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 }
  82
  83 /**
  84  * zero_trun_node_unused - zero out unused fields of an on-flash truncation
  85  *                         node.
  86  * @trun: the truncation node to zero out
  87  */
  88 static inline void zero_trun_node_unused(struct ubifs_trun_node *trun)
  89 {
  90         memset(trun->padding, 0, 12);
  91 }
  92
  93 /**
  94  * reserve_space - reserve space in the journal.
  95  * @c: UBIFS file-system description object
  96  * @jhead: journal head number
  97  * @len: node length
  98  *
  99  * This function reserves space in journal head @head. If the reservation
 100  * succeeded, the journal head stays locked and later has to be unlocked using
 101  * 'release_head()'. 'write_node()' and 'write_head()' functions also unlock
 102  * it. Returns zero in case of success, %-EAGAIN if commit has to be done, and
 103  * other negative error codes in case of other failures.
 104  */
 105 static int reserve_space(struct ubifs_info *c, int jhead, int len)
 106 {
 107         int err = 0, err1, retries = 0, avail, lnum, offs, squeeze;
 108         struct ubifs_wbuf *wbuf = &c->jheads[jhead].wbuf;
 109
 110         /*
 111          * Typically, the base head has smaller nodes written to it, so it is
 112          * better to try to allocate space at the ends of eraseblocks. This is
 113          * what the squeeze parameter does.
 114          */
 115         ubifs_assert(!c->ro_media && !c->ro_mount);
 116         squeeze = (jhead == BASEHD);
 117 again:
 118         mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
 119
 120         if (c->ro_error) {
 121                 err = -EROFS;
 122                 goto out_unlock;
 123         }
 124
 125         avail = c->leb_size - wbuf->offs - wbuf->used;
 126         if (wbuf->lnum != -1 && avail >= len)
 127                 return 0;
 128
 129         /*
 130          * Write buffer wasn't seek'ed or there is no enough space - look for an
 131          * LEB with some empty space.
 132          */
 133         lnum = ubifs_find_free_space(c, len, &offs, squeeze);
 134         if (lnum >= 0)
 135                 goto out;
 136
 137         err = lnum;
 138         if (err != -ENOSPC)
 139                 goto out_unlock;
 140
 141         /*
 142          * No free space, we have to run garbage collector to make
 143          * some. But the write-buffer mutex has to be unlocked because
 144          * GC also takes it.
 145          */
 146         dbg_jnl("no free space in jhead %s, run GC", dbg_jhead(jhead));
 147         mutex_unlock(&wbuf->io_mutex);
 148
 149         lnum = ubifs_garbage_collect(c, 0);
 150         if (lnum < 0) {
 151                 err = lnum;
 152                 if (err != -ENOSPC)
 153                         return err;
 154
 155                 /*
 156                  * GC could not make a free LEB. But someone else may
 157                  * have allocated new bud for this journal head,
 158                  * because we dropped @wbuf->io_mutex, so try once
 159                  * again.
 160                  */
 161                 dbg_jnl("GC couldn't make a free LEB for jhead %s",
 162                         dbg_jhead(jhead));
 163                 if (retries++ < 2) {
 164                         dbg_jnl("retry (%d)", retries);
 165                         goto again;
 166                 }
 167
 168                 dbg_jnl("return -ENOSPC");
 169                 return err;
 170         }
 171
 172         mutex_lock_nested(&wbuf->io_mutex, wbuf->jhead);
 173         dbg_jnl("got LEB %d for jhead %s", lnum, dbg_jhead(jhead));
 174         avail = c->leb_size - wbuf->offs - wbuf->used;
 175
 176         if (wbuf->lnum != -1 && avail >= len) {
 177                 /*
 178                  * Someone else has switched the journal head and we have
 179                  * enough space now. This happens when more than one process is
 180                  * trying to write to the same journal head at the same time.
 181                  */
 182                 dbg_jnl("return LEB %d back, already have LEB %d:%d",
 183                         lnum, wbuf->lnum, wbuf->offs + wbuf->used);
 184                 err = ubifs_return_leb(c, lnum);
 185                 if (err)
 186                         goto out_unlock;
 187                 return 0;
 188         }
 189
 190         offs = 0;
 191
 192 out:
 193         /*
 194          * Make sure we synchronize the write-buffer before we add the new bud
 195          * to the log. Otherwise we may have a power cut after the log
 196          * reference node for the last bud (@lnum) is written but before the
 197          * write-buffer data are written to the next-to-last bud
 198          * (@wbuf->lnum). And the effect would be that the recovery would see
 199          * that there is corruption in the next-to-last bud.
 200          */
 201         err = ubifs_wbuf_sync_nolock(wbuf);
 202         if (err)
 203                 goto out_return;
 204         err = ubifs_add_bud_to_log(c, jhead, lnum, offs);
 205         if (err)
 206                 goto out_return;
 207         err = ubifs_wbuf_seek_nolock(wbuf, lnum, offs);
 208         if (err)
 209                 goto out_unlock;
 210
 211         return 0;
 212
 213 out_unlock:
 214         mutex_unlock(&wbuf->io_mutex);
 215         return err;
 216
 217 out_return:
 218         /* An error occurred and the LEB has to be returned to lprops */
 219         ubifs_assert(err < 0);
 220         err1 = ubifs_return_leb(c, lnum);
 221         if (err1 && err == -EAGAIN)
 222                 /*
 223                  * Return original error code only if it is not %-EAGAIN,
 224                  * which is not really an error. Otherwise, return the error
 225                  * code of 'ubifs_return_leb()'.
 226                  */
 227                 err = err1;
 228         mutex_unlock(&wbuf->io_mutex);
 229         return err;
 230 }
 231
 232 /**
 233  * write_node - write node to a journal head.
 234  * @c: UBIFS file-system description object
 235  * @jhead: journal head
 236  * @node: node to write
 237  * @len: node length
 238  * @lnum: LEB number written is returned here
 239  * @offs: offset written is returned here
 240  *
 241  * This function writes a node to reserved space of journal head @jhead.
 242  * Returns zero in case of success and a negative error code in case of
 243  * failure.
 244  */
 245 static int write_node(struct ubifs_info *c, int jhead, void *node, int len,
 246                       int *lnum, int *offs)
 247 {
 248         struct ubifs_wbuf *wbuf = &c->jheads[jhead].wbuf;
 249
 250         ubifs_assert(jhead != GCHD);
 251
 252         *lnum = c->jheads[jhead].wbuf.lnum;
 253         *offs = c->jheads[jhead].wbuf.offs + c->jheads[jhead].wbuf.used;
 254
 255         dbg_jnl("jhead %s, LEB %d:%d, len %d",
 256                 dbg_jhead(jhead), *lnum, *offs, len);
 257         ubifs_prepare_node(c, node, len, 0);
 258
 259         return ubifs_wbuf_write_nolock(wbuf, node, len);
 260 }
 261
 262 /**
 263  * write_head - write data to a journal head.
 264  * @c: UBIFS file-system description object
 265  * @jhead: journal head
 266  * @buf: buffer to write
 267  * @len: length to write
 268  * @lnum: LEB number written is returned here
 269  * @offs: offset written is returned here
 270  * @sync: non-zero if the write-buffer has to by synchronized
 271  *
 272  * This function is the same as 'write_node()' but it does not assume the
 273  * buffer it is writing is a node, so it does not prepare it (which means
 274  * initializing common header and calculating CRC).
 275  */
 276 static int write_head(struct ubifs_info *c, int jhead, void *buf, int len,
 277                       int *lnum, int *offs, int sync)
 278 {
 279         int err;
 280         struct ubifs_wbuf *wbuf = &c->jheads[jhead].wbuf;
 281
 282         ubifs_assert(jhead != GCHD);
 283
 284         *lnum = c->jheads[jhead].wbuf.lnum;
 285         *offs = c->jheads[jhead].wbuf.offs + c->jheads[jhead].wbuf.used;
 286         dbg_jnl("jhead %s, LEB %d:%d, len %d",
 287                 dbg_jhead(jhead), *lnum, *offs, len);
 288
 289         err = ubifs_wbuf_write_nolock(wbuf, buf, len);
 290         if (err)
 291                 return err;
 292         if (sync)
 293                 err = ubifs_wbuf_sync_nolock(wbuf);
 294         return err;
 295 }
 296
 297 /**
 298  * make_reservation - reserve journal space.
