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->i_uid = current_fsuid();
 108         if (dir->i_mode & S_ISGID) {
 109                 inode->i_gid = dir->i_gid;
 110                 if (S_ISDIR(mode))
 111                         mode |= S_ISGID;
 112         } else
 113                 inode->i_gid = current_fsgid();
 114         inode->i_mode = mode;
 115         inode->i_mtime = inode->i_atime = inode->i_ctime =
 116                          ubifs_current_time(inode);
 117         inode->i_mapping->nrpages = 0;
 118         /* Disable readahead */
 119         inode->i_mapping->backing_dev_info = &c->bdi;
 120
 121         switch (mode & S_IFMT) {
 122         case S_IFREG:
 123                 inode->i_mapping->a_ops = &ubifs_file_address_operations;
 124                 inode->i_op = &ubifs_file_inode_operations;
 125                 inode->i_fop = &ubifs_file_operations;
 126                 break;
 127         case S_IFDIR:
 128                 inode->i_op  = &ubifs_dir_inode_operations;
 129                 inode->i_fop = &ubifs_dir_operations;
 130                 inode->i_size = ui->ui_size = UBIFS_INO_NODE_SZ;
 131                 break;
 132         case S_IFLNK:
 133                 inode->i_op = &ubifs_symlink_inode_operations;
 134                 break;
 135         case S_IFSOCK:
 136         case S_IFIFO:
 137         case S_IFBLK:
 138         case S_IFCHR:
 139                 inode->i_op  = &ubifs_file_inode_operations;
 140                 break;
 141         default:
 142                 BUG();
 143         }
 144
 145         ui->flags = inherit_flags(dir, mode);
 146         ubifs_set_inode_flags(inode);
 147         if (S_ISREG(mode))
 148                 ui->compr_type = c->default_compr;
 149         else
 150                 ui->compr_type = UBIFS_COMPR_NONE;
 151         ui->synced_i_size = 0;
 152
 153         spin_lock(&c->cnt_lock);
 154         /* Inode number overflow is currently not supported */
 155         if (c->highest_inum >= INUM_WARN_WATERMARK) {
 156                 if (c->highest_inum >= INUM_WATERMARK) {
 157                         spin_unlock(&c->cnt_lock);
 158                         ubifs_err("out of inode numbers");
 159                         make_bad_inode(inode);
 160                         iput(inode);
 161                         return ERR_PTR(-EINVAL);
 162                 }
 163                 ubifs_warn("running out of inode numbers (current %lu, max %d)",
 164                            (unsigned long)c->highest_inum, INUM_WATERMARK);
 165         }
 166
 167         inode->i_ino = ++c->highest_inum;
 168         /*
 169          * The creation sequence number remains with this inode for its
 170          * lifetime. All nodes for this inode have a greater sequence number,
 171          * and so it is possible to distinguish obsolete nodes belonging to a
 172          * previous incarnation of the same inode number - for example, for the
 173          * purpose of rebuilding the index.
 174          */
 175         ui->creat_sqnum = ++c->max_sqnum;
 176         spin_unlock(&c->cnt_lock);
 177         return inode;
 178 }
 179
 180 #ifdef CONFIG_UBIFS_FS_DEBUG
 181
 182 static int dbg_check_name(struct ubifs_dent_node *dent, struct qstr *nm)
 183 {
 184         if (!(ubifs_chk_flags & UBIFS_CHK_GEN))
 185                 return 0;
 186         if (le16_to_cpu(dent->nlen) != nm->len)
 187                 return -EINVAL;
 188         if (memcmp(dent->name, nm->name, nm->len))
 189                 return -EINVAL;
 190         return 0;
 191 }
 192
 193 #else
 194
 195 #define dbg_check_name(dent, nm) 0
 196
 197 #endif
 198
 199 static struct dentry *ubifs_lookup(struct inode *dir, struct dentry *dentry,
 200                                    struct nameidata *nd)
 201 {
 202         int err;
 203         union ubifs_key key;
 204         struct inode *inode = NULL;
 205         struct ubifs_dent_node *dent;
 206         struct ubifs_info *c = dir->i_sb->s_fs_info;
 207
 208         dbg_gen("'%.*s' in dir ino %lu",
 209                 dentry->d_name.len, dentry->d_name.name, dir->i_ino);
 210
 211         if (dentry->d_name.len > UBIFS_MAX_NLEN)
 212                 return ERR_PTR(-ENAMETOOLONG);
 213
 214         dent = kmalloc(UBIFS_MAX_DENT_NODE_SZ, GFP_NOFS);
 215         if (!dent)
 216                 return ERR_PTR(-ENOMEM);
 217
 218         dent_key_init(c, &key, dir->i_ino, &dentry->d_name);
 219
 220         err = ubifs_tnc_lookup_nm(c, &key, dent, &dentry->d_name);
 221         if (err) {
 222                 if (err == -ENOENT) {
 223                         dbg_gen("not found");
 224                         goto done;
 225                 }
 226                 goto out;
 227         }
 228
 229         if (dbg_check_name(dent, &dentry->d_name)) {
 230                 err = -EINVAL;
 231                 goto out;
 232         }
 233
 234         inode = ubifs_iget(dir->i_sb, le64_to_cpu(dent->inum));
 235         if (IS_ERR(inode)) {
 236                 /*
 237                  * This should not happen. Probably the file-system needs
 238                  * checking.
 239                  */
 240                 err = PTR_ERR(inode);
 241                 ubifs_err("dead directory entry '%.*s', error %d",
 242                           dentry->d_name.len, dentry->d_name.name, err);
 243                 ubifs_ro_mode(c, err);
 244                 goto out;
 245         }
 246
 247 done:
 248         kfree(dent);
 249         /*
 250          * Note, d_splice_alias() would be required instead if we supported
 251          * NFS.
