff_t old_size, loff_t new_size)
1146 {
1147         union ubifs_key key, to_key;
1148         struct ubifs_ino_node *ino;
1149         struct ubifs_trun_node *trun;
1150         struct ubifs_data_node *uninitialized_var(dn);
1151         int err, dlen, len, lnum, offs, bit, sz, sync = IS_SYNC(inode);
1152         struct ubifs_inode *ui = ubifs_inode(inode);
1153         ino_t inum = inode->i_ino;
1154         unsigned int blk;
1155
1156         dbg_jnl("ino %lu, size %lld -> %lld",
1157                 (unsigned long)inum, old_size, new_size);
1158         ubifs_assert(!ui->data_len);
1159         ubifs_assert(S_ISREG(inode->i_mode));
1160         ubifs_assert(mutex_is_locked(&ui->ui_mutex));
1161
1162         sz = UBIFS_TRUN_NODE_SZ + UBIFS_INO_NODE_SZ +
1163              UBIFS_MAX_DATA_NODE_SZ * WORST_COMPR_FACTOR;
1164         ino = kmalloc(sz, GFP_NOFS);
1165         if (!ino)
1166                 return -ENOMEM;
1167
1168         trun = (void *)ino + UBIFS_INO_NODE_SZ;
1169         trun->ch.node_type = UBIFS_TRUN_NODE;
1170         trun->inum = cpu_to_le32(inum);
1171         trun->old_size = cpu_to_le64(old_size);
1172         trun->new_size = cpu_to_le64(new_size);
1173         zero_trun_node_unused(trun);
1174
1175         dlen = new_size & (UBIFS_BLOCK_SIZE - 1);
1176         if (dlen) {
1177                 /* Get last data block so it can be truncated */
1178                 dn = (void *)trun + UBIFS_TRUN_NODE_SZ;
1179                 blk = new_size >> UBIFS_BLOCK_SHIFT;
1180                 data_key_init(c, &key, inum, blk);
1181                 dbg_jnlk(&key, "last block key ");
1182                 err = ubifs_tnc_lookup(c, &key, dn);
1183                 if (err == -ENOENT)
1184                         dlen = 0; /* Not found (so it is a hole) */
1185                 else if (err)
1186                         goto out_free;
1187                 else {
1188                         if (le32_to_cpu(dn->size) <= dlen)
1189                                 dlen = 0; /* Nothing to do */
1190                         else {
1191                                 int compr_type = le16_to_cpu(dn->compr_type);
1192
1193                                 if (compr_type != UBIFS_COMPR_NONE) {
1194                                         err = recomp_data_node(dn, &dlen);
1195                                         if (err)
1196                                                 goto out_free;
1197                                 } else {
1198                                         dn->size = cpu_to_le32(dlen);
1199                                         dlen += UBIFS_DATA_NODE_SZ;
1200                                 }
1201                                 zero_data_node_unused(dn);
1202                         }
1203                 }
1204         }
1205
1206         /* Must make reservation before allocating sequence numbers */
1207         len = UBIFS_TRUN_NODE_SZ + UBIFS_INO_NODE_SZ;
1208         if (dlen)
1209                 len += dlen;
1210         err = make_reservation(c, BASEHD, len);
1211         if (err)
1212                 goto out_free;
1213
1214         pack_inode(c, ino, inode, 0);
1215         ubifs_prep_grp_node(c, trun, UBIFS_TRUN_NODE_SZ, dlen ? 0 : 1);
1216         if (dlen)
1217                 ubifs_prep_grp_node(c, dn, dlen, 1);
1218
1219         err = write_head(c, BASEHD, ino, len, &lnum, &offs, sync);
1220         if (err)
1221                 goto out_release;
1222         if (!sync)
1223                 ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf, inum);
1224         release_head(c, BASEHD);
1225
1226         if (dlen) {
1227                 sz = offs + UBIFS_INO_NODE_SZ + UBIFS_TRUN_NODE_SZ;
1228                 err = ubifs_tnc_add(c, &key, lnum, sz, dlen);
1229                 if (err)
1230                         goto out_ro;
1231         }
1232
1233         ino_key_init(c, &key, inum);
1234         err = ubifs_tnc_add(c, &key, lnum, offs, UBIFS_INO_NODE_SZ);
1235         if (err)
1236                 goto out_ro;
1237
1238         err = ubifs_add_dirt(c, lnum, UBIFS_TRUN_NODE_SZ);
1239         if (err)
1240                 goto out_ro;
1241
1242         bit = new_size & (UBIFS_BLOCK_SIZE - 1);
1243         blk = (new_size >> UBIFS_BLOCK_SHIFT) + (bit ? 1 : 0);
1244         data_key_init(c, &key, inum, blk);
1245
1246         bit = old_size & (UBIFS_BLOCK_SIZE - 1);
1247         blk = (old_size >> UBIFS_BLOCK_SHIFT) - (bit ? 0 : 1);
1248         data_key_init(c, &to_key, inum, blk);
1249
1250         err = ubifs_tnc_remove_range(c, &key, &to_key);
1251         if (err)
1252                 goto out_ro;
1253
1254         finish_reservation(c);
1255         spin_lock(&ui->ui_lock);
1256         ui->synced_i_size = ui->ui_size;
1257         spin_unlock(&ui->ui_lock);
1258         mark_inode_clean(c, ui);
1259         kfree(ino);
1260         return 0;
1261
1262 out_release:
1263         release_head(c, BASEHD);
1264 out_ro:
1265         ubifs_ro_mode(c, err);
1266         finish_reservation(c);
1267 out_free:
1268         kfree(ino);
1269         return err;
1270 }
1271
1272
1273 /**
1274  * ubifs_jnl_delete_xattr - delete an extended attribute.