 299  * @c: UBIFS file-system description object
 300  * @jhead: journal head
 301  * @len: how many bytes to reserve
 302  *
 303  * This function makes space reservation in journal head @jhead. The function
 304  * takes the commit lock and locks the journal head, and the caller has to
 305  * unlock the head and finish the reservation with 'finish_reservation()'.
 306  * Returns zero in case of success and a negative error code in case of
 307  * failure.
 308  *
 309  * Note, the journal head may be unlocked as soon as the data is written, while
 310  * the commit lock has to be released after the data has been added to the
 311  * TNC.
 312  */
 313 static int make_reservation(struct ubifs_info *c, int jhead, int len)
 314 {
 315         int err, cmt_retries = 0, nospc_retries = 0;
 316
 317 again:
 318         down_read(&c->commit_sem);
 319         err = reserve_space(c, jhead, len);
 320         if (!err)
 321                 return 0;
 322         up_read(&c->commit_sem);
 323
 324         if (err == -ENOSPC) {
 325                 /*
 326                  * GC could not make any progress. We should try to commit
 327                  * once because it could make some dirty space and GC would
 328                  * make progress, so make the error -EAGAIN so that the below
 329                  * will commit and re-try.
 330                  */
 331                 if (nospc_retries++ < 2) {
 332                         dbg_jnl("no space, retry");
 333                         err = -EAGAIN;
 334                 }
 335
 336                 /*
 337                  * This means that the budgeting is incorrect. We always have
 338                  * to be able to write to the media, because all operations are
 339                  * budgeted. Deletions are not budgeted, though, but we reserve
 340                  * an extra LEB for them.
 341                  */
 342         }
 343
 344         if (err != -EAGAIN)
 345                 goto out;
 346
 347         /*
 348          * -EAGAIN means that the journal is full or too large, or the above
 349          * code wants to do one commit. Do this and re-try.
 350          */
 351         if (cmt_retries > 128) {
 352                 /*
 353                  * This should not happen unless the journal size limitations
 354                  * are too tough.
 355                  */
 356                 ubifs_err(c, "stuck in space allocation");
 357                 err = -ENOSPC;
 358                 goto out;
 359         } else if (cmt_retries > 32)
 360                 ubifs_warn(c, "too many space allocation re-tries (%d)",
 361                            cmt_retries);
 362
 363         dbg_jnl("-EAGAIN, commit and retry (retried %d times)",
 364                 cmt_retries);
 365         cmt_retries += 1;
 366
 367         err = ubifs_run_commit(c);
 368         if (err)
 369                 return err;
 370         goto again;
 371
 372 out:
 373         ubifs_err(c, "cannot reserve %d bytes in jhead %d, error %d",
 374                   len, jhead, err);
 375         if (err == -ENOSPC) {
 376                 /* This are some budgeting problems, print useful information */
 377                 down_write(&c->commit_sem);
 378                 dump_stack();
 379                 ubifs_dump_budg(c, &c->bi);
 380                 ubifs_dump_lprops(c);
 381                 cmt_retries = dbg_check_lprops(c);
 382                 up_write(&c->commit_sem);
 383         }
 384         return err;
 385 }
 386
 387 /**
 388  * release_head - release a journal head.
 389  * @c: UBIFS file-system description object
 390  * @jhead: journal head
 391  *
 392  * This function releases journal head @jhead which was locked by
 393  * the 'make_reservation()' function. It has to be called after each successful
 394  * 'make_reservation()' invocation.
 395  */
 396 static inline void release_head(struct ubifs_info *c, int jhead)
 397 {
 398         mutex_unlock(&c->jheads[jhead].wbuf.io_mutex);
 399 }
 400
 401 /**
 402  * finish_reservation - finish a reservation.
 403  * @c: UBIFS file-system description object
 404  *
 405  * This function finishes journal space reservation. It must be called after
 406  * 'make_reservation()'.
 407  */
 408 static void finish_reservation(struct ubifs_info *c)
 409 {
 410         up_read(&c->commit_sem);
 411 }
 412
 413 /**
 414  * get_dent_type - translate VFS inode mode to UBIFS directory entry type.
 415  * @mode: inode mode
 416  */
 417 static int get_dent_type(int mode)
 418 {
 419         switch (mode & S_IFMT) {
 420         case S_IFREG:
 421                 return UBIFS_ITYPE_REG;
 422         case S_IFDIR:
 423                 return UBIFS_ITYPE_DIR;
 424         case S_IFLNK:
 425                 return UBIFS_ITYPE_LNK;
 426         case S_IFBLK:
 427                 return UBIFS_ITYPE_BLK;
 428         case S_IFCHR:
 429                 return UBIFS_ITYPE_CHR;
 430         case S_IFIFO:
 431                 return UBIFS_ITYPE_FIFO;
 432         case S_IFSOCK:
 433                 return UBIFS_ITYPE_SOCK;
 434         default:
 435                 BUG();
 436         }
 437         return 0;
 438 }
 439
 440 /**
 441  * pack_inode - pack an inode node.
 442  * @c: UBIFS file-system description object
 443  * @ino: buffer in which to pack inode node
 444  * @inode: inode to pack
 445  * @last: indicates the last node of the group
 446  */
 447 static void pack_inode(struct ubifs_info *c, struct ubifs_ino_node *ino,
 448                        const struct inode *inode, int last)
 449 {
 450         int data_len = 0, last_reference = !inode->i_nlink;
 451         struct ubifs_inode *ui = ubifs_inode(inode);
 452
 453         ino->ch.node_type = UBIFS_INO_NODE;
 454         ino_key_init_flash(c, &ino->key, inode->i_ino);
 455         ino->creat_sqnum = cpu_to_le64(ui->creat_sqnum);
 456         ino->atime_sec  = cpu_to_le64(inode->i_atime.tv_sec);
 457         ino->atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
 458         ino->ctime_sec  = cpu_to_le64(inode->i_ctime.tv_sec);
 459         ino->ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 460         ino->mtime_sec  = cpu_to_le64(inode->i_mtime.tv_sec);
 461         ino->mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 462         ino->uid   = cpu_to_le32(i_uid_read(inode));
 463         ino->gid   = cpu_to_le32(i_gid_read(inode));
 464         ino->mode  = cpu_to_le32(inode->i_mode);
 465         ino->flags = cpu_to_le32(ui->flags);
 466         ino->size  = cpu_to_le64(ui->ui_size);
 467         ino->nlink = cpu_to_le32(inode->i_nlink);
 468         ino->compr_type  = cpu_to_le16(ui->compr_type);
 469         ino->data_len    = cpu_to_le32(ui->data_len);
 470         ino->xattr_cnt   = cpu_to_le32(ui->xattr_cnt);
 471         ino->xattr_size  = cpu_to_le32(ui->xattr_size);
 472         ino->xattr_names = cpu_to_le32(ui->xattr_names);
 473         zero_ino_node_unused(ino);
 474
 475         /*
 476          * Drop the attached data if this is a deletion inode, the data is not
 477          * needed anymore.
 478          */
 479         if (!last_reference) {
 480                 memcpy(ino->data, ui->data, ui->data_len);
 481                 data_len = ui->data_len;
 482         }
 483
 484         ubifs_prep_grp_node(c, ino, UBIFS_INO_NODE_SZ + data_len, last);
 485 }
 486
 487 /**
 488  * mark_inode_clean - mark UBIFS inode as clean.
 489  * @c: UBIFS file-system description object
 490  * @ui: UBIFS inode to mark as clean
 491  *
 492  * This helper function marks UBIFS inode @ui as clean by cleaning the
 493  * @ui->dirty flag and releasing its budget. Note, VFS may still treat the
 494  * inode as dirty and try to write it back, but 'ubifs_write_inode()' would
 495  * just do nothing.
 496  */
 497 static void mark_inode_clean(struct ubifs_info *c, struct ubifs_inode *ui)
 498 {
 499         if (ui->dirty)
 500                 ubifs_release_dirty_inode_budget(c, ui);
 501         ui->dirty = 0;
 502 }
 503
 504 static void set_dent_cookie(struct ubifs_info *c, struct ubifs_dent_node *dent)
 505 {
 506         if (c->double_hash)
 507                 dent->cookie = prandom_u32();
 508         else
 509                 dent->cookie = 0;
 510 }
 511
 512 /**
 513  * ubifs_jnl_update - update inode.