 252          */
 253         d_add(dentry, inode);
 254         return NULL;
 255
 256 out:
 257         kfree(dent);
 258         return ERR_PTR(err);
 259 }
 260
 261 static int ubifs_create(struct inode *dir, struct dentry *dentry, int mode,
 262                         struct nameidata *nd)
 263 {
 264         struct inode *inode;
 265         struct ubifs_info *c = dir->i_sb->s_fs_info;
 266         int err, sz_change = CALC_DENT_SIZE(dentry->d_name.len);
 267         struct ubifs_budget_req req = { .new_ino = 1, .new_dent = 1,
 268                                         .dirtied_ino = 1 };
 269         struct ubifs_inode *dir_ui = ubifs_inode(dir);
 270
 271         /*
 272          * Budget request settings: new inode, new direntry, changing the
 273          * parent directory inode.
 274          */
 275
 276         dbg_gen("dent '%.*s', mode %#x in dir ino %lu",
 277                 dentry->d_name.len, dentry->d_name.name, mode, dir->i_ino);
 278
 279         err = ubifs_budget_space(c, &req);
 280         if (err)
 281                 return err;
 282
 283         inode = ubifs_new_inode(c, dir, mode);
 284         if (IS_ERR(inode)) {
 285                 err = PTR_ERR(inode);
 286                 goto out_budg;
 287         }
 288
 289         mutex_lock(&dir_ui->ui_mutex);
 290         dir->i_size += sz_change;
 291         dir_ui->ui_size = dir->i_size;
 292         dir->i_mtime = dir->i_ctime = inode->i_ctime;
 293         err = ubifs_jnl_update(c, dir, &dentry->d_name, inode, 0, 0);
 294         if (err)
 295                 goto out_cancel;
 296         mutex_unlock(&dir_ui->ui_mutex);
 297
 298         ubifs_release_budget(c, &req);
 299         insert_inode_hash(inode);
 300         d_instantiate(dentry, inode);
 301         return 0;
 302
 303 out_cancel:
 304         dir->i_size -= sz_change;
 305         dir_ui->ui_size = dir->i_size;
 306         mutex_unlock(&dir_ui->ui_mutex);
 307         make_bad_inode(inode);
 308         iput(inode);
 309 out_budg:
 310         ubifs_release_budget(c, &req);
 311         ubifs_err("cannot create regular file, error %d", err);
 312         return err;
 313 }
 314
 315 /**
 316  * vfs_dent_type - get VFS directory entry type.
 317  * @type: UBIFS directory entry type
 318  *
 319  * This function converts UBIFS directory entry type into VFS directory entry
 320  * type.
 321  */
 322 static unsigned int vfs_dent_type(uint8_t type)
 323 {
 324         switch (type) {
 325         case UBIFS_ITYPE_REG:
 326                 return DT_REG;
 327         case UBIFS_ITYPE_DIR:
 328                 return DT_DIR;
 329         case UBIFS_ITYPE_LNK:
 330                 return DT_LNK;
 331         case UBIFS_ITYPE_BLK:
 332                 return DT_BLK;
 333         case UBIFS_ITYPE_CHR:
 334                 return DT_CHR;
 335         case UBIFS_ITYPE_FIFO:
 336                 return DT_FIFO;
 337         case UBIFS_ITYPE_SOCK:
 338                 return DT_SOCK;
 339         default:
 340                 BUG();
 341         }
 342         return 0;
 343 }
 344
 345 /*
 346  * The classical Unix view for directory is that it is a linear array of
 347  * (name, inode number) entries. Linux/VFS assumes this model as well.
 348  * Particularly, 'readdir()' call wants us to return a directory entry offset
 349  * which later may be used to continue 'readdir()'ing the directory or to
 350  * 'seek()' to that specific direntry. Obviously UBIFS does not really fit this
 351  * model because directory entries are identified by keys, which may collide.
 352  *
 353  * UBIFS uses directory entry hash value for directory offsets, so
 354  * 'seekdir()'/'telldir()' may not always work because of possible key
 355  * collisions. But UBIFS guarantees that consecutive 'readdir()' calls work
 356  * properly by means of saving full directory entry name in the private field
 357  * of the file description object.
 358  *
 359  * This means that UBIFS cannot support NFS which requires full
 360  * 'seekdir()'/'telldir()' support.
 361  */
 362 static int ubifs_readdir(struct file *file, void *dirent, filldir_t filldir)
 363 {
 364         int err, over = 0;
 365         struct qstr nm;
 366         union ubifs_key key;
 367         struct ubifs_dent_node *dent;
 368         struct inode *dir = file->f_path.dentry->d_inode;
 369         struct ubifs_info *c = dir->i_sb->s_fs_info;
 370
 371         dbg_gen("dir ino %lu, f_pos %#llx", dir->i_ino, file->f_pos);
 372
 373         if (file->f_pos > UBIFS_S_KEY_HASH_MASK || file->f_pos == 2)
 374                 /*
 375                  * The directory was seek'ed to a senseless position or there
 376                  * are no more entries.
 377                  */
 378                 return 0;
 379
 380         /* File positions 0 and 1 correspond to "." and ".." */
 381         if (file->f_pos == 0) {
 382                 ubifs_assert(!file->private_data);
 383                 over = filldir(dirent, ".", 1, 0, dir->i_ino, DT_DIR);
 384                 if (over)
 385                         return 0;
 386                 file->f_pos = 1;
 387         }
 388
 389         if (file->f_pos == 1) {
 390                 ubifs_assert(!file->private_data);
 391                 over = filldir(dirent, "..", 2, 1,
 392                                parent_ino(file->f_path.dentry), DT_DIR);
 393                 if (over)
 394                         return 0;
 395
 396                 /* Find the first entry in TNC and save it */
 397                 lowest_dent_key(c, &key, dir->i_ino);
 398                 nm.name = NULL;
 399                 dent = ubifs_tnc_next_ent(c, &key, &nm);
 400                 if (IS_ERR(dent)) {
 401                         err = PTR_ERR(dent);
 402                         goto out;
 403                 }
 404
 405                 file->f_pos = key_hash_flash(c, &dent->key);
 406                 file->private_data = dent;
 407         }
 408
 409         dent = file->private_data;
 410         if (!dent) {
 411                 /*
 412                  * The directory was seek'ed to and is now readdir'ed.
