1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2001,2005 Silicon Graphics, Inc.
6 #include "libxfs_priv.h"
8 #include "xfs_format.h"
9 #include "xfs_log_format.h"
10 #include "xfs_trans_resv.h"
11 #include "xfs_mount.h"
12 #include "xfs_defer.h"
13 #include "xfs_da_format.h"
14 #include "xfs_da_btree.h"
15 #include "xfs_inode.h"
16 #include "xfs_trans.h"
19 #include "xfs_dir2_priv.h"
20 #include "xfs_ialloc.h"
21 #include "xfs_errortag.h"
22 #include "xfs_trace.h"
24 struct xfs_name xfs_name_dotdot
= { (unsigned char *)"..", 2, XFS_DIR3_FT_DIR
};
27 * Convert inode mode to directory entry filetype
33 switch (mode
& S_IFMT
) {
35 return XFS_DIR3_FT_REG_FILE
;
37 return XFS_DIR3_FT_DIR
;
39 return XFS_DIR3_FT_CHRDEV
;
41 return XFS_DIR3_FT_BLKDEV
;
43 return XFS_DIR3_FT_FIFO
;
45 return XFS_DIR3_FT_SOCK
;
47 return XFS_DIR3_FT_SYMLINK
;
49 return XFS_DIR3_FT_UNKNOWN
;
54 * ASCII case-insensitive (ie. A-Z) support for directories that was
58 xfs_ascii_ci_hashname(
59 struct xfs_name
*name
)
64 for (i
= 0, hash
= 0; i
< name
->len
; i
++)
65 hash
= tolower(name
->name
[i
]) ^ rol32(hash
, 7);
71 xfs_ascii_ci_compname(
72 struct xfs_da_args
*args
,
73 const unsigned char *name
,
76 enum xfs_dacmp result
;
79 if (args
->namelen
!= len
)
80 return XFS_CMP_DIFFERENT
;
82 result
= XFS_CMP_EXACT
;
83 for (i
= 0; i
< len
; i
++) {
84 if (args
->name
[i
] == name
[i
])
86 if (tolower(args
->name
[i
]) != tolower(name
[i
]))
87 return XFS_CMP_DIFFERENT
;
88 result
= XFS_CMP_CASE
;
94 static const struct xfs_nameops xfs_ascii_ci_nameops
= {
95 .hashname
= xfs_ascii_ci_hashname
,
96 .compname
= xfs_ascii_ci_compname
,
101 struct xfs_mount
*mp
)
103 struct xfs_da_geometry
*dageo
;
107 ASSERT(mp
->m_sb
.sb_versionnum
& XFS_SB_VERSION_DIRV2BIT
);
108 ASSERT(xfs_dir2_dirblock_bytes(&mp
->m_sb
) <= XFS_MAX_BLOCKSIZE
);
110 mp
->m_dir_inode_ops
= xfs_dir_get_ops(mp
, NULL
);
111 mp
->m_nondir_inode_ops
= xfs_nondir_get_ops(mp
, NULL
);
113 nodehdr_size
= mp
->m_dir_inode_ops
->node_hdr_size
;
114 mp
->m_dir_geo
= kmem_zalloc(sizeof(struct xfs_da_geometry
),
115 KM_SLEEP
| KM_MAYFAIL
);
116 mp
->m_attr_geo
= kmem_zalloc(sizeof(struct xfs_da_geometry
),
117 KM_SLEEP
| KM_MAYFAIL
);
118 if (!mp
->m_dir_geo
|| !mp
->m_attr_geo
) {
119 kmem_free(mp
->m_dir_geo
);
120 kmem_free(mp
->m_attr_geo
);
124 /* set up directory geometry */
125 dageo
= mp
->m_dir_geo
;
126 dageo
->blklog
= mp
->m_sb
.sb_blocklog
+ mp
->m_sb
.sb_dirblklog
;
127 dageo
->fsblog
= mp
->m_sb
.sb_blocklog
;
128 dageo
->blksize
= xfs_dir2_dirblock_bytes(&mp
->m_sb
);
129 dageo
->fsbcount
= 1 << mp
->m_sb
.sb_dirblklog
;
132 * Now we've set up the block conversion variables, we can calculate the
133 * segment block constants using the geometry structure.
