1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4 * Copyright (c) 2013 Red Hat, Inc.
9 #include "xfs_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
13 #include "xfs_mount.h"
14 #include "xfs_da_format.h"
15 #include "xfs_da_btree.h"
16 #include "xfs_inode.h"
18 #include "xfs_dir2_priv.h"
19 #include "xfs_error.h"
20 #include "xfs_trace.h"
22 #include "xfs_trans.h"
25 * Directory file type support functions
27 static unsigned char xfs_dir3_filetype_table
[] = {
28 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
,
29 DT_FIFO
, DT_SOCK
, DT_LNK
, DT_WHT
,
37 if (!xfs_sb_version_hasftype(&mp
->m_sb
))
40 if (filetype
>= XFS_DIR3_FT_MAX
)
43 return xfs_dir3_filetype_table
[filetype
];
48 struct xfs_da_args
*args
,
49 struct dir_context
*ctx
)
51 int i
; /* shortform entry number */
52 struct xfs_inode
*dp
= args
->dp
; /* incore directory inode */
53 xfs_dir2_dataptr_t off
; /* current entry's offset */
54 xfs_dir2_sf_entry_t
*sfep
; /* shortform directory entry */
55 xfs_dir2_sf_hdr_t
*sfp
; /* shortform structure */
56 xfs_dir2_dataptr_t dot_offset
;
57 xfs_dir2_dataptr_t dotdot_offset
;
59 struct xfs_da_geometry
*geo
= args
->geo
;
61 ASSERT(dp
->i_df
.if_flags
& XFS_IFINLINE
);
62 ASSERT(dp
->i_df
.if_bytes
== dp
->i_d
.di_size
);
63 ASSERT(dp
->i_df
.if_u1
.if_data
!= NULL
);
65 sfp
= (xfs_dir2_sf_hdr_t
*)dp
->i_df
.if_u1
.if_data
;
68 * If the block number in the offset is out of range, we're done.
70 if (xfs_dir2_dataptr_to_db(geo
, ctx
->pos
) > geo
->datablk
)
74 * Precalculate offsets for . and .. as we will always need them.
76 * XXX(hch): the second argument is sometimes 0 and sometimes
79 dot_offset
= xfs_dir2_db_off_to_dataptr(geo
, geo
->datablk
,
80 dp
->d_ops
->data_dot_offset
);
81 dotdot_offset
= xfs_dir2_db_off_to_dataptr(geo
, geo
->datablk
,
82 dp
->d_ops
->data_dotdot_offset
);
85 * Put . entry unless we're starting past it.
87 if (ctx
->pos
<= dot_offset
) {
88 ctx
->pos
= dot_offset
& 0x7fffffff;
89 if (!dir_emit(ctx
, ".", 1, dp
->i_ino
, DT_DIR
))
94 * Put .. entry unless we're starting past it.
96 if (ctx
->pos
<= dotdot_offset
) {
97 ino
= dp
->d_ops
->sf_get_parent_ino(sfp
);
98 ctx
->pos
= dotdot_offset
& 0x7fffffff;
99 if (!dir_emit(ctx
, "..", 2, ino
, DT_DIR
))
104 * Loop while there are more entries and put'ing works.
106 sfep
= xfs_dir2_sf_firstentry(sfp
);
107 for (i
= 0; i
< sfp
->count
; i
++) {
110 off
= xfs_dir2_db_off_to_dataptr(geo
, geo
->datablk
,
111 xfs_dir2_sf_get_offset(sfep
));
113 if (ctx
->pos
> off
) {
114 sfep
= dp
->d_ops
->sf_nextentry(sfp
, sfep
);
118 ino
= dp
->d_ops
->sf_get_ino(sfp
, sfep
);
119 filetype
= dp
->d_ops
->sf_get_ftype(sfep
);
120 ctx
->pos
= off
& 0x7fffffff;
121 if (!dir_emit(ctx
, (char *)sfep
->name
, sfep
->namelen
, ino
,
122 xfs_dir3_get_dtype(dp
->i_mount
, filetype
)))
124 sfep
= dp
->d_ops
->sf_nextentry(sfp
, sfep
);
127 ctx
->pos
= xfs_dir2_db_off_to_dataptr(geo
, geo
->datablk
+ 1, 0) &
133 * Readdir for block directories.
136 xfs_dir2_block_getdents(
137 struct xfs_da_args
*args
,
138 struct dir_context
*ctx
)
140 struct xfs_inode
*dp
= args
->dp
; /* incore directory inode */
141 xfs_dir2_data_hdr_t
*hdr
; /* block header */
142 struct xfs_buf
*bp
; /* buffer for block */
143 xfs_dir2_data_entry_t
*dep
; /* block data entry */
144 xfs_dir2_data_unused_t
*dup
; /* block unused entry */
145 char *endptr
; /* end of the data entries */
146 int error
; /* error return value */
147 char *ptr
; /* current data entry */
148 int wantoff
; /* starting block offset */
150 struct xfs_da_geometry
*geo
= args
->geo
;
154 * If the block number in the offset is out of range, we're done.
