libxfs/xfs_dir2.c

   1 /*
   2  * Copyright (c) 2000 Silicon Graphics, Inc.  All Rights Reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms of version 2 of the GNU General Public License as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it would be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
  11  *
  12  * Further, this software is distributed without any warranty that it is
  13  * free of the rightful claim of any third person regarding infringement
  14  * or the like.  Any license provided herein, whether implied or
  15  * otherwise, applies only to this software file.  Patent licenses, if
  16  * any, provided herein do not apply to combinations of this program with
  17  * other software, or any other product whatsoever.
  18  *
  19  * You should have received a copy of the GNU General Public License along
  20  * with this program; if not, write the Free Software Foundation, Inc., 59
  21  * Temple Place - Suite 330, Boston MA 02111-1307, USA.
  22  *
  23  * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
  24  * Mountain View, CA  94043, or:
  25  *
  26  * http://www.sgi.com
  27  *
  28  * For further information regarding this notice, see:
  29  *
  30  * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
  31  */
  32
  33 /*
  34  * XFS v2 directory implmentation.
  35  * Top-level and utility routines.
  36  */
  37
  38 #include <xfs.h>
  39
  40
  41 /*
  42  * Initialize directory-related fields in the mount structure.
  43  */
  44 void
  45 xfs_dir2_mount(
  46         xfs_mount_t     *mp)            /* filesystem mount point */
  47 {
  48         mp->m_dirversion = 2;
  49         ASSERT((1 << (mp->m_sb.sb_blocklog + mp->m_sb.sb_dirblklog)) <=
  50                XFS_MAX_BLOCKSIZE);
  51         mp->m_dirblksize = 1 << (mp->m_sb.sb_blocklog + mp->m_sb.sb_dirblklog);
  52         mp->m_dirblkfsbs = 1 << mp->m_sb.sb_dirblklog;
  53         mp->m_dirdatablk = XFS_DIR2_DB_TO_DA(mp, XFS_DIR2_DATA_FIRSTDB(mp));
  54         mp->m_dirleafblk = XFS_DIR2_DB_TO_DA(mp, XFS_DIR2_LEAF_FIRSTDB(mp));
  55         mp->m_dirfreeblk = XFS_DIR2_DB_TO_DA(mp, XFS_DIR2_FREE_FIRSTDB(mp));
  56         mp->m_da_node_ents =
  57                 (mp->m_dirblksize - (uint)sizeof(xfs_da_node_hdr_t)) /
  58                 (uint)sizeof(xfs_da_node_entry_t);
  59         mp->m_dir_magicpct = (mp->m_dirblksize * 37) / 100;
  60 }
  61
  62 /*
  63  * Initialize a directory with its "." and ".." entries.
  64  */
  65 int                             /* error */
  66 xfs_dir2_init(
  67         xfs_trans_t     *tp,            /* transaction pointer */
  68         xfs_inode_t     *dp,            /* incore directory inode */
  69         xfs_inode_t     *pdp)           /* incore parent directory inode */
  70 {
  71         xfs_da_args_t   args;           /* operation arguments */
  72         int             error;          /* error return value */
  73
  74         bzero((char *)&args, sizeof(args));
  75         args.dp = dp;
  76         args.trans = tp;
  77         ASSERT((dp->i_d.di_mode & IFMT) == IFDIR);
  78         if ((error = xfs_dir_ino_validate(tp->t_mountp, pdp->i_ino))) {
  79                 return error;
  80         }
  81         return xfs_dir2_sf_create(&args, pdp->i_ino);
