fs/ubifs/dir.c

   1 /* * This file is part of UBIFS.
   2  *
   3  * Copyright (C) 2006-2008 Nokia Corporation.
   4  * Copyright (C) 2006, 2007 University of Szeged, Hungary
   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  *          Zoltan Sogor
  22  */
  23
  24 /*
  25  * This file implements directory operations.
  26  *
  27  * All FS operations in this file allocate budget before writing anything to the
  28  * media. If they fail to allocate it, the error is returned. The only
  29  * exceptions are 'ubifs_unlink()' and 'ubifs_rmdir()' which keep working even
  30  * if they unable to allocate the budget, because deletion %-ENOSPC failure is
  31  * not what users are usually ready to get. UBIFS budgeting subsystem has some
  32  * space reserved for these purposes.
  33  *
  34  * All operations in this file write all inodes which they change straight
  35  * away, instead of marking them dirty. For example, 'ubifs_link()' changes
  36  * @i_size of the parent inode and writes the parent inode together with the
  37  * target inode. This was done to simplify file-system recovery which would
  38  * otherwise be very difficult to do. The only exception is rename which marks
  39  * the re-named inode dirty (because its @i_ctime is updated) but does not
  40  * write it, but just marks it as dirty.
  41  */
  42
  43 #include "ubifs.h"
  44
  45 /**
  46  * inherit_flags - inherit flags of the parent inode.
  47  * @dir: parent inode
  48  * @mode: new inode mode flags
  49  *
  50  * This is a helper function for 'ubifs_new_inode()' which inherits flag of the
  51  * parent directory inode @dir. UBIFS inodes inherit the following flags:
  52  * o %UBIFS_COMPR_FL, which is useful to switch compression on/of on
  53  *   sub-directory basis;
  54  * o %UBIFS_SYNC_FL - useful for the same reasons;
  55  * o %UBIFS_DIRSYNC_FL - similar, but relevant only to directories.
  56  *
  57  * This function returns the inherited flags.
  58  */
  59 static int inherit_flags(const struct inode *dir, int mode)
  60 {
  61         int flags;
  62         const struct ubifs_inode *ui = ubifs_inode(dir);
  63
  64         if (!S_ISDIR(dir->i_mode))
  65                 /*
  66                  * The parent is not a directory, which means that an extended
  67                  * attribute inode is being created. No flags.
  68                  */
  69                 return 0;
  70
  71         flags = ui->flags & (UBIFS_COMPR_FL | UBIFS_SYNC_FL | UBIFS_DIRSYNC_FL);
  72         if (!S_ISDIR(mode))
  73                 /* The "DIRSYNC" flag only applies to directories */
  74                 flags &= ~UBIFS_DIRSYNC_FL;
  75         return flags;
  76 }
  77
  78 /**
  79  * ubifs_new_inode - allocate new UBIFS inode object.
  80  * @c: UBIFS file-system description object
  81  * @dir: parent directory inode
  82  * @mode: inode mode flags
  83  *
  84  * This function finds an unused inode number, allocates new inode and
  85  * initializes it. Returns new inode in case of success and an error code in
  86  * case of failure.
  87  */
  88 struct inode *ubifs_new_inode(struct ubifs_info *c, const struct inode *dir,
  89                               int mode)
  90 {
  91         struct inode *inode;
  92         struct ubifs_inode *ui;
  93
  94         inode = new_inode(c->vfs_sb);
  95         ui = ubifs_inode(inode);
  96         if (!inode)
  97                 return ERR_PTR(-ENOMEM);
  98
  99         /*
 100          * Set 'S_NOCMTIME' to prevent VFS form updating [mc]time of inodes and
 101          * marking them dirty in file write path (see 'file_update_time()').
 102          * UBIFS has to fully control "clean <-> dirty" transitions of inodes
 103          * to make budgeting work.
 104          */
 105         inode->i_flags |= (S_NOCMTIME);
 106
 107         inode_init_owner(inode, dir, mode);
 108         inode->i_mtime = inode->i_atime = inode->i_ctime =
 109                          ubifs_current_time(inode);
 110         inode->i_mapping->nrpages = 0;
 111         /* Disable readahead */
 112         inode->i_mapping->backing_dev_info = &c->bdi;
 113
 114         switch (mode & S_IFMT) {
 115         case S_IFREG:
 116                 inode->i_mapping->a_ops = &ubifs_file_address_operations;
 117                 inode->i_op = &ubifs_file_inode_operations;
 118                 inode->i_fop = &ubifs_file_operations;
 119                 break;
 120         case S_IFDIR:
 121                 inode->i_op  = &ubifs_dir_inode_operations;
 122                 inode->i_fop = &ubifs_dir_operations;
 123                 inode->i_size = ui->ui_size = UBIFS_INO_NODE_SZ;
 124                 break;
 125         case S_IFLNK:
 126                 inode->i_op = &ubifs_symlink_inode_operations;
 127                 break;
 128         case S_IFSOCK:
 129         case S_IFIFO:
 130         case S_IFBLK:
 131         case S_IFCHR:
 132                 inode->i_op  = &ubifs_file_inode_operations;
 133                 break;
 134         default:
 135                 BUG();
 136         }
 137
 138         ui->flags = inherit_flags(dir, mode);
 139         ubifs_set_inode_flags(inode);
 140         if (S_ISREG(mode))
 141                 ui->compr_type = c->default_compr;
 142         else
 143                 ui->compr_type = UBIFS_COMPR_NONE;
 144         ui->synced_i_size = 0;
 145
 146         spin_lock(&c->cnt_lock);
 147         /* Inode number overflow is currently not supported */
 148         if (c->highest_inum >= INUM_WARN_WATERMARK) {
 149                 if (c->highest_inum >= INUM_WATERMARK) {
 150                         spin_unlock(&c->cnt_lock);
 151                         ubifs_err("out of inode numbers");
 152                         make_bad_inode(inode);
 153                         iput(inode);
 154                         return ERR_PTR(-EINVAL);
 155                 }
 156                 ubifs_warn("running out of inode numbers (current %lu, max %d)",
 157                            (unsigned long)c->highest_inum, INUM_WATERMARK);
 158         }
 159
 160         inode->i_ino = ++c->highest_inum;
 161         /*
 162          * The creation sequence number remains with this inode for its
 163          * lifetime. All nodes for this inode have a greater sequence number,
 164          * and so it is possible to distinguish obsolete nodes belonging to a
 165          * previous incarnation of the same inode number - for example, for the
 166          * purpose of rebuilding the index.