1275  * @c: UBIFS file-system description object
1276  * @host: host inode
1277  * @inode: extended attribute inode
1278  * @nm: extended attribute entry name
1279  *
1280  * This function delete an extended attribute which is very similar to
1281  * un-linking regular files - it writes a deletion xentry, a deletion inode and
1282  * updates the target inode. Returns zero in case of success and a negative
1283  * error code in case of failure.
1284  */
1285 int ubifs_jnl_delete_xattr(struct ubifs_info *c, const struct inode *host,
1286                            const struct inode *inode, const struct qstr *nm)
1287 {
1288         int err, xlen, hlen, len, lnum, xent_offs, aligned_xlen;
1289         struct ubifs_dent_node *xent;
1290         struct ubifs_ino_node *ino;
1291         union ubifs_key xent_key, key1, key2;
1292         int sync = IS_DIRSYNC(host);
1293         struct ubifs_inode *host_ui = ubifs_inode(host);
1294
1295         dbg_jnl("host %lu, xattr ino %lu, name '%s', data len %d",
1296                 host->i_ino, inode->i_ino, nm->name,
1297                 ubifs_inode(inode)->data_len);
1298         ubifs_assert(inode->i_nlink == 0);
1299         ubifs_assert(mutex_is_locked(&host_ui->ui_mutex));
1300
1301         /*
1302          * Since we are deleting the inode, we do not bother to attach any data
1303          * to it and assume its length is %UBIFS_INO_NODE_SZ.
1304          */
1305         xlen = UBIFS_DENT_NODE_SZ + nm->len + 1;
1306         aligned_xlen = ALIGN(xlen, 8);
1307         hlen = host_ui->data_len + UBIFS_INO_NODE_SZ;
1308         len = aligned_xlen + UBIFS_INO_NODE_SZ + ALIGN(hlen, 8);
1309
1310         xent = kmalloc(len, GFP_NOFS);
1311         if (!xent)
1312                 return -ENOMEM;
1313
1314         /* Make reservation before allocating sequence numbers */
1315         err = make_reservation(c, BASEHD, len);
1316         if (err) {
1317                 kfree(xent);
1318                 return err;
1319         }
1320
1321         xent->ch.node_type = UBIFS_XENT_NODE;
1322         xent_key_init(c, &xent_key, host->i_ino, nm);
1323         key_write(c, &xent_key, xent->key);
1324         xent->inum = 0;
1325         xent->type = get_dent_type(inode->i_mode);
1326         xent->nlen = cpu_to_le16(nm->len);
1327         memcpy(xent->name, nm->name, nm->len);
1328         xent->name[nm->len] = '\0';
1329         zero_dent_node_unused(xent);
1330         ubifs_prep_grp_node(c, xent, xlen, 0);
1331
1332         ino = (void *)xent + aligned_xlen;
1333         pack_inode(c, ino, inode, 0);
1334         ino = (void *)ino + UBIFS_INO_NODE_SZ;
1335         pack_inode(c, ino, host, 1);
1336
1337         err = write_head(c, BASEHD, xent, len, &lnum, &xent_offs, sync);
1338         if (!sync && !err)
1339                 ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf, host->i_ino);
1340         release_head(c, BASEHD);
1341         kfree(xent);
1342         if (err)
1343                 goto out_ro;
1344
1345         /* Remove the extended attribute entry from TNC */
1346         err = ubifs_tnc_remove_nm(c, &xent_key, nm);
1347         if (err)
1348                 goto out_ro;
1349         err = ubifs_add_dirt(c, lnum, xlen);
1350         if (err)
1351                 goto out_ro;
1352
1353         /*
1354          * Remove all nodes belonging to the extended attribute inode from TNC.
