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
);
74 ASSERT(dp
->i_df
.if_bytes
== dp
->i_d
.di_size
);
75 ASSERT(dp
->i_df
.if_u1
.if_data
!= NULL
);
77 sfp
= (xfs_dir2_sf_hdr_t
*)dp
->i_df
.if_u1
.if_data
;
80 * If the block number in the offset is out of range, we're done.
82 if (xfs_dir2_dataptr_to_db(geo
, ctx
->pos
) > geo
->datablk
)
86 * Precalculate offsets for . and .. as we will always need them.
88 * XXX(hch): the second argument is sometimes 0 and sometimes
91 dot_offset
= xfs_dir2_db_off_to_dataptr(geo
, geo
->datablk
,
92 dp
->d_ops
->data_dot_offset
);
93 dotdot_offset
= xfs_dir2_db_off_to_dataptr(geo
, geo
->datablk
,
94 dp
->d_ops
->data_dotdot_offset
);
97 * Put . entry unless we're starting past it.
99 if (ctx
->pos
<= dot_offset
) {
100 ctx
->pos
= dot_offset
& 0x7fffffff;
101 if (!dir_emit(ctx
, ".", 1, dp
->i_ino
, DT_DIR
))
106 * Put .. entry unless we're starting past it.
108 if (ctx
->pos
<= dotdot_offset
) {
109 ino
= dp
->d_ops
->sf_get_parent_ino(sfp
);
110 ctx
->pos
= dotdot_offset
& 0x7fffffff;
111 if (!dir_emit(ctx
, "..", 2, ino
, DT_DIR
))
116 * Loop while there are more entries and put'ing works.
118 sfep
= xfs_dir2_sf_firstentry(sfp
);
119 for (i
= 0; i
< sfp
->count
; i
++) {
122 off
= xfs_dir2_db_off_to_dataptr(geo
, geo
->datablk
,
123 xfs_dir2_sf_get_offset(sfep
));
125 if (ctx
->pos
> off
) {
126 sfep
= dp
->d_ops
->sf_nextentry(sfp
, sfep
);
130 ino
= dp
->d_ops
->sf_get_ino(sfp
, sfep
);
131 filetype
= dp
->d_ops
->sf_get_ftype(sfep
);
132 ctx
->pos
= off
& 0x7fffffff;
133 if (!dir_emit(ctx
, (char *)sfep
->name
, sfep
->namelen
, ino
,
134 xfs_dir3_get_dtype(dp
->i_mount
, filetype
)))
136 sfep
= dp
->d_ops
->sf_nextentry(sfp
, sfep
);
139 ctx
->pos
= xfs_dir2_db_off_to_dataptr(geo
, geo
->datablk
+ 1, 0) &
145 * Readdir for block directories.
148 xfs_dir2_block_getdents(
149 struct xfs_da_args
*args
,
150 struct dir_context
*ctx
)
152 struct xfs_inode
*dp
= args
->dp
; /* incore directory inode */
153 xfs_dir2_data_hdr_t
*hdr
; /* block header */
154 struct xfs_buf
*bp
; /* buffer for block */
155 xfs_dir2_block_tail_t
*btp
; /* block tail */
156 xfs_dir2_data_entry_t
*dep
; /* block data entry */
157 xfs_dir2_data_unused_t
*dup
; /* block unused entry */
158 char *endptr
; /* end of the data entries */
159 int error
; /* error return value */
160 char *ptr
; /* current data entry */
161 int wantoff
; /* starting block offset */
163 struct xfs_da_geometry
*geo
= args
->geo
;
167 * If the block number in the offset is out of range, we're done.
169 if (xfs_dir2_dataptr_to_db(geo
, ctx
->pos
) > geo
->datablk
)
172 lock_mode
= xfs_ilock_data_map_shared(dp
);
173 error
= xfs_dir3_block_read(NULL
, dp
, &bp
);
174 xfs_iunlock(dp
, lock_mode
);
179 * Extract the byte offset we start at from the seek pointer.