  82 }
  83
  84 /*
  85   Enter a name in a directory.
  86  */
  87 STATIC int                                      /* error */
  88 xfs_dir2_createname(
  89         xfs_trans_t             *tp,            /* transaction pointer */
  90         xfs_inode_t             *dp,            /* incore directory inode */
  91         char                    *name,          /* new entry name */
  92         int                     namelen,        /* new entry name length */
  93         xfs_ino_t               inum,           /* new entry inode number */
  94         xfs_fsblock_t           *first,         /* bmap's firstblock */
  95         xfs_bmap_free_t         *flist,         /* bmap's freeblock list */
  96         xfs_extlen_t            total)          /* bmap's total block count */
  97 {
  98         xfs_da_args_t           args;           /* operation arguments */
  99         int                     rval;           /* return value */
 100         int                     v;              /* type-checking value */
 101
 102         ASSERT((dp->i_d.di_mode & IFMT) == IFDIR);
 103         if ((rval = xfs_dir_ino_validate(tp->t_mountp, inum))) {
 104                 return rval;
 105         }
 106         XFS_STATS_INC(xs_dir_create);
 107         /*
 108          * Fill in the arg structure for this request.
 109          */
 110         args.name = name;
 111         args.namelen = namelen;
 112         args.hashval = xfs_da_hashname(name, namelen);
 113         args.inumber = inum;
 114         args.dp = dp;
 115         args.firstblock = first;
 116         args.flist = flist;
 117         args.total = total;
 118         args.whichfork = XFS_DATA_FORK;
 119         args.trans = tp;
 120         args.justcheck = 0;
 121         args.addname = args.oknoent = 1;
 122         /*
 123          * Decide on what work routines to call based on the inode size.
 124          */
 125         if (dp->i_d.di_format == XFS_DINODE_FMT_LOCAL)
 126                 rval = xfs_dir2_sf_addname(&args);
 127         else if ((rval = xfs_dir2_isblock(tp, dp, &v))) {
 128                 return rval;
 129         } else if (v)
 130                 rval = xfs_dir2_block_addname(&args);
 131         else if ((rval = xfs_dir2_isleaf(tp, dp, &v))) {
 132                 return rval;
 133         } else if (v)
 134                 rval = xfs_dir2_leaf_addname(&args);
 135         else
 136                 rval = xfs_dir2_node_addname(&args);
 137         return rval;
 138 }
 139
 140 /*
 141  * Lookup a name in a directory, give back the inode number.
 142  */
 143 STATIC int                              /* error */
 144 xfs_dir2_lookup(
 145         xfs_trans_t     *tp,            /* transaction pointer */
 146         xfs_inode_t     *dp,            /* incore directory inode */
 147         char            *name,          /* lookup name */
 148         int             namelen,        /* lookup name length */
 149         xfs_ino_t       *inum)          /* out: inode number */
 150 {
 151         xfs_da_args_t   args;           /* operation arguments */
 152         int             rval;           /* return value */
 153         int             v;              /* type-checking value */
 154
 155         ASSERT((dp->i_d.di_mode & IFMT) == IFDIR);
 156         if (namelen >= MAXNAMELEN) {
 157                 return XFS_ERROR(EINVAL);
 158         }
 159         XFS_STATS_INC(xs_dir_lookup);
 160         /*
 161          * Fill in the arg structure for this request.
 162          */
 163         args.name = name;
 164         args.namelen = namelen;
 165         args.hashval = xfs_da_hashname(name, namelen);
 166         args.inumber = 0;
 167         args.dp = dp;
 168         args.firstblock = NULL;
 169         args.flist = NULL;
 170         args.total = 0;
 171         args.whichfork = XFS_DATA_FORK;
 172         args.trans = tp;
 173         args.justcheck = args.addname = 0;
 174         args.oknoent = 1;
 175         /*
 176          * Decide on what work routines to call based on the inode size.
 177          */
 178         if (dp->i_d.di_format == XFS_DINODE_FMT_LOCAL)
 179                 rval = xfs_dir2_sf_lookup(&args);
 180         else if ((rval = xfs_dir2_isblock(tp, dp, &v))) {
 181                 return rval;
 182         } else if (v)
 183                 rval = xfs_dir2_block_lookup(&args);
 184         else if ((rval = xfs_dir2_isleaf(tp, dp, &v))) {
 185                 return rval;
 186         } else if (v)
 187                 rval = xfs_dir2_leaf_lookup(&args);
 188         else
 189                 rval = xfs_dir2_node_lookup(&args);
 190         if (rval == EEXIST)
 191                 rval = 0;
 192         if (rval == 0)
 193                 *inum = args.inumber;
 194         return rval;
 195 }
 196
 197 /*
 198  * Remove an entry from a directory.