135 dageo
->datablk
= xfs_dir2_byte_to_da(dageo
, XFS_DIR2_DATA_OFFSET
);
136 dageo
->leafblk
= xfs_dir2_byte_to_da(dageo
, XFS_DIR2_LEAF_OFFSET
);
137 dageo
->freeblk
= xfs_dir2_byte_to_da(dageo
, XFS_DIR2_FREE_OFFSET
);
138 dageo
->node_ents
= (dageo
->blksize
- nodehdr_size
) /
139 (uint
)sizeof(xfs_da_node_entry_t
);
140 dageo
->magicpct
= (dageo
->blksize
* 37) / 100;
142 /* set up attribute geometry - single fsb only */
143 dageo
= mp
->m_attr_geo
;
144 dageo
->blklog
= mp
->m_sb
.sb_blocklog
;
145 dageo
->fsblog
= mp
->m_sb
.sb_blocklog
;
146 dageo
->blksize
= 1 << dageo
->blklog
;
148 dageo
->node_ents
= (dageo
->blksize
- nodehdr_size
) /
149 (uint
)sizeof(xfs_da_node_entry_t
);
150 dageo
->magicpct
= (dageo
->blksize
* 37) / 100;
152 if (xfs_sb_version_hasasciici(&mp
->m_sb
))
153 mp
->m_dirnameops
= &xfs_ascii_ci_nameops
;
155 mp
->m_dirnameops
= &xfs_default_nameops
;
162 struct xfs_mount
*mp
)
164 kmem_free(mp
->m_dir_geo
);
165 kmem_free(mp
->m_attr_geo
);
169 * Return 1 if directory contains only "." and "..".
175 xfs_dir2_sf_hdr_t
*sfp
;
177 ASSERT(S_ISDIR(VFS_I(dp
)->i_mode
));
178 if (dp
->i_d
.di_size
== 0) /* might happen during shutdown. */
180 if (dp
->i_d
.di_size
> XFS_IFORK_DSIZE(dp
))
182 sfp
= (xfs_dir2_sf_hdr_t
*)dp
->i_df
.if_u1
.if_data
;
187 * Validate a given inode number.
190 xfs_dir_ino_validate(
194 bool ino_ok
= xfs_verify_dir_ino(mp
, ino
);
196 if (unlikely(XFS_TEST_ERROR(!ino_ok
, mp
, XFS_ERRTAG_DIR_INO_VALIDATE
))) {
197 xfs_warn(mp
, "Invalid inode number 0x%Lx",
198 (unsigned long long) ino
);
199 XFS_ERROR_REPORT("xfs_dir_ino_validate", XFS_ERRLEVEL_LOW
, mp
);
200 return -EFSCORRUPTED
;
206 * Initialize a directory with its "." and ".." entries.
214 struct xfs_da_args
*args
;
217 ASSERT(S_ISDIR(VFS_I(dp
)->i_mode
));
218 error
= xfs_dir_ino_validate(tp
->t_mountp
, pdp
->i_ino
);
222 args
= kmem_zalloc(sizeof(*args
), KM_SLEEP
| KM_NOFS
);
226 args
->geo
= dp
->i_mount
->m_dir_geo
;
229 error
= xfs_dir2_sf_create(args
, pdp
->i_ino
);
235 * Enter a name in a directory, or check for available space.
236 * If inum is 0, only the available space test is performed.
