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 - lock two UBIFS inodes.
 486  * @inode1: first inode
 487  * @inode2: second inode
 488  */
 489 static void lock_2_inodes(struct inode *inode1, struct inode *inode2)
 490 {
 491         if (inode1->i_ino < inode2->i_ino) {
 492                 mutex_lock_nested(&ubifs_inode(inode1)->ui_mutex, WB_MUTEX_2);
 493                 mutex_lock_nested(&ubifs_inode(inode2)->ui_mutex, WB_MUTEX_3);
 494         } else {
 495                 mutex_lock_nested(&ubifs_inode(inode2)->ui_mutex, WB_MUTEX_2);
 496                 mutex_lock_nested(&ubifs_inode(inode1)->ui_mutex, WB_MUTEX_3);
 497         }
 498 }
 499
 500 /**
 501  * unlock_2_inodes - unlock two UBIFS inodes inodes.
 502  * @inode1: first inode
 503  * @inode2: second inode
 504  */
 505 static void unlock_2_inodes(struct inode *inode1, struct inode *inode2)
 506 {
 507         mutex_unlock(&ubifs_inode(inode1)->ui_mutex);
 508         mutex_unlock(&ubifs_inode(inode2)->ui_mutex);
 509 }
 510
 511 static int ubifs_link(struct dentry *old_dentry, struct inode *dir,
 512                       struct dentry *dentry)
 513 {
 514         struct ubifs_info *c = dir->i_sb->s_fs_info;
 515         struct inode *inode = old_dentry->d_inode;
 516         struct ubifs_inode *ui = ubifs_inode(inode);
 517         struct ubifs_inode *dir_ui = ubifs_inode(dir);
 518         int err, sz_change = CALC_DENT_SIZE(dentry->d_name.len);
 519         struct ubifs_budget_req req = { .new_dent = 1, .dirtied_ino = 2,
 520                                 .dirtied_ino_d = ALIGN(ui->data_len, 8) };
 521
 522         /*
 523          * Budget request settings: new direntry, changing the target inode,
 524          * changing the parent inode.
 525          */
 526
 527         dbg_gen("dent '%.*s' to ino %lu (nlink %d) in dir ino %lu",
 528                 dentry->d_name.len, dentry->d_name.name, inode->i_ino,
 529                 inode->i_nlink, dir->i_ino);
 530         err = dbg_check_synced_i_size(inode);
 531         if (err)
 532                 return err;
 533
 534         err = ubifs_budget_space(c, &req);
 535         if (err)
 536                 return err;
 537
 538         lock_2_inodes(dir, inode);
 539         inc_nlink(inode);
 540         atomic_inc(&inode->i_count);
 541         inode->i_ctime = ubifs_current_time(inode);
 542         dir->i_size += sz_change;
 543         dir_ui->ui_size = dir->i_size;
 544         dir->i_mtime = dir->i_ctime = inode->i_ctime;
 545         err = ubifs_jnl_update(c, dir, &dentry->d_name, inode, 0, 0);
 546         if (err)
 547                 goto out_cancel;
 548         unlock_2_inodes(dir, inode);
 549
 550         ubifs_release_budget(c, &req);
 551         d_instantiate(dentry, inode);
 552         return 0;
 553
 554 out_cancel:
 555         dir->i_size -= sz_change;
 556         dir_ui->ui_size = dir->i_size;
 557         drop_nlink(inode);
 558         unlock_2_inodes(dir, inode);
 559         ubifs_release_budget(c, &req);
 560         iput(inode);
 561         return err;
 562 }
 563
 564 static int ubifs_unlink(struct inode *dir, struct dentry *dentry)
 565 {
 566         struct ubifs_info *c = dir->i_sb->s_fs_info;
 567         struct inode *inode = dentry->d_inode;
 568         struct ubifs_inode *dir_ui = ubifs_inode(dir);
 569         int sz_change = CALC_DENT_SIZE(dentry->d_name.len);
 570         int err, budgeted = 1;
 571         struct ubifs_budget_req req = { .mod_dent = 1, .dirtied_ino = 2 };
 572
 573         /*
 574          * Budget request settings: deletion direntry, deletion inode (+1 for
 575          * @dirtied_ino), changing the parent directory inode. If budgeting
 576          * fails, go ahead anyway because we have extra space reserved for
 577          * deletions.