 413                  * Find the entry corresponding to @file->f_pos or the
 414                  * closest one.
 415                  */
 416                 dent_key_init_hash(c, &key, dir->i_ino, file->f_pos);
 417                 nm.name = NULL;
 418                 dent = ubifs_tnc_next_ent(c, &key, &nm);
 419                 if (IS_ERR(dent)) {
 420                         err = PTR_ERR(dent);
 421                         goto out;
 422                 }
 423                 file->f_pos = key_hash_flash(c, &dent->key);
 424                 file->private_data = dent;
 425         }
 426
 427         while (1) {
 428                 dbg_gen("feed '%s', ino %llu, new f_pos %#x",
 429                         dent->name, (unsigned long long)le64_to_cpu(dent->inum),
 430                         key_hash_flash(c, &dent->key));
 431                 ubifs_assert(le64_to_cpu(dent->ch.sqnum) >
 432                              ubifs_inode(dir)->creat_sqnum);
 433
 434                 nm.len = le16_to_cpu(dent->nlen);
 435                 over = filldir(dirent, dent->name, nm.len, file->f_pos,
 436                                le64_to_cpu(dent->inum),
 437                                vfs_dent_type(dent->type));
 438                 if (over)
 439                         return 0;
 440
 441                 /* Switch to the next entry */
 442                 key_read(c, &dent->key, &key);
 443                 nm.name = dent->name;
 444                 dent = ubifs_tnc_next_ent(c, &key, &nm);
 445                 if (IS_ERR(dent)) {
 446                         err = PTR_ERR(dent);
 447                         goto out;
 448                 }
 449
 450                 kfree(file->private_data);
 451                 file->f_pos = key_hash_flash(c, &dent->key);
 452                 file->private_data = dent;
 453                 cond_resched();
 454         }
 455
 456 out:
 457         if (err != -ENOENT) {
 458                 ubifs_err("cannot find next direntry, error %d", err);
 459                 return err;
 460         }
 461
 462         kfree(file->private_data);
 463         file->private_data = NULL;
 464         file->f_pos = 2;
 465         return 0;
 466 }
 467
 468 /* If a directory is seeked, we have to free saved readdir() state */
 469 static loff_t ubifs_dir_llseek(struct file *file, loff_t offset, int origin)
 470 {
 471         kfree(file->private_data);
 472         file->private_data = NULL;
 473         return generic_file_llseek(file, offset, origin);
 474 }
 475
 476 /* Free saved readdir() state when the directory is closed */
 477 static int ubifs_dir_release(struct inode *dir, struct file *file)
 478 {
 479         kfree(file->private_data);
 480         file->private_data = NULL;
 481         return 0;
 482 }
 483
 484 /**
 485  * lock_2_inodes - a wrapper for locking two UBIFS inodes.
 486  * @inode1: first inode
 487  * @inode2: second inode
 488  *
 489  * We do not implement any tricks to guarantee strict lock ordering, because
 490  * VFS has already done it for us on the @i_mutex. So this is just a simple
 491  * wrapper function.
 492  */
 493 static void lock_2_inodes(struct inode *inode1, struct inode *inode2)
 494 {
 495         mutex_lock_nested(&ubifs_inode(inode1)->ui_mutex, WB_MUTEX_1);
 496         mutex_lock_nested(&ubifs_inode(inode2)->ui_mutex, WB_MUTEX_2);
 497 }
 498
 499 /**
 500  * unlock_2_inodes - a wrapper for unlocking two UBIFS inodes.
 501  * @inode1: first inode
 502  * @inode2: second inode
 503  */
 504 static void unlock_2_inodes(struct inode *inode1, struct inode *inode2)
 505 {
 506         mutex_unlock(&ubifs_inode(inode2)->ui_mutex);
 507         mutex_unlock(&ubifs_inode(inode1)->ui_mutex);
 508 }
 509
 510 static int ubifs_link(struct dentry *old_dentry, struct inode *dir,
 511                       struct dentry *dentry)
 512 {
 513         struct ubifs_info *c = dir->i_sb->s_fs_info;
 514         struct inode *inode = old_dentry->d_inode;
 515         struct ubifs_inode *ui = ubifs_inode(inode);
 516         struct ubifs_inode *dir_ui = ubifs_inode(dir);
 517         int err, sz_change = CALC_DENT_SIZE(dentry->d_name.len);
 518         struct ubifs_budget_req req = { .new_dent = 1, .dirtied_ino = 2,
 519                                 .dirtied_ino_d = ALIGN(ui->data_len, 8) };
 520
 521         /*
 522          * Budget request settings: new direntry, changing the target inode,
 523          * changing the parent inode.
 524          */
 525
 526         dbg_gen("dent '%.*s' to ino %lu (nlink %d) in dir ino %lu",
 527                 dentry->d_name.len, dentry->d_name.name, inode->i_ino,
 528                 inode->i_nlink, dir->i_ino);
 529         ubifs_assert(mutex_is_locked(&dir->i_mutex));
 530         ubifs_assert(mutex_is_locked(&inode->i_mutex));
 531
 532         /*
 533          * Return -ENOENT if we've raced with unlink and i_nlink is 0.  Doing
 534          * otherwise has the potential to corrupt the orphan inode list.