 167          */
 168         ui->creat_sqnum = ++c->max_sqnum;
 169         spin_unlock(&c->cnt_lock);
 170         return inode;
 171 }
 172
 173 #ifdef CONFIG_UBIFS_FS_DEBUG
 174
 175 static int dbg_check_name(struct ubifs_dent_node *dent, struct qstr *nm)
 176 {
 177         if (!(ubifs_chk_flags & UBIFS_CHK_GEN))
 178                 return 0;
 179         if (le16_to_cpu(dent->nlen) != nm->len)
 180                 return -EINVAL;
 181         if (memcmp(dent->name, nm->name, nm->len))
 182                 return -EINVAL;
 183         return 0;
 184 }
 185
 186 #else
 187
 188 #define dbg_check_name(dent, nm) 0
 189
 190 #endif
 191
 192 static struct dentry *ubifs_lookup(struct inode *dir, struct dentry *dentry,
 193                                    struct nameidata *nd)
 194 {
 195         int err;
 196         union ubifs_key key;
 197         struct inode *inode = NULL;
 198         struct ubifs_dent_node *dent;
 199         struct ubifs_info *c = dir->i_sb->s_fs_info;
 200
 201         dbg_gen("'%.*s' in dir ino %lu",
 202                 dentry->d_name.len, dentry->d_name.name, dir->i_ino);
 203
 204         if (dentry->d_name.len > UBIFS_MAX_NLEN)
 205                 return ERR_PTR(-ENAMETOOLONG);
 206
 207         dent = kmalloc(UBIFS_MAX_DENT_NODE_SZ, GFP_NOFS);
 208         if (!dent)
 209                 return ERR_PTR(-ENOMEM);
 210
 211         dent_key_init(c, &key, dir->i_ino, &dentry->d_name);
 212
 213         err = ubifs_tnc_lookup_nm(c, &key, dent, &dentry->d_name);
 214         if (err) {
 215                 if (err == -ENOENT) {
 216                         dbg_gen("not found");
 217                         goto done;
 218                 }
 219                 goto out;
 220         }
 221
 222         if (dbg_check_name(dent, &dentry->d_name)) {
 223                 err = -EINVAL;
 224                 goto out;
 225         }
 226
 227         inode = ubifs_iget(dir->i_sb, le64_to_cpu(dent->inum));
 228         if (IS_ERR(inode)) {
 229                 /*
 230                  * This should not happen. Probably the file-system needs
 231                  * checking.
 232                  */
 233                 err = PTR_ERR(inode);
 234                 ubifs_err("dead directory entry '%.*s', error %d",
 235                           dentry->d_name.len, dentry->d_name.name, err);
 236                 ubifs_ro_mode(c, err);
 237                 goto out;
 238         }
 239
 240 done:
 241         kfree(dent);
 242         /*
 243          * Note, d_splice_alias() would be required instead if we supported
 244          * NFS.
 245          */
 246         d_add(dentry, inode);
 247         return NULL;
 248
 249 out:
 250         kfree(dent);
 251         return ERR_PTR(err);
 252 }
 253
 254 static int ubifs_create(struct inode *dir, struct dentry *dentry, int mode,
 255                         struct nameidata *nd)
 256 {
 257         struct inode *inode;
 258         struct ubifs_info *c = dir->i_sb->s_fs_info;
 259         int err, sz_change = CALC_DENT_SIZE(dentry->d_name.len);
 260         struct ubifs_budget_req req = { .new_ino = 1, .new_dent = 1,
 261                                         .dirtied_ino = 1 };
 262         struct ubifs_inode *dir_ui = ubifs_inode(dir);
 263
 264         /*
 265          * Budget request settings: new inode, new direntry, changing the
 266          * parent directory inode.
 267          */
 268
 269         dbg_gen("dent '%.*s', mode %#x in dir ino %lu",
 270                 dentry->d_name.len, dentry->d_name.name, mode, dir->i_ino);
 271
 272         err = ubifs_budget_space(c, &req);
 273         if (err)
 274                 return err;
 275
 276         inode = ubifs_new_inode(c, dir, mode);
 277         if (IS_ERR(inode)) {
 278                 err = PTR_ERR(inode);
 279                 goto out_budg;
 280         }
 281
 282         mutex_lock(&dir_ui->ui_mutex);
 283         dir->i_size += sz_change;
 284         dir_ui->ui_size = dir->i_size;
 285         dir->i_mtime = dir->i_ctime = inode->i_ctime;
 286         err = ubifs_jnl_update(c, dir, &dentry->d_name, inode, 0, 0);
 287         if (err)
 288                 goto out_cancel;
 289         mutex_unlock(&dir_ui->ui_mutex);
 290
 291         ubifs_release_budget(c, &req);
 292         insert_inode_hash(inode);
 293         d_instantiate(dentry, inode);
 294         return 0;
 295
 296 out_cancel:
 297         dir->i_size -= sz_change;
 298         dir_ui->ui_size = dir->i_size;
 299         mutex_unlock(&dir_ui->ui_mutex);
 300         make_bad_inode(inode);
 301         iput(inode);
 302 out_budg:
 303         ubifs_release_budget(c, &req);
 304         ubifs_err("cannot create regular file, error %d", err);
 305         return err;
 306 }
 307
 308 /**
 309  * vfs_dent_type - get VFS directory entry type.
 310  * @type: UBIFS directory entry type
 311  *
 312  * This function converts UBIFS directory entry type into VFS directory entry
 313  * type.
 314  */
 315 static unsigned int vfs_dent_type(uint8_t type)
 316 {
 317         switch (type) {
 318         case UBIFS_ITYPE_REG:
 319                 return DT_REG;
 320         case UBIFS_ITYPE_DIR:
 321                 return DT_DIR;
 322         case UBIFS_ITYPE_LNK:
 323                 return DT_LNK;
 324         case UBIFS_ITYPE_BLK:
 325                 return DT_BLK;
 326         case UBIFS_ITYPE_CHR:
 327                 return DT_CHR;
 328         case UBIFS_ITYPE_FIFO:
 329                 return DT_FIFO;
 330         case UBIFS_ITYPE_SOCK:
 331                 return DT_SOCK;
 332         default:
 333                 BUG();
 334         }
 335         return 0;
 336 }
 337
 338 /*
 339  * The classical Unix view for directory is that it is a linear array of
 340  * (name, inode number) entries. Linux/VFS assumes this model as well.
