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;
90 struct xfs_da_geometry *geo;
95 ASSERT(dp->i_df.if_flags & XFS_IFINLINE);
97 * Give up if the directory is way too short.
99 if (dp->i_d.di_size < offsetof(xfs_dir2_sf_hdr_t, parent)) {
100 ASSERT(XFS_FORCED_SHUTDOWN(mp));
101 return XFS_ERROR(EIO);
104 ASSERT(dp->i_df.if_bytes == dp->i_d.di_size);
105 ASSERT(dp->i_df.if_u1.if_data != NULL);
107 sfp = (xfs_dir2_sf_hdr_t *)dp->i_df.if_u1.if_data;
109 ASSERT(dp->i_d.di_size >= xfs_dir2_sf_hdr_size(sfp->i8count));
112 * If the block number in the offset is out of range, we're done.
114 if (xfs_dir2_dataptr_to_db(geo, ctx->pos) > geo->datablk)
118 * Precalculate offsets for . and .. as we will always need them.
120 * XXX(hch): the second argument is sometimes 0 and sometimes
123 dot_offset = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
124 dp->d_ops->data_dot_offset);
125 dotdot_offset = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
126 dp->d_ops->data_dotdot_offset);
129 * Put . entry unless we're starting past it.
131 if (ctx->pos <= dot_offset) {
132 ctx->pos = dot_offset & 0x7fffffff;
133 if (!dir_emit(ctx, ".", 1, dp->i_ino, DT_DIR))
138 * Put .. entry unless we're starting past it.
140 if (ctx->pos <= dotdot_offset) {
141 ino = dp->d_ops->sf_get_parent_ino(sfp);
142 ctx->pos = dotdot_offset & 0x7fffffff;
143 if (!dir_emit(ctx, "..", 2, ino, DT_DIR))
148 * Loop while there are more entries and put'ing works.
150 sfep = xfs_dir2_sf_firstentry(sfp);
151 for (i = 0; i < sfp->count; i++) {
154 off = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
155 xfs_dir2_sf_get_offset(sfep));
157 if (ctx->pos > off) {
158 sfep = dp->d_ops->sf_nextentry(sfp, sfep);
162 ino = dp->d_ops->sf_get_ino(sfp, sfep);
163 filetype = dp->d_ops->sf_get_ftype(sfep);
164 ctx->pos = off & 0x7fffffff;
165 if (!dir_emit(ctx, (char *)sfep->name, sfep->namelen, ino,
166 xfs_dir3_get_dtype(mp, filetype)))
168 sfep = dp->d_ops->sf_nextentry(sfp, sfep);
171 ctx->pos = xfs_dir2_db_off_to_dataptr(geo, geo->datablk + 1, 0) &
177 * Readdir for block directories.
180 xfs_dir2_block_getdents(
181 xfs_inode_t *dp, /* incore inode */
182 struct dir_context *ctx)
184 xfs_dir2_data_hdr_t *hdr; /* block header */
185 struct xfs_buf *bp; /* buffer for block */
186 xfs_dir2_block_tail_t *btp; /* block tail */
187 xfs_dir2_data_entry_t *dep; /* block data entry */
188 xfs_dir2_data_unused_t *dup; /* block unused entry */
189 char *endptr; /* end of the data entries */
190 int error; /* error return value */
191 xfs_mount_t *mp; /* filesystem mount point */
192 char *ptr; /* current data entry */
193 int wantoff; /* starting block offset */
195 struct xfs_da_geometry *geo;
200 * If the block number in the offset is out of range, we're done.
202 if (xfs_dir2_dataptr_to_db(geo, ctx->pos) > geo->datablk)
205 error = xfs_dir3_block_read(NULL, dp, &bp);
210 * Extract the byte offset we start at from the seek pointer.
211 * We'll skip entries before this.
213 wantoff = xfs_dir2_dataptr_to_off(geo, ctx->pos);
215 xfs_dir3_data_check(dp, bp);
217 * Set up values for the loop.
219 btp = xfs_dir2_block_tail_p(geo, hdr);
220 ptr = (char *)dp->d_ops->data_entry_p(hdr);
221 endptr = (char *)xfs_dir2_block_leaf_p(btp);
224 * Loop over the data portion of the block.
225 * Each object is a real entry (dep) or an unused one (dup).