 514  * @c: UBIFS file-system description object
 515  * @dir: parent inode or host inode in case of extended attributes
 516  * @nm: directory entry name
 517  * @inode: inode to update
 518  * @deletion: indicates a directory entry deletion i.e unlink or rmdir
 519  * @xent: non-zero if the directory entry is an extended attribute entry
 520  *
 521  * This function updates an inode by writing a directory entry (or extended
 522  * attribute entry), the inode itself, and the parent directory inode (or the
 523  * host inode) to the journal.
 524  *
 525  * The function writes the host inode @dir last, which is important in case of
 526  * extended attributes. Indeed, then we guarantee that if the host inode gets
 527  * synchronized (with 'fsync()'), and the write-buffer it sits in gets flushed,
 528  * the extended attribute inode gets flushed too. And this is exactly what the
 529  * user expects - synchronizing the host inode synchronizes its extended
 530  * attributes. Similarly, this guarantees that if @dir is synchronized, its
 531  * directory entry corresponding to @nm gets synchronized too.
 532  *
 533  * If the inode (@inode) or the parent directory (@dir) are synchronous, this
 534  * function synchronizes the write-buffer.
 535  *
 536  * This function marks the @dir and @inode inodes as clean and returns zero on
 537  * success. In case of failure, a negative error code is returned.
 538  */
 539 int ubifs_jnl_update(struct ubifs_info *c, const struct inode *dir,
 540                      const struct fscrypt_name *nm, const struct inode *inode,
 541                      int deletion, int xent)
 542 {
 543         int err, dlen, ilen, len, lnum, ino_offs, dent_offs;
 544         int aligned_dlen, aligned_ilen, sync = IS_DIRSYNC(dir);
 545         int last_reference = !!(deletion && inode->i_nlink == 0);
 546         struct ubifs_inode *ui = ubifs_inode(inode);
 547         struct ubifs_inode *host_ui = ubifs_inode(dir);
 548         struct ubifs_dent_node *dent;
 549         struct ubifs_ino_node *ino;
 550         union ubifs_key dent_key, ino_key;
 551
 552         ubifs_assert(mutex_is_locked(&host_ui->ui_mutex));
 553
 554         dlen = UBIFS_DENT_NODE_SZ + fname_len(nm) + 1;
 555         ilen = UBIFS_INO_NODE_SZ;
 556
 557         /*
 558          * If the last reference to the inode is being deleted, then there is
 559          * no need to attach and write inode data, it is being deleted anyway.
 560          * And if the inode is being deleted, no need to synchronize
 561          * write-buffer even if the inode is synchronous.
 562          */
 563         if (!last_reference) {
 564                 ilen += ui->data_len;
 565                 sync |= IS_SYNC(inode);
 566         }
 567
 568         aligned_dlen = ALIGN(dlen, 8);
 569         aligned_ilen = ALIGN(ilen, 8);
 570
 571         len = aligned_dlen + aligned_ilen + UBIFS_INO_NODE_SZ;
 572         /* Make sure to also account for extended attributes */
 573         len += host_ui->data_len;
 574
 575         dent = kzalloc(len, GFP_NOFS);
 576         if (!dent)
 577                 return -ENOMEM;
 578
 579         /* Make reservation before allocating sequence numbers */
 580         err = make_reservation(c, BASEHD, len);
 581         if (err)
 582                 goto out_free;
 583
 584         if (!xent) {
 585                 dent->ch.node_type = UBIFS_DENT_NODE;
 586                 if (nm->hash)
 587                         dent_key_init_hash(c, &dent_key, dir->i_ino, nm->hash);
 588                 else
 589                         dent_key_init(c, &dent_key, dir->i_ino, nm);
 590         } else {
 591                 dent->ch.node_type = UBIFS_XENT_NODE;
 592                 xent_key_init(c, &dent_key, dir->i_ino, nm);
 593         }
 594
 595         key_write(c, &dent_key, dent->key);
 596         dent->inum = deletion ? 0 : cpu_to_le64(inode->i_ino);
 597         dent->type = get_dent_type(inode->i_mode);
 598         dent->nlen = cpu_to_le16(fname_len(nm));
 599         memcpy(dent->name, fname_name(nm), fname_len(nm));
 600         dent->name[fname_len(nm)] = '\0';
 601         set_dent_cookie(c, dent);
 602
 603         zero_dent_node_unused(dent);
 604         ubifs_prep_grp_node(c, dent, dlen, 0);
 605
 606         ino = (void *)dent + aligned_dlen;
 607         pack_inode(c, ino, inode, 0);
 608         ino = (void *)ino + aligned_ilen;
 609         pack_inode(c, ino, dir, 1);
 610
 611         if (last_reference) {
 612                 err = ubifs_add_orphan(c, inode->i_ino);
 613                 if (err) {
 614                         release_head(c, BASEHD);
 615                         goto out_finish;
 616                 }
 617                 ui->del_cmtno = c->cmt_no;
 618         }
 619
 620         err = write_head(c, BASEHD, dent, len, &lnum, &dent_offs, sync);
 621         if (err)
 622                 goto out_release;
 623         if (!sync) {
 624                 struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf;
 625
 626                 ubifs_wbuf_add_ino_nolock(wbuf, inode->i_ino);
 627                 ubifs_wbuf_add_ino_nolock(wbuf, dir->i_ino);
 628         }
 629         release_head(c, BASEHD);
 630         kfree(dent);
 631
 632         if (deletion) {
 633                 if (nm->hash)
 634                         err = ubifs_tnc_remove_dh(c, &dent_key, nm->minor_hash);
 635                 else
 636                         err = ubifs_tnc_remove_nm(c, &dent_key, nm);
 637                 if (err)
 638                         goto out_ro;
 639                 err = ubifs_add_dirt(c, lnum, dlen);
 640         } else
 641                 err = ubifs_tnc_add_nm(c, &dent_key, lnum, dent_offs, dlen, nm);
 642         if (err)
 643                 goto out_ro;
 644
 645         /*
 646          * Note, we do not remove the inode from TNC even if the last reference
 647          * to it has just been deleted, because the inode may still be opened.
 648          * Instead, the inode has been added to orphan lists and the orphan
 649          * subsystem will take further care about it.
 650          */
 651         ino_key_init(c, &ino_key, inode->i_ino);
 652         ino_offs = dent_offs + aligned_dlen;
 653         err = ubifs_tnc_add(c, &ino_key, lnum, ino_offs, ilen);
 654         if (err)
 655                 goto out_ro;
 656
 657         ino_key_init(c, &ino_key, dir->i_ino);
 658         ino_offs += aligned_ilen;
 659         err = ubifs_tnc_add(c, &ino_key, lnum, ino_offs,
 660                             UBIFS_INO_NODE_SZ + host_ui->data_len);
 661         if (err)
 662                 goto out_ro;
 663
 664         finish_reservation(c);
 665         spin_lock(&ui->ui_lock);
 666         ui->synced_i_size = ui->ui_size;
 667         spin_unlock(&ui->ui_lock);
 668         if (xent) {
 669                 spin_lock(&host_ui->ui_lock);
 670                 host_ui->synced_i_size = host_ui->ui_size;
 671                 spin_unlock(&host_ui->ui_lock);
 672         }
 673         mark_inode_clean(c, ui);
 674         mark_inode_clean(c, host_ui);
 675         return 0;
 676
 677 out_finish:
 678         finish_reservation(c);
 679 out_free:
 680         kfree(dent);
 681         return err;
 682
 683 out_release:
 684         release_head(c, BASEHD);
 685         kfree(dent);
 686 out_ro:
 687         ubifs_ro_mode(c, err);
 688         if (last_reference)
 689                 ubifs_delete_orphan(c, inode->i_ino);
 690         finish_reservation(c);
 691         return err;
 692 }
 693
 694 /**
 695  * ubifs_jnl_write_data - write a data node to the journal.
 696  * @c: UBIFS file-system description object
 697  * @inode: inode the data node belongs to
 698  * @key: node key
 699  * @buf: buffer to write
 700  * @len: data length (must not exceed %UBIFS_BLOCK_SIZE)
 701  *
 702  * This function writes a data node to the journal. Returns %0 if the data node
 703  * was successfully written, and a negative error code in case of failure.