 341  * Particularly, 'readdir()' call wants us to return a directory entry offset
 342  * which later may be used to continue 'readdir()'ing the directory or to
 343  * 'seek()' to that specific direntry. Obviously UBIFS does not really fit this
 344  * model because directory entries are identified by keys, which may collide.
 345  *
 346  * UBIFS uses directory entry hash value for directory offsets, so
 347  * 'seekdir()'/'telldir()' may not always work because of possible key
 348  * collisions. But UBIFS guarantees that consecutive 'readdir()' calls work
 349  * properly by means of saving full directory entry name in the private field
 350  * of the file description object.
 351  *
 352  * This means that UBIFS cannot support NFS which requires full
 353  * 'seekdir()'/'telldir()' support.
 354  */
 355 static int ubifs_readdir(struct file *file, void *dirent, filldir_t filldir)
 356 {
 357         int err, over = 0;
 358         struct qstr nm;
 359         union ubifs_key key;
 360         struct ubifs_dent_node *dent;
 361         struct inode *dir = file->f_path.dentry->d_inode;
 362         struct ubifs_info *c = dir->i_sb->s_fs_info;
 363
 364         dbg_gen("dir ino %lu, f_pos %#llx", dir->i_ino, file->f_pos);
 365
 366         if (file->f_pos > UBIFS_S_KEY_HASH_MASK || file->f_pos == 2)
 367                 /*
 368                  * The directory was seek'ed to a senseless position or there
 369                  * are no more entries.
 370                  */
 371                 return 0;
 372
 373         /* File positions 0 and 1 correspond to "." and ".." */
 374         if (file->f_pos == 0) {
 375                 ubifs_assert(!file->private_data);
 376                 over = filldir(dirent, ".", 1, 0, dir->i_ino, DT_DIR);
 377                 if (over)
 378                         return 0;
 379                 file->f_pos = 1;
 380         }
 381
 382         if (file->f_pos == 1) {
 383                 ubifs_assert(!file->private_data);
 384                 over = filldir(dirent, "..", 2, 1,
 385                                parent_ino(file->f_path.dentry), DT_DIR);
 386                 if (over)
 387                         return 0;
 388
 389                 /* Find the first entry in TNC and save it */
 390                 lowest_dent_key(c, &key, dir->i_ino);
 391                 nm.name = NULL;
 392                 dent = ubifs_tnc_next_ent(c, &key, &nm);
 393                 if (IS_ERR(dent)) {
 394                         err = PTR_ERR(dent);
 395                         goto out;
 396                 }
 397
 398                 file->f_pos = key_hash_flash(c, &dent->key);
 399                 file->private_data = dent;
 400         }
 401
 402         dent = file->private_data;
 403         if (!dent) {
 404                 /*
 405                  * The directory was seek'ed to and is now readdir'ed.
 406                  * Find the entry corresponding to @file->f_pos or the
 407                  * closest one.
 408                  */
 409                 dent_key_init_hash(c, &key, dir->i_ino, file->f_pos);
 410                 nm.name = NULL;
 411                 dent = ubifs_tnc_next_ent(c, &key, &nm);
 412                 if (IS_ERR(dent)) {
 413                         err = PTR_ERR(dent);
 414                         goto out;
 415                 }
 416                 file->f_pos = key_hash_flash(c, &dent->key);
 417                 file->private_data = dent;
 418         }
 419
 420         while (1) {
 421                 dbg_gen("feed '%s', ino %llu, new f_pos %#x",
 422                         dent->name, (unsigned long long)le64_to_cpu(dent->inum),
 423                         key_hash_flash(c, &dent->key));
 424                 ubifs_assert(le64_to_cpu(dent->ch.sqnum) >
 425                              ubifs_inode(dir)->creat_sqnum);
 426
 427                 nm.len = le16_to_cpu(dent->nlen);
 428                 over = filldir(dirent, dent->name, nm.len, file->f_pos,
 429                                le64_to_cpu(dent->inum),
 430                                vfs_dent_type(dent->type));
 431                 if (over)
 432                         return 0;
 433
 434                 /* Switch to the next entry */
 435                 key_read(c, &dent->key, &key);
 436                 nm.name = dent->name;
 437                 dent = ubifs_tnc_next_ent(c, &key, &nm);
 438                 if (IS_ERR(dent)) {
 439                         err = PTR_ERR(dent);
 440                         goto out;
 441                 }
 442
 443                 kfree(file->private_data);
 444                 file->f_pos = key_hash_flash(c, &dent->key);
 445                 file->private_data = dent;
 446                 cond_resched();
 447         }
 448
 449 out:
 450         if (err != -ENOENT) {
 451                 ubifs_err("cannot find next direntry, error %d", err);
 452                 return err;
 453         }
 454
 455         kfree(file->private_data);
 456         file->private_data = NULL;
 457         file->f_pos = 2;
 458         return 0;
 459 }
 460
 461 /* If a directory is seeked, we have to free saved readdir() state */
 462 static loff_t ubifs_dir_llseek(struct file *file, loff_t offset, int origin)
 463 {
 464         kfree(file->private_data);
 465         file->private_data = NULL;
 466         return generic_file_llseek(file, offset, origin);
 467 }
 468
 469 /* Free saved readdir() state when the directory is closed */
 470 static int ubifs_dir_release(struct inode *dir, struct file *file)
 471 {
 472         kfree(file->private_data);
 473         file->private_data = NULL;
 474         return 0;
 475 }
 476
 477 /**
 478  * lock_2_inodes - a wrapper for locking two UBIFS inodes.
 479  * @inode1: first inode
 480  * @inode2: second inode
 481  *
 482  * We do not implement any tricks to guarantee strict lock ordering, because
 483  * VFS has already done it for us on the @i_mutex. So this is just a simple
 484  * wrapper function.