 535          *
 536          * Indeed, consider a scenario when 'vfs_link(dirA/fileA)' and
 537          * 'vfs_unlink(dirA/fileA, dirB/fileB)' race. 'vfs_link()' does not
 538          * lock 'dirA->i_mutex', so this is possible. Both of the functions
 539          * lock 'fileA->i_mutex' though. Suppose 'vfs_unlink()' wins, and takes
 540          * 'fileA->i_mutex' mutex first. Suppose 'fileA->i_nlink' is 1. In this
 541          * case 'ubifs_unlink()' will drop the last reference, and put 'inodeA'
 542          * to the list of orphans. After this, 'vfs_link()' will link
 543          * 'dirB/fileB' to 'inodeA'. This is a problem because, for example,
 544          * the subsequent 'vfs_unlink(dirB/fileB)' will add the same inode
 545          * to the list of orphans.
 546          */
 547          if (inode->i_nlink == 0)
 548                  return -ENOENT;
 549
 550         err = dbg_check_synced_i_size(inode);
 551         if (err)
 552                 return err;
 553
 554         err = ubifs_budget_space(c, &req);
 555         if (err)
 556                 return err;
 557
 558         lock_2_inodes(dir, inode);
 559         inc_nlink(inode);
 560         atomic_inc(&inode->i_count);
 561         inode->i_ctime = ubifs_current_time(inode);
 562         dir->i_size += sz_change;
 563         dir_ui->ui_size = dir->i_size;
 564         dir->i_mtime = dir->i_ctime = inode->i_ctime;
 565         err = ubifs_jnl_update(c, dir, &dentry->d_name, inode, 0, 0);
 566         if (err)
 567                 goto out_cancel;
 568         unlock_2_inodes(dir, inode);
 569
 570         ubifs_release_budget(c, &req);
 571         d_instantiate(dentry, inode);
 572         return 0;
 573
 574 out_cancel:
 575         dir->i_size -= sz_change;
 576         dir_ui->ui_size = dir->i_size;
 577         drop_nlink(inode);
 578         unlock_2_inodes(dir, inode);
 579         ubifs_release_budget(c, &req);
 580         iput(inode);
 581         return err;
 582 }
 583
 584 static int ubifs_unlink(struct inode *dir, struct dentry *dentry)
 585 {
 586         struct ubifs_info *c = dir->i_sb->s_fs_info;
 587         struct inode *inode = dentry->d_inode;
 588         struct ubifs_inode *dir_ui = ubifs_inode(dir);
 589         int sz_change = CALC_DENT_SIZE(dentry->d_name.len);
 590         int err, budgeted = 1;
 591         struct ubifs_budget_req req = { .mod_dent = 1, .dirtied_ino = 2 };
 592
 593         /*
 594          * Budget request settings: deletion direntry, deletion inode (+1 for
 595          * @dirtied_ino), changing the parent directory inode. If budgeting
 596          * fails, go ahead anyway because we have extra space reserved for
 597          * deletions.
 598          */
 599
 600         dbg_gen("dent '%.*s' from ino %lu (nlink %d) in dir ino %lu",
 601                 dentry->d_name.len, dentry->d_name.name, inode->i_ino,
 602                 inode->i_nlink, dir->i_ino);
 603         ubifs_assert(mutex_is_locked(&dir->i_mutex));
 604         ubifs_assert(mutex_is_locked(&inode->i_mutex));
 605         err = dbg_check_synced_i_size(inode);
 606         if (err)
 607                 return err;
 608
 609         err = ubifs_budget_space(c, &req);
 610         if (err) {
 611                 if (err != -ENOSPC)
 612                         return err;
 613                 budgeted = 0;
 614         }
 615
 616         lock_2_inodes(dir, inode);
 617         inode->i_ctime = ubifs_current_time(dir);
 618         drop_nlink(inode);
 619         dir->i_size -= sz_change;
 620         dir_ui->ui_size = dir->i_size;
 621         dir->i_mtime = dir->i_ctime = inode->i_ctime;
 622         err = ubifs_jnl_update(c, dir, &dentry->d_name, inode, 1, 0);
 623         if (err)
 624                 goto out_cancel;
 625         unlock_2_inodes(dir, inode);
 626
 627         if (budgeted)
 628                 ubifs_release_budget(c, &req);
 629         else {
 630                 /* We've deleted something - clean the "no space" flags */
 631                 c->nospace = c->nospace_rp = 0;
 632                 smp_wmb();
 633         }
 634         return 0;
 635
 636 out_cancel:
 637         dir->i_size += sz_change;
 638         dir_ui->ui_size = dir->i_size;
 639         inc_nlink(inode);
 640         unlock_2_inodes(dir, inode);
 641         if (budgeted)
 642                 ubifs_release_budget(c, &req);
 643         return err;
 644 }
 645
 646 /**
 647  * check_dir_empty - check if a directory is empty or not.
 648  * @c: UBIFS file-system description object
 649  * @dir: VFS inode object of the directory to check
 650  *
 651  * This function checks if directory @dir is empty. Returns zero if the
 652  * directory is empty, %-ENOTEMPTY if it is not, and other negative error codes
 653  * in case of of errors.
