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"
25 #include "xfs_mount.h"
26 #include "xfs_da_format.h"
27 #include "xfs_da_btree.h"
28 #include "xfs_inode.h"
30 #include "xfs_dir2_priv.h"
31 #include "xfs_error.h"
32 #include "xfs_trace.h"
34 #include "xfs_trans.h"
37 * Directory file type support functions
39 static unsigned char xfs_dir3_filetype_table
[] = {
40 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
,
41 DT_FIFO
, DT_SOCK
, DT_LNK
, DT_WHT
,
49 if (!xfs_sb_version_hasftype(&mp
->m_sb
))
52 if (filetype
>= XFS_DIR3_FT_MAX
)
55 return xfs_dir3_filetype_table
[filetype
];
60 struct xfs_da_args
*args
,
61 struct dir_context
*ctx
)
63 int i
; /* shortform entry number */
64 struct xfs_inode
*dp
= args
->dp
; /* incore directory inode */
65 xfs_dir2_dataptr_t off
; /* current entry's offset */
66 xfs_dir2_sf_entry_t
*sfep
; /* shortform directory entry */
67 xfs_dir2_sf_hdr_t
*sfp
; /* shortform structure */
68 xfs_dir2_dataptr_t dot_offset
;
69 xfs_dir2_dataptr_t dotdot_offset
;
71 struct xfs_da_geometry
*geo
= args
->geo
;
73 ASSERT(dp
->i_df
.if_flags
& XFS_IFINLINE
);
75 * Give up if the directory is way too short.
77 if (dp
->i_d
.di_size
< offsetof(xfs_dir2_sf_hdr_t
, parent
)) {
78 ASSERT(XFS_FORCED_SHUTDOWN(dp
->i_mount
));
82 ASSERT(dp
->i_df
.if_bytes
== dp
->i_d
.di_size
);
83 ASSERT(dp
->i_df
.if_u1
.if_data
!= NULL
);
85 sfp
= (xfs_dir2_sf_hdr_t
*)dp
->i_df
.if_u1
.if_data
;
87 ASSERT(dp
->i_d
.di_size
>= xfs_dir2_sf_hdr_size(sfp
->i8count
));
90 * If the block number in the offset is out of range, we're done.
92 if (xfs_dir2_dataptr_to_db(geo
, ctx
->pos
) > geo
->datablk
)
96 * Precalculate offsets for . and .. as we will always need them.
98 * XXX(hch): the second argument is sometimes 0 and sometimes
101 dot_offset
= xfs_dir2_db_off_to_dataptr(geo
, geo
->datablk
,
102 dp
->d_ops
->data_dot_offset
);
103 dotdot_offset
= xfs_dir2_db_off_to_dataptr(geo
, geo
->datablk
,
104 dp
->d_ops
->data_dotdot_offset
);
107 * Put . entry unless we're starting past it.
109 if (ctx
->pos
<= dot_offset
) {
110 ctx
->pos
= dot_offset
& 0x7fffffff;
111 if (!dir_emit(ctx
, ".", 1, dp
->i_ino
, DT_DIR
))
116 * Put .. entry unless we're starting past it.
118 if (ctx
->pos
<= dotdot_offset
) {
119 ino
= dp
->d_ops
->sf_get_parent_ino(sfp
);
120 ctx
->pos
= dotdot_offset
& 0x7fffffff;
121 if (!dir_emit(ctx
, "..", 2, ino
, DT_DIR
))
126 * Loop while there are more entries and put'ing works.
128 sfep
= xfs_dir2_sf_firstentry(sfp
);
129 for (i
= 0; i
< sfp
->count
; i
++) {
132 off
= xfs_dir2_db_off_to_dataptr(geo
, geo
->datablk
,
133 xfs_dir2_sf_get_offset(sfep
));
135 if (ctx
->pos
> off
) {
136 sfep
= dp
->d_ops
->sf_nextentry(sfp
, sfep
);
140 ino
= dp
->d_ops
->sf_get_ino(sfp
, sfep
);
141 filetype
= dp
->d_ops
->sf_get_ftype(sfep
);
142 ctx
->pos
= off
& 0x7fffffff;
143 if (!dir_emit(ctx
, (char *)sfep
->name
, sfep
->namelen
, ino
,
144 xfs_dir3_get_dtype(dp
->i_mount
, filetype
)))
146 sfep
= dp
->d_ops
->sf_nextentry(sfp
, sfep
);
149 ctx
->pos
= xfs_dir2_db_off_to_dataptr(geo
, geo
->datablk
+ 1, 0) &
155 * Readdir for block directories.