180 * We'll skip entries before this.
182 wantoff
= xfs_dir2_dataptr_to_off(geo
, ctx
->pos
);
184 xfs_dir3_data_check(dp
, bp
);
186 * Set up values for the loop.
188 btp
= xfs_dir2_block_tail_p(geo
, hdr
);
189 ptr
= (char *)dp
->d_ops
->data_entry_p(hdr
);
190 endptr
= (char *)xfs_dir2_block_leaf_p(btp
);
193 * Loop over the data portion of the block.
194 * Each object is a real entry (dep) or an unused one (dup).
196 while (ptr
< endptr
) {
199 dup
= (xfs_dir2_data_unused_t
*)ptr
;
203 if (be16_to_cpu(dup
->freetag
) == XFS_DIR2_DATA_FREE_TAG
) {
204 ptr
+= be16_to_cpu(dup
->length
);
208 dep
= (xfs_dir2_data_entry_t
*)ptr
;
211 * Bump pointer for the next iteration.
213 ptr
+= dp
->d_ops
->data_entsize(dep
->namelen
);
215 * The entry is before the desired starting point, skip it.
217 if ((char *)dep
- (char *)hdr
< wantoff
)
220 cook
= xfs_dir2_db_off_to_dataptr(geo
, geo
->datablk
,
221 (char *)dep
- (char *)hdr
);
223 ctx
->pos
= cook
& 0x7fffffff;
224 filetype
= dp
->d_ops
->data_get_ftype(dep
);
226 * If it didn't fit, set the final offset to here & return.
228 if (!dir_emit(ctx
, (char *)dep
->name
, dep
->namelen
,
229 be64_to_cpu(dep
->inumber
),
230 xfs_dir3_get_dtype(dp
->i_mount
, filetype
))) {
231 xfs_trans_brelse(NULL
, bp
);
237 * Reached the end of the block.
238 * Set the offset to a non-existent block 1 and return.
240 ctx
->pos
= xfs_dir2_db_off_to_dataptr(geo
, geo
->datablk
+ 1, 0) &
242 xfs_trans_brelse(NULL
, bp
);
246 struct xfs_dir2_leaf_map_info
{
247 xfs_extlen_t map_blocks
; /* number of fsbs in map */
248 xfs_dablk_t map_off
; /* last mapped file offset */
249 int map_size
; /* total entries in *map */
250 int map_valid
; /* valid entries in *map */
251 int nmap
; /* mappings to ask xfs_bmapi */
252 xfs_dir2_db_t curdb
; /* db for current block */
253 int ra_current
; /* number of read-ahead blks */
254 int ra_index
; /* *map index for read-ahead */
255 int ra_offset
; /* map entry offset for ra */
256 int ra_want
; /* readahead count wanted */
257 struct xfs_bmbt_irec map
[]; /* map vector for blocks */
261 xfs_dir2_leaf_readbuf(
262 struct xfs_da_args
*args
,
264 struct xfs_dir2_leaf_map_info
*mip
,
265 xfs_dir2_off_t
*curoff
,
266 struct xfs_buf
**bpp
,
269 struct xfs_inode
*dp
= args
->dp
;
270 struct xfs_buf
*bp
= NULL
;
271 struct xfs_bmbt_irec
*map
= mip
->map
;
272 struct blk_plug plug
;
277 struct xfs_da_geometry
*geo
= args
->geo
;
280 * If the caller just finished processing a buffer, it will tell us
281 * we need to trim that block out of the mapping now it is done.
284 mip
->map_blocks
-= geo
->fsbcount
;
286 * Loop to get rid of the extents for the
289 for (i
= geo
->fsbcount
; i
> 0; ) {
290 j
= min_t(int, map
->br_blockcount
, i
);
291 map
->br_blockcount
-= j
;
292 map
->br_startblock
+= j
;
293 map
->br_startoff
+= j
;
295 * If mapping is done, pitch it from
298 if (!map
->br_blockcount
&& --mip
->map_valid
)
299 memmove(&map
[0], &map
[1],
300 sizeof(map
[0]) * mip
->map_valid
);
306 * Recalculate the readahead blocks wanted.
