fs/ocfs2/dir.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * dir.c
   5  *
   6  * Creates, reads, walks and deletes directory-nodes
   7  *
   8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9  *
  10  *  Portions of this code from linux/fs/ext3/dir.c
  11  *
  12  *  Copyright (C) 1992, 1993, 1994, 1995
  13  *  Remy Card (card@masi.ibp.fr)
  14  *  Laboratoire MASI - Institut Blaise pascal
  15  *  Universite Pierre et Marie Curie (Paris VI)
  16  *
  17  *   from
  18  *
  19  *   linux/fs/minix/dir.c
  20  *
  21  *   Copyright (C) 1991, 1992 Linux Torvalds
  22  *
  23  * This program is free software; you can redistribute it and/or
  24  * modify it under the terms of the GNU General Public
  25  * License as published by the Free Software Foundation; either
  26  * version 2 of the License, or (at your option) any later version.
  27  *
  28  * This program is distributed in the hope that it will be useful,
  29  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  30  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  31  * General Public License for more details.
  32  *
  33  * You should have received a copy of the GNU General Public
  34  * License along with this program; if not, write to the
  35  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  36  * Boston, MA 021110-1307, USA.
  37  */
  38
  39 #include <linux/fs.h>
  40 #include <linux/types.h>
  41 #include <linux/slab.h>
  42 #include <linux/highmem.h>
  43
  44 #define MLOG_MASK_PREFIX ML_NAMEI
  45 #include <cluster/masklog.h>
  46
  47 #include "ocfs2.h"
  48
  49 #include "alloc.h"
  50 #include "dir.h"
  51 #include "dlmglue.h"
  52 #include "extent_map.h"
  53 #include "file.h"
  54 #include "inode.h"
  55 #include "journal.h"
  56 #include "namei.h"
  57 #include "suballoc.h"
  58 #include "uptodate.h"
  59
  60 #include "buffer_head_io.h"
  61
  62 static unsigned char ocfs2_filetype_table[] = {
  63         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
  64 };
  65
  66 static int ocfs2_extend_dir(struct ocfs2_super *osb,
  67                             struct inode *dir,
  68                             struct buffer_head *parent_fe_bh,
  69                             struct buffer_head **new_de_bh);
  70 /*
  71  * ocfs2_readdir()
  72  *
  73  */
  74 int ocfs2_readdir(struct file * filp, void * dirent, filldir_t filldir)
  75 {
  76         int error = 0;
  77         unsigned long offset, blk;
  78         int i, num, stored;
  79         struct buffer_head * bh, * tmp;
  80         struct ocfs2_dir_entry * de;
  81         int err;
  82         struct inode *inode = filp->f_dentry->d_inode;
  83         struct super_block * sb = inode->i_sb;
  84         int have_disk_lock = 0;
  85
  86         mlog_entry("dirino=%llu\n",
  87                    (unsigned long long)OCFS2_I(inode)->ip_blkno);
  88
  89         stored = 0;
  90         bh = NULL;
  91
  92         error = ocfs2_meta_lock(inode, NULL, NULL, 0);
  93         if (error < 0) {
  94                 if (error != -ENOENT)
  95                         mlog_errno(error);
  96                 /* we haven't got any yet, so propagate the error. */
  97                 stored = error;
  98                 goto bail;
  99         }
 100         have_disk_lock = 1;
 101
 102         offset = filp->f_pos & (sb->s_blocksize - 1);
 103
 104         while (!error && !stored && filp->f_pos < i_size_read(inode)) {
 105                 blk = (filp->f_pos) >> sb->s_blocksize_bits;
 106                 bh = ocfs2_bread(inode, blk, &err, 0);
 107                 if (!bh) {
 108                         mlog(ML_ERROR,
 109                              "directory #%llu contains a hole at offset %lld\n",
 110                              (unsigned long long)OCFS2_I(inode)->ip_blkno,
 111                              filp->f_pos);
 112                         filp->f_pos += sb->s_blocksize - offset;
 113                         continue;
 114                 }
 115
 116                 /*
 117                  * Do the readahead (8k)
 118                  */
 119                 if (!offset) {
 120                         for (i = 16 >> (sb->s_blocksize_bits - 9), num = 0;
 121                              i > 0; i--) {
 122                                 tmp = ocfs2_bread(inode, ++blk, &err, 1);
 123                                 if (tmp)