 485  */
 486 static void lock_2_inodes(struct inode *inode1, struct inode *inode2)
 487 {
 488         mutex_lock_nested(&ubifs_inode(inode1)->ui_mutex, WB_MUTEX_1);
 489         mutex_lock_nested(&ubifs_inode(inode2)->ui_mutex, WB_MUTEX_2);
 490 }
 491
 492 /**
 493  * unlock_2_inodes - a wrapper for unlocking two UBIFS inodes.
 494  * @inode1: first inode
 495  * @inode2: second inode
 496  */
 497 static void unlock_2_inodes(struct inode *inode1, struct inode *inode2)
 498 {
 499         mutex_unlock(&ubifs_inode(inode2)->ui_mutex);
 500         mutex_unlock(&ubifs_inode(inode1)->ui_mutex);
 501 }
 502
 503 static int ubifs_link(struct dentry *old_dentry, struct inode *dir,
 504                       struct dentry *dentry)
 505 {
 506         struct ubifs_info *c = dir->i_sb->s_fs_info;
 507         struct inode *inode = old_dentry->d_inode;
 508         struct ubifs_inode *ui = ubifs_inode(inode);
 509         struct ubifs_inode *dir_ui = ubifs_inode(dir);
 510         int err, sz_change = CALC_DENT_SIZE(dentry->d_name.len);
 511         struct ubifs_budget_req req = { .new_dent = 1, .dirtied_ino = 2,
 512                                 .dirtied_ino_d = ALIGN(ui->data_len, 8) };
 513
 514         /*
 515          * Budget request settings: new direntry, changing the target inode,
 516          * changing the parent inode.
 517          */
 518
 519         dbg_gen("dent '%.*s' to ino %lu (nlink %d) in dir ino %lu",
 520                 dentry->d_name.len, dentry->d_name.name, inode->i_ino,
 521                 inode->i_nlink, dir->i_ino);
 522         ubifs_assert(mutex_is_locked(&dir->i_mutex));
 523         ubifs_assert(mutex_is_locked(&inode->i_mutex));
 524
 525         /*
 526          * Return -ENOENT if we've raced with unlink and i_nlink is 0.  Doing
 527          * otherwise has the potential to corrupt the orphan inode list.
 528          *
 529          * Indeed, consider a scenario when 'vfs_link(dirA/fileA)' and
 530          * 'vfs_unlink(dirA/fileA, dirB/fileB)' race. 'vfs_link()' does not
 531          * lock 'dirA->i_mutex', so this is possible. Both of the functions
 532          * lock 'fileA->i_mutex' though. Suppose 'vfs_unlink()' wins, and takes
 533          * 'fileA->i_mutex' mutex first. Suppose 'fileA->i_nlink' is 1. In this
 534          * case 'ubifs_unlink()' will drop the last reference, and put 'inodeA'
 535          * to the list of orphans. After this, 'vfs_link()' will link
 536          * 'dirB/fileB' to 'inodeA'. This is a problem because, for example,
 537          * the subsequent 'vfs_unlink(dirB/fileB)' will add the same inode
 538          * to the list of orphans.
 539          */
 540          if (inode->i_nlink == 0)
 541                  return -ENOENT;
 542
 543         err = dbg_check_synced_i_size(inode);
 544         if (err)
 545                 return err;
 546
 547         err = ubifs_budget_space(c, &req);
 548         if (err)
 549                 return err;
 550
 551         lock_2_inodes(dir, inode);
 552         inc_nlink(inode);
 553         atomic_inc(&inode->i_count);
 554         inode->i_ctime = ubifs_current_time(inode);
 555         dir->i_size += sz_change;
 556         dir_ui->ui_size = dir->i_size;
 557         dir->i_mtime = dir->i_ctime = inode->i_ctime;
 558         err = ubifs_jnl_update(c, dir, &dentry->d_name, inode, 0, 0);
 559         if (err)
 560                 goto out_cancel;
 561         unlock_2_inodes(dir, inode);
 562
 563         ubifs_release_budget(c, &req);
 564         d_instantiate(dentry, inode);
 565         return 0;
 566
 567 out_cancel:
 568         dir->i_size -= sz_change;
 569         dir_ui->ui_size = dir->i_size;
 570         drop_nlink(inode);
 571         unlock_2_inodes(dir, inode);
 572         ubifs_release_budget(c, &req);
 573         iput(inode);
 574         return err;
 575 }
 576
 577 static int ubifs_unlink(struct inode *dir, struct dentry *dentry)
 578 {
 579         struct ubifs_info *c = dir->i_sb->s_fs_info;
 580         struct inode *inode = dentry->d_inode;
 581         struct ubifs_inode *dir_ui = ubifs_inode(dir);
 582         int sz_change = CALC_DENT_SIZE(dentry->d_name.len);
 583         int err, budgeted = 1;
 584         struct ubifs_budget_req req = { .mod_dent = 1, .dirtied_ino = 2 };
 585
 586         /*
 587          * Budget request settings: deletion direntry, deletion inode (+1 for
 588          * @dirtied_ino), changing the parent directory inode. If budgeting
 589          * fails, go ahead anyway because we have extra space reserved for
 590          * deletions.
 591          */
 592
 593         dbg_gen("dent '%.*s' from ino %lu (nlink %d) in dir ino %lu",
 594                 dentry->d_name.len, dentry->d_name.name, inode->i_ino,
 595                 inode->i_nlink, dir->i_ino);
 596         ubifs_assert(mutex_is_locked(&dir->i_mutex));
 597         ubifs_assert(mutex_is_locked(&inode->i_mutex));
 598         err = dbg_check_synced_i_size(inode);
 599         if (err)
 600                 return err;
 601
 602         err = ubifs_budget_space(c, &req);
 603         if (err) {
 604                 if (err != -ENOSPC)
 605                         return err;
 606                 budgeted = 0;
 607         }
 608
 609         lock_2_inodes(dir, inode);
 610         inode->i_ctime = ubifs_current_time(dir);
 611         drop_nlink(inode);
 612         dir->i_size -= sz_change;
 613         dir_ui->ui_size = dir->i_size;
 614         dir->i_mtime = dir->i_ctime = inode->i_ctime;
 615         err = ubifs_jnl_update(c, dir, &dentry->d_name, inode, 1, 0);
 616         if (err)
 617                 goto out_cancel;
 618         unlock_2_inodes(dir, inode);
 619
 620         if (budgeted)
 621                 ubifs_release_budget(c, &req);
 622         else {
 623                 /* We've deleted something - clean the "no space" flags */
 624                 c->nospace = c->nospace_rp = 0;
 625                 smp_wmb();
 626         }
 627         return 0;
 628
 629 out_cancel:
 630         dir->i_size += sz_change;
 631         dir_ui->ui_size = dir->i_size;
 632         inc_nlink(inode);
 633         unlock_2_inodes(dir, inode);
 634         if (budgeted)
 635                 ubifs_release_budget(c, &req);
 636         return err;
 637 }
 638
 639 /**
 640  * check_dir_empty - check if a directory is empty or not.
