2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * Copyright (c) 2013 Red Hat, Inc.
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
8 * published by the Free Software Foundation.
10 * This program is distributed in the hope that it would be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
27 #include "xfs_mount.h"
28 #include "xfs_da_format.h"
29 #include "xfs_da_btree.h"
30 #include "xfs_inode.h"
32 #include "xfs_dir2_priv.h"
33 #include "xfs_error.h"
34 #include "xfs_trace.h"
36 #include "xfs_trans.h"
37 #include "xfs_dinode.h"
40 * Directory file type support functions
42 static unsigned char xfs_dir3_filetype_table[] = {
43 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK,
44 DT_FIFO, DT_SOCK, DT_LNK, DT_WHT,
52 if (!xfs_sb_version_hasftype(&mp->m_sb))
55 if (filetype >= XFS_DIR3_FT_MAX)
58 return xfs_dir3_filetype_table[filetype];
61 * @mode, if set, indicates that the type field needs to be set up.
62 * This uses the transformation from file mode to DT_* as defined in linux/fs.h
63 * for file type specification. This will be propagated into the directory
64 * structure if appropriate for the given operation and filesystem config.
66 const unsigned char xfs_mode_to_ftype[S_IFMT >> S_SHIFT] = {
67 [0] = XFS_DIR3_FT_UNKNOWN,
68 [S_IFREG >> S_SHIFT] = XFS_DIR3_FT_REG_FILE,
69 [S_IFDIR >> S_SHIFT] = XFS_DIR3_FT_DIR,
70 [S_IFCHR >> S_SHIFT] = XFS_DIR3_FT_CHRDEV,
71 [S_IFBLK >> S_SHIFT] = XFS_DIR3_FT_BLKDEV,
72 [S_IFIFO >> S_SHIFT] = XFS_DIR3_FT_FIFO,
73 [S_IFSOCK >> S_SHIFT] = XFS_DIR3_FT_SOCK,
74 [S_IFLNK >> S_SHIFT] = XFS_DIR3_FT_SYMLINK,
79 xfs_inode_t *dp, /* incore directory inode */
80 struct dir_context *ctx)
82 int i; /* shortform entry number */
83 xfs_mount_t *mp; /* filesystem mount point */
84 xfs_dir2_dataptr_t off; /* current entry's offset */
85 xfs_dir2_sf_entry_t *sfep; /* shortform directory entry */
86 xfs_dir2_sf_hdr_t *sfp; /* shortform structure */
87 xfs_dir2_dataptr_t dot_offset;
88 xfs_dir2_dataptr_t dotdot_offset;
93 ASSERT(dp->i_df.if_flags & XFS_IFINLINE);
95 * Give up if the directory is way too short.
97 if (dp->i_d.di_size < offsetof(xfs_dir2_sf_hdr_t, parent)) {
98 ASSERT(XFS_FORCED_SHUTDOWN(mp));
99 return XFS_ERROR(EIO);
102 ASSERT(dp->i_df.if_bytes == dp->i_d.di_size);
103 ASSERT(dp->i_df.if_u1.if_data != NULL);
105 sfp = (xfs_dir2_sf_hdr_t *)dp->i_df.if_u1.if_data;
107 ASSERT(dp->i_d.di_size >= xfs_dir2_sf_hdr_size(sfp->i8count));
110 * If the block number in the offset is out of range, we're done.
112 if (xfs_dir2_dataptr_to_db(mp, ctx->pos) > mp->m_dirdatablk)
116 * Precalculate offsets for . and .. as we will always need them.
118 * XXX(hch): the second argument is sometimes 0 and sometimes
121 dot_offset = xfs_dir2_db_off_to_dataptr(mp, mp->m_dirdatablk,
122 dp->d_ops->data_dot_offset);
123 dotdot_offset = xfs_dir2_db_off_to_dataptr(mp, mp->m_dirdatablk,
124 dp->d_ops->data_dotdot_offset);
127 * Put . entry unless we're starting past it.