156 if (xfs_dir2_dataptr_to_db(geo
, ctx
->pos
) > geo
->datablk
)
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
);
166 * Extract the byte offset we start at from the seek pointer.
167 * We'll skip entries before this.
169 wantoff
= xfs_dir2_dataptr_to_off(geo
, ctx
->pos
);
171 xfs_dir3_data_check(dp
, bp
);
173 * Set up values for the loop.
175 ptr
= (char *)dp
->d_ops
->data_entry_p(hdr
);
176 endptr
= xfs_dir3_data_endp(geo
, hdr
);
179 * Loop over the data portion of the block.
180 * Each object is a real entry (dep) or an unused one (dup).
182 while (ptr
< endptr
) {
185 dup
= (xfs_dir2_data_unused_t
*)ptr
;
189 if (be16_to_cpu(dup
->freetag
) == XFS_DIR2_DATA_FREE_TAG
) {
190 ptr
+= be16_to_cpu(dup
->length
);
194 dep
= (xfs_dir2_data_entry_t
*)ptr
;
197 * Bump pointer for the next iteration.
199 ptr
+= dp
->d_ops
->data_entsize(dep
->namelen
);
201 * The entry is before the desired starting point, skip it.
203 if ((char *)dep
- (char *)hdr
< wantoff
)
206 cook
= xfs_dir2_db_off_to_dataptr(geo
, geo
->datablk
,
207 (char *)dep
- (char *)hdr
);
209 ctx
->pos
= cook
& 0x7fffffff;
210 filetype
= dp
->d_ops
->data_get_ftype(dep
);
212 * If it didn't fit, set the final offset to here & return.
214 if (!dir_emit(ctx
, (char *)dep
->name
, dep
->namelen
,
215 be64_to_cpu(dep
->inumber
),
216 xfs_dir3_get_dtype(dp
->i_mount
, filetype
))) {
217 xfs_trans_brelse(args
->trans
, bp
);
223 * Reached the end of the block.
224 * Set the offset to a non-existent block 1 and return.
226 ctx
->pos
= xfs_dir2_db_off_to_dataptr(geo
, geo
->datablk
+ 1, 0) &
228 xfs_trans_brelse(args
->trans
, bp
);
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.
238 xfs_dir2_leaf_readbuf(
239 struct xfs_da_args
*args
,
241 xfs_dir2_off_t
*cur_off
,
243 struct xfs_buf
**bpp
)
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
;
255 struct xfs_iext_cursor icur
;
259 if (!(ifp
->if_flags
& XFS_IFEXTENTS
)) {
260 error
= xfs_iread_extents(args
->trans
, dp
, XFS_DATA_FORK
);
266 * Look for mapped directory blocks at or above the current offset.
267 * Truncate down to the nearest directory block to start the scanning
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
))
274 if (map
.br_startoff
>= last_da
)
276 xfs_trim_extent(&map
, map_off
, last_da
- map_off
);
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
)
282 error
= xfs_dir3_data_read(args
->trans
, dp
, map
.br_startoff
, -1, &bp
);
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).
291 ra_want
= howmany(bufsize
+ geo
->blksize
, (1 << geo
->fsblog
));
292 if (*ra_blk
>= last_da
)
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
)
299 if (map
.br_blockcount
< geo
->fsbcount
&&
300 !xfs_iext_next_extent(ifp
, &icur
, &map
))
302 if (map
.br_startoff
>= last_da
)
304 xfs_trim_extent(&map
, next_ra
, last_da
- next_ra
);
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
) {
316 if (next_ra
> *ra_blk
) {
317 xfs_dir3_data_readahead(dp
, next_ra
, -2);
320 ra_want
-= geo
->fsbcount
;
321 next_ra
+= geo
->fsbcount
;
323 if (!xfs_iext_next_extent(ifp
, &icur
, &map
)) {
328 blk_finish_plug(&plug
);
339 * Getdents (readdir) for leaf and node directories.
340 * This reads the data blocks only, so is the same for both forms.
343 xfs_dir2_leaf_getdents(
344 struct xfs_da_args
*args
,
345 struct dir_context
*ctx
,
348 struct xfs_inode
*dp
= args
->dp
;
349 struct xfs_buf
*bp
= NULL
; /* data block buffer */
350 xfs_dir2_data_hdr_t
*hdr
; /* data block header */
351 xfs_dir2_data_entry_t
*dep
; /* data entry */
352 xfs_dir2_data_unused_t
*dup
; /* unused entry */
353 char *ptr
= NULL
; /* pointer to current data */
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 */
360 int error
= 0; /* error return value */
363 * If the offset is at or past the largest allowed value,
364 * give up right away.