 641  * @c: UBIFS file-system description object
 642  * @dir: VFS inode object of the directory to check
 643  *
 644  * This function checks if directory @dir is empty. Returns zero if the
 645  * directory is empty, %-ENOTEMPTY if it is not, and other negative error codes
 646  * in case of of errors.
 647  */
 648 static int check_dir_empty(struct ubifs_info *c, struct inode *dir)
 649 {
 650         struct qstr nm = { .name = NULL };
 651         struct ubifs_dent_node *dent;
 652         union ubifs_key key;
 653         int err;
 654
 655         lowest_dent_key(c, &key, dir->i_ino);
 656         dent = ubifs_tnc_next_ent(c, &key, &nm);
 657         if (IS_ERR(dent)) {
 658                 err = PTR_ERR(dent);
 659                 if (err == -ENOENT)
 660                         err = 0;
 661         } else {
 662                 kfree(dent);
 663                 err = -ENOTEMPTY;
 664         }
 665         return err;
 666 }
 667
 668 static int ubifs_rmdir(struct inode *dir, struct dentry *dentry)
 669 {
 670         struct ubifs_info *c = dir->i_sb->s_fs_info;
 671         struct inode *inode = dentry->d_inode;
 672         int sz_change = CALC_DENT_SIZE(dentry->d_name.len);
 673         int err, budgeted = 1;
 674         struct ubifs_inode *dir_ui = ubifs_inode(dir);
 675         struct ubifs_budget_req req = { .mod_dent = 1, .dirtied_ino = 2 };
 676
 677         /*
 678          * Budget request settings: deletion direntry, deletion inode and
 679          * changing the parent inode. If budgeting fails, go ahead anyway
 680          * because we have extra space reserved for deletions.
 681          */
 682
 683         dbg_gen("directory '%.*s', ino %lu in dir ino %lu", dentry->d_name.len,
 684                 dentry->d_name.name, inode->i_ino, dir->i_ino);
 685         ubifs_assert(mutex_is_locked(&dir->i_mutex));
 686         ubifs_assert(mutex_is_locked(&inode->i_mutex));
 687         err = check_dir_empty(c, dentry->d_inode);
 688         if (err)
 689                 return err;
 690
 691         err = ubifs_budget_space(c, &req);
 692         if (err) {
 693                 if (err != -ENOSPC)
 694                         return err;
 695                 budgeted = 0;
 696         }
 697
 698         lock_2_inodes(dir, inode);
 699         inode->i_ctime = ubifs_current_time(dir);
 700         clear_nlink(inode);
 701         drop_nlink(dir);
 702         dir->i_size -= sz_change;
 703         dir_ui->ui_size = dir->i_size;
 704         dir->i_mtime = dir->i_ctime = inode->i_ctime;
 705         err = ubifs_jnl_update(c, dir, &dentry->d_name, inode, 1, 0);
 706         if (err)
 707                 goto out_cancel;
 708         unlock_2_inodes(dir, inode);
 709
 710         if (budgeted)
 711                 ubifs_release_budget(c, &req);
 712         else {
 713                 /* We've deleted something - clean the "no space" flags */
 714                 c->nospace = c->nospace_rp = 0;
 715                 smp_wmb();
 716         }
 717         return 0;
 718
 719 out_cancel:
 720         dir->i_size += sz_change;
 721         dir_ui->ui_size = dir->i_size;
 722         inc_nlink(dir);
 723         inc_nlink(inode);
 724         inc_nlink(inode);
 725         unlock_2_inodes(dir, inode);
 726         if (budgeted)
 727                 ubifs_release_budget(c, &req);
 728         return err;
 729 }
 730
 731 static int ubifs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 732 {
 733         struct inode *inode;
 734         struct ubifs_inode *dir_ui = ubifs_inode(dir);
 735         struct ubifs_info *c = dir->i_sb->s_fs_info;
 736         int err, sz_change = CALC_DENT_SIZE(dentry->d_name.len);
 737         struct ubifs_budget_req req = { .new_ino = 1, .new_dent = 1 };
 738
 739         /*
 740          * Budget request settings: new inode, new direntry and changing parent
 741          * directory inode.
