linux/fs/xfs/xfs_dir2_readdir.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   4 * Copyright (c) 2013 Red Hat, Inc.
   5 * All Rights Reserved.
   6 */
   7#include "xfs.h"
   8#include "xfs_fs.h"
   9#include "xfs_shared.h"
  10#include "xfs_format.h"
  11#include "xfs_log_format.h"
  12#include "xfs_trans_resv.h"
  13#include "xfs_mount.h"
  14#include "xfs_inode.h"
  15#include "xfs_dir2.h"
  16#include "xfs_dir2_priv.h"
  17#include "xfs_trace.h"
  18#include "xfs_bmap.h"
  19#include "xfs_trans.h"
  20#include "xfs_error.h"
  21
  22/*
  23 * Directory file type support functions
  24 */
  25static unsigned char xfs_dir3_filetype_table[] = {
  26        DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK,
  27        DT_FIFO, DT_SOCK, DT_LNK, DT_WHT,
  28};
  29
  30unsigned char
  31xfs_dir3_get_dtype(
  32        struct xfs_mount        *mp,
  33        uint8_t                 filetype)
  34{
  35        if (!xfs_sb_version_hasftype(&mp->m_sb))
  36                return DT_UNKNOWN;
  37
  38        if (filetype >= XFS_DIR3_FT_MAX)
  39                return DT_UNKNOWN;
  40
  41        return xfs_dir3_filetype_table[filetype];
  42}
  43
  44STATIC int
  45xfs_dir2_sf_getdents(
  46        struct xfs_da_args      *args,
  47        struct dir_context      *ctx)
  48{
  49        int                     i;              /* shortform entry number */
  50        struct xfs_inode        *dp = args->dp; /* incore directory inode */
  51        struct xfs_mount        *mp = dp->i_mount;
  52        xfs_dir2_dataptr_t      off;            /* current entry's offset */
  53        xfs_dir2_sf_entry_t     *sfep;          /* shortform directory entry */
  54        xfs_dir2_sf_hdr_t       *sfp;           /* shortform structure */
  55        xfs_dir2_dataptr_t      dot_offset;
  56        xfs_dir2_dataptr_t      dotdot_offset;
  57        xfs_ino_t               ino;
  58        struct xfs_da_geometry  *geo = args->geo;
  59
  60        ASSERT(dp->i_df.if_flags & XFS_IFINLINE);
  61        ASSERT(dp->i_df.if_bytes == dp->i_d.di_size);
  62        ASSERT(dp->i_df.if_u1.if_data != NULL);
  63
  64        sfp = (xfs_dir2_sf_hdr_t *)dp->i_df.if_u1.if_data;
  65
  66        /*
  67         * If the block number in the offset is out of range, we're done.
  68         */
  69        if (xfs_dir2_dataptr_to_db(geo, ctx->pos) > geo->datablk)
  70                return 0;
  71
  72        /*
  73         * Precalculate offsets for "." and ".." as we will always need them.
  74         * This relies on the fact that directories always start with the
  75         * entries for "." and "..".
  76         */
  77        dot_offset = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
  78                        geo->data_entry_offset);
  79        dotdot_offset = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
  80                        geo->data_entry_offset +
  81                        xfs_dir2_data_entsize(mp, sizeof(".") - 1));
  82
  83        /*
  84         * Put . entry unless we're starting past it.
  85         */
  86        if (ctx->pos <= dot_offset) {
  87                ctx->pos = dot_offset & 0x7fffffff;
  88                if (!dir_emit(ctx, ".", 1, dp->i_ino, DT_DIR))
  89                        return 0;
  90        }
  91
  92        /*
  93         * Put .. entry unless we're starting past it.
  94         */
  95        if (ctx->pos <= dotdot_offset) {
  96                ino = xfs_dir2_sf_get_parent_ino(sfp);
  97                ctx->pos = dotdot_offset & 0x7fffffff;
  98                if (!dir_emit(ctx, "..", 2, ino, DT_DIR))
  99                        return 0;
 100        }
 101
 102        /*
 103         * Loop while there are more entries and put'ing works.
