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 if (dp
->i_d
.di_size
< xfs_dir2_sf_hdr_size(sfp
->i8count
))
91 * If the block number in the offset is out of range, we're done.
93 if (xfs_dir2_dataptr_to_db(geo
, ctx
->pos
) > geo
->datablk
)
97 * Precalculate offsets for . and .. as we will always need them.
99 * XXX(hch): the second argument is sometimes 0 and sometimes
102 dot_offset
= xfs_dir2_db_off_to_dataptr(geo
, geo
->datablk
,
103 dp
->d_ops
->data_dot_offset
);
104 dotdot_offset
= xfs_dir2_db_off_to_dataptr(geo
, geo
->datablk
,
105 dp
->d_ops
->data_dotdot_offset
);
108 * Put . entry unless we're starting past it.
110 if (ctx
->pos
<= dot_offset
) {
111 ctx
->pos
= dot_offset
& 0x7fffffff;
112 if (!dir_emit(ctx
, ".", 1, dp
->i_ino
, DT_DIR
))
117 * Put .. entry unless we're starting past it.
119 if (ctx
->pos
<= dotdot_offset
) {
120 ino
= dp
->d_ops
->sf_get_parent_ino(sfp
);
121 ctx
->pos
= dotdot_offset
& 0x7fffffff;
122 if (!dir_emit(ctx
, "..", 2, ino
, DT_DIR
))
127 * Loop while there are more entries and put'ing works.
129 sfep
= xfs_dir2_sf_firstentry(sfp
);
130 for (i
= 0; i
< sfp
->count
; i
++) {
133 off
= xfs_dir2_db_off_to_dataptr(geo
, geo
->datablk
,
134 xfs_dir2_sf_get_offset(sfep
));
136 if (ctx
->pos
> off
) {
137 sfep
= dp
->d_ops
->sf_nextentry(sfp
, sfep
);
141 ino
= dp
->d_ops
->sf_get_ino(sfp
, sfep
);
142 filetype
= dp
->d_ops
->sf_get_ftype(sfep
);
143 ctx
->pos
= off
& 0x7fffffff;
144 if (!dir_emit(ctx
, (char *)sfep
->name
, sfep
->namelen
, ino
,
145 xfs_dir3_get_dtype(dp
->i_mount
, filetype
)))
147 sfep
= dp
->d_ops
->sf_nextentry(sfp
, sfep
);
150 ctx
->pos
= xfs_dir2_db_off_to_dataptr(geo
, geo
->datablk
+ 1, 0) &
156 * Readdir for block directories.
159 xfs_dir2_block_getdents(
160 struct xfs_da_args
*args
,
161 struct dir_context
*ctx
)
163 struct xfs_inode
*dp
= args
->dp
; /* incore directory inode */
164 xfs_dir2_data_hdr_t
*hdr
; /* block header */
165 struct xfs_buf
*bp
; /* buffer for block */
166 xfs_dir2_block_tail_t
*btp
; /* block tail */
167 xfs_dir2_data_entry_t
*dep
; /* block data entry */
168 xfs_dir2_data_unused_t
*dup
; /* block unused entry */
169 char *endptr
; /* end of the data entries */
170 int error
; /* error return value */
171 char *ptr
; /* current data entry */
172 int wantoff
; /* starting block offset */
174 struct xfs_da_geometry
*geo
= args
->geo
;
178 * If the block number in the offset is out of range, we're done.
180 if (xfs_dir2_dataptr_to_db(geo
, ctx
->pos
) > geo
->datablk
)
183 lock_mode
= xfs_ilock_data_map_shared(dp
);
184 error
= xfs_dir3_block_read(NULL
, dp
, &bp
);
185 xfs_iunlock(dp
, lock_mode
);
190 * Extract the byte offset we start at from the seek pointer.
191 * We'll skip entries before this.
193 wantoff
= xfs_dir2_dataptr_to_off(geo
, ctx
->pos
);
195 xfs_dir3_data_check(dp
, bp
);
197 * Set up values for the loop.
199 btp
= xfs_dir2_block_tail_p(geo
, hdr
);
200 ptr
= (char *)dp
->d_ops
->data_entry_p(hdr
);
201 endptr
= (char *)xfs_dir2_block_leaf_p(btp
);
204 * Loop over the data portion of the block.
205 * Each object is a real entry (dep) or an unused one (dup).
207 while (ptr
< endptr
) {
210 dup
= (xfs_dir2_data_unused_t
*)ptr
;
214 if (be16_to_cpu(dup
->freetag
) == XFS_DIR2_DATA_FREE_TAG
) {
215 ptr
+= be16_to_cpu(dup
->length
);
219 dep
= (xfs_dir2_data_entry_t
*)ptr
;
222 * Bump pointer for the next iteration.