240 struct xfs_trans
*tp
,
241 struct xfs_inode
*dp
,
242 struct xfs_name
*name
,
243 xfs_ino_t inum
, /* new entry inode number */
244 xfs_extlen_t total
) /* bmap's total block count */
246 struct xfs_da_args
*args
;
248 int v
; /* type-checking value */
250 ASSERT(S_ISDIR(VFS_I(dp
)->i_mode
));
253 rval
= xfs_dir_ino_validate(tp
->t_mountp
, inum
);
256 XFS_STATS_INC(dp
->i_mount
, xs_dir_create
);
259 args
= kmem_zalloc(sizeof(*args
), KM_SLEEP
| KM_NOFS
);
263 args
->geo
= dp
->i_mount
->m_dir_geo
;
264 args
->name
= name
->name
;
265 args
->namelen
= name
->len
;
266 args
->filetype
= name
->type
;
267 args
->hashval
= dp
->i_mount
->m_dirnameops
->hashname(name
);
268 args
->inumber
= inum
;
271 args
->whichfork
= XFS_DATA_FORK
;
273 args
->op_flags
= XFS_DA_OP_ADDNAME
| XFS_DA_OP_OKNOENT
;
275 args
->op_flags
|= XFS_DA_OP_JUSTCHECK
;
277 if (dp
->i_d
.di_format
== XFS_DINODE_FMT_LOCAL
) {
278 rval
= xfs_dir2_sf_addname(args
);
282 rval
= xfs_dir2_isblock(args
, &v
);
286 rval
= xfs_dir2_block_addname(args
);
290 rval
= xfs_dir2_isleaf(args
, &v
);
294 rval
= xfs_dir2_leaf_addname(args
);
296 rval
= xfs_dir2_node_addname(args
);
304 * If doing a CI lookup and case-insensitive match, dup actual name into
305 * args.value. Return EEXIST for success (ie. name found) or an error.
308 xfs_dir_cilookup_result(
309 struct xfs_da_args
*args
,
310 const unsigned char *name
,
313 if (args
->cmpresult
== XFS_CMP_DIFFERENT
)
315 if (args
->cmpresult
!= XFS_CMP_CASE
||
316 !(args
->op_flags
& XFS_DA_OP_CILOOKUP
))
319 args
->value
= kmem_alloc(len
, KM_NOFS
| KM_MAYFAIL
);
323 memcpy(args
->value
, name
, len
);
324 args
->valuelen
= len
;
329 * Lookup a name in a directory, give back the inode number.
330 * If ci_name is not NULL, returns the actual name in ci_name if it differs
331 * to name, or ci_name->name is set to NULL for an exact match.
338 struct xfs_name
*name
,
339 xfs_ino_t
*inum
, /* out: inode number */
340 struct xfs_name
*ci_name
) /* out: actual name if CI match */
342 struct xfs_da_args
*args
;
344 int v
; /* type-checking value */
347 ASSERT(S_ISDIR(VFS_I(dp
)->i_mode
));
348 XFS_STATS_INC(dp
->i_mount
, xs_dir_lookup
);
351 * We need to use KM_NOFS here so that lockdep will not throw false
352 * positive deadlock warnings on a non-transactional lookup path. It is
353 * safe to recurse into inode recalim in that case, but lockdep can't
354 * easily be taught about it. Hence KM_NOFS avoids having to add more
355 * lockdep Doing this avoids having to add a bunch of lockdep class
356 * annotations into the reclaim path for the ilock.
358 args
= kmem_zalloc(sizeof(*args
), KM_SLEEP
| KM_NOFS
);
359 args
->geo
= dp
->i_mount
->m_dir_geo
;
360 args
->name
= name
->name
;
361 args
->namelen
= name
->len
;
362 args
->filetype
= name
->type
;
363 args
->hashval
= dp
->i_mount
->m_dirnameops
->hashname(name
);
365 args
->whichfork
= XFS_DATA_FORK
;
367 args
->op_flags
= XFS_DA_OP_OKNOENT
;
369 args
->op_flags
|= XFS_DA_OP_CILOOKUP
;
371 lock_mode
= xfs_ilock_data_map_shared(dp
);
372 if (dp
->i_d
.di_format
== XFS_DINODE_FMT_LOCAL
) {
373 rval
= xfs_dir2_sf_lookup(args
);
377 rval
= xfs_dir2_isblock(args
, &v
);
381 rval
= xfs_dir2_block_lookup(args
);
385 rval
= xfs_dir2_isleaf(args
, &v
);
389 rval
= xfs_dir2_leaf_lookup(args
);
391 rval
= xfs_dir2_node_lookup(args
);
397 *inum
= args
->inumber
;
399 ci_name
->name
= args
->value
;
400 ci_name
->len
= args
->valuelen
;
404 xfs_iunlock(dp
, lock_mode
);
410 * Remove an entry from a directory.
