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, last_ra_blk = 0;
  78         int i, 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         unsigned int ra_sectors = 16;
  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_nolock;
  99         }
 100
 101         offset = filp->f_pos & (sb->s_blocksize - 1);
 102
 103         while (!error && !stored && filp->f_pos < i_size_read(inode)) {
 104                 blk = (filp->f_pos) >> sb->s_blocksize_bits;
 105                 bh = ocfs2_bread(inode, blk, &err, 0);
 106                 if (!bh) {
 107                         mlog(ML_ERROR,
 108                              "directory #%llu contains a hole at offset %lld\n",
 109                              (unsigned long long)OCFS2_I(inode)->ip_blkno,
 110                              filp->f_pos);
 111                         filp->f_pos += sb->s_blocksize - offset;
 112                         continue;
 113                 }
 114
 115                 /* The idea here is to begin with 8k read-ahead and to stay
 116                  * 4k ahead of our current position.
 117                  *
 118                  * TODO: Use the pagecache for this. We just need to
 119                  * make sure it's cluster-safe... */
 120                 if (!last_ra_blk
 121                     || (((last_ra_blk - blk) << 9) <= (ra_sectors / 2))) {
 122                         for (i = ra_sectors >> (sb->s_blocksize_bits - 9);
 123                              i > 0; i--) {
 124                                 tmp = ocfs2_bread(inode, ++blk, &err, 1);
 125                                 if (tmp)
 126                                         brelse(tmp);
 127                         }
 128                         last_ra_blk = blk;
 129                         ra_sectors = 8;
 130                 }
 131
 132 revalidate:
 133                 /* If the dir block has changed since the last call to
 134                  * readdir(2), then we might be pointing to an invalid
 135                  * dirent right now.  Scan from the start of the block
 136                  * to make sure. */
 137                 if (filp->f_version != inode->i_version) {
 138                         for (i = 0; i < sb->s_blocksize && i < offset; ) {
 139                                 de = (struct ocfs2_dir_entry *) (bh->b_data + i);
 140                                 /* It's too expensive to do a full
 141                                  * dirent test each time round this
 142                                  * loop, but we do have to test at
 143                                  * least that it is non-zero.  A
 144                                  * failure will be detected in the
 145                                  * dirent test below. */
 146                                 if (le16_to_cpu(de->rec_len) <
 147                                     OCFS2_DIR_REC_LEN(1))
 148                                         break;
 149                                 i += le16_to_cpu(de->rec_len);
 150                         }
 151                         offset = i;
 152                         filp->f_pos = (filp->f_pos & ~(sb->s_blocksize - 1))
 153                                 | offset;
 154                         filp->f_version = inode->i_version;
 155                 }
 156
 157                 while (!error && filp->f_pos < i_size_read(inode)
 158                        && offset < sb->s_blocksize) {
 159                         de = (struct ocfs2_dir_entry *) (bh->b_data + offset);
 160                         if (!ocfs2_check_dir_entry(inode, de, bh, offset)) {
 161                                 /* On error, skip the f_pos to the
 162                                    next block. */
 163                                 filp->f_pos = (filp->f_pos |
 164                                                (sb->s_blocksize - 1)) + 1;
 165                                 brelse(bh);
 166                                 goto bail;
 167                         }
 168                         offset += le16_to_cpu(de->rec_len);
 169                         if (le64_to_cpu(de->inode)) {
 170                                 /* We might block in the next section
 171                                  * if the data destination is
 172                                  * currently swapped out.  So, use a
 173                                  * version stamp to detect whether or
 174                                  * not the directory has been modified
 175                                  * during the copy operation.