 654  */
 655 static int check_dir_empty(struct ubifs_info *c, struct inode *dir)
 656 {
 657         struct qstr nm = { .name = NULL };
 658         struct ubifs_dent_node *dent;
 659         union ubifs_key key;
 660         int err;
 661
 662         lowest_dent_key(c, &key, dir->i_ino);
 663         dent = ubifs_tnc_next_ent(c, &key, &nm);
 664         if (IS_ERR(dent)) {
 665                 err = PTR_ERR(dent);
 666                 if (err == -ENOENT)
 667                         err = 0;
 668         } else {
 669                 kfree(dent);
 670                 err = -ENOTEMPTY;
 671         }
 672         return err;
 673 }
 674
 675 static int ubifs_rmdir(struct inode *dir, struct dentry *dentry)
 676 {
 677         struct ubifs_info *c = dir->i_sb->s_fs_info;
 678         struct inode *inode = dentry->d_inode;
 679         int sz_change = CALC_DENT_SIZE(dentry->d_name.len);
 680         int err, budgeted = 1;
 681         struct ubifs_inode *dir_ui = ubifs_inode(dir);
 682         struct ubifs_budget_req req = { .mod_dent = 1, .dirtied_ino = 2 };
 683
 684         /*
 685          * Budget request settings: deletion direntry, deletion inode and
 686          * changing the parent inode. If budgeting fails, go ahead anyway
 687          * because we have extra space reserved for deletions.
 688          */
 689
 690         dbg_gen("directory '%.*s', ino %lu in dir ino %lu", dentry->d_name.len,
 691                 dentry->d_name.name, inode->i_ino, dir->i_ino);
 692         ubifs_assert(mutex_is_locked(&dir->i_mutex));
 693         ubifs_assert(mutex_is_locked(&inode->i_mutex));
 694         err = check_dir_empty(c, dentry->d_inode);
 695         if (err)
 696                 return err;
 697
 698         err = ubifs_budget_space(c, &req);
 699         if (err) {
 700                 if (err != -ENOSPC)
 701                         return err;
 702                 budgeted = 0;
 703         }
 704
 705         lock_2_inodes(dir, inode);
 706         inode->i_ctime = ubifs_current_time(dir);
 707         clear_nlink(inode);
 708         drop_nlink(dir);
 709         dir->i_size -= sz_change;
 710         dir_ui->ui_size = dir->i_size;
 711         dir->i_mtime = dir->i_ctime = inode->i_ctime;
 712         err = ubifs_jnl_update(c, dir, &dentry->d_name, inode, 1, 0);
 713         if (err)
 714                 goto out_cancel;
 715         unlock_2_inodes(dir, inode);
 716
 717         if (budgeted)
 718                 ubifs_release_budget(c, &req);
 719         else {
 720                 /* We've deleted something - clean the "no space" flags */
 721                 c->nospace = c->nospace_rp = 0;
 722                 smp_wmb();
 723         }
 724         return 0;
 725
 726 out_cancel:
 727         dir->i_size += sz_change;
 728         dir_ui->ui_size = dir->i_size;
 729         inc_nlink(dir);
 730         inc_nlink(inode);
 731         inc_nlink(inode);
 732         unlock_2_inodes(dir, inode);
 733         if (budgeted)
 734                 ubifs_release_budget(c, &req);
 735         return err;
 736 }
 737
 738 static int ubifs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 739 {
 740         struct inode *inode;
 741         struct ubifs_inode *dir_ui = ubifs_inode(dir);
 742         struct ubifs_info *c = dir->i_sb->s_fs_info;
 743         int err, sz_change = CALC_DENT_SIZE(dentry->d_name.len);
 744         struct ubifs_budget_req req = { .new_ino = 1, .new_dent = 1 };
 745
 746         /*
 747          * Budget request settings: new inode, new direntry and changing parent
 748          * directory inode.
 749          */
 750
 751         dbg_gen("dent '%.*s', mode %#x in dir ino %lu",
 752                 dentry->d_name.len, dentry->d_name.name, mode, dir->i_ino);
 753
 754         err = ubifs_budget_space(c, &req);
 755         if (err)
 756                 return err;
 757
 758         inode = ubifs_new_inode(c, dir, S_IFDIR | mode);
 759         if (IS_ERR(inode)) {
 760                 err = PTR_ERR(inode);
 761                 goto out_budg;
 762         }
 763
 764         mutex_lock(&dir_ui->ui_mutex);
 765         insert_inode_hash(inode);
 766         inc_nlink(inode);
 767         inc_nlink(dir);
 768         dir->i_size += sz_change;
 769         dir_ui->ui_size = dir->i_size;
 770         dir->i_mtime = dir->i_ctime = inode->i_ctime;
 771         err = ubifs_jnl_update(c, dir, &dentry->d_name, inode, 0, 0);
 772         if (err) {
 773                 ubifs_err("cannot create directory, error %d", err);
 774                 goto out_cancel;
 775         }
 776         mutex_unlock(&dir_ui->ui_mutex);
 777
 778         ubifs_release_budget(c, &req);
 779         d_instantiate(dentry, inode);
 780         return 0;
 781
 782 out_cancel:
 783         dir->i_size -= sz_change;
 784         dir_ui->ui_size = dir->i_size;
 785         drop_nlink(dir);
 786         mutex_unlock(&dir_ui->ui_mutex);
 787         make_bad_inode(inode);
 788         iput(inode);
 789 out_budg:
 790         ubifs_release_budget(c, &req);
 791         return err;
 792 }
 793
 794 static int ubifs_mknod(struct inode *dir, struct dentry *dentry,
 795                        int mode, dev_t rdev)
 796 {
 797         struct inode *inode;
 798         struct ubifs_inode *ui;
 799         struct ubifs_inode *dir_ui = ubifs_inode(dir);
 800         struct ubifs_info *c = dir->i_sb->s_fs_info;
 801         union ubifs_dev_desc *dev = NULL;
 802         int sz_change = CALC_DENT_SIZE(dentry->d_name.len);
 803         int err, devlen = 0;
 804         struct ubifs_budget_req req = { .new_ino = 1, .new_dent = 1,
 805                                         .new_ino_d = ALIGN(devlen, 8),
 806                                         .dirtied_ino = 1 };
 807
 808         /*
 809          * Budget request settings: new inode, new direntry and changing parent
 810          * directory inode.