129 if (ctx->pos <= dot_offset) {
130 ctx->pos = dot_offset & 0x7fffffff;
131 if (!dir_emit(ctx, ".", 1, dp->i_ino, DT_DIR))
136 * Put .. entry unless we're starting past it.
138 if (ctx->pos <= dotdot_offset) {
139 ino = dp->d_ops->sf_get_parent_ino(sfp);
140 ctx->pos = dotdot_offset & 0x7fffffff;
141 if (!dir_emit(ctx, "..", 2, ino, DT_DIR))
146 * Loop while there are more entries and put'ing works.
148 sfep = xfs_dir2_sf_firstentry(sfp);
149 for (i = 0; i < sfp->count; i++) {
152 off = xfs_dir2_db_off_to_dataptr(mp, mp->m_dirdatablk,
153 xfs_dir2_sf_get_offset(sfep));
155 if (ctx->pos > off) {
156 sfep = dp->d_ops->sf_nextentry(sfp, sfep);
160 ino = dp->d_ops->sf_get_ino(sfp, sfep);
161 filetype = dp->d_ops->sf_get_ftype(sfep);
162 ctx->pos = off & 0x7fffffff;
163 if (!dir_emit(ctx, (char *)sfep->name, sfep->namelen, ino,
164 xfs_dir3_get_dtype(mp, filetype)))
166 sfep = dp->d_ops->sf_nextentry(sfp, sfep);
169 ctx->pos = xfs_dir2_db_off_to_dataptr(mp, mp->m_dirdatablk + 1, 0) &
175 * Readdir for block directories.
178 xfs_dir2_block_getdents(
179 xfs_inode_t *dp, /* incore inode */
180 struct dir_context *ctx)
182 xfs_dir2_data_hdr_t *hdr; /* block header */
183 struct xfs_buf *bp; /* buffer for block */
184 xfs_dir2_block_tail_t *btp; /* block tail */
185 xfs_dir2_data_entry_t *dep; /* block data entry */
186 xfs_dir2_data_unused_t *dup; /* block unused entry */
187 char *endptr; /* end of the data entries */
188 int error; /* error return value */
189 xfs_mount_t *mp; /* filesystem mount point */
190 char *ptr; /* current data entry */
191 int wantoff; /* starting block offset */
196 * If the block number in the offset is out of range, we're done.
198 if (xfs_dir2_dataptr_to_db(mp, ctx->pos) > mp->m_dirdatablk)
201 error = xfs_dir3_block_read(NULL, dp, &bp);
206 * Extract the byte offset we start at from the seek pointer.
207 * We'll skip entries before this.
209 wantoff = xfs_dir2_dataptr_to_off(mp, ctx->pos);
211 xfs_dir3_data_check(dp, bp);
213 * Set up values for the loop.
215 btp = xfs_dir2_block_tail_p(mp, hdr);
216 ptr = (char *)dp->d_ops->data_entry_p(hdr);
217 endptr = (char *)xfs_dir2_block_leaf_p(btp);
220 * Loop over the data portion of the block.
221 * Each object is a real entry (dep) or an unused one (dup).
223 while (ptr < endptr) {
226 dup = (xfs_dir2_data_unused_t *)ptr;
230 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
231 ptr += be16_to_cpu(dup->length);
235 dep = (xfs_dir2_data_entry_t *)ptr;
238 * Bump pointer for the next iteration.
240 ptr += dp->d_ops->data_entsize(dep->namelen);
242 * The entry is before the desired starting point, skip it.
244 if ((char *)dep - (char *)hdr < wantoff)
247 cook = xfs_dir2_db_off_to_dataptr(mp, mp->m_dirdatablk,
248 (char *)dep - (char *)hdr);
250 ctx->pos = cook & 0x7fffffff;
251 filetype = dp->d_ops->data_get_ftype(dep);
253 * If it didn't fit, set the final offset to here & return.
255 if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
256 be64_to_cpu(dep->inumber),
257 xfs_dir3_get_dtype(mp, filetype))) {
258 xfs_trans_brelse(NULL, bp);
264 * Reached the end of the block.