 199  */
 200 STATIC int                              /* error */
 201 xfs_dir2_removename(
 202         xfs_trans_t     *tp,            /* transaction pointer */
 203         xfs_inode_t     *dp,            /* incore directory inode */
 204         char            *name,          /* name of entry to remove */
 205         int             namelen,        /* name length of entry to remove */
 206         xfs_ino_t       ino,            /* inode number of entry to remove */
 207         xfs_fsblock_t   *first,         /* bmap's firstblock */
 208         xfs_bmap_free_t *flist,         /* bmap's freeblock list */
 209         xfs_extlen_t    total)          /* bmap's total block count */
 210 {
 211         xfs_da_args_t   args;           /* operation arguments */
 212         int             rval;           /* return value */
 213         int             v;              /* type-checking value */
 214
 215         ASSERT((dp->i_d.di_mode & IFMT) == IFDIR);
 216         XFS_STATS_INC(xs_dir_remove);
 217         /*
 218          * Fill in the arg structure for this request.
 219          */
 220         args.name = name;
 221         args.namelen = namelen;
 222         args.hashval = xfs_da_hashname(name, namelen);
 223         args.inumber = ino;
 224         args.dp = dp;
 225         args.firstblock = first;
 226         args.flist = flist;
 227         args.total = total;
 228         args.whichfork = XFS_DATA_FORK;
 229         args.trans = tp;
 230         args.justcheck = args.addname = args.oknoent = 0;
 231         /*
 232          * Decide on what work routines to call based on the inode size.
 233          */
 234         if (dp->i_d.di_format == XFS_DINODE_FMT_LOCAL)
 235                 rval = xfs_dir2_sf_removename(&args);
 236         else if ((rval = xfs_dir2_isblock(tp, dp, &v))) {
 237                 return rval;
 238         } else if (v)
 239                 rval = xfs_dir2_block_removename(&args);
 240         else if ((rval = xfs_dir2_isleaf(tp, dp, &v))) {
 241                 return rval;
 242         } else if (v)
 243                 rval = xfs_dir2_leaf_removename(&args);
 244         else
 245                 rval = xfs_dir2_node_removename(&args);
 246         return rval;
 247 }
 248
 249 /*
 250  * Replace the inode number of a directory entry.
 251  */
 252 STATIC int                              /* error */
 253 xfs_dir2_replace(
 254         xfs_trans_t     *tp,            /* transaction pointer */
 255         xfs_inode_t     *dp,            /* incore directory inode */
 256         char            *name,          /* name of entry to replace */
 257         int             namelen,        /* name length of entry to replace */
 258         xfs_ino_t       inum,           /* new inode number */
 259         xfs_fsblock_t   *first,         /* bmap's firstblock */
 260         xfs_bmap_free_t *flist,         /* bmap's freeblock list */
 261         xfs_extlen_t    total)          /* bmap's total block count */
 262 {
 263         xfs_da_args_t   args;           /* operation arguments */
 264         int             rval;           /* return value */
 265         int             v;              /* type-checking value */
 266
 267         ASSERT((dp->i_d.di_mode & IFMT) == IFDIR);
 268         if (namelen >= MAXNAMELEN) {
 269                 return XFS_ERROR(EINVAL);
 270         }
 271         if ((rval = xfs_dir_ino_validate(tp->t_mountp, inum))) {
 272                 return rval;
 273         }
 274         /*
 275          * Fill in the arg structure for this request.