 811          */
 812
 813         dbg_gen("dent '%.*s' in dir ino %lu",
 814                 dentry->d_name.len, dentry->d_name.name, dir->i_ino);
 815
 816         if (!new_valid_dev(rdev))
 817                 return -EINVAL;
 818
 819         if (S_ISBLK(mode) || S_ISCHR(mode)) {
 820                 dev = kmalloc(sizeof(union ubifs_dev_desc), GFP_NOFS);
 821                 if (!dev)
 822                         return -ENOMEM;
 823                 devlen = ubifs_encode_dev(dev, rdev);
 824         }
 825
 826         err = ubifs_budget_space(c, &req);
 827         if (err) {
 828                 kfree(dev);
 829                 return err;
 830         }
 831
 832         inode = ubifs_new_inode(c, dir, mode);
 833         if (IS_ERR(inode)) {
 834                 kfree(dev);
 835                 err = PTR_ERR(inode);
 836                 goto out_budg;
 837         }
 838
 839         init_special_inode(inode, inode->i_mode, rdev);
 840         inode->i_size = ubifs_inode(inode)->ui_size = devlen;
 841         ui = ubifs_inode(inode);
 842         ui->data = dev;
 843         ui->data_len = devlen;
 844
 845         mutex_lock(&dir_ui->ui_mutex);
 846         dir->i_size += sz_change;
 847         dir_ui->ui_size = dir->i_size;
 848         dir->i_mtime = dir->i_ctime = inode->i_ctime;
 849         err = ubifs_jnl_update(c, dir, &dentry->d_name, inode, 0, 0);
 850         if (err)
 851                 goto out_cancel;
 852         mutex_unlock(&dir_ui->ui_mutex);
 853
 854         ubifs_release_budget(c, &req);
 855         insert_inode_hash(inode);
 856         d_instantiate(dentry, inode);
 857         return 0;
 858
 859 out_cancel:
 860         dir->i_size -= sz_change;
 861         dir_ui->ui_size = dir->i_size;
 862         mutex_unlock(&dir_ui->ui_mutex);
 863         make_bad_inode(inode);
 864         iput(inode);
 865 out_budg:
 866         ubifs_release_budget(c, &req);
 867         return err;
 868 }
 869
 870 static int ubifs_symlink(struct inode *dir, struct dentry *dentry,
 871                          const char *symname)
 872 {
 873         struct inode *inode;
 874         struct ubifs_inode *ui;
 875         struct ubifs_inode *dir_ui = ubifs_inode(dir);
 876         struct ubifs_info *c = dir->i_sb->s_fs_info;
 877         int err, len = strlen(symname);
 878         int sz_change = CALC_DENT_SIZE(dentry->d_name.len);
 879         struct ubifs_budget_req req = { .new_ino = 1, .new_dent = 1,
 880                                         .new_ino_d = ALIGN(len, 8),
 881                                         .dirtied_ino = 1 };
 882
 883         /*
 884          * Budget request settings: new inode, new direntry and changing parent
 885          * directory inode.
 886          */
 887
 888         dbg_gen("dent '%.*s', target '%s' in dir ino %lu", dentry->d_name.len,
 889                 dentry->d_name.name, symname, dir->i_ino);
 890
 891         if (len > UBIFS_MAX_INO_DATA)
 892                 return -ENAMETOOLONG;
 893
 894         err = ubifs_budget_space(c, &req);
 895         if (err)
 896                 return err;
 897
 898         inode = ubifs_new_inode(c, dir, S_IFLNK | S_IRWXUGO);
 899         if (IS_ERR(inode)) {
 900                 err = PTR_ERR(inode);
 901                 goto out_budg;
 902         }
 903
 904         ui = ubifs_inode(inode);
 905         ui->data = kmalloc(len + 1, GFP_NOFS);
 906         if (!ui->data) {
 907                 err = -ENOMEM;
 908                 goto out_inode;
 909         }
 910
 911         memcpy(ui->data, symname, len);
 912         ((char *)ui->data)[len] = '\0';
 913         /*
 914          * The terminating zero byte is not written to the flash media and it
 915          * is put just to make later in-memory string processing simpler. Thus,
 916          * data length is @len, not @len + %1.
 917          */
 918         ui->data_len = len;
 919         inode->i_size = ubifs_inode(inode)->ui_size = len;
 920
 921         mutex_lock(&dir_ui->ui_mutex);
 922         dir->i_size += sz_change;
 923         dir_ui->ui_size = dir->i_size;
 924         dir->i_mtime = dir->i_ctime = inode->i_ctime;
 925         err = ubifs_jnl_update(c, dir, &dentry->d_name, inode, 0, 0);
 926         if (err)
 927                 goto out_cancel;
 928         mutex_unlock(&dir_ui->ui_mutex);
 929
 930         ubifs_release_budget(c, &req);
 931         insert_inode_hash(inode);
 932         d_instantiate(dentry, inode);
 933         return 0;
 934
 935 out_cancel:
 936         dir->i_size -= sz_change;
 937         dir_ui->ui_size = dir->i_size;
 938         mutex_unlock(&dir_ui->ui_mutex);
 939 out_inode:
 940         make_bad_inode(inode);
 941         iput(inode);
 942 out_budg:
 943         ubifs_release_budget(c, &req);
 944         return err;
 945 }
 946
 947 /**
 948  * lock_3_inodes - a wrapper for locking three UBIFS inodes.
 949  * @inode1: first inode
 950  * @inode2: second inode
 951  * @inode3: third inode
 952  *
 953  * This function is used for 'ubifs_rename()' and @inode1 may be the same as
 954  * @inode2 whereas @inode3 may be %NULL.
 955  *
 956  * We do not implement any tricks to guarantee strict lock ordering, because
 957  * VFS has already done it for us on the @i_mutex. So this is just a simple
 958  * wrapper function.