 742          */
 743
 744         dbg_gen("dent '%.*s', mode %#x in dir ino %lu",
 745                 dentry->d_name.len, dentry->d_name.name, mode, dir->i_ino);
 746
 747         err = ubifs_budget_space(c, &req);
 748         if (err)
 749                 return err;
 750
 751         inode = ubifs_new_inode(c, dir, S_IFDIR | mode);
 752         if (IS_ERR(inode)) {
 753                 err = PTR_ERR(inode);
 754                 goto out_budg;
 755         }
 756
 757         mutex_lock(&dir_ui->ui_mutex);
 758         insert_inode_hash(inode);
 759         inc_nlink(inode);
 760         inc_nlink(dir);
 761         dir->i_size += sz_change;
 762         dir_ui->ui_size = dir->i_size;
 763         dir->i_mtime = dir->i_ctime = inode->i_ctime;
 764         err = ubifs_jnl_update(c, dir, &dentry->d_name, inode, 0, 0);
 765         if (err) {
 766                 ubifs_err("cannot create directory, error %d", err);
 767                 goto out_cancel;
 768         }
 769         mutex_unlock(&dir_ui->ui_mutex);
 770
 771         ubifs_release_budget(c, &req);
 772         d_instantiate(dentry, inode);
 773         return 0;
 774
 775 out_cancel:
 776         dir->i_size -= sz_change;
 777         dir_ui->ui_size = dir->i_size;
 778         drop_nlink(dir);
 779         mutex_unlock(&dir_ui->ui_mutex);
 780         make_bad_inode(inode);
 781         iput(inode);
 782 out_budg:
 783         ubifs_release_budget(c, &req);
 784         return err;
 785 }
 786
 787 static int ubifs_mknod(struct inode *dir, struct dentry *dentry,
 788                        int mode, dev_t rdev)
 789 {
 790         struct inode *inode;
 791         struct ubifs_inode *ui;
 792         struct ubifs_inode *dir_ui = ubifs_inode(dir);
 793         struct ubifs_info *c = dir->i_sb->s_fs_info;
 794         union ubifs_dev_desc *dev = NULL;
 795         int sz_change = CALC_DENT_SIZE(dentry->d_name.len);
 796         int err, devlen = 0;
 797         struct ubifs_budget_req req = { .new_ino = 1, .new_dent = 1,
 798                                         .new_ino_d = ALIGN(devlen, 8),
 799                                         .dirtied_ino = 1 };
 800
 801         /*
 802          * Budget request settings: new inode, new direntry and changing parent
 803          * directory inode.
 804          */
 805
 806         dbg_gen("dent '%.*s' in dir ino %lu",
 807                 dentry->d_name.len, dentry->d_name.name, dir->i_ino);
 808
 809         if (!new_valid_dev(rdev))
 810                 return -EINVAL;
 811
 812         if (S_ISBLK(mode) || S_ISCHR(mode)) {
 813                 dev = kmalloc(sizeof(union ubifs_dev_desc), GFP_NOFS);
 814                 if (!dev)
 815                         return -ENOMEM;
 816                 devlen = ubifs_encode_dev(dev, rdev);
 817         }
 818
 819         err = ubifs_budget_space(c, &req);
 820         if (err) {
 821                 kfree(dev);
 822                 return err;
 823         }
 824
 825         inode = ubifs_new_inode(c, dir, mode);
 826         if (IS_ERR(inode)) {
 827                 kfree(dev);
 828                 err = PTR_ERR(inode);
 829                 goto out_budg;
 830         }
 831
 832         init_special_inode(inode, inode->i_mode, rdev);
 833         inode->i_size = ubifs_inode(inode)->ui_size = devlen;
 834         ui = ubifs_inode(inode);
 835         ui->data = dev;
 836         ui->data_len = devlen;
 837
 838         mutex_lock(&dir_ui->ui_mutex);
 839         dir->i_size += sz_change;
 840         dir_ui->ui_size = dir->i_size;
 841         dir->i_mtime = dir->i_ctime = inode->i_ctime;
 842         err = ubifs_jnl_update(c, dir, &dentry->d_name, inode, 0, 0);
 843         if (err)
 844                 goto out_cancel;
 845         mutex_unlock(&dir_ui->ui_mutex);
 846
 847         ubifs_release_budget(c, &req);
 848         insert_inode_hash(inode);
 849         d_instantiate(dentry, inode);
 850         return 0;
 851
 852 out_cancel:
 853         dir->i_size -= sz_change;
 854         dir_ui->ui_size = dir->i_size;
 855         mutex_unlock(&dir_ui->ui_mutex);
 856         make_bad_inode(inode);
 857         iput(inode);
 858 out_budg:
 859         ubifs_release_budget(c, &req);
 860         return err;
 861 }
 862
 863 static int ubifs_symlink(struct inode *dir, struct dentry *dentry,
 864                          const char *symname)
 865 {
 866         struct inode *inode;
 867         struct ubifs_inode *ui;
 868         struct ubifs_inode *dir_ui = ubifs_inode(dir);
 869         struct ubifs_info *c = dir->i_sb->s_fs_info;
 870         int err, len = strlen(symname);
 871         int sz_change = CALC_DENT_SIZE(dentry->d_name.len);
 872         struct ubifs_budget_req req = { .new_ino = 1, .new_dent = 1,
 873                                         .new_ino_d = ALIGN(len, 8),
 874                                         .dirtied_ino = 1 };
 875
 876         /*
 877          * Budget request settings: new inode, new direntry and changing parent
 878          * directory inode.
 879          */
 880
 881         dbg_gen("dent '%.*s', target '%s' in dir ino %lu", dentry->d_name.len,
 882                 dentry->d_name.name, symname, dir->i_ino);
 883
 884         if (len > UBIFS_MAX_INO_DATA)
 885                 return -ENAMETOOLONG;
 886
 887         err = ubifs_budget_space(c, &req);
 888         if (err)
 889                 return err;
 890
 891         inode = ubifs_new_inode(c, dir, S_IFLNK | S_IRWXUGO);
 892         if (IS_ERR(inode)) {
 893                 err = PTR_ERR(inode);
 894                 goto out_budg;
 895         }
 896
 897         ui = ubifs_inode(inode);
 898         ui->data = kmalloc(len + 1, GFP_NOFS);
 899         if (!ui->data) {
 900                 err = -ENOMEM;
 901                 goto out_inode;
 902         }
 903
 904         memcpy(ui->data, symname, len);
 905         ((char *)ui->data)[len] = '\0';
 906         /*
 907          * The terminating zero byte is not written to the flash media and it
 908          * is put just to make later in-memory string processing simpler. Thus,
 909          * data length is @len, not @len + %1.
 910          */
 911         ui->data_len = len;
 912         inode->i_size = ubifs_inode(inode)->ui_size = len;
 913
 914         mutex_lock(&dir_ui->ui_mutex);
 915         dir->i_size += sz_change;
 916         dir_ui->ui_size = dir->i_size;
 917         dir->i_mtime = dir->i_ctime = inode->i_ctime;
 918         err = ubifs_jnl_update(c, dir, &dentry->d_name, inode, 0, 0);
 919         if (err)
 920                 goto out_cancel;
 921         mutex_unlock(&dir_ui->ui_mutex);
 922
 923         ubifs_release_budget(c, &req);
 924         insert_inode_hash(inode);
 925         d_instantiate(dentry, inode);
 926         return 0;
 927
 928 out_cancel:
 929         dir->i_size -= sz_change;
 930         dir_ui->ui_size = dir->i_size;
 931         mutex_unlock(&dir_ui->ui_mutex);
 932 out_inode:
 933         make_bad_inode(inode);
 934         iput(inode);
 935 out_budg:
 936         ubifs_release_budget(c, &req);
 937         return err;
 938 }
 939
 940 /**
 941  * lock_3_inodes - a wrapper for locking three UBIFS inodes.