224 ptr
+= dp
->d_ops
->data_entsize(dep
->namelen
);
226 * The entry is before the desired starting point, skip it.
228 if ((char *)dep
- (char *)hdr
< wantoff
)
231 cook
= xfs_dir2_db_off_to_dataptr(geo
, geo
->datablk
,
232 (char *)dep
- (char *)hdr
);
234 ctx
->pos
= cook
& 0x7fffffff;
235 filetype
= dp
->d_ops
->data_get_ftype(dep
);
237 * If it didn't fit, set the final offset to here & return.
239 if (!dir_emit(ctx
, (char *)dep
->name
, dep
->namelen
,
240 be64_to_cpu(dep
->inumber
),
241 xfs_dir3_get_dtype(dp
->i_mount
, filetype
))) {
242 xfs_trans_brelse(NULL
, bp
);
248 * Reached the end of the block.
249 * Set the offset to a non-existent block 1 and return.
251 ctx
->pos
= xfs_dir2_db_off_to_dataptr(geo
, geo
->datablk
+ 1, 0) &
253 xfs_trans_brelse(NULL
, bp
);
257 struct xfs_dir2_leaf_map_info
{
258 xfs_extlen_t map_blocks
; /* number of fsbs in map */
259 xfs_dablk_t map_off
; /* last mapped file offset */
260 int map_size
; /* total entries in *map */
261 int map_valid
; /* valid entries in *map */
262 int nmap
; /* mappings to ask xfs_bmapi */
263 xfs_dir2_db_t curdb
; /* db for current block */
264 int ra_current
; /* number of read-ahead blks */
265 int ra_index
; /* *map index for read-ahead */
266 int ra_offset
; /* map entry offset for ra */
267 int ra_want
; /* readahead count wanted */
268 struct xfs_bmbt_irec map
[]; /* map vector for blocks */
272 xfs_dir2_leaf_readbuf(
273 struct xfs_da_args
*args
,
275 struct xfs_dir2_leaf_map_info
*mip
,
276 xfs_dir2_off_t
*curoff
,
277 struct xfs_buf
**bpp
,
280 struct xfs_inode
*dp
= args
->dp
;
281 struct xfs_buf
*bp
= NULL
;
282 struct xfs_bmbt_irec
*map
= mip
->map
;
283 struct blk_plug plug
;
288 struct xfs_da_geometry
*geo
= args
->geo
;
291 * If the caller just finished processing a buffer, it will tell us
292 * we need to trim that block out of the mapping now it is done.
295 mip
->map_blocks
-= geo
->fsbcount
;
297 * Loop to get rid of the extents for the
300 for (i
= geo
->fsbcount
; i
> 0; ) {
301 j
= min_t(int, map
->br_blockcount
, i
);
302 map
->br_blockcount
-= j
;
303 map
->br_startblock
+= j
;
304 map
->br_startoff
+= j
;
306 * If mapping is done, pitch it from
309 if (!map
->br_blockcount
&& --mip
->map_valid
)
310 memmove(&map
[0], &map
[1],
311 sizeof(map
[0]) * mip
->map_valid
);
317 * Recalculate the readahead blocks wanted.
319 mip
->ra_want
= howmany(bufsize
+ geo
->blksize
, (1 << geo
->fsblog
)) - 1;
320 ASSERT(mip
->ra_want
>= 0);
323 * If we don't have as many as we want, and we haven't
324 * run out of data blocks, get some more mappings.
326 if (1 + mip
->ra_want
> mip
->map_blocks
&&
327 mip
->map_off
< xfs_dir2_byte_to_da(geo
, XFS_DIR2_LEAF_OFFSET
)) {
329 * Get more bmaps, fill in after the ones
330 * we already have in the table.
332 mip
->nmap
= mip
->map_size
- mip
->map_valid
;
333 error
= xfs_bmapi_read(dp
, mip
->map_off
,
334 xfs_dir2_byte_to_da(geo
, XFS_DIR2_LEAF_OFFSET
) -
336 &map
[mip
->map_valid
], &mip
->nmap
, 0);
339 * Don't know if we should ignore this or try to return an
340 * error. The trouble with returning errors is that readdir
341 * will just stop without actually passing the error through.
347 * If we got all the mappings we asked for, set the final map
348 * offset based on the last bmap value received. Otherwise,
349 * we've reached the end.