 176                                  */
 177                                 unsigned long version = filp->f_version;
 178                                 unsigned char d_type = DT_UNKNOWN;
 179
 180                                 if (de->file_type < OCFS2_FT_MAX)
 181                                         d_type = ocfs2_filetype_table[de->file_type];
 182                                 error = filldir(dirent, de->name,
 183                                                 de->name_len,
 184                                                 filp->f_pos,
 185                                                 ino_from_blkno(sb, le64_to_cpu(de->inode)),
 186                                                 d_type);
 187                                 if (error)
 188                                         break;
 189                                 if (version != filp->f_version)
 190                                         goto revalidate;
 191                                 stored ++;
 192                         }
 193                         filp->f_pos += le16_to_cpu(de->rec_len);
 194                 }
 195                 offset = 0;
 196                 brelse(bh);
 197         }
 198
 199         stored = 0;
 200 bail:
 201         ocfs2_meta_unlock(inode, 0);
 202
 203 bail_nolock:
 204         mlog_exit(stored);
 205
 206         return stored;
 207 }
 208
 209 /*
 210  * NOTE: this should always be called with parent dir i_mutex taken.
 211  */
 212 int ocfs2_find_files_on_disk(const char *name,
 213                              int namelen,
 214                              u64 *blkno,
 215                              struct inode *inode,
 216                              struct buffer_head **dirent_bh,
 217                              struct ocfs2_dir_entry **dirent)
 218 {
 219         int status = -ENOENT;
 220
 221         mlog_entry("(name=%.*s, blkno=%p, inode=%p, dirent_bh=%p, dirent=%p)\n",
 222                    namelen, name, blkno, inode, dirent_bh, dirent);
 223
 224         *dirent_bh = ocfs2_find_entry(name, namelen, inode, dirent);
 225         if (!*dirent_bh || !*dirent) {
 226                 status = -ENOENT;
 227                 goto leave;
 228         }
 229
 230         *blkno = le64_to_cpu((*dirent)->inode);
 231
 232         status = 0;
 233 leave:
 234         if (status < 0) {
 235                 *dirent = NULL;
 236                 if (*dirent_bh) {
 237                         brelse(*dirent_bh);
 238                         *dirent_bh = NULL;
 239                 }
 240         }
 241
 242         mlog_exit(status);
 243         return status;
 244 }
 245
 246 /* Check for a name within a directory.
 247  *
 248  * Return 0 if the name does not exist
 249  * Return -EEXIST if the directory contains the name
 250  *
 251  * Callers should have i_mutex + a cluster lock on dir
 252  */
 253 int ocfs2_check_dir_for_entry(struct inode *dir,
 254                               const char *name,
 255                               int namelen)
 256 {
 257         int ret;
 258         struct buffer_head *dirent_bh = NULL;
 259         struct ocfs2_dir_entry *dirent = NULL;
 260
 261         mlog_entry("dir %llu, name '%.*s'\n",
 262                    (unsigned long long)OCFS2_I(dir)->ip_blkno, namelen, name);
 263
 264         ret = -EEXIST;
 265         dirent_bh = ocfs2_find_entry(name, namelen, dir, &dirent);
 266         if (dirent_bh)
 267                 goto bail;
 268
 269         ret = 0;
 270 bail:
 271         if (dirent_bh)
 272                 brelse(dirent_bh);
 273
 274         mlog_exit(ret);
 275         return ret;
 276 }
 277
 278 /*
 279  * routine to check that the specified directory is empty (for rmdir)
 280  */
 281 int ocfs2_empty_dir(struct inode *inode)
 282 {
 283         unsigned long offset;
 284         struct buffer_head * bh;
 285         struct ocfs2_dir_entry * de, * de1;
 286         struct super_block * sb;
 287         int err;
 288
 289         sb = inode->i_sb;
 290         if ((i_size_read(inode) <
 291              (OCFS2_DIR_REC_LEN(1) + OCFS2_DIR_REC_LEN(2))) ||
 292             !(bh = ocfs2_bread(inode, 0, &err, 0))) {
 293                 mlog(ML_ERROR, "bad directory (dir #%llu) - no data block\n",
 294                      (unsigned long long)OCFS2_I(inode)->ip_blkno);
 295                 return 1;
 296         }
 297
 298         de = (struct ocfs2_dir_entry *) bh->b_data;
 299         de1 = (struct ocfs2_dir_entry *)