158 xfs_dir2_block_getdents(
159 struct xfs_da_args
*args
,
160 struct dir_context
*ctx
)
162 struct xfs_inode
*dp
= args
->dp
; /* incore directory inode */
163 xfs_dir2_data_hdr_t
*hdr
; /* block header */
164 struct xfs_buf
*bp
; /* buffer for block */
165 xfs_dir2_block_tail_t
*btp
; /* block tail */
166 xfs_dir2_data_entry_t
*dep
; /* block data entry */
167 xfs_dir2_data_unused_t
*dup
; /* block unused entry */
168 char *endptr
; /* end of the data entries */
169 int error
; /* error return value */
170 char *ptr
; /* current data entry */
171 int wantoff
; /* starting block offset */
173 struct xfs_da_geometry
*geo
= args
->geo
;
177 * If the block number in the offset is out of range, we're done.
179 if (xfs_dir2_dataptr_to_db(geo
, ctx
->pos
) > geo
->datablk
)
182 lock_mode
= xfs_ilock_data_map_shared(dp
);
183 error
= xfs_dir3_block_read(NULL
, dp
, &bp
);
184 xfs_iunlock(dp
, lock_mode
);
189 * Extract the byte offset we start at from the seek pointer.
190 * We'll skip entries before this.
192 wantoff
= xfs_dir2_dataptr_to_off(geo
, ctx
->pos
);
194 xfs_dir3_data_check(dp
, bp
);
196 * Set up values for the loop.
198 btp
= xfs_dir2_block_tail_p(geo
, hdr
);
199 ptr
= (char *)dp
->d_ops
->data_entry_p(hdr
);
200 endptr
= (char *)xfs_dir2_block_leaf_p(btp
);
203 * Loop over the data portion of the block.
204 * Each object is a real entry (dep) or an unused one (dup).
206 while (ptr
< endptr
) {
209 dup
= (xfs_dir2_data_unused_t
*)ptr
;
213 if (be16_to_cpu(dup
->freetag
) == XFS_DIR2_DATA_FREE_TAG
) {
214 ptr
+= be16_to_cpu(dup
->length
);
218 dep
= (xfs_dir2_data_entry_t
*)ptr
;
221 * Bump pointer for the next iteration.
223 ptr
+= dp
->d_ops
->data_entsize(dep
->namelen
);
225 * The entry is before the desired starting point, skip it.
227 if ((char *)dep
- (char *)hdr
< wantoff
)
230 cook
= xfs_dir2_db_off_to_dataptr(geo
, geo
->datablk
,
231 (char *)dep
- (char *)hdr
);
233 ctx
->pos
= cook
& 0x7fffffff;
234 filetype
= dp
->d_ops
->data_get_ftype(dep
);
236 * If it didn't fit, set the final offset to here & return.
238 if (!dir_emit(ctx
, (char *)dep
->name
, dep
->namelen
,
239 be64_to_cpu(dep
->inumber
),
240 xfs_dir3_get_dtype(dp
->i_mount
, filetype
))) {
241 xfs_trans_brelse(NULL
, bp
);
247 * Reached the end of the block.
248 * Set the offset to a non-existent block 1 and return.
250 ctx
->pos
= xfs_dir2_db_off_to_dataptr(geo
, geo
->datablk
+ 1, 0) &
252 xfs_trans_brelse(NULL
, bp
);
256 struct xfs_dir2_leaf_map_info
{
257 xfs_extlen_t map_blocks
; /* number of fsbs in map */
258 xfs_dablk_t map_off
; /* last mapped file offset */
259 int map_size
; /* total entries in *map */
260 int map_valid
; /* valid entries in *map */
261 int nmap
; /* mappings to ask xfs_bmapi */
262 xfs_dir2_db_t curdb
; /* db for current block */
263 int ra_current
; /* number of read-ahead blks */
264 int ra_index
; /* *map index for read-ahead */
265 int ra_offset
; /* map entry offset for ra */
266 int ra_want
; /* readahead count wanted */
267 struct xfs_bmbt_irec map
[]; /* map vector for blocks */
271 xfs_dir2_leaf_readbuf(
272 struct xfs_da_args
*args
,
274 struct xfs_dir2_leaf_map_info
*mip
,
275 xfs_dir2_off_t
*curoff
,
276 struct xfs_buf
**bpp
,
279 struct xfs_inode
*dp
= args
->dp
;
280 struct xfs_buf
*bp
= NULL
;
281 struct xfs_bmbt_irec
*map
= mip
->map
;
282 struct blk_plug plug
;
287 struct xfs_da_geometry
*geo
= args
->geo
;
290 * If the caller just finished processing a buffer, it will tell us
291 * we need to trim that block out of the mapping now it is done.