 104         */
 105        sfep = xfs_dir2_sf_firstentry(sfp);
 106        for (i = 0; i < sfp->count; i++) {
 107                uint8_t filetype;
 108
 109                off = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
 110                                xfs_dir2_sf_get_offset(sfep));
 111
 112                if (ctx->pos > off) {
 113                        sfep = xfs_dir2_sf_nextentry(mp, sfp, sfep);
 114                        continue;
 115                }
 116
 117                ino = xfs_dir2_sf_get_ino(mp, sfp, sfep);
 118                filetype = xfs_dir2_sf_get_ftype(mp, sfep);
 119                ctx->pos = off & 0x7fffffff;
 120                if (XFS_IS_CORRUPT(dp->i_mount,
 121                                   !xfs_dir2_namecheck(sfep->name,
 122                                                       sfep->namelen)))
 123                        return -EFSCORRUPTED;
 124                if (!dir_emit(ctx, (char *)sfep->name, sfep->namelen, ino,
 125                            xfs_dir3_get_dtype(mp, filetype)))
 126                        return 0;
 127                sfep = xfs_dir2_sf_nextentry(mp, sfp, sfep);
 128        }
 129
 130        ctx->pos = xfs_dir2_db_off_to_dataptr(geo, geo->datablk + 1, 0) &
 131                                                                0x7fffffff;
 132        return 0;
 133}
 134
 135/*
 136 * Readdir for block directories.
 137 */
 138STATIC int
 139xfs_dir2_block_getdents(
 140        struct xfs_da_args      *args,
 141        struct dir_context      *ctx)
 142{
 143        struct xfs_inode        *dp = args->dp; /* incore directory inode */
 144        struct xfs_buf          *bp;            /* buffer for block */
 145        int                     error;          /* error return value */
 146        int                     wantoff;        /* starting block offset */
 147        xfs_off_t               cook;
 148        struct xfs_da_geometry  *geo = args->geo;
 149        int                     lock_mode;
 150        unsigned int            offset;
 151        unsigned int            end;
 152
 153        /*
 154         * If the block number in the offset is out of range, we're done.
 155         */
 156        if (xfs_dir2_dataptr_to_db(geo, ctx->pos) > geo->datablk)
 157                return 0;
 158
 159        lock_mode = xfs_ilock_data_map_shared(dp);
 160        error = xfs_dir3_block_read(args->trans, dp, &bp);
 161        xfs_iunlock(dp, lock_mode);
 162        if (error)
 163                return error;
 164
 165        /*
 166         * Extract the byte offset we start at from the seek pointer.
 167         * We'll skip entries before this.
 168         */
 169        wantoff = xfs_dir2_dataptr_to_off(geo, ctx->pos);
 170        xfs_dir3_data_check(dp, bp);
 171
 172        /*
 173         * Loop over the data portion of the block.
 174         * Each object is a real entry (dep) or an unused one (dup).
 175         */
 176        offset = geo->data_entry_offset;
 177        end = xfs_dir3_data_end_offset(geo, bp->b_addr);
 178        while (offset < end) {
 179                struct xfs_dir2_data_unused     *dup = bp->b_addr + offset;
 180                struct xfs_dir2_data_entry      *dep = bp->b_addr + offset;
 181                uint8_t filetype;
 182
 183                /*
 184                 * Unused, skip it.
 185                 */
 186                if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 187                        offset += be16_to_cpu(dup->length);
 188                        continue;
 189                }
 190
 191                /*
 192                 * Bump pointer for the next iteration.
 193                 */
 194                offset += xfs_dir2_data_entsize(dp->i_mount, dep->namelen);
 195
 196                /*
 197                 * The entry is before the desired starting point, skip it.
 198                 */
 199                if (offset < wantoff)
 200                        continue;
 201
 202                cook = xfs_dir2_db_off_to_dataptr(geo, geo->datablk, offset);
 203
 204                ctx->pos = cook & 0x7fffffff;
 205                filetype = xfs_dir2_data_get_ftype(dp->i_mount, dep);
 206                /*
 207                 * If it didn't fit, set the final offset to here & return.