227 while (ptr < endptr) {
230 dup = (xfs_dir2_data_unused_t *)ptr;
234 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
235 ptr += be16_to_cpu(dup->length);
239 dep = (xfs_dir2_data_entry_t *)ptr;
242 * Bump pointer for the next iteration.
244 ptr += dp->d_ops->data_entsize(dep->namelen);
246 * The entry is before the desired starting point, skip it.
248 if ((char *)dep - (char *)hdr < wantoff)
251 cook = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
252 (char *)dep - (char *)hdr);
254 ctx->pos = cook & 0x7fffffff;
255 filetype = dp->d_ops->data_get_ftype(dep);
257 * If it didn't fit, set the final offset to here & return.
259 if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
260 be64_to_cpu(dep->inumber),
261 xfs_dir3_get_dtype(mp, filetype))) {
262 xfs_trans_brelse(NULL, bp);
268 * Reached the end of the block.
269 * Set the offset to a non-existent block 1 and return.
271 ctx->pos = xfs_dir2_db_off_to_dataptr(geo, geo->datablk + 1, 0) &
273 xfs_trans_brelse(NULL, bp);
277 struct xfs_dir2_leaf_map_info {
278 xfs_extlen_t map_blocks; /* number of fsbs in map */
279 xfs_dablk_t map_off; /* last mapped file offset */
280 int map_size; /* total entries in *map */
281 int map_valid; /* valid entries in *map */
282 int nmap; /* mappings to ask xfs_bmapi */
283 xfs_dir2_db_t curdb; /* db for current block */
284 int ra_current; /* number of read-ahead blks */
285 int ra_index; /* *map index for read-ahead */
286 int ra_offset; /* map entry offset for ra */
287 int ra_want; /* readahead count wanted */
288 struct xfs_bmbt_irec map[]; /* map vector for blocks */
292 xfs_dir2_leaf_readbuf(
293 struct xfs_inode *dp,
295 struct xfs_dir2_leaf_map_info *mip,
296 xfs_dir2_off_t *curoff,
297 struct xfs_buf **bpp)
299 struct xfs_mount *mp = dp->i_mount;
300 struct xfs_buf *bp = *bpp;
301 struct xfs_bmbt_irec *map = mip->map;
302 struct blk_plug plug;
307 struct xfs_da_geometry *geo = mp->m_dir_geo;
310 * If we have a buffer, we need to release it and
311 * take it out of the mapping.
315 xfs_trans_brelse(NULL, bp);
317 mip->map_blocks -= geo->fsbcount;
319 * Loop to get rid of the extents for the
322 for (i = geo->fsbcount; i > 0; ) {
323 j = min_t(int, map->br_blockcount, i);
324 map->br_blockcount -= j;
325 map->br_startblock += j;
326 map->br_startoff += j;
328 * If mapping is done, pitch it from
331 if (!map->br_blockcount && --mip->map_valid)
332 memmove(&map[0], &map[1],
333 sizeof(map[0]) * mip->map_valid);
339 * Recalculate the readahead blocks wanted.
341 mip->ra_want = howmany(bufsize + geo->blksize,
342 mp->m_sb.sb_blocksize) - 1;
343 ASSERT(mip->ra_want >= 0);
346 * If we don't have as many as we want, and we haven't
347 * run out of data blocks, get some more mappings.
349 if (1 + mip->ra_want > mip->map_blocks &&
350 mip->map_off < xfs_dir2_byte_to_da(geo, XFS_DIR2_LEAF_OFFSET)) {
352 * Get more bmaps, fill in after the ones
353 * we already have in the table.
355 mip->nmap = mip->map_size - mip->map_valid;
356 error = xfs_bmapi_read(dp, mip->map_off,
357 xfs_dir2_byte_to_da(geo, XFS_DIR2_LEAF_OFFSET) -
359 &map[mip->map_valid], &mip->nmap, 0);
362 * Don't know if we should ignore this or try to return an
363 * error. The trouble with returning errors is that readdir
364 * will just stop without actually passing the error through.
370 * If we got all the mappings we asked for, set the final map
371 * offset based on the last bmap value received. Otherwise,
372 * we've reached the end.