294 mip
->map_blocks
-= geo
->fsbcount
;
296 * Loop to get rid of the extents for the
299 for (i
= geo
->fsbcount
; i
> 0; ) {
300 j
= min_t(int, map
->br_blockcount
, i
);
301 map
->br_blockcount
-= j
;
302 map
->br_startblock
+= j
;
303 map
->br_startoff
+= j
;
305 * If mapping is done, pitch it from
308 if (!map
->br_blockcount
&& --mip
->map_valid
)
309 memmove(&map
[0], &map
[1],
310 sizeof(map
[0]) * mip
->map_valid
);
316 * Recalculate the readahead blocks wanted.
318 mip
->ra_want
= howmany(bufsize
+ geo
->blksize
, (1 << geo
->fsblog
)) - 1;
319 ASSERT(mip
->ra_want
>= 0);
322 * If we don't have as many as we want, and we haven't
323 * run out of data blocks, get some more mappings.
325 if (1 + mip
->ra_want
> mip
->map_blocks
&&
326 mip
->map_off
< xfs_dir2_byte_to_da(geo
, XFS_DIR2_LEAF_OFFSET
)) {
328 * Get more bmaps, fill in after the ones
329 * we already have in the table.
331 mip
->nmap
= mip
->map_size
- mip
->map_valid
;
332 error
= xfs_bmapi_read(dp
, mip
->map_off
,
333 xfs_dir2_byte_to_da(geo
, XFS_DIR2_LEAF_OFFSET
) -
335 &map
[mip
->map_valid
], &mip
->nmap
, 0);
338 * Don't know if we should ignore this or try to return an
339 * error. The trouble with returning errors is that readdir
340 * will just stop without actually passing the error through.
346 * If we got all the mappings we asked for, set the final map
347 * offset based on the last bmap value received. Otherwise,
348 * we've reached the end.
350 if (mip
->nmap
== mip
->map_size
- mip
->map_valid
) {
351 i
= mip
->map_valid
+ mip
->nmap
- 1;
352 mip
->map_off
= map
[i
].br_startoff
+ map
[i
].br_blockcount
;
354 mip
->map_off
= xfs_dir2_byte_to_da(geo
,
355 XFS_DIR2_LEAF_OFFSET
);
358 * Look for holes in the mapping, and eliminate them. Count up
361 for (i
= mip
->map_valid
; i
< mip
->map_valid
+ mip
->nmap
; ) {
362 if (map
[i
].br_startblock
== HOLESTARTBLOCK
) {
364 length
= mip
->map_valid
+ mip
->nmap
- i
;
366 memmove(&map
[i
], &map
[i
+ 1],
367 sizeof(map
[i
]) * length
);
369 mip
->map_blocks
+= map
[i
].br_blockcount
;
373 mip
->map_valid
+= mip
->nmap
;
377 * No valid mappings, so no more data blocks.
379 if (!mip
->map_valid
) {
380 *curoff
= xfs_dir2_da_to_byte(geo
, mip
->map_off
);
385 * Read the directory block starting at the first mapping.
387 mip
->curdb
= xfs_dir2_da_to_db(geo
, map
->br_startoff
);
388 error
= xfs_dir3_data_read(NULL
, dp
, map
->br_startoff
,
389 map
->br_blockcount
>= geo
->fsbcount
?
390 XFS_FSB_TO_DADDR(dp
->i_mount
, map
->br_startblock
) :
393 * Should just skip over the data block instead of giving up.
399 * Adjust the current amount of read-ahead: we just read a block that
403 mip
->ra_current
-= geo
->fsbcount
;
406 * Do we need more readahead?