 578          */
 579
 580         dbg_gen("dent '%.*s' from ino %lu (nlink %d) in dir ino %lu",
 581                 dentry->d_name.len, dentry->d_name.name, inode->i_ino,
 582                 inode->i_nlink, dir->i_ino);
 583         err = dbg_check_synced_i_size(inode);
 584         if (err)
 585                 return err;
 586
 587         err = ubifs_budget_space(c, &req);
 588         if (err) {
 589                 if (err != -ENOSPC)
 590                         return err;
 591                 budgeted = 0;
 592         }
 593
 594         lock_2_inodes(dir, inode);
 595         inode->i_ctime = ubifs_current_time(dir);
 596         drop_nlink(inode);
 597         dir->i_size -= sz_change;
 598         dir_ui->ui_size = dir->i_size;
 599         dir->i_mtime = dir->i_ctime = inode->i_ctime;
 600         err = ubifs_jnl_update(c, dir, &dentry->d_name, inode, 1, 0);
 601         if (err)
 602                 goto out_cancel;
 603         unlock_2_inodes(dir, inode);
 604
 605         if (budgeted)
 606                 ubifs_release_budget(c, &req);
 607         else {
 608                 /* We've deleted something - clean the "no space" flags */
 609                 c->nospace = c->nospace_rp = 0;
 610                 smp_wmb();
 611         }
 612         return 0;
 613
 614 out_cancel:
 615         dir->i_size += sz_change;
 616         dir_ui->ui_size = dir->i_size;
 617         inc_nlink(inode);
 618         unlock_2_inodes(dir, inode);
 619         if (budgeted)
 620                 ubifs_release_budget(c, &req);
 621         return err;
 622 }
 623
 624 /**
 625  * check_dir_empty - check if a directory is empty or not.
 626  * @c: UBIFS file-system description object
 627  * @dir: VFS inode object of the directory to check
 628  *
 629  * This function checks if directory @dir is empty. Returns zero if the
 630  * directory is empty, %-ENOTEMPTY if it is not, and other negative error codes
 631  * in case of of errors.
 632  */
 633 static int check_dir_empty(struct ubifs_info *c, struct inode *dir)
 634 {
 635         struct qstr nm = { .name = NULL };
 636         struct ubifs_dent_node *dent;
 637         union ubifs_key key;
 638         int err;
 639
 640         lowest_dent_key(c, &key, dir->i_ino);
 641         dent = ubifs_tnc_next_ent(c, &key, &nm);
 642         if (IS_ERR(dent)) {
 643                 err = PTR_ERR(dent);
 644                 if (err == -ENOENT)
 645                         err = 0;
 646         } else {
 647                 kfree(dent);
 648                 err = -ENOTEMPTY;
 649         }
 650         return err;
 651 }
 652
 653 static int ubifs_rmdir(struct inode *dir, struct dentry *dentry)
 654 {
 655         struct ubifs_info *c = dir->i_sb->s_fs_info;
 656         struct inode *inode = dentry->d_inode;
 657         int sz_change = CALC_DENT_SIZE(dentry->d_name.len);
 658         int err, budgeted = 1;
 659         struct ubifs_inode *dir_ui = ubifs_inode(dir);
 660         struct ubifs_budget_req req = { .mod_dent = 1, .dirtied_ino = 2 };
 661
 662         /*
 663          * Budget request settings: deletion direntry, deletion inode and
 664          * changing the parent inode. If budgeting fails, go ahead anyway
 665          * because we have extra space reserved for deletions.
 666          */
 667
 668         dbg_gen("directory '%.*s', ino %lu in dir ino %lu", dentry->d_name.len,
 669                 dentry->d_name.name, inode->i_ino, dir->i_ino);
 670
 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->nospace = c->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 - lock three UBIFS inodes for rename.
 926  * @inode1: first inode
 927  * @inode2: second inode
 928  * @inode3: third inode
 929  *
 930  * For 'ubifs_rename()', @inode1 may be the same as @inode2 whereas @inode3 may
 931  * be null.