351 if (mip
->nmap
== mip
->map_size
- mip
->map_valid
) {
352 i
= mip
->map_valid
+ mip
->nmap
- 1;
353 mip
->map_off
= map
[i
].br_startoff
+ map
[i
].br_blockcount
;
355 mip
->map_off
= xfs_dir2_byte_to_da(geo
,
356 XFS_DIR2_LEAF_OFFSET
);
359 * Look for holes in the mapping, and eliminate them. Count up
362 for (i
= mip
->map_valid
; i
< mip
->map_valid
+ mip
->nmap
; ) {
363 if (map
[i
].br_startblock
== HOLESTARTBLOCK
) {
365 length
= mip
->map_valid
+ mip
->nmap
- i
;
367 memmove(&map
[i
], &map
[i
+ 1],
368 sizeof(map
[i
]) * length
);
370 mip
->map_blocks
+= map
[i
].br_blockcount
;
374 mip
->map_valid
+= mip
->nmap
;
378 * No valid mappings, so no more data blocks.
380 if (!mip
->map_valid
) {
381 *curoff
= xfs_dir2_da_to_byte(geo
, mip
->map_off
);
386 * Read the directory block starting at the first mapping.
388 mip
->curdb
= xfs_dir2_da_to_db(geo
, map
->br_startoff
);
389 error
= xfs_dir3_data_read(NULL
, dp
, map
->br_startoff
,
390 map
->br_blockcount
>= geo
->fsbcount
?
391 XFS_FSB_TO_DADDR(dp
->i_mount
, map
->br_startblock
) :
394 * Should just skip over the data block instead of giving up.
400 * Adjust the current amount of read-ahead: we just read a block that
404 mip
->ra_current
-= geo
->fsbcount
;
407 * Do we need more readahead?
409 blk_start_plug(&plug
);
410 for (mip
->ra_index
= mip
->ra_offset
= i
= 0;
411 mip
->ra_want
> mip
->ra_current
&& i
< mip
->map_blocks
;
412 i
+= geo
->fsbcount
) {
413 ASSERT(mip
->ra_index
< mip
->map_valid
);
415 * Read-ahead a contiguous directory block.
417 if (i
> mip
->ra_current
&&
418 map
[mip
->ra_index
].br_blockcount
>= geo
->fsbcount
) {
419 xfs_dir3_data_readahead(dp
,
420 map
[mip
->ra_index
].br_startoff
+ mip
->ra_offset
,
421 XFS_FSB_TO_DADDR(dp
->i_mount
,
422 map
[mip
->ra_index
].br_startblock
+
428 * Read-ahead a non-contiguous directory block. This doesn't
429 * use our mapping, but this is a very rare case.
431 else if (i
> mip
->ra_current
) {
432 xfs_dir3_data_readahead(dp
,
433 map
[mip
->ra_index
].br_startoff
+
439 * Advance offset through the mapping table.
441 for (j
= 0; j
< geo
->fsbcount
; j
+= length
) {
443 * The rest of this extent but not more than a dir
446 length
= min_t(int, geo
->fsbcount
,
447 map
[mip
->ra_index
].br_blockcount
-
449 mip
->ra_offset
+= length
;
452 * Advance to the next mapping if this one is used up.
454 if (mip
->ra_offset
== map
[mip
->ra_index
].br_blockcount
) {
460 blk_finish_plug(&plug
);
468 * Getdents (readdir) for leaf and node directories.
469 * This reads the data blocks only, so is the same for both forms.
472 xfs_dir2_leaf_getdents(
473 struct xfs_da_args
*args
,
474 struct dir_context
*ctx
,
477 struct xfs_inode
*dp
= args
->dp
;
478 struct xfs_buf
*bp
= NULL
; /* data block buffer */
479 xfs_dir2_data_hdr_t
*hdr
; /* data block header */
480 xfs_dir2_data_entry_t
*dep
; /* data entry */
481 xfs_dir2_data_unused_t
*dup
; /* unused entry */
482 int error
= 0; /* error return value */
483 int length
; /* temporary length value */
484 int byteoff
; /* offset in current block */
485 xfs_dir2_off_t curoff
; /* current overall offset */
486 xfs_dir2_off_t newoff
; /* new curoff after new blk */
487 char *ptr
= NULL
; /* pointer to current data */
488 struct xfs_dir2_leaf_map_info
*map_info
;
489 struct xfs_da_geometry
*geo
= args
->geo
;
492 * If the offset is at or past the largest allowed value,
493 * give up right away.
495 if (ctx
->pos
>= XFS_DIR2_MAX_DATAPTR
)
499 * Set up to bmap a number of blocks based on the caller's
500 * buffer size, the directory block size, and the filesystem
503 length
= howmany(bufsize
+ geo
->blksize
, (1 << geo
->fsblog
));
504 map_info
= kmem_zalloc(offsetof(struct xfs_dir2_leaf_map_info
, map
) +
505 (length
* sizeof(struct xfs_bmbt_irec
)),
507 map_info
->map_size
= length
;
510 * Inside the loop we keep the main offset value as a byte offset
511 * in the directory file.