 704  */
 705 int ubifs_jnl_write_data(struct ubifs_info *c, const struct inode *inode,
 706                          const union ubifs_key *key, const void *buf, int len)
 707 {
 708         struct ubifs_data_node *data;
 709         int err, lnum, offs, compr_type, out_len, compr_len;
 710         int dlen = COMPRESSED_DATA_NODE_BUF_SZ, allocated = 1;
 711         struct ubifs_inode *ui = ubifs_inode(inode);
 712         bool encrypted = ubifs_crypt_is_encrypted(inode);
 713
 714         dbg_jnlk(key, "ino %lu, blk %u, len %d, key ",
 715                 (unsigned long)key_inum(c, key), key_block(c, key), len);
 716         ubifs_assert(len <= UBIFS_BLOCK_SIZE);
 717
 718         if (encrypted)
 719                 dlen += UBIFS_CIPHER_BLOCK_SIZE;
 720
 721         data = kmalloc(dlen, GFP_NOFS | __GFP_NOWARN);
 722         if (!data) {
 723                 /*
 724                  * Fall-back to the write reserve buffer. Note, we might be
 725                  * currently on the memory reclaim path, when the kernel is
 726                  * trying to free some memory by writing out dirty pages. The
 727                  * write reserve buffer helps us to guarantee that we are
 728                  * always able to write the data.
 729                  */
 730                 allocated = 0;
 731                 mutex_lock(&c->write_reserve_mutex);
 732                 data = c->write_reserve_buf;
 733         }
 734
 735         data->ch.node_type = UBIFS_DATA_NODE;
 736         key_write(c, key, &data->key);
 737         data->size = cpu_to_le32(len);
 738
 739         if (!(ui->flags & UBIFS_COMPR_FL))
 740                 /* Compression is disabled for this inode */
 741                 compr_type = UBIFS_COMPR_NONE;
 742         else
 743                 compr_type = ui->compr_type;
 744
 745         out_len = compr_len = dlen - UBIFS_DATA_NODE_SZ;
 746         ubifs_compress(c, buf, len, &data->data, &compr_len, &compr_type);
 747         ubifs_assert(compr_len <= UBIFS_BLOCK_SIZE);
 748
 749         if (encrypted) {
 750                 err = ubifs_encrypt(inode, data, compr_len, &out_len, key_block(c, key));
 751                 if (err)
 752                         goto out_free;
 753
 754         } else {
 755                 data->compr_size = 0;
 756                 out_len = compr_len;
 757         }
 758
 759         dlen = UBIFS_DATA_NODE_SZ + out_len;
 760         data->compr_type = cpu_to_le16(compr_type);
 761
 762         /* Make reservation before allocating sequence numbers */
 763         err = make_reservation(c, DATAHD, dlen);
 764         if (err)
 765                 goto out_free;
 766
 767         err = write_node(c, DATAHD, data, dlen, &lnum, &offs);
 768         if (err)
 769                 goto out_release;
 770         ubifs_wbuf_add_ino_nolock(&c->jheads[DATAHD].wbuf, key_inum(c, key));
 771         release_head(c, DATAHD);
 772
 773         err = ubifs_tnc_add(c, key, lnum, offs, dlen);
 774         if (err)
 775                 goto out_ro;
 776
 777         finish_reservation(c);
 778         if (!allocated)
 779                 mutex_unlock(&c->write_reserve_mutex);
 780         else
 781                 kfree(data);
 782         return 0;
 783
 784 out_release:
 785         release_head(c, DATAHD);
 786 out_ro:
 787         ubifs_ro_mode(c, err);
 788         finish_reservation(c);
 789 out_free:
 790         if (!allocated)
 791                 mutex_unlock(&c->write_reserve_mutex);
 792         else
 793                 kfree(data);
 794         return err;
 795 }
 796
 797 /**
 798  * ubifs_jnl_write_inode - flush inode to the journal.
 799  * @c: UBIFS file-system description object
 800  * @inode: inode to flush
 801  *
 802  * This function writes inode @inode to the journal. If the inode is
 803  * synchronous, it also synchronizes the write-buffer. Returns zero in case of
 804  * success and a negative error code in case of failure.
 805  */
 806 int ubifs_jnl_write_inode(struct ubifs_info *c, const struct inode *inode)
 807 {
 808         int err, lnum, offs;
 809         struct ubifs_ino_node *ino;
 810         struct ubifs_inode *ui = ubifs_inode(inode);
 811         int sync = 0, len = UBIFS_INO_NODE_SZ, last_reference = !inode->i_nlink;
 812
 813         dbg_jnl("ino %lu, nlink %u", inode->i_ino, inode->i_nlink);
 814
 815         /*
 816          * If the inode is being deleted, do not write the attached data. No
 817          * need to synchronize the write-buffer either.
 818          */
 819         if (!last_reference) {
 820                 len += ui->data_len;
 821                 sync = IS_SYNC(inode);
 822         }
 823         ino = kmalloc(len, GFP_NOFS);
 824         if (!ino)
 825                 return -ENOMEM;
 826
 827         /* Make reservation before allocating sequence numbers */
 828         err = make_reservation(c, BASEHD, len);
 829         if (err)
 830                 goto out_free;
 831
 832         pack_inode(c, ino, inode, 1);
 833         err = write_head(c, BASEHD, ino, len, &lnum, &offs, sync);
 834         if (err)
 835                 goto out_release;
 836         if (!sync)
 837                 ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf,
 838                                           inode->i_ino);
 839         release_head(c, BASEHD);
 840
 841         if (last_reference) {
 842                 err = ubifs_tnc_remove_ino(c, inode->i_ino);
 843                 if (err)
 844                         goto out_ro;
 845                 ubifs_delete_orphan(c, inode->i_ino);
 846                 err = ubifs_add_dirt(c, lnum, len);
 847         } else {
 848                 union ubifs_key key;
 849
 850                 ino_key_init(c, &key, inode->i_ino);
 851                 err = ubifs_tnc_add(c, &key, lnum, offs, len);
 852         }
 853         if (err)
 854                 goto out_ro;
 855
 856         finish_reservation(c);
 857         spin_lock(&ui->ui_lock);
 858         ui->synced_i_size = ui->ui_size;
 859         spin_unlock(&ui->ui_lock);
 860         kfree(ino);
 861         return 0;
 862
 863 out_release:
 864         release_head(c, BASEHD);
 865 out_ro:
 866         ubifs_ro_mode(c, err);
 867         finish_reservation(c);
 868 out_free:
 869         kfree(ino);
 870         return err;
 871 }
 872
 873 /**
 874  * ubifs_jnl_delete_inode - delete an inode.
 875  * @c: UBIFS file-system description object
 876  * @inode: inode to delete
 877  *
 878  * This function deletes inode @inode which includes removing it from orphans,
 879  * deleting it from TNC and, in some cases, writing a deletion inode to the
 880  * journal.
 881  *
 882  * When regular file inodes are unlinked or a directory inode is removed, the
 883  * 'ubifs_jnl_update()' function writes a corresponding deletion inode and
 884  * direntry to the media, and adds the inode to orphans. After this, when the
 885  * last reference to this inode has been dropped, this function is called. In
 886  * general, it has to write one more deletion inode to the media, because if
 887  * a commit happened between 'ubifs_jnl_update()' and
 888  * 'ubifs_jnl_delete_inode()', the deletion inode is not in the journal
 889  * anymore, and in fact it might not be on the flash anymore, because it might
 890  * have been garbage-collected already. And for optimization reasons UBIFS does
 891  * not read the orphan area if it has been unmounted cleanly, so it would have
 892  * no indication in the journal that there is a deleted inode which has to be
 893  * removed from TNC.
 894  *
 895  * However, if there was no commit between 'ubifs_jnl_update()' and
 896  * 'ubifs_jnl_delete_inode()', then there is no need to write the deletion
 897  * inode to the media for the second time. And this is quite a typical case.
 898  *
 899  * This function returns zero in case of success and a negative error code in
 900  * case of failure.
 901  */
 902 int ubifs_jnl_delete_inode(struct ubifs_info *c, const struct inode *inode)
 903 {
 904         int err;
 905         struct ubifs_inode *ui = ubifs_inode(inode);
 906
 907         ubifs_assert(inode->i_nlink == 0);
 908
 909         if (ui->del_cmtno != c->cmt_no)
 910                 /* A commit happened for sure */
 911                 return ubifs_jnl_write_inode(c, inode);
 912
 913         down_read(&c->commit_sem);
 914         /*
 915          * Check commit number again, because the first test has been done
 916          * without @c->commit_sem, so a commit might have happened.