 208                 */
 209                if (XFS_IS_CORRUPT(dp->i_mount,
 210                                   !xfs_dir2_namecheck(dep->name,
 211                                                       dep->namelen))) {
 212                        error = -EFSCORRUPTED;
 213                        goto out_rele;
 214                }
 215                if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
 216                            be64_to_cpu(dep->inumber),
 217                            xfs_dir3_get_dtype(dp->i_mount, filetype)))
 218                        goto out_rele;
 219        }
 220
 221        /*
 222         * Reached the end of the block.
 223         * Set the offset to a non-existent block 1 and return.
 224         */
 225        ctx->pos = xfs_dir2_db_off_to_dataptr(geo, geo->datablk + 1, 0) &
 226                                                                0x7fffffff;
 227out_rele:
 228        xfs_trans_brelse(args->trans, bp);
 229        return error;
 230}
 231
 232/*
 233 * Read a directory block and initiate readahead for blocks beyond that.
 234 * We maintain a sliding readahead window of the remaining space in the
 235 * buffer rounded up to the nearest block.
 236 */
 237STATIC int
 238xfs_dir2_leaf_readbuf(
 239        struct xfs_da_args      *args,
 240        size_t                  bufsize,
 241        xfs_dir2_off_t          *cur_off,
 242        xfs_dablk_t             *ra_blk,
 243        struct xfs_buf          **bpp)
 244{
 245        struct xfs_inode        *dp = args->dp;
 246        struct xfs_buf          *bp = NULL;
 247        struct xfs_da_geometry  *geo = args->geo;
 248        struct xfs_ifork        *ifp = XFS_IFORK_PTR(dp, XFS_DATA_FORK);
 249        struct xfs_bmbt_irec    map;
 250        struct blk_plug         plug;
 251        xfs_dir2_off_t          new_off;
 252        xfs_dablk_t             next_ra;
 253        xfs_dablk_t             map_off;
 254        xfs_dablk_t             last_da;
 255        struct xfs_iext_cursor  icur;
 256        int                     ra_want;
 257        int                     error = 0;
 258
 259        if (!(ifp->if_flags & XFS_IFEXTENTS)) {
 260                error = xfs_iread_extents(args->trans, dp, XFS_DATA_FORK);
 261                if (error)
 262                        goto out;
 263        }
 264
 265        /*
 266         * Look for mapped directory blocks at or above the current offset.
 267         * Truncate down to the nearest directory block to start the scanning
 268         * operation.
 269         */
 270        last_da = xfs_dir2_byte_to_da(geo, XFS_DIR2_LEAF_OFFSET);
 271        map_off = xfs_dir2_db_to_da(geo, xfs_dir2_byte_to_db(geo, *cur_off));
 272        if (!xfs_iext_lookup_extent(dp, ifp, map_off, &icur, &map))
 273                goto out;
 274        if (map.br_startoff >= last_da)
 275                goto out;
 276        xfs_trim_extent(&map, map_off, last_da - map_off);
 277
 278        /* Read the directory block of that first mapping. */
 279        new_off = xfs_dir2_da_to_byte(geo, map.br_startoff);
 280        if (new_off > *cur_off)
 281                *cur_off = new_off;
 282        error = xfs_dir3_data_read(args->trans, dp, map.br_startoff, 0, &bp);
 283        if (error)
 284                goto out;
 285
 286        /*
 287         * Start readahead for the next bufsize's worth of dir data blocks.
 288         * We may have already issued readahead for some of that range;
 289         * ra_blk tracks the last block we tried to read(ahead).