 300                         ((char *)de + le16_to_cpu(de->rec_len));
 301         if ((le64_to_cpu(de->inode) != OCFS2_I(inode)->ip_blkno) ||
 302                         !le64_to_cpu(de1->inode) ||
 303                         strcmp(".", de->name) ||
 304                         strcmp("..", de1->name)) {
 305                 mlog(ML_ERROR, "bad directory (dir #%llu) - no `.' or `..'\n",
 306                      (unsigned long long)OCFS2_I(inode)->ip_blkno);
 307                 brelse(bh);
 308                 return 1;
 309         }
 310         offset = le16_to_cpu(de->rec_len) + le16_to_cpu(de1->rec_len);
 311         de = (struct ocfs2_dir_entry *)((char *)de1 + le16_to_cpu(de1->rec_len));
 312         while (offset < i_size_read(inode) ) {
 313                 if (!bh || (void *)de >= (void *)(bh->b_data + sb->s_blocksize)) {
 314                         brelse(bh);
 315                         bh = ocfs2_bread(inode,
 316                                          offset >> sb->s_blocksize_bits, &err, 0);
 317                         if (!bh) {
 318                                 mlog(ML_ERROR, "dir %llu has a hole at %lu\n",
 319                                      (unsigned long long)OCFS2_I(inode)->ip_blkno, offset);
 320                                 offset += sb->s_blocksize;
 321                                 continue;
 322                         }
 323                         de = (struct ocfs2_dir_entry *) bh->b_data;
 324                 }
 325                 if (!ocfs2_check_dir_entry(inode, de, bh, offset)) {
 326                         brelse(bh);
 327                         return 1;
 328                 }
 329                 if (le64_to_cpu(de->inode)) {
 330                         brelse(bh);
 331                         return 0;
 332                 }
 333                 offset += le16_to_cpu(de->rec_len);
 334                 de = (struct ocfs2_dir_entry *)
 335                         ((char *)de + le16_to_cpu(de->rec_len));
 336         }
 337         brelse(bh);
 338         return 1;
 339 }
 340
 341 /* returns a bh of the 1st new block in the allocation. */
 342 int ocfs2_do_extend_dir(struct super_block *sb,
 343                         struct ocfs2_journal_handle *handle,
 344                         struct inode *dir,
 345                         struct buffer_head *parent_fe_bh,
 346                         struct ocfs2_alloc_context *data_ac,
 347                         struct ocfs2_alloc_context *meta_ac,
 348                         struct buffer_head **new_bh)
 349 {
 350         int status;
 351         int extend;
 352         u64 p_blkno;
 353
 354         spin_lock(&OCFS2_I(dir)->ip_lock);
 355         extend = (i_size_read(dir) == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters));
 356         spin_unlock(&OCFS2_I(dir)->ip_lock);
 357
 358         if (extend) {
 359                 status = ocfs2_do_extend_allocation(OCFS2_SB(sb), dir, 1,
 360                                                     parent_fe_bh, handle,
 361                                                     data_ac, meta_ac, NULL);
 362                 BUG_ON(status == -EAGAIN);
 363                 if (status < 0) {
 364                         mlog_errno(status);
 365                         goto bail;
 366                 }
 367         }
 368
 369         status = ocfs2_extent_map_get_blocks(dir, (dir->i_blocks >>
 370                                                    (sb->s_blocksize_bits - 9)),
 371                                              1, &p_blkno, NULL);
 372         if (status < 0) {
 373                 mlog_errno(status);
 374                 goto bail;
 375         }
 376
 377         *new_bh = sb_getblk(sb, p_blkno);
 378         if (!*new_bh) {
 379                 status = -EIO;
 380                 mlog_errno(status);
 381                 goto bail;
 382         }
 383         status = 0;
 384 bail:
 385         mlog_exit(status);
 386         return status;
 387 }
 388
 389 /* assumes you already have a cluster lock on the directory. */
 390 static int ocfs2_extend_dir(struct ocfs2_super *osb,
 391                             struct inode *dir,
 392                             struct buffer_head *parent_fe_bh,
 393                             struct buffer_head **new_de_bh)
 394 {
 395         int status = 0;
 396         int credits, num_free_extents;
 397         loff_t dir_i_size;
 398         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 399         struct ocfs2_alloc_context *data_ac = NULL;