 917          */
 918         if (ui->del_cmtno != c->cmt_no) {
 919                 up_read(&c->commit_sem);
 920                 return ubifs_jnl_write_inode(c, inode);
 921         }
 922
 923         err = ubifs_tnc_remove_ino(c, inode->i_ino);
 924         if (err)
 925                 ubifs_ro_mode(c, err);
 926         else
 927                 ubifs_delete_orphan(c, inode->i_ino);
 928         up_read(&c->commit_sem);
 929         return err;
 930 }
 931
 932 /**
 933  * ubifs_jnl_xrename - cross rename two directory entries.
 934  * @c: UBIFS file-system description object
 935  * @fst_dir: parent inode of 1st directory entry to exchange
 936  * @fst_inode: 1st inode to exchange
 937  * @fst_nm: name of 1st inode to exchange
 938  * @snd_dir: parent inode of 2nd directory entry to exchange
 939  * @snd_inode: 2nd inode to exchange
 940  * @snd_nm: name of 2nd inode to exchange
 941  * @sync: non-zero if the write-buffer has to be synchronized
 942  *
 943  * This function implements the cross rename operation which may involve
 944  * writing 2 inodes and 2 directory entries. It marks the written inodes as clean
 945  * and returns zero on success. In case of failure, a negative error code is
 946  * returned.
 947  */
 948 int ubifs_jnl_xrename(struct ubifs_info *c, const struct inode *fst_dir,
 949                       const struct inode *fst_inode,
 950                       const struct fscrypt_name *fst_nm,
 951                       const struct inode *snd_dir,
 952                       const struct inode *snd_inode,
 953                       const struct fscrypt_name *snd_nm, int sync)
 954 {
 955         union ubifs_key key;
 956         struct ubifs_dent_node *dent1, *dent2;
 957         int err, dlen1, dlen2, lnum, offs, len, plen = UBIFS_INO_NODE_SZ;
 958         int aligned_dlen1, aligned_dlen2;
 959         int twoparents = (fst_dir != snd_dir);
 960         void *p;
 961
 962         ubifs_assert(ubifs_inode(fst_dir)->data_len == 0);
 963         ubifs_assert(ubifs_inode(snd_dir)->data_len == 0);
 964         ubifs_assert(mutex_is_locked(&ubifs_inode(fst_dir)->ui_mutex));
 965         ubifs_assert(mutex_is_locked(&ubifs_inode(snd_dir)->ui_mutex));
 966
 967         dlen1 = UBIFS_DENT_NODE_SZ + fname_len(snd_nm) + 1;
 968         dlen2 = UBIFS_DENT_NODE_SZ + fname_len(fst_nm) + 1;
 969         aligned_dlen1 = ALIGN(dlen1, 8);
 970         aligned_dlen2 = ALIGN(dlen2, 8);
 971
 972         len = aligned_dlen1 + aligned_dlen2 + ALIGN(plen, 8);
 973         if (twoparents)
 974                 len += plen;
 975
 976         dent1 = kzalloc(len, GFP_NOFS);
 977         if (!dent1)
 978                 return -ENOMEM;
 979
 980         /* Make reservation before allocating sequence numbers */
 981         err = make_reservation(c, BASEHD, len);
 982         if (err)
 983                 goto out_free;
 984
 985         /* Make new dent for 1st entry */
 986         dent1->ch.node_type = UBIFS_DENT_NODE;
 987         dent_key_init_flash(c, &dent1->key, snd_dir->i_ino, snd_nm);
 988         dent1->inum = cpu_to_le64(fst_inode->i_ino);
 989         dent1->type = get_dent_type(fst_inode->i_mode);
 990         dent1->nlen = cpu_to_le16(fname_len(snd_nm));
 991         memcpy(dent1->name, fname_name(snd_nm), fname_len(snd_nm));
 992         dent1->name[fname_len(snd_nm)] = '\0';
 993         set_dent_cookie(c, dent1);
 994         zero_dent_node_unused(dent1);
 995         ubifs_prep_grp_node(c, dent1, dlen1, 0);
 996
 997         /* Make new dent for 2nd entry */
 998         dent2 = (void *)dent1 + aligned_dlen1;
 999         dent2->ch.node_type = UBIFS_DENT_NODE;
1000         dent_key_init_flash(c, &dent2->key, fst_dir->i_ino, fst_nm);
1001         dent2->inum = cpu_to_le64(snd_inode->i_ino);
1002         dent2->type = get_dent_type(snd_inode->i_mode);
1003         dent2->nlen = cpu_to_le16(fname_len(fst_nm));
1004         memcpy(dent2->name, fname_name(fst_nm), fname_len(fst_nm));
1005         dent2->name[fname_len(fst_nm)] = '\0';
1006         set_dent_cookie(c, dent2);
1007         zero_dent_node_unused(dent2);
1008         ubifs_prep_grp_node(c, dent2, dlen2, 0);
1009
1010         p = (void *)dent2 + aligned_dlen2;
1011         if (!twoparents)
1012                 pack_inode(c, p, fst_dir, 1);
1013         else {
1014                 pack_inode(c, p, fst_dir, 0);
1015                 p += ALIGN(plen, 8);
1016                 pack_inode(c, p, snd_dir, 1);
1017         }
1018
1019         err = write_head(c, BASEHD, dent1, 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, fst_dir->i_ino);
1026                 ubifs_wbuf_add_ino_nolock(wbuf, snd_dir->i_ino);
1027         }
1028         release_head(c, BASEHD);
1029
1030         dent_key_init(c, &key, snd_dir->i_ino, snd_nm);
1031         err = ubifs_tnc_add_nm(c, &key, lnum, offs, dlen1, snd_nm);
1032         if (err)
1033                 goto out_ro;
1034
1035         offs += aligned_dlen1;
1036         dent_key_init(c, &key, fst_dir->i_ino, fst_nm);
1037         err = ubifs_tnc_add_nm(c, &key, lnum, offs, dlen2, fst_nm);
1038         if (err)
1039                 goto out_ro;
1040
1041         offs += aligned_dlen2;
1042
1043         ino_key_init(c, &key, fst_dir->i_ino);
1044         err = ubifs_tnc_add(c, &key, lnum, offs, plen);
1045         if (err)
1046                 goto out_ro;
1047
1048         if (twoparents) {
1049                 offs += ALIGN(plen, 8);
1050                 ino_key_init(c, &key, snd_dir->i_ino);
1051                 err = ubifs_tnc_add(c, &key, lnum, offs, plen);
1052                 if (err)
1053                         goto out_ro;
1054         }
1055
1056         finish_reservation(c);
1057
1058         mark_inode_clean(c, ubifs_inode(fst_dir));
1059         if (twoparents)
1060                 mark_inode_clean(c, ubifs_inode(snd_dir));
1061         kfree(dent1);
1062         return 0;
1063
1064 out_release:
1065         release_head(c, BASEHD);
1066 out_ro:
1067         ubifs_ro_mode(c, err);
1068         finish_reservation(c);
1069 out_free:
1070         kfree(dent1);
1071         return err;
1072 }
1073
1074 /**
1075  * ubifs_jnl_rename - rename a directory entry.
1076  * @c: UBIFS file-system description object
1077  * @old_dir: parent inode of directory entry to rename
1078  * @old_dentry: directory entry to rename
1079  * @new_dir: parent inode of directory entry to rename
1080  * @new_dentry: new directory entry (or directory entry to replace)
1081  * @sync: non-zero if the write-buffer has to be synchronized
1082  *
1083  * This function implements the re-name operation which may involve writing up
1084  * to 4 inodes and 2 directory entries. It marks the written inodes as clean
1085  * and returns zero on success. In case of failure, a negative error code is
1086  * returned.