265 * Set the offset to a non-existent block 1 and return.
267 ctx->pos = xfs_dir2_db_off_to_dataptr(mp, mp->m_dirdatablk + 1, 0) &
269 xfs_trans_brelse(NULL, bp);
273 struct xfs_dir2_leaf_map_info {
274 xfs_extlen_t map_blocks; /* number of fsbs in map */
275 xfs_dablk_t map_off; /* last mapped file offset */
276 int map_size; /* total entries in *map */
277 int map_valid; /* valid entries in *map */
278 int nmap; /* mappings to ask xfs_bmapi */
279 xfs_dir2_db_t curdb; /* db for current block */
280 int ra_current; /* number of read-ahead blks */
281 int ra_index; /* *map index for read-ahead */
282 int ra_offset; /* map entry offset for ra */
283 int ra_want; /* readahead count wanted */
284 struct xfs_bmbt_irec map[]; /* map vector for blocks */
288 xfs_dir2_leaf_readbuf(
289 struct xfs_inode *dp,
291 struct xfs_dir2_leaf_map_info *mip,
292 xfs_dir2_off_t *curoff,
293 struct xfs_buf **bpp)
295 struct xfs_mount *mp = dp->i_mount;
296 struct xfs_buf *bp = *bpp;
297 struct xfs_bmbt_irec *map = mip->map;
298 struct blk_plug plug;
305 * If we have a buffer, we need to release it and
306 * take it out of the mapping.
310 xfs_trans_brelse(NULL, bp);
312 mip->map_blocks -= mp->m_dirblkfsbs;
314 * Loop to get rid of the extents for the
317 for (i = mp->m_dirblkfsbs; i > 0; ) {
318 j = min_t(int, map->br_blockcount, i);
319 map->br_blockcount -= j;
320 map->br_startblock += j;
321 map->br_startoff += j;
323 * If mapping is done, pitch it from
326 if (!map->br_blockcount && --mip->map_valid)
327 memmove(&map[0], &map[1],
328 sizeof(map[0]) * mip->map_valid);
334 * Recalculate the readahead blocks wanted.
336 mip->ra_want = howmany(bufsize + mp->m_dirblksize,
337 mp->m_sb.sb_blocksize) - 1;
338 ASSERT(mip->ra_want >= 0);
341 * If we don't have as many as we want, and we haven't
342 * run out of data blocks, get some more mappings.
344 if (1 + mip->ra_want > mip->map_blocks &&
345 mip->map_off < xfs_dir2_byte_to_da(mp, XFS_DIR2_LEAF_OFFSET)) {
347 * Get more bmaps, fill in after the ones
348 * we already have in the table.
350 mip->nmap = mip->map_size - mip->map_valid;
351 error = xfs_bmapi_read(dp, mip->map_off,
352 xfs_dir2_byte_to_da(mp, XFS_DIR2_LEAF_OFFSET) -
354 &map[mip->map_valid], &mip->nmap, 0);
357 * Don't know if we should ignore this or try to return an
358 * error. The trouble with returning errors is that readdir
359 * will just stop without actually passing the error through.
365 * If we got all the mappings we asked for, set the final map
366 * offset based on the last bmap value received. Otherwise,
367 * we've reached the end.
369 if (mip->nmap == mip->map_size - mip->map_valid) {
370 i = mip->map_valid + mip->nmap - 1;
371 mip->map_off = map[i].br_startoff + map[i].br_blockcount;
373 mip->map_off = xfs_dir2_byte_to_da(mp,
374 XFS_DIR2_LEAF_OFFSET);
377 * Look for holes in the mapping, and eliminate them. Count up
380 for (i = mip->map_valid; i < mip->map_valid + mip->nmap; ) {
381 if (map[i].br_startblock == HOLESTARTBLOCK) {
383 length = mip->map_valid + mip->nmap - i;
385 memmove(&map[i], &map[i + 1],
386 sizeof(map[i]) * length);
388 mip->map_blocks += map[i].br_blockcount;
392 mip->map_valid += mip->nmap;
396 * No valid mappings, so no more data blocks.
