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 struct xfs_da_args *args,
80 struct dir_context *ctx)
82 int i; /* shortform entry number */
83 struct xfs_inode *dp = args->dp; /* incore directory inode */
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 = args->geo;
92 ASSERT(dp->i_df.if_flags & XFS_IFINLINE);
94 * Give up if the directory is way too short.
96 if (dp->i_d.di_size < offsetof(xfs_dir2_sf_hdr_t, parent)) {
97 ASSERT(XFS_FORCED_SHUTDOWN(dp->i_mount));
101 ASSERT(dp->i_df.if_bytes == dp->i_d.di_size);
102 ASSERT(dp->i_df.if_u1.if_data != NULL);
104 sfp = (xfs_dir2_sf_hdr_t *)dp->i_df.if_u1.if_data;
106 ASSERT(dp->i_d.di_size >= xfs_dir2_sf_hdr_size(sfp->i8count));
109 * If the block number in the offset is out of range, we're done.
111 if (xfs_dir2_dataptr_to_db(geo, ctx->pos) > geo->datablk)
115 * Precalculate offsets for . and .. as we will always need them.
117 * XXX(hch): the second argument is sometimes 0 and sometimes
120 dot_offset = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
121 dp->d_ops->data_dot_offset);
122 dotdot_offset = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
123 dp->d_ops->data_dotdot_offset);
126 * Put . entry unless we're starting past it.
128 if (ctx->pos <= dot_offset) {
129 ctx->pos = dot_offset & 0x7fffffff;
130 if (!dir_emit(ctx, ".", 1, dp->i_ino, DT_DIR))
135 * Put .. entry unless we're starting past it.
137 if (ctx->pos <= dotdot_offset) {
138 ino = dp->d_ops->sf_get_parent_ino(sfp);
139 ctx->pos = dotdot_offset & 0x7fffffff;
140 if (!dir_emit(ctx, "..", 2, ino, DT_DIR))
145 * Loop while there are more entries and put'ing works.
147 sfep = xfs_dir2_sf_firstentry(sfp);
148 for (i = 0; i < sfp->count; i++) {
151 off = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
152 xfs_dir2_sf_get_offset(sfep));
154 if (ctx->pos > off) {
155 sfep = dp->d_ops->sf_nextentry(sfp, sfep);
159 ino = dp->d_ops->sf_get_ino(sfp, sfep);
160 filetype = dp->d_ops->sf_get_ftype(sfep);
161 ctx->pos = off & 0x7fffffff;
162 if (!dir_emit(ctx, (char *)sfep->name, sfep->namelen, ino,
163 xfs_dir3_get_dtype(dp->i_mount, filetype)))
165 sfep = dp->d_ops->sf_nextentry(sfp, sfep);
168 ctx->pos = xfs_dir2_db_off_to_dataptr(geo, geo->datablk + 1, 0) &
174 * Readdir for block directories.
177 xfs_dir2_block_getdents(
178 struct xfs_da_args *args,
179 struct dir_context *ctx)
181 struct xfs_inode *dp = args->dp; /* incore directory inode */
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 char *ptr; /* current data entry */
190 int wantoff; /* starting block offset */
192 struct xfs_da_geometry *geo = args->geo;
195 * If the block number in the offset is out of range, we're done.
197 if (xfs_dir2_dataptr_to_db(geo, ctx->pos) > geo->datablk)
200 error = xfs_dir3_block_read(NULL, dp, &bp);
205 * Extract the byte offset we start at from the seek pointer.
206 * We'll skip entries before this.
208 wantoff = xfs_dir2_dataptr_to_off(geo, ctx->pos);
210 xfs_dir3_data_check(dp, bp);
212 * Set up values for the loop.
214 btp = xfs_dir2_block_tail_p(geo, hdr);
215 ptr = (char *)dp->d_ops->data_entry_p(hdr);
216 endptr = (char *)xfs_dir2_block_leaf_p(btp);
219 * Loop over the data portion of the block.
220 * Each object is a real entry (dep) or an unused one (dup).
222 while (ptr < endptr) {
225 dup = (xfs_dir2_data_unused_t *)ptr;
229 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
230 ptr += be16_to_cpu(dup->length);
234 dep = (xfs_dir2_data_entry_t *)ptr;
237 * Bump pointer for the next iteration.