513 curoff
= xfs_dir2_dataptr_to_byte(ctx
->pos
);
516 * Force this conversion through db so we truncate the offset
517 * down to get the start of the data block.
519 map_info
->map_off
= xfs_dir2_db_to_da(geo
,
520 xfs_dir2_byte_to_db(geo
, curoff
));
523 * Loop over directory entries until we reach the end offset.
524 * Get more blocks and readahead as necessary.
526 while (curoff
< XFS_DIR2_LEAF_OFFSET
) {
530 * If we have no buffer, or we're off the end of the
531 * current buffer, need to get another one.
533 if (!bp
|| ptr
>= (char *)bp
->b_addr
+ geo
->blksize
) {
535 bool trim_map
= false;
538 xfs_trans_brelse(NULL
, bp
);
543 lock_mode
= xfs_ilock_data_map_shared(dp
);
544 error
= xfs_dir2_leaf_readbuf(args
, bufsize
, map_info
,
545 &curoff
, &bp
, trim_map
);
546 xfs_iunlock(dp
, lock_mode
);
547 if (error
|| !map_info
->map_valid
)
551 * Having done a read, we need to set a new offset.
553 newoff
= xfs_dir2_db_off_to_byte(geo
,
556 * Start of the current block.
561 * Make sure we're in the right block.
563 else if (curoff
> newoff
)
564 ASSERT(xfs_dir2_byte_to_db(geo
, curoff
) ==
567 xfs_dir3_data_check(dp
, bp
);
569 * Find our position in the block.
571 ptr
= (char *)dp
->d_ops
->data_entry_p(hdr
);
572 byteoff
= xfs_dir2_byte_to_off(geo
, curoff
);
574 * Skip past the header.
577 curoff
+= dp
->d_ops
->data_entry_offset
;
579 * Skip past entries until we reach our offset.
582 while ((char *)ptr
- (char *)hdr
< byteoff
) {
583 dup
= (xfs_dir2_data_unused_t
*)ptr
;
585 if (be16_to_cpu(dup
->freetag
)
586 == XFS_DIR2_DATA_FREE_TAG
) {
588 length
= be16_to_cpu(dup
->length
);
592 dep
= (xfs_dir2_data_entry_t
*)ptr
;
594 dp
->d_ops
->data_entsize(dep
->namelen
);
598 * Now set our real offset.
601 xfs_dir2_db_off_to_byte(geo
,
602 xfs_dir2_byte_to_db(geo
, curoff
),
603 (char *)ptr
- (char *)hdr
);
604 if (ptr
>= (char *)hdr
+ geo
->blksize
) {
610 * We have a pointer to an entry.
613 dup
= (xfs_dir2_data_unused_t
*)ptr
;
615 * No, it's unused, skip over it.
617 if (be16_to_cpu(dup
->freetag
) == XFS_DIR2_DATA_FREE_TAG
) {
618 length
= be16_to_cpu(dup
->length
);
624 dep
= (xfs_dir2_data_entry_t
*)ptr
;
625 length
= dp
->d_ops
->data_entsize(dep
->namelen
);
626 filetype
= dp
->d_ops
->data_get_ftype(dep
);
628 ctx
->pos
= xfs_dir2_byte_to_dataptr(curoff
) & 0x7fffffff;
629 if (!dir_emit(ctx
, (char *)dep
->name
, dep
->namelen
,
630 be64_to_cpu(dep
->inumber
),
631 xfs_dir3_get_dtype(dp
->i_mount
, filetype
)))
635 * Advance to next entry in the block.
639 /* bufsize may have just been a guess; don't go negative */
640 bufsize
= bufsize
> length
? bufsize
- length
: 0;
644 * All done. Set output offset value to current offset.
646 if (curoff
> xfs_dir2_dataptr_to_byte(XFS_DIR2_MAX_DATAPTR
))
647 ctx
->pos
= XFS_DIR2_MAX_DATAPTR
& 0x7fffffff;
649 ctx
->pos
= xfs_dir2_byte_to_dataptr(curoff
) & 0x7fffffff;
652 xfs_trans_brelse(NULL
, bp
);
661 struct xfs_inode
*dp
,
662 struct dir_context
*ctx
,
665 struct xfs_da_args args
= { NULL
};
669 trace_xfs_readdir(dp
);
671 if (XFS_FORCED_SHUTDOWN(dp
->i_mount
))
674 ASSERT(S_ISDIR(VFS_I(dp
)->i_mode
));
675 XFS_STATS_INC(dp
->i_mount
, xs_dir_getdents
);
678 args
.geo
= dp
->i_mount
->m_dir_geo
;
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
);