 932  */
 933 static void lock_3_inodes(struct inode *inode1, struct inode *inode2,
 934                           struct inode *inode3)
 935 {
 936         struct inode *i1, *i2, *i3;
 937
 938         if (!inode3) {
 939                 if (inode1 != inode2) {
 940                         lock_2_inodes(inode1, inode2);
 941                         return;
 942                 }
 943                 mutex_lock_nested(&ubifs_inode(inode1)->ui_mutex, WB_MUTEX_1);
 944                 return;
 945         }
 946
 947         if (inode1 == inode2) {
 948                 lock_2_inodes(inode1, inode3);
 949                 return;
 950         }
 951
 952         /* 3 different inodes */
 953         if (inode1 < inode2) {
 954                 i3 = inode2;
 955                 if (inode1 < inode3) {
 956                         i1 = inode1;
 957                         i2 = inode3;
 958                 } else {
 959                         i1 = inode3;
 960                         i2 = inode1;
 961                 }
 962         } else {
 963                 i3 = inode1;
 964                 if (inode2 < inode3) {
 965                         i1 = inode2;
 966                         i2 = inode3;
 967                 } else {
 968                         i1 = inode3;
 969                         i2 = inode2;
 970                 }
 971         }
 972         mutex_lock_nested(&ubifs_inode(i1)->ui_mutex, WB_MUTEX_1);
 973         lock_2_inodes(i2, i3);
 974 }
 975
 976 /**
 977  * unlock_3_inodes - unlock three UBIFS inodes for rename.
 978  * @inode1: first inode
 979  * @inode2: second inode
 980  * @inode3: third inode
 981  */
 982 static void unlock_3_inodes(struct inode *inode1, struct inode *inode2,
 983                             struct inode *inode3)
 984 {
 985         mutex_unlock(&ubifs_inode(inode1)->ui_mutex);
 986         if (inode1 != inode2)
 987                 mutex_unlock(&ubifs_inode(inode2)->ui_mutex);
 988         if (inode3)
 989                 mutex_unlock(&ubifs_inode(inode3)->ui_mutex);
 990 }
 991
 992 static int ubifs_rename(struct inode *old_dir, struct dentry *old_dentry,
 993                         struct inode *new_dir, struct dentry *new_dentry)
 994 {
 995         struct ubifs_info *c = old_dir->i_sb->s_fs_info;
 996         struct inode *old_inode = old_dentry->d_inode;
 997         struct inode *new_inode = new_dentry->d_inode;
 998         struct ubifs_inode *old_inode_ui = ubifs_inode(old_inode);
 999         int err, release, sync = 0, move = (new_dir != old_dir);
1000         int is_dir = S_ISDIR(old_inode->i_mode);
1001         int unlink = !!new_inode;
1002         int new_sz = CALC_DENT_SIZE(new_dentry->d_name.len);
1003         int old_sz = CALC_DENT_SIZE(old_dentry->d_name.len);
1004         struct ubifs_budget_req req = { .new_dent = 1, .mod_dent = 1,
1005                                         .dirtied_ino = 3 };
1006         struct ubifs_budget_req ino_req = { .dirtied_ino = 1,
1007                         .dirtied_ino_d = ALIGN(old_inode_ui->data_len, 8) };
1008         struct timespec time;
1009
1010         /*
1011          * Budget request settings: deletion direntry, new direntry, removing
1012          * the old inode, and changing old and new parent directory inodes.
1013          *
1014          * However, this operation also marks the target inode as dirty and
1015          * does not write it, so we allocate budget for the target inode
1016          * separately.