408 blk_start_plug(&plug
);
409 for (mip
->ra_index
= mip
->ra_offset
= i
= 0;
410 mip
->ra_want
> mip
->ra_current
&& i
< mip
->map_blocks
;
411 i
+= geo
->fsbcount
) {
412 ASSERT(mip
->ra_index
< mip
->map_valid
);
414 * Read-ahead a contiguous directory block.
416 if (i
> mip
->ra_current
&&
417 map
[mip
->ra_index
].br_blockcount
>= geo
->fsbcount
) {
418 xfs_dir3_data_readahead(dp
,
419 map
[mip
->ra_index
].br_startoff
+ mip
->ra_offset
,
420 XFS_FSB_TO_DADDR(dp
->i_mount
,
421 map
[mip
->ra_index
].br_startblock
+
427 * Read-ahead a non-contiguous directory block. This doesn't
428 * use our mapping, but this is a very rare case.
430 else if (i
> mip
->ra_current
) {
431 xfs_dir3_data_readahead(dp
,
432 map
[mip
->ra_index
].br_startoff
+
438 * Advance offset through the mapping table.
440 for (j
= 0; j
< geo
->fsbcount
; j
+= length
) {
442 * The rest of this extent but not more than a dir
445 length
= min_t(int, geo
->fsbcount
,
446 map
[mip
->ra_index
].br_blockcount
-
448 mip
->ra_offset
+= length
;
451 * Advance to the next mapping if this one is used up.
453 if (mip
->ra_offset
== map
[mip
->ra_index
].br_blockcount
) {
459 blk_finish_plug(&plug
);
467 * Getdents (readdir) for leaf and node directories.
468 * This reads the data blocks only, so is the same for both forms.
471 xfs_dir2_leaf_getdents(
472 struct xfs_da_args
*args
,
473 struct dir_context
*ctx
,
476 struct xfs_inode
*dp
= args
->dp
;
477 struct xfs_buf
*bp
= NULL
; /* data block buffer */
478 xfs_dir2_data_hdr_t
*hdr
; /* data block header */
479 xfs_dir2_data_entry_t
*dep
; /* data entry */
480 xfs_dir2_data_unused_t
*dup
; /* unused entry */
481 int error
= 0; /* error return value */
482 int length
; /* temporary length value */
483 int byteoff
; /* offset in current block */
484 xfs_dir2_off_t curoff
; /* current overall offset */
485 xfs_dir2_off_t newoff
; /* new curoff after new blk */
486 char *ptr
= NULL
; /* pointer to current data */
487 struct xfs_dir2_leaf_map_info
*map_info
;
488 struct xfs_da_geometry
*geo
= args
->geo
;
491 * If the offset is at or past the largest allowed value,
492 * give up right away.
494 if (ctx
->pos
>= XFS_DIR2_MAX_DATAPTR
)
498 * Set up to bmap a number of blocks based on the caller's
499 * buffer size, the directory block size, and the filesystem
502 length
= howmany(bufsize
+ geo
->blksize
, (1 << geo
->fsblog
));
503 map_info
= kmem_zalloc(offsetof(struct xfs_dir2_leaf_map_info
, map
) +
504 (length
* sizeof(struct xfs_bmbt_irec
)),
506 map_info
->map_size
= length
;
509 * Inside the loop we keep the main offset value as a byte offset
510 * in the directory file.
512 curoff
= xfs_dir2_dataptr_to_byte(ctx
->pos
);
515 * Force this conversion through db so we truncate the offset
516 * down to get the start of the data block.
518 map_info
->map_off
= xfs_dir2_db_to_da(geo
,
519 xfs_dir2_byte_to_db(geo
, curoff
));
522 * Loop over directory entries until we reach the end offset.
523 * Get more blocks and readahead as necessary.
525 while (curoff
< XFS_DIR2_LEAF_OFFSET
) {
529 * If we have no buffer, or we're off the end of the
530 * current buffer, need to get another one.
532 if (!bp
|| ptr
>= (char *)bp
->b_addr
+ geo
->blksize
) {
534 bool trim_map
= false;
537 xfs_trans_brelse(NULL
, bp
);
542 lock_mode
= xfs_ilock_data_map_shared(dp
);
543 error
= xfs_dir2_leaf_readbuf(args
, bufsize
, map_info
,
544 &curoff
, &bp
, trim_map
);
545 xfs_iunlock(dp
, lock_mode
);
546 if (error
|| !map_info
->map_valid
)
550 * Having done a read, we need to set a new offset.