239 ptr += dp->d_ops->data_entsize(dep->namelen);
241 * The entry is before the desired starting point, skip it.
243 if ((char *)dep - (char *)hdr < wantoff)
246 cook = xfs_dir2_db_off_to_dataptr(geo, geo->datablk,
247 (char *)dep - (char *)hdr);
249 ctx->pos = cook & 0x7fffffff;
250 filetype = dp->d_ops->data_get_ftype(dep);
252 * If it didn't fit, set the final offset to here & return.
254 if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
255 be64_to_cpu(dep->inumber),
256 xfs_dir3_get_dtype(dp->i_mount, filetype))) {
257 xfs_trans_brelse(NULL, bp);
263 * Reached the end of the block.
264 * Set the offset to a non-existent block 1 and return.
266 ctx->pos = xfs_dir2_db_off_to_dataptr(geo, geo->datablk + 1, 0) &
268 xfs_trans_brelse(NULL, bp);
272 struct xfs_dir2_leaf_map_info {
273 xfs_extlen_t map_blocks; /* number of fsbs in map */
274 xfs_dablk_t map_off; /* last mapped file offset */
275 int map_size; /* total entries in *map */
276 int map_valid; /* valid entries in *map */
277 int nmap; /* mappings to ask xfs_bmapi */
278 xfs_dir2_db_t curdb; /* db for current block */
279 int ra_current; /* number of read-ahead blks */
280 int ra_index; /* *map index for read-ahead */
281 int ra_offset; /* map entry offset for ra */
282 int ra_want; /* readahead count wanted */
283 struct xfs_bmbt_irec map[]; /* map vector for blocks */
287 xfs_dir2_leaf_readbuf(
288 struct xfs_da_args *args,
290 struct xfs_dir2_leaf_map_info *mip,
291 xfs_dir2_off_t *curoff,
292 struct xfs_buf **bpp)
294 struct xfs_inode *dp = args->dp;
295 struct xfs_buf *bp = *bpp;
296 struct xfs_bmbt_irec *map = mip->map;
297 struct blk_plug plug;
302 struct xfs_da_geometry *geo = args->geo;
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 -= geo->fsbcount;
314 * Loop to get rid of the extents for the
317 for (i = geo->fsbcount; 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 + geo->blksize, (1 << geo->fsblog)) - 1;
337 ASSERT(mip->ra_want >= 0);
340 * If we don't have as many as we want, and we haven't
341 * run out of data blocks, get some more mappings.
343 if (1 + mip->ra_want > mip->map_blocks &&
344 mip->map_off < xfs_dir2_byte_to_da(geo, XFS_DIR2_LEAF_OFFSET)) {
346 * Get more bmaps, fill in after the ones
347 * we already have in the table.
349 mip->nmap = mip->map_size - mip->map_valid;
350 error = xfs_bmapi_read(dp, mip->map_off,
351 xfs_dir2_byte_to_da(geo, XFS_DIR2_LEAF_OFFSET) -
353 &map[mip->map_valid], &mip->nmap, 0);
356 * Don't know if we should ignore this or try to return an
357 * error. The trouble with returning errors is that readdir
358 * will just stop without actually passing the error through.
364 * If we got all the mappings we asked for, set the final map
365 * offset based on the last bmap value received. Otherwise,
366 * we've reached the end.
368 if (mip->nmap == mip->map_size - mip->map_valid) {
369 i = mip->map_valid + mip->nmap - 1;
370 mip->map_off = map[i].br_startoff + map[i].br_blockcount;
372 mip->map_off = xfs_dir2_byte_to_da(geo,
373 XFS_DIR2_LEAF_OFFSET);
376 * Look for holes in the mapping, and eliminate them. Count up
379 for (i = mip->map_valid; i < mip->map_valid + mip->nmap; ) {
380 if (map[i].br_startblock == HOLESTARTBLOCK) {
382 length = mip->map_valid + mip->nmap - i;
384 memmove(&map[i], &map[i + 1],
385 sizeof(map[i]) * length);
387 mip->map_blocks += map[i].br_blockcount;
391 mip->map_valid += mip->nmap;
395 * No valid mappings, so no more data blocks.