366 if (ctx
->pos
>= XFS_DIR2_MAX_DATAPTR
)
370 * Inside the loop we keep the main offset value as a byte offset
371 * in the directory file.
373 curoff
= xfs_dir2_dataptr_to_byte(ctx
->pos
);
376 * Loop over directory entries until we reach the end offset.
377 * Get more blocks and readahead as necessary.
379 while (curoff
< XFS_DIR2_LEAF_OFFSET
) {
383 * If we have no buffer, or we're off the end of the
384 * current buffer, need to get another one.
386 if (!bp
|| ptr
>= (char *)bp
->b_addr
+ geo
->blksize
) {
388 xfs_trans_brelse(args
->trans
, bp
);
392 lock_mode
= xfs_ilock_data_map_shared(dp
);
393 error
= xfs_dir2_leaf_readbuf(args
, bufsize
, &curoff
,
395 xfs_iunlock(dp
, lock_mode
);
400 xfs_dir3_data_check(dp
, bp
);
402 * Find our position in the block.
404 ptr
= (char *)dp
->d_ops
->data_entry_p(hdr
);
405 byteoff
= xfs_dir2_byte_to_off(geo
, curoff
);
407 * Skip past the header.
410 curoff
+= dp
->d_ops
->data_entry_offset
;
412 * Skip past entries until we reach our offset.
415 while ((char *)ptr
- (char *)hdr
< byteoff
) {
416 dup
= (xfs_dir2_data_unused_t
*)ptr
;
418 if (be16_to_cpu(dup
->freetag
)
419 == XFS_DIR2_DATA_FREE_TAG
) {
421 length
= be16_to_cpu(dup
->length
);
425 dep
= (xfs_dir2_data_entry_t
*)ptr
;
427 dp
->d_ops
->data_entsize(dep
->namelen
);
431 * Now set our real offset.
434 xfs_dir2_db_off_to_byte(geo
,
435 xfs_dir2_byte_to_db(geo
, curoff
),
436 (char *)ptr
- (char *)hdr
);
437 if (ptr
>= (char *)hdr
+ geo
->blksize
) {
443 * We have a pointer to an entry.
446 dup
= (xfs_dir2_data_unused_t
*)ptr
;
448 * No, it's unused, skip over it.
450 if (be16_to_cpu(dup
->freetag
) == XFS_DIR2_DATA_FREE_TAG
) {
451 length
= be16_to_cpu(dup
->length
);
457 dep
= (xfs_dir2_data_entry_t
*)ptr
;
458 length
= dp
->d_ops
->data_entsize(dep
->namelen
);
459 filetype
= dp
->d_ops
->data_get_ftype(dep
);
461 ctx
->pos
= xfs_dir2_byte_to_dataptr(curoff
) & 0x7fffffff;
462 if (!dir_emit(ctx
, (char *)dep
->name
, dep
->namelen
,
463 be64_to_cpu(dep
->inumber
),
464 xfs_dir3_get_dtype(dp
->i_mount
, filetype
)))
468 * Advance to next entry in the block.
472 /* bufsize may have just been a guess; don't go negative */
473 bufsize
= bufsize
> length
? bufsize
- length
: 0;
477 * All done. Set output offset value to current offset.
479 if (curoff
> xfs_dir2_dataptr_to_byte(XFS_DIR2_MAX_DATAPTR
))
480 ctx
->pos
= XFS_DIR2_MAX_DATAPTR
& 0x7fffffff;
482 ctx
->pos
= xfs_dir2_byte_to_dataptr(curoff
) & 0x7fffffff;
484 xfs_trans_brelse(args
->trans
, bp
);
491 * If supplied, the transaction collects locked dir buffers to avoid
492 * nested buffer deadlocks. This function does not dirty the
493 * transaction. The caller should ensure that the inode is locked
494 * before calling this function.
498 struct xfs_trans
*tp
,
499 struct xfs_inode
*dp
,
500 struct dir_context
*ctx
,
503 struct xfs_da_args args
= { NULL
};
507 trace_xfs_readdir(dp
);
509 if (XFS_FORCED_SHUTDOWN(dp
->i_mount
))
512 ASSERT(S_ISDIR(VFS_I(dp
)->i_mode
));
513 XFS_STATS_INC(dp
->i_mount
, xs_dir_getdents
);
516 args
.geo
= dp
->i_mount
->m_dir_geo
;
519 if (dp
->i_d
.di_format
== XFS_DINODE_FMT_LOCAL
)
520 rval
= xfs_dir2_sf_getdents(&args
, ctx
);
521 else if ((rval
= xfs_dir2_isblock(&args
, &v
)))
524 rval
= xfs_dir2_block_getdents(&args
, ctx
);
526 rval
= xfs_dir2_leaf_getdents(&args
, ctx
, bufsize
);