1017          */
1018
1019         dbg_gen("dent '%.*s' ino %lu in dir ino %lu to dent '%.*s' in "
1020                 "dir ino %lu", old_dentry->d_name.len, old_dentry->d_name.name,
1021                 old_inode->i_ino, old_dir->i_ino, new_dentry->d_name.len,
1022                 new_dentry->d_name.name, new_dir->i_ino);
1023
1024         if (unlink && is_dir) {
1025                 err = check_dir_empty(c, new_inode);
1026                 if (err)
1027                         return err;
1028         }
1029
1030         err = ubifs_budget_space(c, &req);
1031         if (err)
1032                 return err;
1033         err = ubifs_budget_space(c, &ino_req);
1034         if (err) {
1035                 ubifs_release_budget(c, &req);
1036                 return err;
1037         }
1038
1039         lock_3_inodes(old_dir, new_dir, new_inode);
1040
1041         /*
1042          * Like most other Unix systems, set the @i_ctime for inodes on a
1043          * rename.
1044          */
1045         time = ubifs_current_time(old_dir);
1046         old_inode->i_ctime = time;
1047
1048         /* We must adjust parent link count when renaming directories */
1049         if (is_dir) {
1050                 if (move) {
1051                         /*
1052                          * @old_dir loses a link because we are moving
1053                          * @old_inode to a different directory.
1054                          */
1055                         drop_nlink(old_dir);
1056                         /*
1057                          * @new_dir only gains a link if we are not also
1058                          * overwriting an existing directory.
1059                          */
1060                         if (!unlink)
1061                                 inc_nlink(new_dir);
1062                 } else {
1063                         /*
1064                          * @old_inode is not moving to a different directory,
1065                          * but @old_dir still loses a link if we are
1066                          * overwriting an existing directory.
1067                          */
1068                         if (unlink)
1069                                 drop_nlink(old_dir);
1070                 }
1071         }
1072
1073         old_dir->i_size -= old_sz;
1074         ubifs_inode(old_dir)->ui_size = old_dir->i_size;
1075         old_dir->i_mtime = old_dir->i_ctime = time;
1076         new_dir->i_mtime = new_dir->i_ctime = time;
1077
1078         /*
1079          * And finally, if we unlinked a direntry which happened to have the
1080          * same name as the moved direntry, we have to decrement @i_nlink of
1081          * the unlinked inode and change its ctime.
1082          */
1083         if (unlink) {
1084                 /*
1085                  * Directories cannot have hard-links, so if this is a
1086                  * directory, decrement its @i_nlink twice because an empty
1087                  * directory has @i_nlink 2.
1088                  */
1089                 if (is_dir)
1090                         drop_nlink(new_inode);
1091                 new_inode->i_ctime = time;
1092                 drop_nlink(new_inode);
1093         } else {
1094                 new_dir->i_size += new_sz;
1095                 ubifs_inode(new_dir)->ui_size = new_dir->i_size;
1096         }
1097
1098         /*
1099          * Do not ask 'ubifs_jnl_rename()' to flush write-buffer if @old_inode
1100          * is dirty, because this will be done later on at the end of
1101          * 'ubifs_rename()'.
1102          */
1103         if (IS_SYNC(old_inode)) {
1104                 sync = IS_DIRSYNC(old_dir) || IS_DIRSYNC(new_dir);
1105                 if (unlink && IS_SYNC(new_inode))
1106                         sync = 1;
1107         }
1108         err = ubifs_jnl_rename(c, old_dir, old_dentry, new_dir, new_dentry,
1109                                sync);
1110         if (err)
1111                 goto out_cancel;
1112
1113         unlock_3_inodes(old_dir, new_dir, new_inode);
1114         ubifs_release_budget(c, &req);
1115
1116         mutex_lock(&old_inode_ui->ui_mutex);
1117         release = old_inode_ui->dirty;
1118         mark_inode_dirty_sync(old_inode);
1119         mutex_unlock(&old_inode_ui->ui_mutex);
1120
1121         if (release)
1122                 ubifs_release_budget(c, &ino_req);
1123         if (IS_SYNC(old_inode))