397 if (!mip->map_valid) {
398 *curoff = xfs_dir2_da_to_byte(geo, mip->map_off);
403 * Read the directory block starting at the first mapping.
405 mip->curdb = xfs_dir2_da_to_db(geo, map->br_startoff);
406 error = xfs_dir3_data_read(NULL, dp, map->br_startoff,
407 map->br_blockcount >= geo->fsbcount ?
408 XFS_FSB_TO_DADDR(dp->i_mount, map->br_startblock) :
411 * Should just skip over the data block instead of giving up.
417 * Adjust the current amount of read-ahead: we just read a block that
421 mip->ra_current -= geo->fsbcount;
424 * Do we need more readahead?
426 blk_start_plug(&plug);
427 for (mip->ra_index = mip->ra_offset = i = 0;
428 mip->ra_want > mip->ra_current && i < mip->map_blocks;
429 i += geo->fsbcount) {
430 ASSERT(mip->ra_index < mip->map_valid);
432 * Read-ahead a contiguous directory block.
434 if (i > mip->ra_current &&
435 map[mip->ra_index].br_blockcount >= geo->fsbcount) {
436 xfs_dir3_data_readahead(dp,
437 map[mip->ra_index].br_startoff + mip->ra_offset,
438 XFS_FSB_TO_DADDR(dp->i_mount,
439 map[mip->ra_index].br_startblock +
445 * Read-ahead a non-contiguous directory block. This doesn't
446 * use our mapping, but this is a very rare case.
448 else if (i > mip->ra_current) {
449 xfs_dir3_data_readahead(dp,
450 map[mip->ra_index].br_startoff +
456 * Advance offset through the mapping table.
458 for (j = 0; j < geo->fsbcount; j += length ) {
460 * The rest of this extent but not more than a dir
463 length = min_t(int, geo->fsbcount,
464 map[mip->ra_index].br_blockcount -
466 mip->ra_offset += length;
469 * Advance to the next mapping if this one is used up.
471 if (mip->ra_offset == map[mip->ra_index].br_blockcount) {
477 blk_finish_plug(&plug);
485 * Getdents (readdir) for leaf and node directories.
486 * This reads the data blocks only, so is the same for both forms.
489 xfs_dir2_leaf_getdents(
490 struct xfs_da_args *args,
491 struct dir_context *ctx,
494 struct xfs_inode *dp = args->dp;
495 struct xfs_buf *bp = NULL; /* data block buffer */
496 xfs_dir2_data_hdr_t *hdr; /* data block header */
497 xfs_dir2_data_entry_t *dep; /* data entry */
498 xfs_dir2_data_unused_t *dup; /* unused entry */
499 int error = 0; /* error return value */
500 int length; /* temporary length value */
501 int byteoff; /* offset in current block */
502 xfs_dir2_off_t curoff; /* current overall offset */
503 xfs_dir2_off_t newoff; /* new curoff after new blk */
504 char *ptr = NULL; /* pointer to current data */
505 struct xfs_dir2_leaf_map_info *map_info;
506 struct xfs_da_geometry *geo = args->geo;
509 * If the offset is at or past the largest allowed value,
510 * give up right away.
512 if (ctx->pos >= XFS_DIR2_MAX_DATAPTR)
516 * Set up to bmap a number of blocks based on the caller's
517 * buffer size, the directory block size, and the filesystem
520 length = howmany(bufsize + geo->blksize, (1 << geo->fsblog));
521 map_info = kmem_zalloc(offsetof(struct xfs_dir2_leaf_map_info, map) +
522 (length * sizeof(struct xfs_bmbt_irec)),
524 map_info->map_size = length;
527 * Inside the loop we keep the main offset value as a byte offset
528 * in the directory file.
530 curoff = xfs_dir2_dataptr_to_byte(ctx->pos);
533 * Force this conversion through db so we truncate the offset
534 * down to get the start of the data block.
536 map_info->map_off = xfs_dir2_db_to_da(geo,
537 xfs_dir2_byte_to_db(geo, curoff));
540 * Loop over directory entries until we reach the end offset.