 959  */
 960 static void lock_3_inodes(struct inode *inode1, struct inode *inode2,
 961                           struct inode *inode3)
 962 {
 963         mutex_lock_nested(&ubifs_inode(inode1)->ui_mutex, WB_MUTEX_1);
 964         if (inode2 != inode1)
 965                 mutex_lock_nested(&ubifs_inode(inode2)->ui_mutex, WB_MUTEX_2);
 966         if (inode3)
 967                 mutex_lock_nested(&ubifs_inode(inode3)->ui_mutex, WB_MUTEX_3);
 968 }
 969
 970 /**
 971  * unlock_3_inodes - a wrapper for unlocking three UBIFS inodes for rename.
 972  * @inode1: first inode
 973  * @inode2: second inode
 974  * @inode3: third inode
 975  */
 976 static void unlock_3_inodes(struct inode *inode1, struct inode *inode2,
 977                             struct inode *inode3)
 978 {
 979         if (inode3)
 980                 mutex_unlock(&ubifs_inode(inode3)->ui_mutex);
 981         if (inode1 != inode2)
 982                 mutex_unlock(&ubifs_inode(inode2)->ui_mutex);
 983         mutex_unlock(&ubifs_inode(inode1)->ui_mutex);
 984 }
 985
 986 static int ubifs_rename(struct inode *old_dir, struct dentry *old_dentry,
 987                         struct inode *new_dir, struct dentry *new_dentry)
 988 {
 989         struct ubifs_info *c = old_dir->i_sb->s_fs_info;
 990         struct inode *old_inode = old_dentry->d_inode;
 991         struct inode *new_inode = new_dentry->d_inode;
 992         struct ubifs_inode *old_inode_ui = ubifs_inode(old_inode);
 993         int err, release, sync = 0, move = (new_dir != old_dir);
 994         int is_dir = S_ISDIR(old_inode->i_mode);
 995         int unlink = !!new_inode;
 996         int new_sz = CALC_DENT_SIZE(new_dentry->d_name.len);
 997         int old_sz = CALC_DENT_SIZE(old_dentry->d_name.len);
 998         struct ubifs_budget_req req = { .new_dent = 1, .mod_dent = 1,
 999                                         .dirtied_ino = 3 };
1000         struct ubifs_budget_req ino_req = { .dirtied_ino = 1,
1001                         .dirtied_ino_d = ALIGN(old_inode_ui->data_len, 8) };
1002         struct timespec time;
1003
1004         /*
1005          * Budget request settings: deletion direntry, new direntry, removing
1006          * the old inode, and changing old and new parent directory inodes.
1007          *
1008          * However, this operation also marks the target inode as dirty and
1009          * does not write it, so we allocate budget for the target inode
1010          * separately.
1011          */
1012
1013         dbg_gen("dent '%.*s' ino %lu in dir ino %lu to dent '%.*s' in "
1014                 "dir ino %lu", old_dentry->d_name.len, old_dentry->d_name.name,
1015                 old_inode->i_ino, old_dir->i_ino, new_dentry->d_name.len,
1016                 new_dentry->d_name.name, new_dir->i_ino);
1017         ubifs_assert(mutex_is_locked(&old_dir->i_mutex));
1018         ubifs_assert(mutex_is_locked(&new_dir->i_mutex));
1019         if (unlink)
1020                 ubifs_assert(mutex_is_locked(&new_inode->i_mutex));
1021
1022
1023         if (unlink && is_dir) {
1024                 err = check_dir_empty(c, new_inode);
1025                 if (err)
1026                         return err;
1027         }
1028
1029         err = ubifs_budget_space(c, &req);
1030         if (err)
1031                 return err;
1032         err = ubifs_budget_space(c, &ino_req);
1033         if (err) {
1034                 ubifs_release_budget(c, &req);
1035                 return err;
1036         }
1037
1038         lock_3_inodes(old_dir, new_dir, new_inode);
1039
1040         /*
1041          * Like most other Unix systems, set the @i_ctime for inodes on a
1042          * rename.
1043          */
1044         time = ubifs_current_time(old_dir);
1045         old_inode->i_ctime = time;
1046
1047         /* We must adjust parent link count when renaming directories */
1048         if (is_dir) {
1049                 if (move) {
1050                         /*
1051                          * @old_dir loses a link because we are moving
1052                          * @old_inode to a different directory.
1053                          */
1054                         drop_nlink(old_dir);
1055                         /*
1056                          * @new_dir only gains a link if we are not also
1057                          * overwriting an existing directory.
1058                          */
1059                         if (!unlink)
1060                                 inc_nlink(new_dir);
1061                 } else {
1062                         /*
1063                          * @old_inode is not moving to a different directory,
1064                          * but @old_dir still loses a link if we are
1065                          * overwriting an existing directory.
1066                          */
1067                         if (unlink)
1068                                 drop_nlink(old_dir);
1069                 }
1070         }
1071
1072         old_dir->i_size -= old_sz;
1073         ubifs_inode(old_dir)->ui_size = old_dir->i_size;
1074         old_dir->i_mtime = old_dir->i_ctime = time;
1075         new_dir->i_mtime = new_dir->i_ctime = time;
1076
1077         /*
1078          * And finally, if we unlinked a direntry which happened to have the
1079          * same name as the moved direntry, we have to decrement @i_nlink of
1080          * the unlinked inode and change its ctime.
1081          */
1082         if (unlink) {
1083                 /*
1084                  * Directories cannot have hard-links, so if this is a
1085                  * directory, decrement its @i_nlink twice because an empty
1086                  * directory has @i_nlink 2.