 124                                         brelse(tmp);
 125                         }
 126                 }
 127
 128 revalidate:
 129                 /* If the dir block has changed since the last call to
 130                  * readdir(2), then we might be pointing to an invalid
 131                  * dirent right now.  Scan from the start of the block
 132                  * to make sure. */
 133                 if (filp->f_version != inode->i_version) {
 134                         for (i = 0; i < sb->s_blocksize && i < offset; ) {
 135                                 de = (struct ocfs2_dir_entry *) (bh->b_data + i);
 136                                 /* It's too expensive to do a full
 137                                  * dirent test each time round this
 138                                  * loop, but we do have to test at
 139                                  * least that it is non-zero.  A
 140                                  * failure will be detected in the
 141                                  * dirent test below. */
 142                                 if (le16_to_cpu(de->rec_len) <
 143                                     OCFS2_DIR_REC_LEN(1))
 144                                         break;
 145                                 i += le16_to_cpu(de->rec_len);
 146                         }
 147                         offset = i;
 148                         filp->f_pos = (filp->f_pos & ~(sb->s_blocksize - 1))
 149                                 | offset;
 150                         filp->f_version = inode->i_version;
 151                 }
 152
 153                 while (!error && filp->f_pos < i_size_read(inode)
 154                        && offset < sb->s_blocksize) {
 155                         de = (struct ocfs2_dir_entry *) (bh->b_data + offset);
 156                         if (!ocfs2_check_dir_entry(inode, de, bh, offset)) {
 157                                 /* On error, skip the f_pos to the
 158                                    next block. */
 159                                 filp->f_pos = (filp->f_pos |
 160                                                (sb->s_blocksize - 1)) + 1;
 161                                 brelse(bh);
 162                                 goto bail;
 163                         }
 164                         offset += le16_to_cpu(de->rec_len);
 165                         if (le64_to_cpu(de->inode)) {
 166                                 /* We might block in the next section
 167                                  * if the data destination is
 168                                  * currently swapped out.  So, use a
 169                                  * version stamp to detect whether or
 170                                  * not the directory has been modified
 171                                  * during the copy operation.
 172                                  */
 173                                 unsigned long version = filp->f_version;
 174                                 unsigned char d_type = DT_UNKNOWN;
 175
 176                                 if (de->file_type < OCFS2_FT_MAX)
 177                                         d_type = ocfs2_filetype_table[de->file_type];
 178                                 error = filldir(dirent, de->name,
 179                                                 de->name_len,
 180                                                 filp->f_pos,
 181                                                 ino_from_blkno(sb, le64_to_cpu(de->inode)),
 182                                                 d_type);
 183                                 if (error)
 184                                         break;
 185                                 if (version != filp->f_version)
 186                                         goto revalidate;
 187                                 stored ++;
 188                         }
 189                         filp->f_pos += le16_to_cpu(de->rec_len);
 190                 }
 191                 offset = 0;
 192                 brelse(bh);
 193         }
 194
 195         stored = 0;
 196 bail:
 197         if (have_disk_lock)
 198                 ocfs2_meta_unlock(inode, 0);
 199
 200         mlog_exit(stored);
 201
 202         return stored;
 203 }
 204
 205 /*
 206  * NOTE: this should always be called with parent dir i_mutex taken.