1087  */
1088 int ubifs_jnl_rename(struct ubifs_info *c, const struct inode *old_dir,
1089                      const struct inode *old_inode,
1090                      const struct fscrypt_name *old_nm,
1091                      const struct inode *new_dir,
1092                      const struct inode *new_inode,
1093                      const struct fscrypt_name *new_nm,
1094                      const struct inode *whiteout, int sync)
1095 {
1096         void *p;
1097         union ubifs_key key;
1098         struct ubifs_dent_node *dent, *dent2;
1099         int err, dlen1, dlen2, ilen, lnum, offs, len;
1100         int aligned_dlen1, aligned_dlen2, plen = UBIFS_INO_NODE_SZ;
1101         int last_reference = !!(new_inode && new_inode->i_nlink == 0);
1102         int move = (old_dir != new_dir);
1103         struct ubifs_inode *uninitialized_var(new_ui);
1104
1105         ubifs_assert(ubifs_inode(old_dir)->data_len == 0);
1106         ubifs_assert(ubifs_inode(new_dir)->data_len == 0);
1107         ubifs_assert(mutex_is_locked(&ubifs_inode(old_dir)->ui_mutex));
1108         ubifs_assert(mutex_is_locked(&ubifs_inode(new_dir)->ui_mutex));
1109
1110         dlen1 = UBIFS_DENT_NODE_SZ + fname_len(new_nm) + 1;
1111         dlen2 = UBIFS_DENT_NODE_SZ + fname_len(old_nm) + 1;
1112         if (new_inode) {
1113                 new_ui = ubifs_inode(new_inode);
1114                 ubifs_assert(mutex_is_locked(&new_ui->ui_mutex));
1115                 ilen = UBIFS_INO_NODE_SZ;
1116                 if (!last_reference)
1117                         ilen += new_ui->data_len;
1118         } else
1119                 ilen = 0;
1120
1121         aligned_dlen1 = ALIGN(dlen1, 8);
1122         aligned_dlen2 = ALIGN(dlen2, 8);
1123         len = aligned_dlen1 + aligned_dlen2 + ALIGN(ilen, 8) + ALIGN(plen, 8);
1124         if (move)
1125                 len += plen;
1126         dent = kzalloc(len, GFP_NOFS);
1127         if (!dent)
1128                 return -ENOMEM;
1129
1130         /* Make reservation before allocating sequence numbers */
1131         err = make_reservation(c, BASEHD, len);
1132         if (err)
1133                 goto out_free;
1134
1135         /* Make new dent */
1136         dent->ch.node_type = UBIFS_DENT_NODE;
1137         dent_key_init_flash(c, &dent->key, new_dir->i_ino, new_nm);
1138         dent->inum = cpu_to_le64(old_inode->i_ino);
1139         dent->type = get_dent_type(old_inode->i_mode);
1140         dent->nlen = cpu_to_le16(fname_len(new_nm));
1141         memcpy(dent->name, fname_name(new_nm), fname_len(new_nm));
1142         dent->name[fname_len(new_nm)] = '\0';
1143         set_dent_cookie(c, dent);
1144         zero_dent_node_unused(dent);
1145         ubifs_prep_grp_node(c, dent, dlen1, 0);
1146
1147         dent2 = (void *)dent + aligned_dlen1;
1148         dent2->ch.node_type = UBIFS_DENT_NODE;
1149         dent_key_init_flash(c, &dent2->key, old_dir->i_ino, old_nm);
1150
1151         if (whiteout) {
1152                 dent2->inum = cpu_to_le64(whiteout->i_ino);
1153                 dent2->type = get_dent_type(whiteout->i_mode);
1154         } else {
1155                 /* Make deletion dent */
1156                 dent2->inum = 0;
1157                 dent2->type = DT_UNKNOWN;
1158         }
1159         dent2->nlen = cpu_to_le16(fname_len(old_nm));
1160         memcpy(dent2->name, fname_name(old_nm), fname_len(old_nm));
1161         dent2->name[fname_len(old_nm)] = '\0';
1162         set_dent_cookie(c, dent2);
1163         zero_dent_node_unused(dent2);
1164         ubifs_prep_grp_node(c, dent2, dlen2, 0);
1165
1166         p = (void *)dent2 + aligned_dlen2;
1167         if (new_inode) {
1168                 pack_inode(c, p, new_inode, 0);
1169                 p += ALIGN(ilen, 8);
1170         }
1171
1172         if (!move)
1173                 pack_inode(c, p, old_dir, 1);
1174         else {
1175                 pack_inode(c, p, old_dir, 0);
1176                 p += ALIGN(plen, 8);
1177                 pack_inode(c, p, new_dir, 1);
1178         }
1179
1180         if (last_reference) {
1181                 err = ubifs_add_orphan(c, new_inode->i_ino);
1182                 if (err) {
1183                         release_head(c, BASEHD);
1184                         goto out_finish;
1185                 }
1186                 new_ui->del_cmtno = c->cmt_no;
1187         }
1188
1189         err = write_head(c, BASEHD, dent, len, &lnum, &offs, sync);
1190         if (err)
1191                 goto out_release;
1192         if (!sync) {
1193                 struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf;
1194
1195                 ubifs_wbuf_add_ino_nolock(wbuf, new_dir->i_ino);
1196                 ubifs_wbuf_add_ino_nolock(wbuf, old_dir->i_ino);
1197                 if (new_inode)
1198                         ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf,
1199                                                   new_inode->i_ino);
1200         }
1201         release_head(c, BASEHD);
1202
1203         dent_key_init(c, &key, new_dir->i_ino, new_nm);
1204         err = ubifs_tnc_add_nm(c, &key, lnum, offs, dlen1, new_nm);
1205         if (err)
1206                 goto out_ro;
1207
1208         offs += aligned_dlen1;
1209         if (whiteout) {
1210                 dent_key_init(c, &key, old_dir->i_ino, old_nm);
1211                 err = ubifs_tnc_add_nm(c, &key, lnum, offs, dlen2, old_nm);
1212                 if (err)
1213                         goto out_ro;
1214
1215                 ubifs_delete_orphan(c, whiteout->i_ino);
1216         } else {
1217                 err = ubifs_add_dirt(c, lnum, dlen2);
1218                 if (err)
1219                         goto out_ro;
1220
1221                 dent_key_init(c, &key, old_dir->i_ino, old_nm);
1222                 err = ubifs_tnc_remove_nm(c, &key, old_nm);
1223                 if (err)
1224                         goto out_ro;
1225         }
1226
1227         offs += aligned_dlen2;
1228         if (new_inode) {
1229                 ino_key_init(c, &key, new_inode->i_ino);
1230                 err = ubifs_tnc_add(c, &key, lnum, offs, ilen);
1231                 if (err)
1232                         goto out_ro;
1233                 offs += ALIGN(ilen, 8);
1234         }
1235
1236         ino_key_init(c, &key, old_dir->i_ino);
1237         err = ubifs_tnc_add(c, &key, lnum, offs, plen);
1238         if (err)
1239                 goto out_ro;
1240
1241         if (move) {
1242                 offs += ALIGN(plen, 8);
1243                 ino_key_init(c, &key, new_dir->i_ino);
1244                 err = ubifs_tnc_add(c, &key, lnum, offs, plen);
1245                 if (err)
1246                         goto out_ro;
1247         }
1248
1249         finish_reservation(c);
1250         if (new_inode) {
1251                 mark_inode_clean(c, new_ui);
1252                 spin_lock(&new_ui->ui_lock);
1253                 new_ui->synced_i_size = new_ui->ui_size;
1254                 spin_unlock(&new_ui->ui_lock);
1255         }
1256         mark_inode_clean(c, ubifs_inode(old_dir));
1257         if (move)
1258                 mark_inode_clean(c, ubifs_inode(new_dir));
1259         kfree(dent);
1260         return 0;
1261
1262 out_release:
1263         release_head(c, BASEHD);
1264 out_ro:
1265         ubifs_ro_mode(c, err);
1266         if (last_reference)
1267                 ubifs_delete_orphan(c, new_inode->i_ino);
1268 out_finish:
1269         finish_reservation(c);
1270 out_free:
1271         kfree(dent);
1272         return err;
1273 }
1274
1275 /**
1276  * truncate_data_node - re-compress/encrypt a truncated data node.
1277  * @c: UBIFS file-system description object
1278  * @inode: inode which referes to the data node
1279  * @block: data block number
1280  * @dn: data node to re-compress
1281  * @new_len: new length
1282  *
1283  * This function is used when an inode is truncated and the last data node of
1284  * the inode has to be re-compressed/encrypted and re-written.