 942  * @inode1: first inode
 943  * @inode2: second inode
 944  * @inode3: third inode
 945  *
 946  * This function is used for 'ubifs_rename()' and @inode1 may be the same as
 947  * @inode2 whereas @inode3 may be %NULL.
 948  *
 949  * We do not implement any tricks to guarantee strict lock ordering, because
 950  * VFS has already done it for us on the @i_mutex. So this is just a simple
 951  * wrapper function.
 952  */
 953 static void lock_3_inodes(struct inode *inode1, struct inode *inode2,
 954                           struct inode *inode3)
 955 {
 956         mutex_lock_nested(&ubifs_inode(inode1)->ui_mutex, WB_MUTEX_1);
 957         if (inode2 != inode1)
 958                 mutex_lock_nested(&ubifs_inode(inode2)->ui_mutex, WB_MUTEX_2);
 959         if (inode3)
 960                 mutex_lock_nested(&ubifs_inode(inode3)->ui_mutex, WB_MUTEX_3);
 961 }
 962
 963 /**
 964  * unlock_3_inodes - a wrapper for unlocking three UBIFS inodes for rename.
 965  * @inode1: first inode
 966  * @inode2: second inode
 967  * @inode3: third inode
 968  */
 969 static void unlock_3_inodes(struct inode *inode1, struct inode *inode2,
 970                             struct inode *inode3)
 971 {
 972         if (inode3)
 973                 mutex_unlock(&ubifs_inode(inode3)->ui_mutex);
 974         if (inode1 != inode2)
 975                 mutex_unlock(&ubifs_inode(inode2)->ui_mutex);
 976         mutex_unlock(&ubifs_inode(inode1)->ui_mutex);
 977 }
 978
 979 static int ubifs_rename(struct inode *old_dir, struct dentry *old_dentry,
 980                         struct inode *new_dir, struct dentry *new_dentry)
 981 {
 982         struct ubifs_info *c = old_dir->i_sb->s_fs_info;
 983         struct inode *old_inode = old_dentry->d_inode;
 984         struct inode *new_inode = new_dentry->d_inode;
 985         struct ubifs_inode *old_inode_ui = ubifs_inode(old_inode);
 986         int err, release, sync = 0, move = (new_dir != old_dir);
 987         int is_dir = S_ISDIR(old_inode->i_mode);
 988         int unlink = !!new_inode;
 989         int new_sz = CALC_DENT_SIZE(new_dentry->d_name.len);
 990         int old_sz = CALC_DENT_SIZE(old_dentry->d_name.len);
 991         struct ubifs_budget_req req = { .new_dent = 1, .mod_dent = 1,
 992                                         .dirtied_ino = 3 };
 993         struct ubifs_budget_req ino_req = { .dirtied_ino = 1,
 994                         .dirtied_ino_d = ALIGN(old_inode_ui->data_len, 8) };
 995         struct timespec time;
 996
 997         /*
 998          * Budget request settings: deletion direntry, new direntry, removing
 999          * the old inode, and changing old and new parent directory inodes.
1000          *
1001          * However, this operation also marks the target inode as dirty and
1002          * does not write it, so we allocate budget for the target inode
1003          * separately.
1004          */
1005
1006         dbg_gen("dent '%.*s' ino %lu in dir ino %lu to dent '%.*s' in "
1007                 "dir ino %lu", old_dentry->d_name.len, old_dentry->d_name.name,
1008                 old_inode->i_ino, old_dir->i_ino, new_dentry->d_name.len,
1009                 new_dentry->d_name.name, new_dir->i_ino);
1010         ubifs_assert(mutex_is_locked(&old_dir->i_mutex));
1011         ubifs_assert(mutex_is_locked(&new_dir->i_mutex));
1012         if (unlink)
1013                 ubifs_assert(mutex_is_locked(&new_inode->i_mutex));
1014
1015
1016         if (unlink && is_dir) {
1017                 err = check_dir_empty(c, new_inode);
1018                 if (err)
1019                         return err;
1020         }
1021
1022         err = ubifs_budget_space(c, &req);
1023         if (err)
1024                 return err;
1025         err = ubifs_budget_space(c, &ino_req);
1026         if (err) {
1027                 ubifs_release_budget(c, &req);
1028                 return err;
1029         }
1030
1031         lock_3_inodes(old_dir, new_dir, new_inode);
1032
1033         /*
1034          * Like most other Unix systems, set the @i_ctime for inodes on a
1035          * rename.
1036          */
1037         time = ubifs_current_time(old_dir);
1038         old_inode->i_ctime = time;
1039
1040         /* We must adjust parent link count when renaming directories */
1041         if (is_dir) {
1042                 if (move) {
1043                         /*
1044                          * @old_dir loses a link because we are moving
1045                          * @old_inode to a different directory.
1046                          */
1047                         drop_nlink(old_dir);
1048                         /*
1049                          * @new_dir only gains a link if we are not also
1050                          * overwriting an existing directory.
1051                          */
1052                         if (!unlink)
1053                                 inc_nlink(new_dir);
1054                 } else {
1055                         /*
1056                          * @old_inode is not moving to a different directory,
1057                          * but @old_dir still loses a link if we are
1058                          * overwriting an existing directory.
1059                          */
1060                         if (unlink)
1061                                 drop_nlink(old_dir);
1062                 }
1063         }
1064
1065         old_dir->i_size -= old_sz;
1066         ubifs_inode(old_dir)->ui_size = old_dir->i_size;
1067         old_dir->i_mtime = old_dir->i_ctime = time;
1068         new_dir->i_mtime = new_dir->i_ctime = time;
1069
1070         /*
1071          * And finally, if we unlinked a direntry which happened to have the
1072          * same name as the moved direntry, we have to decrement @i_nlink of
1073          * the unlinked inode and change its ctime.