 276          */
 277         args.name = name;
 278         args.namelen = namelen;
 279         args.hashval = xfs_da_hashname(name, namelen);
 280         args.inumber = inum;
 281         args.dp = dp;
 282         args.firstblock = first;
 283         args.flist = flist;
 284         args.total = total;
 285         args.whichfork = XFS_DATA_FORK;
 286         args.trans = tp;
 287         args.justcheck = args.addname = args.oknoent = 0;
 288         /*
 289          * Decide on what work routines to call based on the inode size.
 290          */
 291         if (dp->i_d.di_format == XFS_DINODE_FMT_LOCAL)
 292                 rval = xfs_dir2_sf_replace(&args);
 293         else if ((rval = xfs_dir2_isblock(tp, dp, &v))) {
 294                 return rval;
 295         } else if (v)
 296                 rval = xfs_dir2_block_replace(&args);
 297         else if ((rval = xfs_dir2_isleaf(tp, dp, &v))) {
 298                 return rval;
 299         } else if (v)
 300                 rval = xfs_dir2_leaf_replace(&args);
 301         else
 302                 rval = xfs_dir2_node_replace(&args);
 303         return rval;
 304 }
 305
 306 /*
 307  * Utility routines.
 308  */
 309
 310 /*
 311  * Add a block to the directory.
 312  * This routine is for data and free blocks, not leaf/node blocks
 313  * which are handled by xfs_da_grow_inode.
 314  */
 315 int                                     /* error */
 316 xfs_dir2_grow_inode(
 317         xfs_da_args_t   *args,          /* operation arguments */
 318         int             space,          /* v2 dir's space XFS_DIR2_xxx_SPACE */
 319         xfs_dir2_db_t   *dbp)           /* out: block number added */
 320 {
 321         xfs_fileoff_t   bno;            /* directory offset of new block */
 322         int             count;          /* count of filesystem blocks */
 323         xfs_inode_t     *dp;            /* incore directory inode */
 324         int             error;          /* error return value */
 325         int             got;            /* blocks actually mapped */
 326         int             i;              /* temp mapping index */
 327         xfs_bmbt_irec_t map;            /* single structure for bmap */
 328         int             mapi;           /* mapping index */
 329         xfs_bmbt_irec_t *mapp;          /* bmap mapping structure(s) */
 330         xfs_mount_t     *mp;            /* filesystem mount point */
 331         int             nmap;           /* number of bmap entries */
 332         xfs_trans_t     *tp;            /* transaction pointer */
 333
 334         xfs_dir2_trace_args_s("grow_inode", args, space);
 335         dp = args->dp;
 336         tp = args->trans;
 337         mp = dp->i_mount;
 338         /*
 339          * Set lowest possible block in the space requested.
 340          */
 341         bno = XFS_B_TO_FSBT(mp, space * XFS_DIR2_SPACE_SIZE);
 342         count = mp->m_dirblkfsbs;
 343         /*
 344          * Find the first hole for our block.
 345          */
 346         if ((error = xfs_bmap_first_unused(tp, dp, count, &bno, XFS_DATA_FORK))) {
 347                 return error;
 348         }
 349         nmap = 1;
 350         ASSERT(args->firstblock != NULL);
 351         /*
 352          * Try mapping the new block contiguously (one extent).
 353          */
 354         if ((error = xfs_bmapi(tp, dp, bno, count,
 355                         XFS_BMAPI_WRITE|XFS_BMAPI_METADATA|XFS_BMAPI_CONTIG,
 356                         args->firstblock, args->total, &map, &nmap,
 357                         args->flist))) {
 358                 return error;
 359         }
 360         ASSERT(nmap <= 1);
 361         /*
 362          * Got it in 1.
 363          */
 364         if (nmap == 1) {
 365                 mapp = &map;
 366                 mapi = 1;
 367         }
 368         /*
 369          * Didn't work and this is a multiple-fsb directory block.
 370          * Try again with contiguous flag turned on.
 371          */
 372         else if (nmap == 0 && count > 1) {
 373                 xfs_fileoff_t   b;      /* current file offset */
 374
 375                 /*
 376                  * Space for maximum number of mappings.