308 mip
->ra_want
= howmany(bufsize
+ geo
->blksize
, (1 << geo
->fsblog
)) - 1;
309 ASSERT(mip
->ra_want
>= 0);
312 * If we don't have as many as we want, and we haven't
313 * run out of data blocks, get some more mappings.
315 if (1 + mip
->ra_want
> mip
->map_blocks
&&
316 mip
->map_off
< xfs_dir2_byte_to_da(geo
, XFS_DIR2_LEAF_OFFSET
)) {
318 * Get more bmaps, fill in after the ones
319 * we already have in the table.
321 mip
->nmap
= mip
->map_size
- mip
->map_valid
;
322 error
= xfs_bmapi_read(dp
, mip
->map_off
,
323 xfs_dir2_byte_to_da(geo
, XFS_DIR2_LEAF_OFFSET
) -
325 &map
[mip
->map_valid
], &mip
->nmap
, 0);
328 * Don't know if we should ignore this or try to return an
329 * error. The trouble with returning errors is that readdir
330 * will just stop without actually passing the error through.
336 * If we got all the mappings we asked for, set the final map
337 * offset based on the last bmap value received. Otherwise,
338 * we've reached the end.
340 if (mip
->nmap
== mip
->map_size
- mip
->map_valid
) {
341 i
= mip
->map_valid
+ mip
->nmap
- 1;
342 mip
->map_off
= map
[i
].br_startoff
+ map
[i
].br_blockcount
;
344 mip
->map_off
= xfs_dir2_byte_to_da(geo
,
345 XFS_DIR2_LEAF_OFFSET
);
348 * Look for holes in the mapping, and eliminate them. Count up
351 for (i
= mip
->map_valid
; i
< mip
->map_valid
+ mip
->nmap
; ) {
352 if (map
[i
].br_startblock
== HOLESTARTBLOCK
) {
354 length
= mip
->map_valid
+ mip
->nmap
- i
;
356 memmove(&map
[i
], &map
[i
+ 1],
357 sizeof(map
[i
]) * length
);
359 mip
->map_blocks
+= map
[i
].br_blockcount
;
363 mip
->map_valid
+= mip
->nmap
;
367 * No valid mappings, so no more data blocks.
369 if (!mip
->map_valid
) {
370 *curoff
= xfs_dir2_da_to_byte(geo
, mip
->map_off
);
375 * Read the directory block starting at the first mapping.
377 mip
->curdb
= xfs_dir2_da_to_db(geo
, map
->br_startoff
);
378 error
= xfs_dir3_data_read(NULL
, dp
, map
->br_startoff
,
379 map
->br_blockcount
>= geo
->fsbcount
?
380 XFS_FSB_TO_DADDR(dp
->i_mount
, map
->br_startblock
) :
383 * Should just skip over the data block instead of giving up.
389 * Adjust the current amount of read-ahead: we just read a block that
393 mip
->ra_current
-= geo
->fsbcount
;
396 * Do we need more readahead?
397 * Each loop tries to process 1 full dir blk; last may be partial.
399 blk_start_plug(&plug
);
400 for (mip
->ra_index
= mip
->ra_offset
= i
= 0;
401 mip
->ra_want
> mip
->ra_current
&& i
< mip
->map_blocks
;
402 i
+= geo
->fsbcount
) {
403 ASSERT(mip
->ra_index
< mip
->map_valid
);
405 * Read-ahead a contiguous directory block.
407 if (i
> mip
->ra_current
&&
408 (map
[mip
->ra_index
].br_blockcount
- mip
->ra_offset
) >=
410 xfs_dir3_data_readahead(dp
,
411 map
[mip
->ra_index
].br_startoff
+ mip
->ra_offset
,
412 XFS_FSB_TO_DADDR(dp
->i_mount
,
413 map
[mip
->ra_index
].br_startblock
+
419 * Read-ahead a non-contiguous directory block. This doesn't
420 * use our mapping, but this is a very rare case.
