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