 377                  */
 378                 mapp = kmem_alloc(sizeof(*mapp) * count, KM_SLEEP);
 379                 /*
 380                  * Iterate until we get to the end of our block.
 381                  */
 382                 for (b = bno, mapi = 0; b < bno + count; ) {
 383                         int     c;      /* current fsb count */
 384
 385                         /*
 386                          * Can't map more than MAX_NMAP at once.
 387                          */
 388                         nmap = MIN(XFS_BMAP_MAX_NMAP, count);
 389                         c = (int)(bno + count - b);
 390                         if ((error = xfs_bmapi(tp, dp, b, c,
 391                                         XFS_BMAPI_WRITE|XFS_BMAPI_METADATA,
 392                                         args->firstblock, args->total,
 393                                         &mapp[mapi], &nmap, args->flist))) {
 394                                 kmem_free(mapp, sizeof(*mapp) * count);
 395                                 return error;
 396                         }
 397                         if (nmap < 1)
 398                                 break;
 399                         /*
 400                          * Add this bunch into our table, go to the next offset.
 401                          */
 402                         mapi += nmap;
 403                         b = mapp[mapi - 1].br_startoff +
 404                             mapp[mapi - 1].br_blockcount;
 405                 }
 406         }
 407         /*
 408          * Didn't work.
 409          */
 410         else {
 411                 mapi = 0;
 412                 mapp = NULL;
 413         }
 414         /*
 415          * See how many fsb's we got.
 416          */
 417         for (i = 0, got = 0; i < mapi; i++)
 418                 got += mapp[i].br_blockcount;
 419         /*
 420          * Didn't get enough fsb's, or the first/last block's are wrong.
 421          */
 422         if (got != count || mapp[0].br_startoff != bno ||
 423             mapp[mapi - 1].br_startoff + mapp[mapi - 1].br_blockcount !=
 424             bno + count) {
 425                 if (mapp != &map)
 426                         kmem_free(mapp, sizeof(*mapp) * count);
 427                 return XFS_ERROR(ENOSPC);
 428         }
 429         /*
 430          * Done with the temporary mapping table.
 431          */
 432         if (mapp != &map)
 433                 kmem_free(mapp, sizeof(*mapp) * count);
 434         *dbp = XFS_DIR2_DA_TO_DB(mp, (xfs_dablk_t)bno);
 435         /*
 436          * Update file's size if this is the data space and it grew.
 437          */
 438         if (space == XFS_DIR2_DATA_SPACE) {
 439                 xfs_fsize_t     size;           /* directory file (data) size */
 440
 441                 size = XFS_FSB_TO_B(mp, bno + count);
 442                 if (size > dp->i_d.di_size) {
 443                         dp->i_d.di_size = size;
 444                         xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
 445                 }
 446         }
 447         return 0;
 448 }
 449
 450 /*
 451  * See if the directory is a single-block form directory.
 452  */
 453 int                                     /* error */
 454 xfs_dir2_isblock(
 455         xfs_trans_t     *tp,            /* transaction pointer */
 456         xfs_inode_t     *dp,            /* incore directory inode */
 457         int             *vp)            /* out: 1 is block, 0 is not block */
 458 {
 459         xfs_fileoff_t   last;           /* last file offset */
 460         xfs_mount_t     *mp;            /* filesystem mount point */
 461         int             rval;           /* return value */
 462
 463         mp = dp->i_mount;
 464         if ((rval = xfs_bmap_last_offset(tp, dp, &last, XFS_DATA_FORK))) {
 465                 return rval;
 466         }
 467         rval = XFS_FSB_TO_B(mp, last) == mp->m_dirblksize;
 468         ASSERT(rval == 0 || dp->i_d.di_size == mp->m_dirblksize);
 469         *vp = rval;
 470         return 0;
 471 }
 472
 473 /*
 474  * See if the directory is a single-leaf form directory.