 207  */
 208 int ocfs2_find_files_on_disk(const char *name,
 209                              int namelen,
 210                              u64 *blkno,
 211                              struct inode *inode,
 212                              struct buffer_head **dirent_bh,
 213                              struct ocfs2_dir_entry **dirent)
 214 {
 215         int status = -ENOENT;
 216
 217         mlog_entry("(name=%.*s, blkno=%p, inode=%p, dirent_bh=%p, dirent=%p)\n",
 218                    namelen, name, blkno, inode, dirent_bh, dirent);
 219
 220         *dirent_bh = ocfs2_find_entry(name, namelen, inode, dirent);
 221         if (!*dirent_bh || !*dirent) {
 222                 status = -ENOENT;
 223                 goto leave;
 224         }
 225
 226         *blkno = le64_to_cpu((*dirent)->inode);
 227
 228         status = 0;
 229 leave:
 230         if (status < 0) {
 231                 *dirent = NULL;
 232                 if (*dirent_bh) {
 233                         brelse(*dirent_bh);
 234                         *dirent_bh = NULL;
 235                 }
 236         }
 237
 238         mlog_exit(status);
 239         return status;
 240 }
 241
 242 /* Check for a name within a directory.
 243  *
 244  * Return 0 if the name does not exist
 245  * Return -EEXIST if the directory contains the name
 246  *
 247  * Callers should have i_mutex + a cluster lock on dir
 248  */
 249 int ocfs2_check_dir_for_entry(struct inode *dir,
 250                               const char *name,
 251                               int namelen)
 252 {
 253         int ret;
 254         struct buffer_head *dirent_bh = NULL;
 255         struct ocfs2_dir_entry *dirent = NULL;
 256
 257         mlog_entry("dir %llu, name '%.*s'\n",
 258                    (unsigned long long)OCFS2_I(dir)->ip_blkno, namelen, name);
 259
 260         ret = -EEXIST;
 261         dirent_bh = ocfs2_find_entry(name, namelen, dir, &dirent);
 262         if (dirent_bh)
 263                 goto bail;
 264
 265         ret = 0;
 266 bail:
 267         if (dirent_bh)
 268                 brelse(dirent_bh);
 269
 270         mlog_exit(ret);
 271         return ret;
 272 }
 273
 274 /*
 275  * routine to check that the specified directory is empty (for rmdir)
 276  */
 277 int ocfs2_empty_dir(struct inode *inode)
 278 {
 279         unsigned long offset;
 280         struct buffer_head * bh;
 281         struct ocfs2_dir_entry * de, * de1;
 282         struct super_block * sb;
 283         int err;
 284
 285         sb = inode->i_sb;
 286         if ((i_size_read(inode) <
 287              (OCFS2_DIR_REC_LEN(1) + OCFS2_DIR_REC_LEN(2))) ||
 288             !(bh = ocfs2_bread(inode, 0, &err, 0))) {
 289                 mlog(ML_ERROR, "bad directory (dir #%llu) - no data block\n",
 290                      (unsigned long long)OCFS2_I(inode)->ip_blkno);
 291                 return 1;
 292         }
 293
 294         de = (struct ocfs2_dir_entry *) bh->b_data;
 295         de1 = (struct ocfs2_dir_entry *)
 296                         ((char *)de + le16_to_cpu(de->rec_len));
 297         if ((le64_to_cpu(de->inode) != OCFS2_I(inode)->ip_blkno) ||
 298                         !le64_to_cpu(de1->inode) ||
 299                         strcmp(".", de->name) ||
 300                         strcmp("..", de1->name)) {
 301                 mlog(ML_ERROR, "bad directory (dir #%llu) - no `.' or `..'\n",
 302                      (unsigned long long)OCFS2_I(inode)->ip_blkno);
 303                 brelse(bh);
 304                 return 1;
 305         }
 306         offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
 307         de = (struct ocfs2_dir_entry *)((char *)de1 + le16_to_cpu(de1->rec_len));
 308         while (offset < i_size_read(inode) ) {
 309                 if (!bh || (void *)de >= (void *)(bh->b_data + sb->s_blocksize)) {
 310                         brelse(bh);
 311                         bh = ocfs2_bread(inode,
 312                                          offset >> sb->s_blocksize_bits, &err, 0);