422 else if (i
> mip
->ra_current
) {
423 xfs_dir3_data_readahead(dp
,
424 map
[mip
->ra_index
].br_startoff
+
430 * Advance offset through the mapping table, processing a full
431 * dir block even if it is fragmented into several extents.
432 * But stop if we have consumed all valid mappings, even if
433 * it's not yet a full directory block.
436 j
< geo
->fsbcount
&& mip
->ra_index
< mip
->map_valid
;
439 * The rest of this extent but not more than a dir
442 length
= min_t(int, geo
->fsbcount
- j
,
443 map
[mip
->ra_index
].br_blockcount
-
445 mip
->ra_offset
+= length
;
448 * Advance to the next mapping if this one is used up.
450 if (mip
->ra_offset
== map
[mip
->ra_index
].br_blockcount
) {
456 blk_finish_plug(&plug
);
464 * Getdents (readdir) for leaf and node directories.
465 * This reads the data blocks only, so is the same for both forms.
468 xfs_dir2_leaf_getdents(
469 struct xfs_da_args
*args
,
470 struct dir_context
*ctx
,
473 struct xfs_inode
*dp
= args
->dp
;
474 struct xfs_buf
*bp
= NULL
; /* data block buffer */
475 xfs_dir2_data_hdr_t
*hdr
; /* data block header */
476 xfs_dir2_data_entry_t
*dep
; /* data entry */
477 xfs_dir2_data_unused_t
*dup
; /* unused entry */
478 int error
= 0; /* error return value */
479 int length
; /* temporary length value */
480 int byteoff
; /* offset in current block */
481 xfs_dir2_off_t curoff
; /* current overall offset */
482 xfs_dir2_off_t newoff
; /* new curoff after new blk */
483 char *ptr
= NULL
; /* pointer to current data */
484 struct xfs_dir2_leaf_map_info
*map_info
;
485 struct xfs_da_geometry
*geo
= args
->geo
;
488 * If the offset is at or past the largest allowed value,
489 * give up right away.
491 if (ctx
->pos
>= XFS_DIR2_MAX_DATAPTR
)
495 * Set up to bmap a number of blocks based on the caller's
496 * buffer size, the directory block size, and the filesystem
499 length
= howmany(bufsize
+ geo
->blksize
, (1 << geo
->fsblog
));
500 map_info
= kmem_zalloc(offsetof(struct xfs_dir2_leaf_map_info
, map
) +
501 (length
* sizeof(struct xfs_bmbt_irec
)),
503 map_info
->map_size
= length
;
506 * Inside the loop we keep the main offset value as a byte offset
507 * in the directory file.
509 curoff
= xfs_dir2_dataptr_to_byte(ctx
->pos
);
512 * Force this conversion through db so we truncate the offset
513 * down to get the start of the data block.
515 map_info
->map_off
= xfs_dir2_db_to_da(geo
,
516 xfs_dir2_byte_to_db(geo
, curoff
));
519 * Loop over directory entries until we reach the end offset.
520 * Get more blocks and readahead as necessary.
522 while (curoff
< XFS_DIR2_LEAF_OFFSET
) {
526 * If we have no buffer, or we're off the end of the
527 * current buffer, need to get another one.
529 if (!bp
|| ptr
>= (char *)bp
->b_addr
+ geo
->blksize
) {
531 bool trim_map
= false;
534 xfs_trans_brelse(NULL
, bp
);
539 lock_mode
= xfs_ilock_data_map_shared(dp
);
540 error
= xfs_dir2_leaf_readbuf(args
, bufsize
, map_info
,
541 &curoff
, &bp
, trim_map
);
542 xfs_iunlock(dp
, lock_mode
);
543 if (error
|| !map_info
->map_valid
)
547 * Having done a read, we need to set a new offset.