541 * Get more blocks and readahead as necessary.
543 while (curoff < XFS_DIR2_LEAF_OFFSET) {
547 * If we have no buffer, or we're off the end of the
548 * current buffer, need to get another one.
550 if (!bp || ptr >= (char *)bp->b_addr + geo->blksize) {
552 error = xfs_dir2_leaf_readbuf(args, bufsize, map_info,
554 if (error || !map_info->map_valid)
558 * Having done a read, we need to set a new offset.
560 newoff = xfs_dir2_db_off_to_byte(geo,
563 * Start of the current block.
568 * Make sure we're in the right block.
570 else if (curoff > newoff)
571 ASSERT(xfs_dir2_byte_to_db(geo, curoff) ==
574 xfs_dir3_data_check(dp, bp);
576 * Find our position in the block.
578 ptr = (char *)dp->d_ops->data_entry_p(hdr);
579 byteoff = xfs_dir2_byte_to_off(geo, curoff);
581 * Skip past the header.
584 curoff += dp->d_ops->data_entry_offset;
586 * Skip past entries until we reach our offset.
589 while ((char *)ptr - (char *)hdr < byteoff) {
590 dup = (xfs_dir2_data_unused_t *)ptr;
592 if (be16_to_cpu(dup->freetag)
593 == XFS_DIR2_DATA_FREE_TAG) {
595 length = be16_to_cpu(dup->length);
599 dep = (xfs_dir2_data_entry_t *)ptr;
601 dp->d_ops->data_entsize(dep->namelen);
605 * Now set our real offset.
608 xfs_dir2_db_off_to_byte(geo,
609 xfs_dir2_byte_to_db(geo, curoff),
610 (char *)ptr - (char *)hdr);
611 if (ptr >= (char *)hdr + geo->blksize) {
617 * We have a pointer to an entry.
620 dup = (xfs_dir2_data_unused_t *)ptr;
622 * No, it's unused, skip over it.
624 if (be16_to_cpu(dup->freetag) == XFS_DIR2_DATA_FREE_TAG) {
625 length = be16_to_cpu(dup->length);
631 dep = (xfs_dir2_data_entry_t *)ptr;
632 length = dp->d_ops->data_entsize(dep->namelen);
633 filetype = dp->d_ops->data_get_ftype(dep);
635 ctx->pos = xfs_dir2_byte_to_dataptr(curoff) & 0x7fffffff;
636 if (!dir_emit(ctx, (char *)dep->name, dep->namelen,
637 be64_to_cpu(dep->inumber),
638 xfs_dir3_get_dtype(dp->i_mount, filetype)))
642 * Advance to next entry in the block.
646 /* bufsize may have just been a guess; don't go negative */
647 bufsize = bufsize > length ? bufsize - length : 0;
651 * All done. Set output offset value to current offset.
653 if (curoff > xfs_dir2_dataptr_to_byte(XFS_DIR2_MAX_DATAPTR))
654 ctx->pos = XFS_DIR2_MAX_DATAPTR & 0x7fffffff;
656 ctx->pos = xfs_dir2_byte_to_dataptr(curoff) & 0x7fffffff;
659 xfs_trans_brelse(NULL, bp);
668 struct xfs_inode *dp,
669 struct dir_context *ctx,
672 struct xfs_da_args args = { NULL };
677 trace_xfs_readdir(dp);
679 if (XFS_FORCED_SHUTDOWN(dp->i_mount))
682 ASSERT(S_ISDIR(dp->i_d.di_mode));
683 XFS_STATS_INC(xs_dir_getdents);
686 args.geo = dp->i_mount->m_dir_geo;
688 lock_mode = xfs_ilock_data_map_shared(dp);
689 if (dp->i_d.di_format == XFS_DINODE_FMT_LOCAL)
690 rval = xfs_dir2_sf_getdents(&args, ctx);
691 else if ((rval = xfs_dir2_isblock(&args, &v)))
694 rval = xfs_dir2_block_getdents(&args, ctx);
696 rval = xfs_dir2_leaf_getdents(&args, ctx, bufsize);
697 xfs_iunlock(dp, lock_mode);