374 if (mip->nmap == mip->map_size - mip->map_valid) {
375 i = mip->map_valid + mip->nmap - 1;
376 mip->map_off = map[i].br_startoff + map[i].br_blockcount;
378 mip->map_off = xfs_dir2_byte_to_da(geo,
379 XFS_DIR2_LEAF_OFFSET);
382 * Look for holes in the mapping, and eliminate them. Count up
385 for (i = mip->map_valid; i < mip->map_valid + mip->nmap; ) {
386 if (map[i].br_startblock == HOLESTARTBLOCK) {
388 length = mip->map_valid + mip->nmap - i;
390 memmove(&map[i], &map[i + 1],
391 sizeof(map[i]) * length);
393 mip->map_blocks += map[i].br_blockcount;
397 mip->map_valid += mip->nmap;
401 * No valid mappings, so no more data blocks.
403 if (!mip->map_valid) {
404 *curoff = xfs_dir2_da_to_byte(geo, mip->map_off);
409 * Read the directory block starting at the first mapping.
411 mip->curdb = xfs_dir2_da_to_db(geo, map->br_startoff);
412 error = xfs_dir3_data_read(NULL, dp, map->br_startoff,
413 map->br_blockcount >= geo->fsbcount ?
414 XFS_FSB_TO_DADDR(mp, map->br_startblock) : -1, &bp);
417 * Should just skip over the data block instead of giving up.
423 * Adjust the current amount of read-ahead: we just read a block that
427 mip->ra_current -= geo->fsbcount;
430 * Do we need more readahead?
432 blk_start_plug(&plug);
433 for (mip->ra_index = mip->ra_offset = i = 0;
434 mip->ra_want > mip->ra_current && i < mip->map_blocks;
435 i += geo->fsbcount) {
436 ASSERT(mip->ra_index < mip->map_valid);
438 * Read-ahead a contiguous directory block.
440 if (i > mip->ra_current &&
441 map[mip->ra_index].br_blockcount >= geo->fsbcount) {
442 xfs_dir3_data_readahead(dp,
443 map[mip->ra_index].br_startoff + mip->ra_offset,
445 map[mip->ra_index].br_startblock +
451 * Read-ahead a non-contiguous directory block. This doesn't
452 * use our mapping, but this is a very rare case.
454 else if (i > mip->ra_current) {
455 xfs_dir3_data_readahead(dp,
456 map[mip->ra_index].br_startoff +
462 * Advance offset through the mapping table.
464 for (j = 0; j < geo->fsbcount; j += length ) {
466 * The rest of this extent but not more than a dir
469 length = min_t(int, geo->fsbcount,
470 map[mip->ra_index].br_blockcount -
472 mip->ra_offset += length;
475 * Advance to the next mapping if this one is used up.
477 if (mip->ra_offset == map[mip->ra_index].br_blockcount) {
483 blk_finish_plug(&plug);
491 * Getdents (readdir) for leaf and node directories.
492 * This reads the data blocks only, so is the same for both forms.
495 xfs_dir2_leaf_getdents(
496 xfs_inode_t *dp, /* incore directory inode */
497 struct dir_context *ctx,
500 struct xfs_buf *bp = NULL; /* data block buffer */
501 xfs_dir2_data_hdr_t *hdr; /* data block header */
502 xfs_dir2_data_entry_t *dep; /* data entry */
503 xfs_dir2_data_unused_t *dup; /* unused entry */
504 int error = 0; /* error return value */
505 int length; /* temporary length value */
506 xfs_mount_t *mp; /* filesystem mount point */
507 int byteoff; /* offset in current block */
508 xfs_dir2_off_t curoff; /* current overall offset */
509 xfs_dir2_off_t newoff; /* new curoff after new blk */
510 char *ptr = NULL; /* pointer to current data */
511 struct xfs_dir2_leaf_map_info *map_info;
512 struct xfs_da_geometry *geo;
515 * If the offset is at or past the largest allowed value,
516 * give up right away.
518 if (ctx->pos >= XFS_DIR2_MAX_DATAPTR)
525 * Set up to bmap a number of blocks based on the caller's
526 * buffer size, the directory block size, and the filesystem
529 length = howmany(bufsize + geo->blksize, mp->m_sb.sb_blocksize);
530 map_info = kmem_zalloc(offsetof(struct xfs_dir2_leaf_map_info, map) +
531 (length * sizeof(struct xfs_bmbt_irec)),
533 map_info->map_size = length;
536 * Inside the loop we keep the main offset value as a byte offset
537 * in the directory file.