 400         struct ocfs2_alloc_context *meta_ac = NULL;
 401         struct ocfs2_journal_handle *handle = NULL;
 402         struct buffer_head *new_bh = NULL;
 403         struct ocfs2_dir_entry * de;
 404         struct super_block *sb = osb->sb;
 405
 406         mlog_entry_void();
 407
 408         dir_i_size = i_size_read(dir);
 409         mlog(0, "extending dir %llu (i_size = %lld)\n",
 410              (unsigned long long)OCFS2_I(dir)->ip_blkno, dir_i_size);
 411
 412         handle = ocfs2_alloc_handle(osb);
 413         if (handle == NULL) {
 414                 status = -ENOMEM;
 415                 mlog_errno(status);
 416                 goto bail;
 417         }
 418
 419         /* dir->i_size is always block aligned. */
 420         spin_lock(&OCFS2_I(dir)->ip_lock);
 421         if (dir_i_size == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters)) {
 422                 spin_unlock(&OCFS2_I(dir)->ip_lock);
 423                 num_free_extents = ocfs2_num_free_extents(osb, dir, fe);
 424                 if (num_free_extents < 0) {
 425                         status = num_free_extents;
 426                         mlog_errno(status);
 427                         goto bail;
 428                 }
 429
 430                 if (!num_free_extents) {
 431                         status = ocfs2_reserve_new_metadata(osb, handle,
 432                                                             fe, &meta_ac);
 433                         if (status < 0) {
 434                                 if (status != -ENOSPC)
 435                                         mlog_errno(status);
 436                                 goto bail;
 437                         }
 438                 }
 439
 440                 status = ocfs2_reserve_clusters(osb, handle, 1, &data_ac);
 441                 if (status < 0) {
 442                         if (status != -ENOSPC)
 443                                 mlog_errno(status);
 444                         goto bail;
 445                 }
 446
 447                 credits = ocfs2_calc_extend_credits(sb, fe, 1);
 448         } else {
 449                 spin_unlock(&OCFS2_I(dir)->ip_lock);
 450                 credits = OCFS2_SIMPLE_DIR_EXTEND_CREDITS;
 451         }
 452
 453         handle = ocfs2_start_trans(osb, handle, credits);
 454         if (IS_ERR(handle)) {
 455                 status = PTR_ERR(handle);
 456                 handle = NULL;
 457                 mlog_errno(status);
 458                 goto bail;
 459         }
 460
 461         status = ocfs2_do_extend_dir(osb->sb, handle, dir, parent_fe_bh,
 462                                      data_ac, meta_ac, &new_bh);
 463         if (status < 0) {
 464                 mlog_errno(status);
 465                 goto bail;
 466         }
 467
 468         ocfs2_set_new_buffer_uptodate(dir, new_bh);
 469
 470         status = ocfs2_journal_access(handle, dir, new_bh,
 471                                       OCFS2_JOURNAL_ACCESS_CREATE);
 472         if (status < 0) {
 473                 mlog_errno(status);
 474                 goto bail;
 475         }
 476         memset(new_bh->b_data, 0, sb->s_blocksize);
 477         de = (struct ocfs2_dir_entry *) new_bh->b_data;
 478         de->inode = 0;
 479         de->rec_len = cpu_to_le16(sb->s_blocksize);
 480         status = ocfs2_journal_dirty(handle, new_bh);
 481         if (status < 0) {
 482                 mlog_errno(status);
 483                 goto bail;
 484         }
 485
 486         dir_i_size += dir->i_sb->s_blocksize;
 487         i_size_write(dir, dir_i_size);
 488         dir->i_blocks = ocfs2_align_bytes_to_sectors(dir_i_size);
 489         status = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
 490         if (status < 0) {
 491                 mlog_errno(status);
 492                 goto bail;
 493         }
 494
 495         *new_de_bh = new_bh;
 496         get_bh(*new_de_bh);
 497 bail:
 498         if (handle)
 499                 ocfs2_commit_trans(handle);
 500
 501         if (data_ac)
 502                 ocfs2_free_alloc_context(data_ac);
 503         if (meta_ac)
 504                 ocfs2_free_alloc_context(meta_ac);
 505
 506         if (new_bh)
 507                 brelse(new_bh);
 508
 509         mlog_exit(status);
 510         return status;
 511 }
 512
 513 /*
 514  * Search the dir for a good spot, extending it if necessary. The
 515  * block containing an appropriate record is returned in ret_de_bh.