1285  */
1286 static int truncate_data_node(const struct ubifs_info *c, const struct inode *inode,
1287                               unsigned int block, struct ubifs_data_node *dn,
1288                               int *new_len)
1289 {
1290         void *buf;
1291         int err, dlen, compr_type, out_len, old_dlen;
1292
1293         out_len = le32_to_cpu(dn->size);
1294         buf = kmalloc(out_len * WORST_COMPR_FACTOR, GFP_NOFS);
1295         if (!buf)
1296                 return -ENOMEM;
1297
1298         dlen = old_dlen = le32_to_cpu(dn->ch.len) - UBIFS_DATA_NODE_SZ;
1299         compr_type = le16_to_cpu(dn->compr_type);
1300
1301         if (ubifs_crypt_is_encrypted(inode)) {
1302                 err = ubifs_decrypt(inode, dn, &dlen, block);
1303                 if (err)
1304                         goto out;
1305         }
1306
1307         if (compr_type == UBIFS_COMPR_NONE) {
1308                 out_len = *new_len;
1309         } else {
1310                 err = ubifs_decompress(c, &dn->data, dlen, buf, &out_len, compr_type);
1311                 if (err)
1312                         goto out;
1313
1314                 ubifs_compress(c, buf, *new_len, &dn->data, &out_len, &compr_type);
1315         }
1316
1317         if (ubifs_crypt_is_encrypted(inode)) {
1318                 err = ubifs_encrypt(inode, dn, out_len, &old_dlen, block);
1319                 if (err)
1320                         goto out;
1321
1322                 out_len = old_dlen;
1323         } else {
1324                 dn->compr_size = 0;
1325         }
1326
1327         ubifs_assert(out_len <= UBIFS_BLOCK_SIZE);
1328         dn->compr_type = cpu_to_le16(compr_type);
1329         dn->size = cpu_to_le32(*new_len);
1330         *new_len = UBIFS_DATA_NODE_SZ + out_len;
1331         err = 0;
1332 out:
1333         kfree(buf);
1334         return err;
1335 }
1336
1337 /**
1338  * ubifs_jnl_truncate - update the journal for a truncation.
1339  * @c: UBIFS file-system description object
1340  * @inode: inode to truncate
1341  * @old_size: old size
1342  * @new_size: new size
1343  *
1344  * When the size of a file decreases due to truncation, a truncation node is
1345  * written, the journal tree is updated, and the last data block is re-written
1346  * if it has been affected. The inode is also updated in order to synchronize
1347  * the new inode size.
1348  *
1349  * This function marks the inode as clean and returns zero on success. In case
1350  * of failure, a negative error code is returned.
1351  */
1352 int ubifs_jnl_truncate(struct ubifs_info *c, const struct inode *inode,
1353                        loff_t old_size, loff_t new_size)
1354 {
1355         union ubifs_key key, to_key;
1356         struct ubifs_ino_node *ino;
1357         struct ubifs_trun_node *trun;
1358         struct ubifs_data_node *uninitialized_var(dn);
1359         int err, dlen, len, lnum, offs, bit, sz, sync = IS_SYNC(inode);
1360         struct ubifs_inode *ui = ubifs_inode(inode);
1361         ino_t inum = inode->i_ino;
1362         unsigned int blk;
1363
1364         dbg_jnl("ino %lu, size %lld -> %lld",
1365                 (unsigned long)inum, old_size, new_size);
1366         ubifs_assert(!ui->data_len);
1367         ubifs_assert(S_ISREG(inode->i_mode));
1368         ubifs_assert(mutex_is_locked(&ui->ui_mutex));
1369
1370         sz = UBIFS_TRUN_NODE_SZ + UBIFS_INO_NODE_SZ +
1371              UBIFS_MAX_DATA_NODE_SZ * WORST_COMPR_FACTOR;
1372         ino = kmalloc(sz, GFP_NOFS);
1373         if (!ino)
1374                 return -ENOMEM;
1375
1376         trun = (void *)ino + UBIFS_INO_NODE_SZ;
1377         trun->ch.node_type = UBIFS_TRUN_NODE;
1378         trun->inum = cpu_to_le32(inum);
1379         trun->old_size = cpu_to_le64(old_size);
1380         trun->new_size = cpu_to_le64(new_size);
1381         zero_trun_node_unused(trun);
1382
1383         dlen = new_size & (UBIFS_BLOCK_SIZE - 1);
1384         if (dlen) {
1385                 /* Get last data block so it can be truncated */
1386                 dn = (void *)trun + UBIFS_TRUN_NODE_SZ;
1387                 blk = new_size >> UBIFS_BLOCK_SHIFT;
1388                 data_key_init(c, &key, inum, blk);
1389                 dbg_jnlk(&key, "last block key ");
1390                 err = ubifs_tnc_lookup(c, &key, dn);
1391                 if (err == -ENOENT)
1392                         dlen = 0; /* Not found (so it is a hole) */
1393                 else if (err)
1394                         goto out_free;
1395                 else {
1396                         int dn_len = le32_to_cpu(dn->size);
1397
1398                         if (dn_len <= 0 || dn_len > UBIFS_BLOCK_SIZE) {
1399                                 ubifs_err(c, "bad data node (block %u, inode %lu)",
1400                                           blk, inode->i_ino);
1401                                 ubifs_dump_node(c, dn);
1402                                 goto out_free;
1403                         }
1404
1405                         if (dn_len <= dlen)
1406                                 dlen = 0; /* Nothing to do */
1407                         else {
1408                                 err = truncate_data_node(c, inode, blk, dn, &dlen);
1409                                 if (err)
1410                                         goto out_free;
1411                         }
1412                 }
1413         }
1414
1415         /* Must make reservation before allocating sequence numbers */
1416         len = UBIFS_TRUN_NODE_SZ + UBIFS_INO_NODE_SZ;
1417         if (dlen)
1418                 len += dlen;
1419         err = make_reservation(c, BASEHD, len);
1420         if (err)
1421                 goto out_free;
1422
1423         pack_inode(c, ino, inode, 0);
1424         ubifs_prep_grp_node(c, trun, UBIFS_TRUN_NODE_SZ, dlen ? 0 : 1);
1425         if (dlen)
1426                 ubifs_prep_grp_node(c, dn, dlen, 1);
1427
1428         err = write_head(c, BASEHD, ino, len, &lnum, &offs, sync);
1429         if (err)
1430                 goto out_release;
1431         if (!sync)
1432                 ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf, inum);
1433         release_head(c, BASEHD);
1434
1435         if (dlen) {
1436                 sz = offs + UBIFS_INO_NODE_SZ + UBIFS_TRUN_NODE_SZ;
1437                 err = ubifs_tnc_add(c, &key, lnum, sz, dlen);
1438                 if (err)
1439                         goto out_ro;
1440         }
1441
1442         ino_key_init(c, &key, inum);
1443         err = ubifs_tnc_add(c, &key, lnum, offs, UBIFS_INO_NODE_SZ);
1444         if (err)
1445                 goto out_ro;
1446
1447         err = ubifs_add_dirt(c, lnum, UBIFS_TRUN_NODE_SZ);
1448         if (err)
1449                 goto out_ro;
1450
1451         bit = new_size & (UBIFS_BLOCK_SIZE - 1);
1452         blk = (new_size >> UBIFS_BLOCK_SHIFT) + (bit ? 1 : 0);
1453         data_key_init(c, &key, inum, blk);
1454
1455         bit = old_size & (UBIFS_BLOCK_SIZE - 1);
1456         blk = (old_size >> UBIFS_BLOCK_SHIFT) - (bit ? 0 : 1);
1457         data_key_init(c, &to_key, inum, blk);
1458
1459         err = ubifs_tnc_remove_range(c, &key, &to_key);
1460         if (err)
1461                 goto out_ro;
1462
1463         finish_reservation(c);
1464         spin_lock(&ui->ui_lock);
1465         ui->synced_i_size = ui->ui_size;
1466         spin_unlock(&ui->ui_lock);
1467         mark_inode_clean(c, ui);
1468         kfree(ino);
1469         return 0;
1470
1471 out_release:
1472         release_head(c, BASEHD);
1473 out_ro:
1474         ubifs_ro_mode(c, err);
1475         finish_reservation(c);
1476 out_free:
1477         kfree(ino);
1478         return err;
1479 }
1480
1481
1482 /**
1483  * ubifs_jnl_delete_xattr - delete an extended attribute.