 475  */
 476 int                                     /* error */
 477 xfs_dir2_isleaf(
 478         xfs_trans_t     *tp,            /* transaction pointer */
 479         xfs_inode_t     *dp,            /* incore directory inode */
 480         int             *vp)            /* out: 1 is leaf, 0 is not leaf */
 481 {
 482         xfs_fileoff_t   last;           /* last file offset */
 483         xfs_mount_t     *mp;            /* filesystem mount point */
 484         int             rval;           /* return value */
 485
 486         mp = dp->i_mount;
 487         if ((rval = xfs_bmap_last_offset(tp, dp, &last, XFS_DATA_FORK))) {
 488                 return rval;
 489         }
 490         *vp = last == mp->m_dirleafblk + (1 << mp->m_sb.sb_dirblklog);
 491         return 0;
 492 }
 493
 494 /*
 495  * Remove the given block from the directory.
 496  * This routine is used for data and free blocks, leaf/node are done
 497  * by xfs_da_shrink_inode.
 498  */
 499 int
 500 xfs_dir2_shrink_inode(
 501         xfs_da_args_t   *args,          /* operation arguments */
 502         xfs_dir2_db_t   db,             /* directory block number */
 503         xfs_dabuf_t     *bp)            /* block's buffer */
 504 {
 505         xfs_fileoff_t   bno;            /* directory file offset */
 506         xfs_dablk_t     da;             /* directory file offset */
 507         int             done;           /* bunmap is finished */
 508         xfs_inode_t     *dp;            /* incore directory inode */
 509         int             error;          /* error return value */
 510         xfs_mount_t     *mp;            /* filesystem mount point */
 511         xfs_trans_t     *tp;            /* transaction pointer */
 512
 513         xfs_dir2_trace_args_db("shrink_inode", args, db, bp);
 514         dp = args->dp;
 515         mp = dp->i_mount;
 516         tp = args->trans;
 517         da = XFS_DIR2_DB_TO_DA(mp, db);
 518         /*
 519          * Unmap the fsblock(s).
 520          */
 521         if ((error = xfs_bunmapi(tp, dp, da, mp->m_dirblkfsbs,
 522                         XFS_BMAPI_METADATA, 0, args->firstblock, args->flist,
 523                         &done))) {
 524                 /*
 525                  * ENOSPC actually can happen if we're in a removename with
 526                  * no space reservation, and the resulting block removal
 527                  * would cause a bmap btree split or conversion from extents
 528                  * to btree.  This can only happen for un-fragmented
 529                  * directory blocks, since you need to be punching out
 530                  * the middle of an extent.
 531                  * In this case we need to leave the block in the file,
 532                  * and not binval it.
 533                  * So the block has to be in a consistent empty state
 534                  * and appropriately logged.
 535                  * We don't free up the buffer, the caller can tell it
 536                  * hasn't happened since it got an error back.
 537                  */
 538                 return error;
 539         }
 540         ASSERT(done);
 541         /*
 542          * Invalidate the buffer from the transaction.
 543          */
 544         xfs_da_binval(tp, bp);
 545         /*
 546          * If it's not a data block, we're done.
 547          */
 548         if (db >= XFS_DIR2_LEAF_FIRSTDB(mp))
 549                 return 0;
 550         /*
 551          * If the block isn't the last one in the directory, we're done.
 552          */
 553         if (dp->i_d.di_size > XFS_DIR2_DB_OFF_TO_BYTE(mp, db + 1, 0))
 554                 return 0;
 555         bno = da;
 556         if ((error = xfs_bmap_last_before(tp, dp, &bno, XFS_DATA_FORK))) {
 557                 /*
 558                  * This can't really happen unless there's kernel corruption.
 559                  */
 560                 return error;
 561         }
 562         if (db == mp->m_dirdatablk)
 563                 ASSERT(bno == 0);
 564         else
 565                 ASSERT(bno > 0);
 566         /*
 567          * Set the size to the new last block.
 568          */
 569         dp->i_d.di_size = XFS_FSB_TO_B(mp, bno);
 570         xfs_trans_log_inode(tp, dp, XFS_ILOG_CORE);
 571         return 0;
 572 }