 290         */
 291        ra_want = howmany(bufsize + geo->blksize, (1 << geo->fsblog));
 292        if (*ra_blk >= last_da)
 293                goto out;
 294        else if (*ra_blk == 0)
 295                *ra_blk = map.br_startoff;
 296        next_ra = map.br_startoff + geo->fsbcount;
 297        if (next_ra >= last_da)
 298                goto out_no_ra;
 299        if (map.br_blockcount < geo->fsbcount &&
 300            !xfs_iext_next_extent(ifp, &icur, &map))
 301                goto out_no_ra;
 302        if (map.br_startoff >= last_da)
 303                goto out_no_ra;
 304        xfs_trim_extent(&map, next_ra, last_da - next_ra);
 305
 306        /* Start ra for each dir (not fs) block that has a mapping. */
 307        blk_start_plug(&plug);
 308        while (ra_want > 0) {
 309                next_ra = roundup((xfs_dablk_t)map.br_startoff, geo->fsbcount);
 310                while (ra_want > 0 &&
 311                       next_ra < map.br_startoff + map.br_blockcount) {
 312                        if (next_ra >= last_da) {
 313                                *ra_blk = last_da;
 314                                break;
 315                        }
 316                        if (next_ra > *ra_blk) {
 317                                xfs_dir3_data_readahead(dp, next_ra,
 318                                                        XFS_DABUF_MAP_HOLE_OK);
 319                                *ra_blk = next_ra;
 320                        }
 321                        ra_want -= geo->fsbcount;
 322                        next_ra += geo->fsbcount;
 323                }
 324                if (!xfs_iext_next_extent(ifp, &icur, &map)) {
 325                        *ra_blk = last_da;
 326                        break;
 327                }
 328        }
 329        blk_finish_plug(&plug);
 330
 331out:
 332        *bpp = bp;
 333        return error;
 334out_no_ra:
 335        *ra_blk = last_da;
 336        goto out;
 337}
 338
 339/*
 340 * Getdents (readdir) for leaf and node directories.
 341 * This reads the data blocks only, so is the same for both forms.
 342 */
 343STATIC int
 344xfs_dir2_leaf_getdents(
 345        struct xfs_da_args      *args,
 346        struct dir_context      *ctx,
 347        size_t                  bufsize)
 348{
 349        struct xfs_inode        *dp = args->dp;
 350        struct xfs_mount        *mp = dp->i_mount;
 351        struct xfs_buf          *bp = NULL;     /* data block buffer */
 352        xfs_dir2_data_entry_t   *dep;           /* data entry */
 353        xfs_dir2_data_unused_t  *dup;           /* unused entry */
 354        struct xfs_da_geometry  *geo = args->geo;
 355        xfs_dablk_t             rablk = 0;      /* current readahead block */
 356        xfs_dir2_off_t          curoff;         /* current overall offset */
 357        int                     length;         /* temporary length value */
 358        int                     byteoff;        /* offset in current block */
 359        int                     lock_mode;
 360        unsigned int            offset = 0;
 361        int                     error = 0;      /* error return value */
 362
 363        /*
 364         * If the offset is at or past the largest allowed value,
 365         * give up right away.
 366         */
 367        if (ctx->pos >= XFS_DIR2_MAX_DATAPTR)
 368                return 0;
 369
 370        /*
 371         * Inside the loop we keep the main offset value as a byte offset
 372         * in the directory file.
 373         */
 374        curoff = xfs_dir2_dataptr_to_byte(ctx->pos);
 375
 376        /*
 377         * Loop over directory entries until we reach the end offset.
 378         * Get more blocks and readahead as necessary.
 379         */
 380        while (curoff < XFS_DIR2_LEAF_OFFSET) {
 381                uint8_t filetype;
 382
 383                /*
 384                 * If we have no buffer, or we're off the end of the
 385                 * current buffer, need to get another one.
 386                 */
 387                if (!bp || offset >= geo->blksize) {
 388                        if (bp) {
 389                                xfs_trans_brelse(args->trans, bp);
 390                                bp = NULL;
 391                        }
 392
 393                        lock_mode = xfs_ilock_data_map_shared(dp);
 394                        error = xfs_dir2_leaf_readbuf(args, bufsize, &curoff,
 395                                        &rablk, &bp);
 396                        xfs_iunlock(dp, lock_mode);
 397                        if (error || !bp)
 398                                break;
 399
 400                        xfs_dir3_data_check(dp, bp);
 401                        /*
 402                         * Find our position in the block.
 403                         */
 404                        offset = geo->data_entry_offset;
 405                        byteoff = xfs_dir2_byte_to_off(geo, curoff);
 406                        /*
 407                         * Skip past the header.
 408                         */
 409                        if (byteoff == 0)
 410                                curoff += geo->data_entry_offset;
 411                        /*
 412                         * Skip past entries until we reach our offset.