1124                 err = old_inode->i_sb->s_op->write_inode(old_inode, 1);
1125         return err;
1126
1127 out_cancel:
1128         if (unlink) {
1129                 if (is_dir)
1130                         inc_nlink(new_inode);
1131                 inc_nlink(new_inode);
1132         } else {
1133                 new_dir->i_size -= new_sz;
1134                 ubifs_inode(new_dir)->ui_size = new_dir->i_size;
1135         }
1136         old_dir->i_size += old_sz;
1137         ubifs_inode(old_dir)->ui_size = old_dir->i_size;
1138         if (is_dir) {
1139                 if (move) {
1140                         inc_nlink(old_dir);
1141                         if (!unlink)
1142                                 drop_nlink(new_dir);
1143                 } else {
1144                         if (unlink)
1145                                 inc_nlink(old_dir);
1146                 }
1147         }
1148         unlock_3_inodes(old_dir, new_dir, new_inode);
1149         ubifs_release_budget(c, &ino_req);
1150         ubifs_release_budget(c, &req);
1151         return err;
1152 }
1153
1154 int ubifs_getattr(struct vfsmount *mnt, struct dentry *dentry,
1155                   struct kstat *stat)
1156 {
1157         loff_t size;
1158         struct inode *inode = dentry->d_inode;
1159         struct ubifs_inode *ui = ubifs_inode(inode);
1160
1161         mutex_lock(&ui->ui_mutex);
1162         stat->dev = inode->i_sb->s_dev;
1163         stat->ino = inode->i_ino;
1164         stat->mode = inode->i_mode;
1165         stat->nlink = inode->i_nlink;
1166         stat->uid = inode->i_uid;
1167         stat->gid = inode->i_gid;
1168         stat->rdev = inode->i_rdev;
1169         stat->atime = inode->i_atime;
1170         stat->mtime = inode->i_mtime;
1171         stat->ctime = inode->i_ctime;
1172         stat->blksize = UBIFS_BLOCK_SIZE;
1173         stat->size = ui->ui_size;
1174
1175         /*
1176          * Unfortunately, the 'stat()' system call was designed for block
1177          * device based file systems, and it is not appropriate for UBIFS,
1178          * because UBIFS does not have notion of "block". For example, it is
1179          * difficult to tell how many block a directory takes - it actually
1180          * takes less than 300 bytes, but we have to round it to block size,
1181          * which introduces large mistake. This makes utilities like 'du' to
1182          * report completely senseless numbers. This is the reason why UBIFS
1183          * goes the same way as JFFS2 - it reports zero blocks for everything
1184          * but regular files, which makes more sense than reporting completely
1185          * wrong sizes.
1186          */
1187         if (S_ISREG(inode->i_mode)) {
1188                 size = ui->xattr_size;
1189                 size += stat->size;
1190                 size = ALIGN(size, UBIFS_BLOCK_SIZE);
1191                 /*
1192                  * Note, user-space expects 512-byte blocks count irrespectively
1193                  * of what was reported in @stat->size.
1194                  */
1195                 stat->blocks = size >> 9;
1196         } else
1197                 stat->blocks = 0;
1198         mutex_unlock(&ui->ui_mutex);
1199         return 0;
1200 }
1201
1202 struct inode_operations ubifs_dir_inode_operations = {
1203         .lookup      = ubifs_lookup,
1204         .create      = ubifs_create,
1205         .link        = ubifs_link,
1206         .symlink     = ubifs_symlink,
1207         .unlink      = ubifs_unlink,
1208         .mkdir       = ubifs_mkdir,
1209         .rmdir       = ubifs_rmdir,
1210         .mknod       = ubifs_mknod,
1211         .rename      = ubifs_rename,
1212         .setattr     = ubifs_setattr,
1213         .getattr     = ubifs_getattr,
1214 #ifdef CONFIG_UBIFS_FS_XATTR
1215         .setxattr    = ubifs_setxattr,
1216         .getxattr    = ubifs_getxattr,
1217         .listxattr   = ubifs_listxattr,
1218         .removexattr = ubifs_removexattr,
1219 #endif
1220 };
1221
1222 struct file_operations ubifs_dir_operations = {
1223         .llseek         = ubifs_dir_llseek,
1224         .release        = ubifs_dir_release,
1225         .read           = generic_read_dir,
1226         .readdir        = ubifs_readdir,
1227         .fsync          = ubifs_fsync,
1228         .unlocked_ioctl = ubifs_ioctl,
1229 #ifdef CONFIG_COMPAT
1230         .compat_ioctl   = ubifs_compat_ioctl,
1231 #endif
1232 };