 313                         if (!bh) {
 314                                 mlog(ML_ERROR, "dir %llu has a hole at %lu\n",
 315                                      (unsigned long long)OCFS2_I(inode)->ip_blkno, offset);
 316                                 offset += sb->s_blocksize;
 317                                 continue;
 318                         }
 319                         de = (struct ocfs2_dir_entry *) bh->b_data;
 320                 }
 321                 if (!ocfs2_check_dir_entry(inode, de, bh, offset)) {
 322                         brelse(bh);
 323                         return 1;
 324                 }
 325                 if (le64_to_cpu(de->inode)) {
 326                         brelse(bh);
 327                         return 0;
 328                 }
 329                 offset += le16_to_cpu(de->rec_len);
 330                 de = (struct ocfs2_dir_entry *)
 331                         ((char *)de + le16_to_cpu(de->rec_len));
 332         }
 333         brelse(bh);
 334         return 1;
 335 }
 336
 337 /* returns a bh of the 1st new block in the allocation. */
 338 int ocfs2_do_extend_dir(struct super_block *sb,
 339                         struct ocfs2_journal_handle *handle,
 340                         struct inode *dir,
 341                         struct buffer_head *parent_fe_bh,
 342                         struct ocfs2_alloc_context *data_ac,
 343                         struct ocfs2_alloc_context *meta_ac,
 344                         struct buffer_head **new_bh)
 345 {
 346         int status;
 347         int extend;
 348         u64 p_blkno;
 349
 350         spin_lock(&OCFS2_I(dir)->ip_lock);
 351         extend = (i_size_read(dir) == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters));
 352         spin_unlock(&OCFS2_I(dir)->ip_lock);
 353
 354         if (extend) {
 355                 status = ocfs2_do_extend_allocation(OCFS2_SB(sb), dir, 1,
 356                                                     parent_fe_bh, handle,
 357                                                     data_ac, meta_ac, NULL);
 358                 BUG_ON(status == -EAGAIN);
 359                 if (status < 0) {
 360                         mlog_errno(status);
 361                         goto bail;
 362                 }
 363         }
 364
 365         status = ocfs2_extent_map_get_blocks(dir, (dir->i_blocks >>
 366                                                    (sb->s_blocksize_bits - 9)),
 367                                              1, &p_blkno, NULL);
 368         if (status < 0) {
 369                 mlog_errno(status);
 370                 goto bail;
 371         }
 372
 373         *new_bh = sb_getblk(sb, p_blkno);
 374         if (!*new_bh) {
 375                 status = -EIO;
 376                 mlog_errno(status);
 377                 goto bail;
 378         }
 379         status = 0;
 380 bail:
 381         mlog_exit(status);
 382         return status;
 383 }
 384
 385 /* assumes you already have a cluster lock on the directory. */
 386 static int ocfs2_extend_dir(struct ocfs2_super *osb,
 387                             struct inode *dir,
 388                             struct buffer_head *parent_fe_bh,
 389                             struct buffer_head **new_de_bh)
 390 {
 391         int status = 0;
 392         int credits, num_free_extents;
 393         loff_t dir_i_size;
 394         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 395         struct ocfs2_alloc_context *data_ac = NULL;
 396         struct ocfs2_alloc_context *meta_ac = NULL;
 397         struct ocfs2_journal_handle *handle = NULL;
 398         struct buffer_head *new_bh = NULL;
 399         struct ocfs2_dir_entry * de;
 400         struct super_block *sb = osb->sb;
 401
 402         mlog_entry_void();
 403
 404         dir_i_size = i_size_read(dir);
 405         mlog(0, "extending dir %llu (i_size = %lld)\n",
 406              (unsigned long long)OCFS2_I(dir)->ip_blkno, dir_i_size);