1074          */
1075         if (unlink) {
1076                 /*
1077                  * Directories cannot have hard-links, so if this is a
1078                  * directory, decrement its @i_nlink twice because an empty
1079                  * directory has @i_nlink 2.
1080                  */
1081                 if (is_dir)
1082                         drop_nlink(new_inode);
1083                 new_inode->i_ctime = time;
1084                 drop_nlink(new_inode);
1085         } else {
1086                 new_dir->i_size += new_sz;
1087                 ubifs_inode(new_dir)->ui_size = new_dir->i_size;
1088         }
1089
1090         /*
1091          * Do not ask 'ubifs_jnl_rename()' to flush write-buffer if @old_inode
1092          * is dirty, because this will be done later on at the end of
1093          * 'ubifs_rename()'.
1094          */
1095         if (IS_SYNC(old_inode)) {
1096                 sync = IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir);
1097                 if (unlink && IS_SYNC(new_inode))
1098                         sync = 1;
1099         }
1100         err = ubifs_jnl_rename(c, old_dir, old_dentry, new_dir, new_dentry,
1101                                sync);
1102         if (err)
1103                 goto out_cancel;
1104
1105         unlock_3_inodes(old_dir, new_dir, new_inode);
1106         ubifs_release_budget(c, &req);
1107
1108         mutex_lock(&old_inode_ui->ui_mutex);
1109         release = old_inode_ui->dirty;
1110         mark_inode_dirty_sync(old_inode);
1111         mutex_unlock(&old_inode_ui->ui_mutex);
1112
1113         if (release)
1114                 ubifs_release_budget(c, &ino_req);
1115         if (IS_SYNC(old_inode))
1116                 err = old_inode->i_sb->s_op->write_inode(old_inode, NULL);
1117         return err;
1118
1119 out_cancel:
1120         if (unlink) {
1121                 if (is_dir)
1122                         inc_nlink(new_inode);
1123                 inc_nlink(new_inode);
1124         } else {
1125                 new_dir->i_size -= new_sz;
1126                 ubifs_inode(new_dir)->ui_size = new_dir->i_size;
1127         }
1128         old_dir->i_size += old_sz;
1129         ubifs_inode(old_dir)->ui_size = old_dir->i_size;
1130         if (is_dir) {
1131                 if (move) {
1132                         inc_nlink(old_dir);
1133                         if (!unlink)
1134                                 drop_nlink(new_dir);
1135                 } else {
1136                         if (unlink)
1137                                 inc_nlink(old_dir);
1138                 }
1139         }
1140         unlock_3_inodes(old_dir, new_dir, new_inode);
1141         ubifs_release_budget(c, &ino_req);
1142         ubifs_release_budget(c, &req);
1143         return err;
1144 }
1145
1146 int ubifs_getattr(struct vfsmount *mnt, struct dentry *dentry,
1147                   struct kstat *stat)
1148 {
1149         loff_t size;
1150         struct inode *inode = dentry->d_inode;
1151         struct ubifs_inode *ui = ubifs_inode(inode);
1152
1153         mutex_lock(&ui->ui_mutex);
1154         stat->dev = inode->i_sb->s_dev;
1155         stat->ino = inode->i_ino;
1156         stat->mode = inode->i_mode;
1157         stat->nlink = inode->i_nlink;
1158         stat->uid = inode->i_uid;
1159         stat->gid = inode->i_gid;
1160         stat->rdev = inode->i_rdev;
1161         stat->atime = inode->i_atime;
1162         stat->mtime = inode->i_mtime;
1163         stat->ctime = inode->i_ctime;
1164         stat->blksize = UBIFS_BLOCK_SIZE;
1165         stat->size = ui->ui_size;
1166
1167         /*
1168          * Unfortunately, the 'stat()' system call was designed for block
1169          * device based file systems, and it is not appropriate for UBIFS,
1170          * because UBIFS does not have notion of "block". For example, it is
1171          * difficult to tell how many block a directory takes - it actually
1172          * takes less than 300 bytes, but we have to round it to block size,
1173          * which introduces large mistake. This makes utilities like 'du' to
1174          * report completely senseless numbers. This is the reason why UBIFS
1175          * goes the same way as JFFS2 - it reports zero blocks for everything
1176          * but regular files, which makes more sense than reporting completely
1177          * wrong sizes.
1178          */
1179         if (S_ISREG(inode->i_mode)) {
1180                 size = ui->xattr_size;
1181                 size += stat->size;
1182                 size = ALIGN(size, UBIFS_BLOCK_SIZE);
1183                 /*
1184                  * Note, user-space expects 512-byte blocks count irrespectively
1185                  * of what was reported in @stat->size.
1186                  */
1187                 stat->blocks = size >> 9;
1188         } else
1189                 stat->blocks = 0;
1190         mutex_unlock(&ui->ui_mutex);
1191         return 0;
1192 }
1193
1194 const struct inode_operations ubifs_dir_inode_operations = {
1195         .lookup      = ubifs_lookup,
1196         .create      = ubifs_create,
1197         .link        = ubifs_link,
1198         .symlink     = ubifs_symlink,
1199         .unlink      = ubifs_unlink,
1200         .mkdir       = ubifs_mkdir,
1201         .rmdir       = ubifs_rmdir,
1202         .mknod       = ubifs_mknod,
1203         .rename      = ubifs_rename,
1204         .setattr     = ubifs_setattr,
1205         .getattr     = ubifs_getattr,
1206 #ifdef CONFIG_UBIFS_FS_XATTR
1207         .setxattr    = ubifs_setxattr,
1208         .getxattr    = ubifs_getxattr,
1209         .listxattr   = ubifs_listxattr,
1210         .removexattr = ubifs_removexattr,
1211 #endif
1212 };
1213
1214 const struct file_operations ubifs_dir_operations = {
1215         .llseek         = ubifs_dir_llseek,
1216         .release        = ubifs_dir_release,
1217         .read           = generic_read_dir,
1218         .readdir        = ubifs_readdir,
1219         .fsync          = ubifs_fsync,
1220         .unlocked_ioctl = ubifs_ioctl,
1221 #ifdef CONFIG_COMPAT
1222         .compat_ioctl   = ubifs_compat_ioctl,
1223 #endif
1224 };