 516  */
 517 int ocfs2_prepare_dir_for_insert(struct ocfs2_super *osb,
 518                                  struct inode *dir,
 519                                  struct buffer_head *parent_fe_bh,
 520                                  const char *name,
 521                                  int namelen,
 522                                  struct buffer_head **ret_de_bh)
 523 {
 524         unsigned long offset;
 525         struct buffer_head * bh = NULL;
 526         unsigned short rec_len;
 527         struct ocfs2_dinode *fe;
 528         struct ocfs2_dir_entry *de;
 529         struct super_block *sb;
 530         int status;
 531
 532         mlog_entry_void();
 533
 534         mlog(0, "getting ready to insert namelen %d into dir %llu\n",
 535              namelen, (unsigned long long)OCFS2_I(dir)->ip_blkno);
 536
 537         BUG_ON(!S_ISDIR(dir->i_mode));
 538         fe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 539         BUG_ON(le64_to_cpu(fe->i_size) != i_size_read(dir));
 540
 541         sb = dir->i_sb;
 542
 543         if (!namelen) {
 544                 status = -EINVAL;
 545                 mlog_errno(status);
 546                 goto bail;
 547         }
 548
 549         bh = ocfs2_bread(dir, 0, &status, 0);
 550         if (!bh) {
 551                 mlog_errno(status);
 552                 goto bail;
 553         }
 554
 555         rec_len = OCFS2_DIR_REC_LEN(namelen);
 556         offset = 0;
 557         de = (struct ocfs2_dir_entry *) bh->b_data;
 558         while (1) {
 559                 if ((char *)de >= sb->s_blocksize + bh->b_data) {
 560                         brelse(bh);
 561                         bh = NULL;
 562
 563                         if (i_size_read(dir) <= offset) {
 564                                 status = ocfs2_extend_dir(osb,
 565                                                           dir,
 566                                                           parent_fe_bh,
 567                                                           &bh);
 568                                 if (status < 0) {
 569                                         mlog_errno(status);
 570                                         goto bail;
 571                                 }
 572                                 BUG_ON(!bh);
 573                                 *ret_de_bh = bh;
 574                                 get_bh(*ret_de_bh);
 575                                 goto bail;
 576                         }
 577                         bh = ocfs2_bread(dir,
 578                                          offset >> sb->s_blocksize_bits,
 579                                          &status,
 580                                          0);
 581                         if (!bh) {
 582                                 mlog_errno(status);
 583                                 goto bail;
 584                         }
 585                         /* move to next block */
 586                         de = (struct ocfs2_dir_entry *) bh->b_data;
 587                 }
 588                 if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
 589                         status = -ENOENT;
 590                         goto bail;
 591                 }
 592                 if (ocfs2_match(namelen, name, de)) {
 593                         status = -EEXIST;
 594                         goto bail;
 595                 }
 596                 if (((le64_to_cpu(de->inode) == 0) &&
 597                      (le16_to_cpu(de->rec_len) >= rec_len)) ||
 598                     (le16_to_cpu(de->rec_len) >=
 599                      (OCFS2_DIR_REC_LEN(de->name_len) + rec_len))) {
 600                         /* Ok, we found a spot. Return this bh and let
 601                          * the caller actually fill it in. */
 602                         *ret_de_bh = bh;
 603                         get_bh(*ret_de_bh);
 604                         status = 0;
 605                         goto bail;
 606                 }
 607                 offset += le16_to_cpu(de->rec_len);
 608                 de = (struct ocfs2_dir_entry *)((char *) de + le16_to_cpu(de->rec_len));
 609         }
 610
 611         status = 0;
 612 bail:
 613         if (bh)
 614                 brelse(bh);
 615
 616         mlog_exit(status);
 617         return status;
 618 }