398 if (!mip->map_valid) {
399 *curoff = xfs_dir2_da_to_byte(mp, mip->map_off);
404 * Read the directory block starting at the first mapping.
406 mip->curdb = xfs_dir2_da_to_db(mp, map->br_startoff);
407 error = xfs_dir3_data_read(NULL, dp, map->br_startoff,
408 map->br_blockcount >= mp->m_dirblkfsbs ?
409 XFS_FSB_TO_DADDR(mp, map->br_startblock) : -1, &bp);
412 * Should just skip over the data block instead of giving up.
418 * Adjust the current amount of read-ahead: we just read a block that
422 mip->ra_current -= mp->m_dirblkfsbs;
425 * Do we need more readahead?
427 blk_start_plug(&plug);
428 for (mip->ra_index = mip->ra_offset = i = 0;
429 mip->ra_want > mip->ra_current && i < mip->map_blocks;
430 i += mp->m_dirblkfsbs) {
431 ASSERT(mip->ra_index < mip->map_valid);
433 * Read-ahead a contiguous directory block.
435 if (i > mip->ra_current &&
436 map[mip->ra_index].br_blockcount >= mp->m_dirblkfsbs) {
437 xfs_dir3_data_readahead(NULL, dp,
438 map[mip->ra_index].br_startoff + mip->ra_offset,
440 map[mip->ra_index].br_startblock +
446 * Read-ahead a non-contiguous directory block. This doesn't
447 * use our mapping, but this is a very rare case.
449 else if (i > mip->ra_current) {
450 xfs_dir3_data_readahead(NULL, dp,
451 map[mip->ra_index].br_startoff +
457 * Advance offset through the mapping table.
459 for (j = 0; j < mp->m_dirblkfsbs; j++) {
461 * The rest of this extent but not more than a dir
464 length = min_t(int, mp->m_dirblkfsbs,
465 map[mip->ra_index].br_blockcount -
468 mip->ra_offset += length;
471 * Advance to the next mapping if this one is used up.
473 if (mip->ra_offset == map[mip->ra_index].br_blockcount) {
479 blk_finish_plug(&plug);
487 * Getdents (readdir) for leaf and node directories.
488 * This reads the data blocks only, so is the same for both forms.
491 xfs_dir2_leaf_getdents(
492 xfs_inode_t *dp, /* incore directory inode */
493 struct dir_context *ctx,
496 struct xfs_buf *bp = NULL; /* data block buffer */
497 xfs_dir2_data_hdr_t *hdr; /* data block header */
498 xfs_dir2_data_entry_t *dep; /* data entry */
499 xfs_dir2_data_unused_t *dup; /* unused entry */
500 int error = 0; /* error return value */
501 int length; /* temporary length value */
502 xfs_mount_t *mp; /* filesystem mount point */
503 int byteoff; /* offset in current block */
504 xfs_dir2_off_t curoff; /* current overall offset */
505 xfs_dir2_off_t newoff; /* new curoff after new blk */
506 char *ptr = NULL; /* pointer to current data */
507 struct xfs_dir2_leaf_map_info *map_info;
510 * If the offset is at or past the largest allowed value,
511 * give up right away.
513 if (ctx->pos >= XFS_DIR2_MAX_DATAPTR)
519 * Set up to bmap a number of blocks based on the caller's
520 * buffer size, the directory block size, and the filesystem
523 length = howmany(bufsize + mp->m_dirblksize,
524 mp->m_sb.sb_blocksize);
525 map_info = kmem_zalloc(offsetof(struct xfs_dir2_leaf_map_info, map) +
526 (length * sizeof(struct xfs_bmbt_irec)),
528 map_info->map_size = length;
531 * Inside the loop we keep the main offset value as a byte offset
532 * in the directory file.
534 curoff = xfs_dir2_dataptr_to_byte(mp, ctx->pos);
537 * Force this conversion through db so we truncate the offset
538 * down to get the start of the data block.