 407
 408         handle = ocfs2_alloc_handle(osb);
 409         if (handle == NULL) {
 410                 status = -ENOMEM;
 411                 mlog_errno(status);
 412                 goto bail;
 413         }
 414
 415         /* dir->i_size is always block aligned. */
 416         spin_lock(&OCFS2_I(dir)->ip_lock);
 417         if (dir_i_size == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters)) {
 418                 spin_unlock(&OCFS2_I(dir)->ip_lock);
 419                 num_free_extents = ocfs2_num_free_extents(osb, dir, fe);
 420                 if (num_free_extents < 0) {
 421                         status = num_free_extents;
 422                         mlog_errno(status);
 423                         goto bail;
 424                 }
 425
 426                 if (!num_free_extents) {
 427                         status = ocfs2_reserve_new_metadata(osb, handle,
 428                                                             fe, &meta_ac);
 429                         if (status < 0) {
 430                                 if (status != -ENOSPC)
 431                                         mlog_errno(status);
 432                                 goto bail;
 433                         }
 434                 }
 435
 436                 status = ocfs2_reserve_clusters(osb, handle, 1, &data_ac);
 437                 if (status < 0) {
 438                         if (status != -ENOSPC)
 439                                 mlog_errno(status);
 440                         goto bail;
 441                 }
 442
 443                 credits = ocfs2_calc_extend_credits(sb, fe, 1);
 444         } else {
 445                 spin_unlock(&OCFS2_I(dir)->ip_lock);
 446                 credits = OCFS2_SIMPLE_DIR_EXTEND_CREDITS;
 447         }
 448
 449         handle = ocfs2_start_trans(osb, handle, credits);
 450         if (IS_ERR(handle)) {
 451                 status = PTR_ERR(handle);
 452                 handle = NULL;
 453                 mlog_errno(status);
 454                 goto bail;
 455         }
 456
 457         status = ocfs2_do_extend_dir(osb->sb, handle, dir, parent_fe_bh,
 458                                      data_ac, meta_ac, &new_bh);
 459         if (status < 0) {
 460                 mlog_errno(status);
 461                 goto bail;
 462         }
 463
 464         ocfs2_set_new_buffer_uptodate(dir, new_bh);
 465
 466         status = ocfs2_journal_access(handle, dir, new_bh,
 467                                       OCFS2_JOURNAL_ACCESS_CREATE);
 468         if (status < 0) {
 469                 mlog_errno(status);
 470                 goto bail;
 471         }
 472         memset(new_bh->b_data, 0, sb->s_blocksize);
 473         de = (struct ocfs2_dir_entry *) new_bh->b_data;
 474         de->inode = 0;
 475         de->rec_len = cpu_to_le16(sb->s_blocksize);
 476         status = ocfs2_journal_dirty(handle, new_bh);
 477         if (status < 0) {
 478                 mlog_errno(status);
 479                 goto bail;
 480         }
 481
 482         dir_i_size += dir->i_sb->s_blocksize;
 483         i_size_write(dir, dir_i_size);
 484         dir->i_blocks = ocfs2_align_bytes_to_sectors(dir_i_size);
 485         status = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
 486         if (status < 0) {
 487                 mlog_errno(status);
 488                 goto bail;
 489         }
 490
 491         *new_de_bh = new_bh;
 492         get_bh(*new_de_bh);
 493 bail:
 494         if (handle)
 495                 ocfs2_commit_trans(handle);
 496
 497         if (data_ac)
 498                 ocfs2_free_alloc_context(data_ac);
 499         if (meta_ac)
 500                 ocfs2_free_alloc_context(meta_ac);
 501
 502         if (new_bh)
 503                 brelse(new_bh);
 504
 505         mlog_exit(status);
 506         return status;
 507 }
 508
 509 /*
 510  * Search the dir for a good spot, extending it if necessary. The
 511  * block containing an appropriate record is returned in ret_de_bh.