552 newoff
= xfs_dir2_db_off_to_byte(geo
,
555 * Start of the current block.
560 * Make sure we're in the right block.
562 else if (curoff
> newoff
)
563 ASSERT(xfs_dir2_byte_to_db(geo
, curoff
) ==
566 xfs_dir3_data_check(dp
, bp
);
568 * Find our position in the block.
570 ptr
= (char *)dp
->d_ops
->data_entry_p(hdr
);
571 byteoff
= xfs_dir2_byte_to_off(geo
, curoff
);
573 * Skip past the header.
576 curoff
+= dp
->d_ops
->data_entry_offset
;
578 * Skip past entries until we reach our offset.
581 while ((char *)ptr
- (char *)hdr
< byteoff
) {
582 dup
= (xfs_dir2_data_unused_t
*)ptr
;
584 if (be16_to_cpu(dup
->freetag
)
585 == XFS_DIR2_DATA_FREE_TAG
) {
587 length
= be16_to_cpu(dup
->length
);
591 dep
= (xfs_dir2_data_entry_t
*)ptr
;
593 dp
->d_ops
->data_entsize(dep
->namelen
);
597 * Now set our real offset.
600 xfs_dir2_db_off_to_byte(geo
,
601 xfs_dir2_byte_to_db(geo
, curoff
),
602 (char *)ptr
- (char *)hdr
);
603 if (ptr
>= (char *)hdr
+ geo
->blksize
) {
609 * We have a pointer to an entry.
612 dup
= (xfs_dir2_data_unused_t
*)ptr
;
614 * No, it's unused, skip over it.
616 if (be16_to_cpu(dup
->freetag
) == XFS_DIR2_DATA_FREE_TAG
) {
617 length
= be16_to_cpu(dup
->length
);
623 dep
= (xfs_dir2_data_entry_t
*)ptr
;
624 length
= dp
->d_ops
->data_entsize(dep
->namelen
);
625 filetype
= dp
->d_ops
->data_get_ftype(dep
);
627 ctx
->pos
= xfs_dir2_byte_to_dataptr(curoff
) & 0x7fffffff;
628 if (!dir_emit(ctx
, (char *)dep
->name
, dep
->namelen
,
629 be64_to_cpu(dep
->inumber
),
630 xfs_dir3_get_dtype(dp
->i_mount
, filetype
)))
634 * Advance to next entry in the block.
638 /* bufsize may have just been a guess; don't go negative */
639 bufsize
= bufsize
> length
? bufsize
- length
: 0;
643 * All done. Set output offset value to current offset.
645 if (curoff
> xfs_dir2_dataptr_to_byte(XFS_DIR2_MAX_DATAPTR
))
646 ctx
->pos
= XFS_DIR2_MAX_DATAPTR
& 0x7fffffff;
648 ctx
->pos
= xfs_dir2_byte_to_dataptr(curoff
) & 0x7fffffff;
651 xfs_trans_brelse(NULL
, bp
);
660 struct xfs_inode
*dp
,
661 struct dir_context
*ctx
,
664 struct xfs_da_args args
= { NULL
};
668 trace_xfs_readdir(dp
);
670 if (XFS_FORCED_SHUTDOWN(dp
->i_mount
))
673 ASSERT(S_ISDIR(VFS_I(dp
)->i_mode
));
674 XFS_STATS_INC(dp
->i_mount
, xs_dir_getdents
);
677 args
.geo
= dp
->i_mount
->m_dir_geo
;
679 xfs_ilock(dp
, XFS_IOLOCK_SHARED
);
680 if (dp
->i_d
.di_format
== XFS_DINODE_FMT_LOCAL
)
681 rval
= xfs_dir2_sf_getdents(&args
, ctx
);
682 else if ((rval
= xfs_dir2_isblock(&args
, &v
)))
685 rval
= xfs_dir2_block_getdents(&args
, ctx
);
687 rval
= xfs_dir2_leaf_getdents(&args
, ctx
, bufsize
);
688 xfs_iunlock(dp
, XFS_IOLOCK_SHARED
);