539 curoff = xfs_dir2_dataptr_to_byte(ctx->pos);
542 * Force this conversion through db so we truncate the offset
543 * down to get the start of the data block.
545 map_info->map_off = xfs_dir2_db_to_da(geo,
546 xfs_dir2_byte_to_db(geo, curoff));
549 * Loop over directory entries until we reach the end offset.
550 * Get more blocks and readahead as necessary.
552 while (curoff < XFS_DIR2_LEAF_OFFSET) {
556 * If we have no buffer, or we're off the end of the
557 * current buffer, need to get another one.
559 if (!bp || ptr >= (char *)bp->b_addr + geo->blksize) {
561 error = xfs_dir2_leaf_readbuf(dp, bufsize, map_info,
563 if (error || !map_info->map_valid)
567 * Having done a read, we need to set a new offset.
569 newoff = xfs_dir2_db_off_to_byte(mp->m_dir_geo,
572 * Start of the current block.
577 * Make sure we're in the right block.
579 else if (curoff > newoff)
580 ASSERT(xfs_dir2_byte_to_db(geo, curoff) ==
583 xfs_dir3_data_check(dp, bp);
585 * Find our position in the block.
587 ptr = (char *)dp->d_ops->data_entry_p(hdr);
588 byteoff = xfs_dir2_byte_to_off(mp->m_dir_geo, curoff);
590 * Skip past the header.
593 curoff += dp->d_ops->data_entry_offset;
595 * Skip past entries until we reach our offset.
598 while ((char *)ptr - (char *)hdr < byteoff) {
599 dup = (xfs_dir2_data_unused_t *)ptr;
601 if (be16_to_cpu(dup->freetag)
602 == XFS_DIR2_DATA_FREE_TAG) {
604 length = be16_to_cpu(dup->length);
608 dep = (xfs_dir2_data_entry_t *)ptr;
610 dp->d_ops->data_entsize(dep->namelen);
614 * Now set our real offset.
617 xfs_dir2_db_off_to_byte(geo,
618 xfs_dir2_byte_to_db(geo, curoff),
619 (char *)ptr - (char *)hdr);
620 if (ptr >= (char *)hdr + geo->blksize) {
626 * We have a pointer to an entry.
629 dup = (xfs_dir2_data_unused_t *)ptr;
631 * No, it's unused, skip over it.
633 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
634 length = be16_to_cpu(dup->length);
640 dep = (xfs_dir2_data_entry_t *)ptr;
641 length = dp->d_ops->data_entsize(dep->namelen);
642 filetype = dp->d_ops->data_get_ftype(dep);
644 ctx->pos = xfs_dir2_byte_to_dataptr(curoff) & 0x7fffffff;
645 if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
646 be64_to_cpu(dep->inumber),
647 xfs_dir3_get_dtype(mp, filetype)))
651 * Advance to next entry in the block.
655 /* bufsize may have just been a guess; don't go negative */
656 bufsize = bufsize > length ? bufsize - length : 0;
660 * All done. Set output offset value to current offset.
662 if (curoff > xfs_dir2_dataptr_to_byte(XFS_DIR2_MAX_DATAPTR))
663 ctx->pos = XFS_DIR2_MAX_DATAPTR & 0x7fffffff;
665 ctx->pos = xfs_dir2_byte_to_dataptr(curoff) & 0x7fffffff;
668 xfs_trans_brelse(NULL, bp);
678 struct dir_context *ctx,
681 int rval; /* return value */
682 int v; /* type-checking value */
685 trace_xfs_readdir(dp);
687 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
688 return XFS_ERROR(EIO);
690 ASSERT(S_ISDIR(dp->i_d.di_mode));
691 XFS_STATS_INC(xs_dir_getdents);
693 lock_mode = xfs_ilock_data_map_shared(dp);
694 if (dp->i_d.di_format == XFS_DINODE_FMT_LOCAL)
695 rval = xfs_dir2_sf_getdents(dp, ctx);
696 else if ((rval = xfs_dir2_isblock(dp, &v)))
699 rval = xfs_dir2_block_getdents(dp, ctx);
701 rval = xfs_dir2_leaf_getdents(dp, ctx, bufsize);
702 xfs_iunlock(dp, lock_mode);