540 map_info->map_off = xfs_dir2_db_to_da(mp,
541 xfs_dir2_byte_to_db(mp, curoff));
544 * Loop over directory entries until we reach the end offset.
545 * Get more blocks and readahead as necessary.
547 while (curoff < XFS_DIR2_LEAF_OFFSET) {
551 * If we have no buffer, or we're off the end of the
552 * current buffer, need to get another one.
554 if (!bp || ptr >= (char *)bp->b_addr + mp->m_dirblksize) {
556 error = xfs_dir2_leaf_readbuf(dp, bufsize, map_info,
558 if (error || !map_info->map_valid)
562 * Having done a read, we need to set a new offset.
564 newoff = xfs_dir2_db_off_to_byte(mp, map_info->curdb, 0);
566 * Start of the current block.
571 * Make sure we're in the right block.
573 else if (curoff > newoff)
574 ASSERT(xfs_dir2_byte_to_db(mp, curoff) ==
577 xfs_dir3_data_check(dp, bp);
579 * Find our position in the block.
581 ptr = (char *)dp->d_ops->data_entry_p(hdr);
582 byteoff = xfs_dir2_byte_to_off(mp, curoff);
584 * Skip past the header.
587 curoff += dp->d_ops->data_entry_offset;
589 * Skip past entries until we reach our offset.
592 while ((char *)ptr - (char *)hdr < byteoff) {
593 dup = (xfs_dir2_data_unused_t *)ptr;
595 if (be16_to_cpu(dup->freetag)
596 == XFS_DIR2_DATA_FREE_TAG) {
598 length = be16_to_cpu(dup->length);
602 dep = (xfs_dir2_data_entry_t *)ptr;
604 dp->d_ops->data_entsize(dep->namelen);
608 * Now set our real offset.
611 xfs_dir2_db_off_to_byte(mp,
612 xfs_dir2_byte_to_db(mp, curoff),
613 (char *)ptr - (char *)hdr);
614 if (ptr >= (char *)hdr + mp->m_dirblksize) {
620 * We have a pointer to an entry.
623 dup = (xfs_dir2_data_unused_t *)ptr;
625 * No, it's unused, skip over it.
627 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
628 length = be16_to_cpu(dup->length);
634 dep = (xfs_dir2_data_entry_t *)ptr;
635 length = dp->d_ops->data_entsize(dep->namelen);
636 filetype = dp->d_ops->data_get_ftype(dep);
638 ctx->pos = xfs_dir2_byte_to_dataptr(mp, curoff) & 0x7fffffff;
639 if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
640 be64_to_cpu(dep->inumber),
641 xfs_dir3_get_dtype(mp, filetype)))
645 * Advance to next entry in the block.
649 /* bufsize may have just been a guess; don't go negative */
650 bufsize = bufsize > length ? bufsize - length : 0;
654 * All done. Set output offset value to current offset.
656 if (curoff > xfs_dir2_dataptr_to_byte(mp, XFS_DIR2_MAX_DATAPTR))
657 ctx->pos = XFS_DIR2_MAX_DATAPTR & 0x7fffffff;
659 ctx->pos = xfs_dir2_byte_to_dataptr(mp, curoff) & 0x7fffffff;
662 xfs_trans_brelse(NULL, bp);
672 struct dir_context *ctx,
675 int rval; /* return value */
676 int v; /* type-checking value */
679 trace_xfs_readdir(dp);
681 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
682 return XFS_ERROR(EIO);
684 ASSERT(S_ISDIR(dp->i_d.di_mode));
685 XFS_STATS_INC(xs_dir_getdents);
687 lock_mode = xfs_ilock_data_map_shared(dp);
688 if (dp->i_d.di_format == XFS_DINODE_FMT_LOCAL)
689 rval = xfs_dir2_sf_getdents(dp, ctx);
690 else if ((rval = xfs_dir2_isblock(NULL, dp, &v)))
693 rval = xfs_dir2_block_getdents(dp, ctx);
695 rval = xfs_dir2_leaf_getdents(dp, ctx, bufsize);
696 xfs_iunlock(dp, lock_mode);