 512  */
 513 int ocfs2_prepare_dir_for_insert(struct ocfs2_super *osb,
 514                                  struct inode *dir,
 515                                  struct buffer_head *parent_fe_bh,
 516                                  const char *name,
 517                                  int namelen,
 518                                  struct buffer_head **ret_de_bh)
 519 {
 520         unsigned long offset;
 521         struct buffer_head * bh = NULL;
 522         unsigned short rec_len;
 523         struct ocfs2_dinode *fe;
 524         struct ocfs2_dir_entry *de;
 525         struct super_block *sb;
 526         int status;
 527
 528         mlog_entry_void();
 529
 530         mlog(0, "getting ready to insert namelen %d into dir %llu\n",
 531              namelen, (unsigned long long)OCFS2_I(dir)->ip_blkno);
 532
 533         BUG_ON(!S_ISDIR(dir->i_mode));
 534         fe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 535         BUG_ON(le64_to_cpu(fe->i_size) != i_size_read(dir));
 536
 537         sb = dir->i_sb;
 538
 539         if (!namelen) {
 540                 status = -EINVAL;
 541                 mlog_errno(status);
 542                 goto bail;
 543         }
 544
 545         bh = ocfs2_bread(dir, 0, &status, 0);
 546         if (!bh) {
 547                 mlog_errno(status);
 548                 goto bail;
 549         }
 550
 551         rec_len = OCFS2_DIR_REC_LEN(namelen);
 552         offset = 0;
 553         de = (struct ocfs2_dir_entry *) bh->b_data;
 554         while (1) {
 555                 if ((char *)de >= sb->s_blocksize + bh->b_data) {
 556                         brelse(bh);
 557                         bh = NULL;
 558
 559                         if (i_size_read(dir) <= offset) {
 560                                 status = ocfs2_extend_dir(osb,
 561                                                           dir,
 562                                                           parent_fe_bh,
 563                                                           &bh);
 564                                 if (status < 0) {
 565                                         mlog_errno(status);
 566                                         goto bail;
 567                                 }
 568                                 BUG_ON(!bh);
 569                                 *ret_de_bh = bh;
 570                                 get_bh(*ret_de_bh);
 571                                 goto bail;
 572                         }
 573                         bh = ocfs2_bread(dir,
 574                                          offset >> sb->s_blocksize_bits,
 575                                          &status,
 576                                          0);
 577                         if (!bh) {
 578                                 mlog_errno(status);
 579                                 goto bail;
 580                         }
 581                         /* move to next block */
 582                         de = (struct ocfs2_dir_entry *) bh->b_data;
 583                 }
 584                 if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
 585                         status = -ENOENT;
 586                         goto bail;
 587                 }
 588                 if (ocfs2_match(namelen, name, de)) {
 589                         status = -EEXIST;
 590                         goto bail;
 591                 }
 592                 if (((le64_to_cpu(de->inode) == 0) &&
 593                      (le16_to_cpu(de->rec_len) >= rec_len)) ||
 594                     (le16_to_cpu(de->rec_len) >=
 595                      (OCFS2_DIR_REC_LEN(de->name_len) + rec_len))) {
 596                         /* Ok, we found a spot. Return this bh and let
 597                          * the caller actually fill it in. */
 598                         *ret_de_bh = bh;
 599                         get_bh(*ret_de_bh);
 600                         status = 0;
 601                         goto bail;
 602                 }
 603                 offset += le16_to_cpu(de->rec_len);
 604                 de = (struct ocfs2_dir_entry *)((char *) de + le16_to_cpu(de->rec_len));
 605         }
 606
 607         status = 0;
 608 bail:
 609         if (bh)
 610                 brelse(bh);
 611
 612         mlog_exit(status);
 613         return status;
 614 }