 413                         */
 414                        else {
 415                                while (offset < byteoff) {
 416                                        dup = bp->b_addr + offset;
 417
 418                                        if (be16_to_cpu(dup->freetag)
 419                                                  == XFS_DIR2_DATA_FREE_TAG) {
 420
 421                                                length = be16_to_cpu(dup->length);
 422                                                offset += length;
 423                                                continue;
 424                                        }
 425                                        dep = bp->b_addr + offset;
 426                                        length = xfs_dir2_data_entsize(mp,
 427                                                        dep->namelen);
 428                                        offset += length;
 429                                }
 430                                /*
 431                                 * Now set our real offset.
 432                                 */
 433                                curoff =
 434                                        xfs_dir2_db_off_to_byte(geo,
 435                                            xfs_dir2_byte_to_db(geo, curoff),
 436                                            offset);
 437                                if (offset >= geo->blksize)
 438                                        continue;
 439                        }
 440                }
 441
 442                /*
 443                 * We have a pointer to an entry.  Is it a live one?
 444                 */
 445                dup = bp->b_addr + offset;
 446
 447                /*
 448                 * No, it's unused, skip over it.
 449                 */
 450                if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
 451                        length = be16_to_cpu(dup->length);
 452                        offset += length;
 453                        curoff += length;
 454                        continue;
 455                }
 456
 457                dep = bp->b_addr + offset;
 458                length = xfs_dir2_data_entsize(mp, dep->namelen);
 459                filetype = xfs_dir2_data_get_ftype(mp, dep);
 460
 461                ctx->pos = xfs_dir2_byte_to_dataptr(curoff) & 0x7fffffff;
 462                if (XFS_IS_CORRUPT(dp->i_mount,
 463                                   !xfs_dir2_namecheck(dep->name,
 464                                                       dep->namelen))) {
 465                        error = -EFSCORRUPTED;
 466                        break;
 467                }
 468                if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
 469                            be64_to_cpu(dep->inumber),
 470                            xfs_dir3_get_dtype(dp->i_mount, filetype)))
 471                        break;
 472
 473                /*
 474                 * Advance to next entry in the block.
 475                 */
 476                offset += length;
 477                curoff += length;
 478                /* bufsize may have just been a guess; don't go negative */
 479                bufsize = bufsize > length ? bufsize - length : 0;
 480        }
 481
 482        /*
 483         * All done.  Set output offset value to current offset.
 484         */
 485        if (curoff > xfs_dir2_dataptr_to_byte(XFS_DIR2_MAX_DATAPTR))
 486                ctx->pos = XFS_DIR2_MAX_DATAPTR & 0x7fffffff;
 487        else
 488                ctx->pos = xfs_dir2_byte_to_dataptr(curoff) & 0x7fffffff;
 489        if (bp)
 490                xfs_trans_brelse(args->trans, bp);
 491        return error;
 492}
 493
 494/*
 495 * Read a directory.
 496 *
 497 * If supplied, the transaction collects locked dir buffers to avoid
 498 * nested buffer deadlocks.  This function does not dirty the
 499 * transaction.  The caller should ensure that the inode is locked
 500 * before calling this function.
 501 */
 502int
 503xfs_readdir(
 504        struct xfs_trans        *tp,
 505        struct xfs_inode        *dp,
 506        struct dir_context      *ctx,
 507        size_t                  bufsize)
 508{
 509        struct xfs_da_args      args = { NULL };
 510        int                     rval;
 511        int                     v;
 512
 513        trace_xfs_readdir(dp);
 514
 515        if (XFS_FORCED_SHUTDOWN(dp->i_mount))
 516                return -EIO;
 517
 518        ASSERT(S_ISDIR(VFS_I(dp)->i_mode));
 519        XFS_STATS_INC(dp->i_mount, xs_dir_getdents);
 520
 521        args.dp = dp;
 522        args.geo = dp->i_mount->m_dir_geo;
 523        args.trans = tp;
 524
 525        if (dp->i_d.di_format == XFS_DINODE_FMT_LOCAL)
 526                rval = xfs_dir2_sf_getdents(&args, ctx);
 527        else if ((rval = xfs_dir2_isblock(&args, &v)))
 528                ;
 529        else if (v)
 530                rval = xfs_dir2_block_getdents(&args, ctx);
 531        else
 532                rval = xfs_dir2_leaf_getdents(&args, ctx, bufsize);
 533
 534        return rval;
 535}
 536