549 newoff
= xfs_dir2_db_off_to_byte(geo
,
552 * Start of the current block.
557 * Make sure we're in the right block.
559 else if (curoff
> newoff
)
560 ASSERT(xfs_dir2_byte_to_db(geo
, curoff
) ==
563 xfs_dir3_data_check(dp
, bp
);
565 * Find our position in the block.
567 ptr
= (char *)dp
->d_ops
->data_entry_p(hdr
);
568 byteoff
= xfs_dir2_byte_to_off(geo
, curoff
);
570 * Skip past the header.
573 curoff
+= dp
->d_ops
->data_entry_offset
;
575 * Skip past entries until we reach our offset.
578 while ((char *)ptr
- (char *)hdr
< byteoff
) {
579 dup
= (xfs_dir2_data_unused_t
*)ptr
;
581 if (be16_to_cpu(dup
->freetag
)
582 == XFS_DIR2_DATA_FREE_TAG
) {
584 length
= be16_to_cpu(dup
->length
);
588 dep
= (xfs_dir2_data_entry_t
*)ptr
;
590 dp
->d_ops
->data_entsize(dep
->namelen
);
594 * Now set our real offset.
597 xfs_dir2_db_off_to_byte(geo
,
598 xfs_dir2_byte_to_db(geo
, curoff
),
599 (char *)ptr
- (char *)hdr
);
600 if (ptr
>= (char *)hdr
+ geo
->blksize
) {
606 * We have a pointer to an entry.
609 dup
= (xfs_dir2_data_unused_t
*)ptr
;
611 * No, it's unused, skip over it.
613 if (be16_to_cpu(dup
->freetag
) == XFS_DIR2_DATA_FREE_TAG
) {
614 length
= be16_to_cpu(dup
->length
);
620 dep
= (xfs_dir2_data_entry_t
*)ptr
;
621 length
= dp
->d_ops
->data_entsize(dep
->namelen
);
622 filetype
= dp
->d_ops
->data_get_ftype(dep
);
624 ctx
->pos
= xfs_dir2_byte_to_dataptr(curoff
) & 0x7fffffff;
625 if (!dir_emit(ctx
, (char *)dep
->name
, dep
->namelen
,
626 be64_to_cpu(dep
->inumber
),
627 xfs_dir3_get_dtype(dp
->i_mount
, filetype
)))
631 * Advance to next entry in the block.
635 /* bufsize may have just been a guess; don't go negative */
636 bufsize
= bufsize
> length
? bufsize
- length
: 0;
640 * All done. Set output offset value to current offset.
642 if (curoff
> xfs_dir2_dataptr_to_byte(XFS_DIR2_MAX_DATAPTR
))
643 ctx
->pos
= XFS_DIR2_MAX_DATAPTR
& 0x7fffffff;
645 ctx
->pos
= xfs_dir2_byte_to_dataptr(curoff
) & 0x7fffffff;
648 xfs_trans_brelse(NULL
, bp
);
657 struct xfs_inode
*dp
,
658 struct dir_context
*ctx
,
661 struct xfs_da_args args
= { NULL
};
665 trace_xfs_readdir(dp
);
667 if (XFS_FORCED_SHUTDOWN(dp
->i_mount
))
670 ASSERT(S_ISDIR(VFS_I(dp
)->i_mode
));
671 XFS_STATS_INC(dp
->i_mount
, xs_dir_getdents
);
674 args
.geo
= dp
->i_mount
->m_dir_geo
;
676 if (dp
->i_d
.di_format
== XFS_DINODE_FMT_LOCAL
)
677 rval
= xfs_dir2_sf_getdents(&args
, ctx
);
678 else if ((rval
= xfs_dir2_isblock(&args
, &v
)))
681 rval
= xfs_dir2_block_getdents(&args
, ctx
);
683 rval
= xfs_dir2_leaf_getdents(&args
, ctx
, bufsize
);