1484  * @c: UBIFS file-system description object
1485  * @host: host inode
1486  * @inode: extended attribute inode
1487  * @nm: extended attribute entry name
1488  *
1489  * This function delete an extended attribute which is very similar to
1490  * un-linking regular files - it writes a deletion xentry, a deletion inode and
1491  * updates the target inode. Returns zero in case of success and a negative
1492  * error code in case of failure.
1493  */
1494 int ubifs_jnl_delete_xattr(struct ubifs_info *c, const struct inode *host,
1495                            const struct inode *inode,
1496                            const struct fscrypt_name *nm)
1497 {
1498         int err, xlen, hlen, len, lnum, xent_offs, aligned_xlen;
1499         struct ubifs_dent_node *xent;
1500         struct ubifs_ino_node *ino;
1501         union ubifs_key xent_key, key1, key2;
1502         int sync = IS_DIRSYNC(host);
1503         struct ubifs_inode *host_ui = ubifs_inode(host);
1504
1505         ubifs_assert(inode->i_nlink == 0);
1506         ubifs_assert(mutex_is_locked(&host_ui->ui_mutex));
1507
1508         /*
1509          * Since we are deleting the inode, we do not bother to attach any data
1510          * to it and assume its length is %UBIFS_INO_NODE_SZ.
1511          */
1512         xlen = UBIFS_DENT_NODE_SZ + fname_len(nm) + 1;
1513         aligned_xlen = ALIGN(xlen, 8);
1514         hlen = host_ui->data_len + UBIFS_INO_NODE_SZ;
1515         len = aligned_xlen + UBIFS_INO_NODE_SZ + ALIGN(hlen, 8);
1516
1517         xent = kzalloc(len, GFP_NOFS);
1518         if (!xent)
1519                 return -ENOMEM;
1520
1521         /* Make reservation before allocating sequence numbers */
1522         err = make_reservation(c, BASEHD, len);
1523         if (err) {
1524                 kfree(xent);
1525                 return err;
1526         }
1527
1528         xent->ch.node_type = UBIFS_XENT_NODE;
1529         xent_key_init(c, &xent_key, host->i_ino, nm);
1530         key_write(c, &xent_key, xent->key);
1531         xent->inum = 0;
1532         xent->type = get_dent_type(inode->i_mode);
1533         xent->nlen = cpu_to_le16(fname_len(nm));
1534         memcpy(xent->name, fname_name(nm), fname_len(nm));
1535         xent->name[fname_len(nm)] = '\0';
1536         zero_dent_node_unused(xent);
1537         ubifs_prep_grp_node(c, xent, xlen, 0);
1538
1539         ino = (void *)xent + aligned_xlen;
1540         pack_inode(c, ino, inode, 0);
1541         ino = (void *)ino + UBIFS_INO_NODE_SZ;
1542         pack_inode(c, ino, host, 1);
1543
1544         err = write_head(c, BASEHD, xent, len, &lnum, &xent_offs, sync);
1545         if (!sync && !err)
1546                 ubifs_wbuf_add_ino_nolock(&c->jheads[BASEHD].wbuf, host->i_ino);
1547         release_head(c, BASEHD);
1548         kfree(xent);
1549         if (err)
1550                 goto out_ro;
1551
1552         /* Remove the extended attribute entry from TNC */
1553         err = ubifs_tnc_remove_nm(c, &xent_key, nm);
1554         if (err)
1555                 goto out_ro;
1556         err = ubifs_add_dirt(c, lnum, xlen);
1557         if (err)
1558                 goto out_ro;
1559
1560         /*
1561          * Remove all nodes belonging to the extended attribute inode from TNC.
1562          * Well, there actually must be only one node - the inode itself.
1563          */
1564         lowest_ino_key(c, &key1, inode->i_ino);
1565         highest_ino_key(c, &key2, inode->i_ino);
1566         err = ubifs_tnc_remove_range(c, &key1, &key2);
1567         if (err)
1568                 goto out_ro;
1569         err = ubifs_add_dirt(c, lnum, UBIFS_INO_NODE_SZ);
1570         if (err)
1571                 goto out_ro;
1572
1573         /* And update TNC with the new host inode position */
1574         ino_key_init(c, &key1, host->i_ino);
1575         err = ubifs_tnc_add(c, &key1, lnum, xent_offs + len - hlen, hlen);
1576         if (err)
1577                 goto out_ro;
1578
1579         finish_reservation(c);
1580         spin_lock(&host_ui->ui_lock);
1581         host_ui->synced_i_size = host_ui->ui_size;
1582         spin_unlock(&host_ui->ui_lock);
1583         mark_inode_clean(c, host_ui);
1584         return 0;
1585
1586 out_ro:
1587         ubifs_ro_mode(c, err);
1588         finish_reservation(c);
1589         return err;
1590 }
1591
1592 /**
1593  * ubifs_jnl_change_xattr - change an extended attribute.
1594  * @c: UBIFS file-system description object
1595  * @inode: extended attribute inode
1596  * @host: host inode
1597  *
1598  * This function writes the updated version of an extended attribute inode and
1599  * the host inode to the journal (to the base head). The host inode is written
1600  * after the extended attribute inode in order to guarantee that the extended
1601  * attribute will be flushed when the inode is synchronized by 'fsync()' and
1602  * consequently, the write-buffer is synchronized. This function returns zero
1603  * in case of success and a negative error code in case of failure.
1604  */
1605 int ubifs_jnl_change_xattr(struct ubifs_info *c, const struct inode *inode,
1606                            const struct inode *host)
1607 {
1608         int err, len1, len2, aligned_len, aligned_len1, lnum, offs;
1609         struct ubifs_inode *host_ui = ubifs_inode(host);
1610         struct ubifs_ino_node *ino;
1611         union ubifs_key key;
1612         int sync = IS_DIRSYNC(host);
1613
1614         dbg_jnl("ino %lu, ino %lu", host->i_ino, inode->i_ino);
1615         ubifs_assert(host->i_nlink > 0);
1616         ubifs_assert(inode->i_nlink > 0);
1617         ubifs_assert(mutex_is_locked(&host_ui->ui_mutex));
1618
1619         len1 = UBIFS_INO_NODE_SZ + host_ui->data_len;
1620         len2 = UBIFS_INO_NODE_SZ + ubifs_inode(inode)->data_len;
1621         aligned_len1 = ALIGN(len1, 8);
1622         aligned_len = aligned_len1 + ALIGN(len2, 8);
1623
1624         ino = kzalloc(aligned_len, GFP_NOFS);
1625         if (!ino)
1626                 return -ENOMEM;
1627
1628         /* Make reservation before allocating sequence numbers */
1629         err = make_reservation(c, BASEHD, aligned_len);
1630         if (err)
1631                 goto out_free;
1632
1633         pack_inode(c, ino, host, 0);
1634         pack_inode(c, (void *)ino + aligned_len1, inode, 1);
1635
1636         err = write_head(c, BASEHD, ino, aligned_len, &lnum, &offs, 0);
1637         if (!sync && !err) {
1638                 struct ubifs_wbuf *wbuf = &c->jheads[BASEHD].wbuf;
1639
1640                 ubifs_wbuf_add_ino_nolock(wbuf, host->i_ino);
1641                 ubifs_wbuf_add_ino_nolock(wbuf, inode->i_ino);
1642         }
1643         release_head(c, BASEHD);
1644         if (err)
1645                 goto out_ro;
1646
1647         ino_key_init(c, &key, host->i_ino);
1648         err = ubifs_tnc_add(c, &key, lnum, offs, len1);
1649         if (err)
1650                 goto out_ro;
1651
1652         ino_key_init(c, &key, inode->i_ino);
1653         err = ubifs_tnc_add(c, &key, lnum, offs + aligned_len1, len2);
1654         if (err)
1655                 goto out_ro;
1656
1657         finish_reservation(c);
1658         spin_lock(&host_ui->ui_lock);
1659         host_ui->synced_i_size = host_ui->ui_size;
1660         spin_unlock(&host_ui->ui_lock);
1661         mark_inode_clean(c, host_ui);
1662         kfree(ino);
1663         return 0;
1664
1665 out_ro:
1666         ubifs_ro_mode(c, err);
1667         finish_reservation(c);
1668 out_free:
1669         kfree(ino);
1670         return err;
1671 }
1672