1355          * Well, there actually must be only one node - the inode itself.
1356          */
1357         lowest_ino_key(c, &key1, inode->i_ino);
1358         highest_ino_key(c, &key2, inode->i_ino);
1359         err = ubifs_tnc_remove_range(c, &key1, &key2);
1360         if (err)
1361                 goto out_ro;
1362         err = ubifs_add_dirt(c, lnum, UBIFS_INO_NODE_SZ);
1363         if (err)
1364                 goto out_ro;
1365
1366         /* And update TNC with the new host inode position */
1367         ino_key_init(c, &key1, host->i_ino);
1368         err = ubifs_tnc_add(c, &key1, lnum, xent_offs + len - hlen, hlen);
1369         if (err)
1370                 goto out_ro;
1371
1372         finish_reservation(c);
1373         spin_lock(&host_ui->ui_lock);
1374         host_ui->synced_i_size = host_ui->ui_size;
1375         spin_unlock(&host_ui->ui_lock);
1376         mark_inode_clean(c, host_ui);
1377         return 0;
1378
1379 out_ro:
1380         ubifs_ro_mode(c, err);
1381         finish_reservation(c);
1382         return err;
1383 }
1384
1385 /**
1386  * ubifs_jnl_change_xattr - change an extended attribute.
1387  * @c: UBIFS file-system description object
1388  * @inode: extended attribute inode
1389  * @host: host inode
1390  *
1391  * This function writes the updated version of an extended attribute inode and
1392  * the host inode to the journal (to the base head). The host inode is written
1393  * after the extended attribute inode in order to guarantee that the extended
1394  * attribute will be flushed when the inode is synchronized by 'fsync()' and
1395  * consequently, the write-buffer is synchronized. This function returns zero
1396  * in case of success and a negative error code in case of failure.
1397  */
1398 int ubifs_jnl_change_xattr(struct ubifs_info *c, const struct inode *inode,
1399                            const struct inode *host)
1400 {
1401         int err, len1, len2, aligned_len, aligned_len1, lnum, offs;
1402         struct ubifs_inode *host_ui = ubifs_inode(host);
1403         struct ubifs_ino_node *ino;
1404         union ubifs_key key;
1405         int sync = IS_DIRSYNC(host);
1406
1407         dbg_jnl("ino %lu, ino %lu", host->i_ino, inode->i_ino);
1408         ubifs_assert(host->i_nlink > 0);
1409         ubifs_assert(inode->i_nlink > 0);
1410         ubifs_assert(mutex_is_locked(&host_ui->ui_mutex));
1411
1412         len1 = UBIFS_INO_NODE_SZ + host_ui->data_len;
1413         len2 = UBIFS_INO_NODE_SZ + ubifs_inode(inode)->data_len;
1414         aligned_len1 = ALIGN(len1, 8);
1415         aligned_len = aligned_len1 + ALIGN(len2, 8);
1416
1417         ino = kmalloc(aligned_len, GFP_NOFS);
1418         if (!ino)
1419                 return -ENOMEM;
1420
1421         /* Make reservation before allocating sequence numbers */
1422         err = make_reservation(c, BASEHD, aligned_len);
1423         if (err)
1424                 goto out_free;
1425
1426         pack_inode(c, ino, host, 0);
1427         pack_inode(c, (void *)ino + aligned_len1, inode, 1);
1428
1429         err = write_head(c, BASEHD, ino, aligned_len, &lnum, &offs, 0);
1430         if (!sync && !err) {
1431                 struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf;
1432
1433                 ubifs_wbuf_add_ino_nolock(wbuf, host->i_ino);
1434                 ubifs_wbuf_add_ino_nolock(wbuf, inode->i_ino);
1435         }
1436         release_head(c, BASEHD);
1437         if (err)
1438                 goto out_ro;
1439
1440         ino_key_init(c, &key, host->i_ino);
1441         err = ubifs_tnc_add(c, &key, lnum, offs, len1);
1442         if (err)
1443                 goto out_ro;
1444
1445         ino_key_init(c, &key, inode->i_ino);
1446         err = ubifs_tnc_add(c, &key, lnum, offs + aligned_len1, len2);
1447         if (err)
1448                 goto out_ro;
1449
1450         finish_reservation(c);
1451         spin_lock(&host_ui->ui_lock);
1452         host_ui->synced_i_size = host_ui->ui_size;
1453         spin_unlock(&host_ui->ui_lock);
1454         mark_inode_clean(c, host_ui);
1455         kfree(ino);
1456         return 0;
1457
1458 out_ro:
1459         ubifs_ro_mode(c, err);
1460         finish_reservation(c);
1461 out_free:
1462         kfree(ino);
1463         return err;
1464 }
1465