1087                  */
1088                 if (is_dir)
1089                         drop_nlink(new_inode);
1090                 new_inode->i_ctime = time;
1091                 drop_nlink(new_inode);
1092         } else {
1093                 new_dir->i_size += new_sz;
1094                 ubifs_inode(new_dir)->ui_size = new_dir->i_size;
1095         }
1096
1097         /*
1098          * Do not ask 'ubifs_jnl_rename()' to flush write-buffer if @old_inode
1099          * is dirty, because this will be done later on at the end of
1100          * 'ubifs_rename()'.
1101          */
1102         if (IS_SYNC(old_inode)) {
1103                 sync = IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir);
1104                 if (unlink && IS_SYNC(new_inode))
1105                         sync = 1;
1106         }
1107         err = ubifs_jnl_rename(c, old_dir, old_dentry, new_dir, new_dentry,
1108                                sync);
1109         if (err)
1110                 goto out_cancel;
1111
1112         unlock_3_inodes(old_dir, new_dir, new_inode);
1113         ubifs_release_budget(c, &req);
1114
1115         mutex_lock(&old_inode_ui->ui_mutex);
1116         release = old_inode_ui->dirty;
1117         mark_inode_dirty_sync(old_inode);
1118         mutex_unlock(&old_inode_ui->ui_mutex);
1119
1120         if (release)
1121                 ubifs_release_budget(c, &ino_req);
1122         if (IS_SYNC(old_inode))
1123                 err = old_inode->i_sb->s_op->write_inode(old_inode, 1);
1124         return err;
1125
1126 out_cancel:
1127         if (unlink) {
1128                 if (is_dir)
1129                         inc_nlink(new_inode);
1130                 inc_nlink(new_inode);
1131         } else {
1132                 new_dir->i_size -= new_sz;
1133                 ubifs_inode(new_dir)->ui_size = new_dir->i_size;
1134         }
1135         old_dir->i_size += old_sz;
1136         ubifs_inode(old_dir)->ui_size = old_dir->i_size;
1137         if (is_dir) {
1138                 if (move) {
1139                         inc_nlink(old_dir);
1140                         if (!unlink)
1141                                 drop_nlink(new_dir);
1142                 } else {
1143                         if (unlink)
1144                                 inc_nlink(old_dir);
1145                 }
1146         }
1147         unlock_3_inodes(old_dir, new_dir, new_inode);
1148         ubifs_release_budget(c, &ino_req);
1149         ubifs_release_budget(c, &req);
1150         return err;
1151 }
1152
1153 int ubifs_getattr(struct vfsmount *mnt, struct dentry *dentry,
1154                   struct kstat *stat)
1155 {
1156         loff_t size;
1157         struct inode *inode = dentry->d_inode;
1158         struct ubifs_inode *ui = ubifs_inode(inode);
1159
1160         mutex_lock(&ui->ui_mutex);
1161         stat->dev = inode->i_sb->s_dev;
1162         stat->ino = inode->i_ino;
1163         stat->mode = inode->i_mode;
1164         stat->nlink = inode->i_nlink;
1165         stat->uid = inode->i_uid;
1166         stat->gid = inode->i_gid;
1167         stat->rdev = inode->i_rdev;
1168         stat->atime = inode->i_atime;
1169         stat->mtime = inode->i_mtime;
1170         stat->ctime = inode->i_ctime;
1171         stat->blksize = UBIFS_BLOCK_SIZE;
1172         stat->size = ui->ui_size;
1173
1174         /*
1175          * Unfortunately, the 'stat()' system call was designed for block
1176          * device based file systems, and it is not appropriate for UBIFS,
1177          * because UBIFS does not have notion of "block". For example, it is
1178          * difficult to tell how many block a directory takes - it actually
1179          * takes less than 300 bytes, but we have to round it to block size,
1180          * which introduces large mistake. This makes utilities like 'du' to
1181          * report completely senseless numbers. This is the reason why UBIFS
1182          * goes the same way as JFFS2 - it reports zero blocks for everything
1183          * but regular files, which makes more sense than reporting completely
1184          * wrong sizes.
1185          */
1186         if (S_ISREG(inode->i_mode)) {
1187                 size = ui->xattr_size;
1188                 size += stat->size;
1189                 size = ALIGN(size, UBIFS_BLOCK_SIZE);
1190                 /*
1191                  * Note, user-space expects 512-byte blocks count irrespectively
1192                  * of what was reported in @stat->size.
1193                  */
1194                 stat->blocks = size >> 9;
1195         } else
1196                 stat->blocks = 0;
1197         mutex_unlock(&ui->ui_mutex);
1198         return 0;
1199 }
1200
1201 const struct inode_operations ubifs_dir_inode_operations = {
1202         .lookup      = ubifs_lookup,
1203         .create      = ubifs_create,
1204         .link        = ubifs_link,
1205         .symlink     = ubifs_symlink,
1206         .unlink      = ubifs_unlink,
1207         .mkdir       = ubifs_mkdir,
1208         .rmdir       = ubifs_rmdir,
1209         .mknod       = ubifs_mknod,
1210         .rename      = ubifs_rename,
1211         .setattr     = ubifs_setattr,
1212         .getattr     = ubifs_getattr,
1213 #ifdef CONFIG_UBIFS_FS_XATTR
1214         .setxattr    = ubifs_setxattr,
1215         .getxattr    = ubifs_getxattr,
1216         .listxattr   = ubifs_listxattr,
1217         .removexattr = ubifs_removexattr,
1218 #endif
1219 };
1220
1221 const struct file_operations ubifs_dir_operations = {
1222         .llseek         = ubifs_dir_llseek,
1223         .release        = ubifs_dir_release,
1224         .read           = generic_read_dir,
1225         .readdir        = ubifs_readdir,
1226         .fsync          = ubifs_fsync,
1227         .unlocked_ioctl = ubifs_ioctl,
1228 #ifdef CONFIG_COMPAT
1229         .compat_ioctl   = ubifs_compat_ioctl,
1230 #endif
1231 };