414 struct xfs_trans
*tp
,
415 struct xfs_inode
*dp
,
416 struct xfs_name
*name
,
418 xfs_extlen_t total
) /* bmap's total block count */
420 struct xfs_da_args
*args
;
422 int v
; /* type-checking value */
424 ASSERT(S_ISDIR(VFS_I(dp
)->i_mode
));
425 XFS_STATS_INC(dp
->i_mount
, xs_dir_remove
);
427 args
= kmem_zalloc(sizeof(*args
), KM_SLEEP
| KM_NOFS
);
431 args
->geo
= dp
->i_mount
->m_dir_geo
;
432 args
->name
= name
->name
;
433 args
->namelen
= name
->len
;
434 args
->filetype
= name
->type
;
435 args
->hashval
= dp
->i_mount
->m_dirnameops
->hashname(name
);
439 args
->whichfork
= XFS_DATA_FORK
;
442 if (dp
->i_d
.di_format
== XFS_DINODE_FMT_LOCAL
) {
443 rval
= xfs_dir2_sf_removename(args
);
447 rval
= xfs_dir2_isblock(args
, &v
);
451 rval
= xfs_dir2_block_removename(args
);
455 rval
= xfs_dir2_isleaf(args
, &v
);
459 rval
= xfs_dir2_leaf_removename(args
);
461 rval
= xfs_dir2_node_removename(args
);
468 * Replace the inode number of a directory entry.
472 struct xfs_trans
*tp
,
473 struct xfs_inode
*dp
,
474 struct xfs_name
*name
, /* name of entry to replace */
475 xfs_ino_t inum
, /* new inode number */
476 xfs_extlen_t total
) /* bmap's total block count */
478 struct xfs_da_args
*args
;
480 int v
; /* type-checking value */
482 ASSERT(S_ISDIR(VFS_I(dp
)->i_mode
));
484 rval
= xfs_dir_ino_validate(tp
->t_mountp
, inum
);
488 args
= kmem_zalloc(sizeof(*args
), KM_SLEEP
| KM_NOFS
);
492 args
->geo
= dp
->i_mount
->m_dir_geo
;
493 args
->name
= name
->name
;
494 args
->namelen
= name
->len
;
495 args
->filetype
= name
->type
;
496 args
->hashval
= dp
->i_mount
->m_dirnameops
->hashname(name
);
497 args
->inumber
= inum
;
500 args
->whichfork
= XFS_DATA_FORK
;
503 if (dp
->i_d
.di_format
== XFS_DINODE_FMT_LOCAL
) {
504 rval
= xfs_dir2_sf_replace(args
);
508 rval
= xfs_dir2_isblock(args
, &v
);
512 rval
= xfs_dir2_block_replace(args
);
516 rval
= xfs_dir2_isleaf(args
, &v
);
520 rval
= xfs_dir2_leaf_replace(args
);
522 rval
= xfs_dir2_node_replace(args
);
529 * See if this entry can be added to the directory without allocating space.
535 struct xfs_name
*name
) /* name of entry to add */
537 return xfs_dir_createname(tp
, dp
, name
, 0, 0);
545 * Add a block to the directory.
547 * This routine is for data and free blocks, not leaf/node blocks which are
548 * handled by xfs_da_grow_inode.
552 struct xfs_da_args
*args
,
553 int space
, /* v2 dir's space XFS_DIR2_xxx_SPACE */
554 xfs_dir2_db_t
*dbp
) /* out: block number added */
556 struct xfs_inode
*dp
= args
->dp
;
557 struct xfs_mount
*mp
= dp
->i_mount
;
558 xfs_fileoff_t bno
; /* directory offset of new block */
559 int count
; /* count of filesystem blocks */
562 trace_xfs_dir2_grow_inode(args
, space
);
565 * Set lowest possible block in the space requested.
567 bno
= XFS_B_TO_FSBT(mp
, space
* XFS_DIR2_SPACE_SIZE
);
568 count
= args
->geo
->fsbcount
;
570 error
= xfs_da_grow_inode_int(args
, &bno
, count
);
574 *dbp
= xfs_dir2_da_to_db(args
->geo
, (xfs_dablk_t
)bno
);
577 * Update file's size if this is the data space and it grew.
579 if (space
== XFS_DIR2_DATA_SPACE
) {
580 xfs_fsize_t size
; /* directory file (data) size */
582 size
= XFS_FSB_TO_B(mp
, bno
+ count
);
583 if (size
> dp
->i_d
.di_size
) {
584 dp
->i_d
.di_size
= size
;
585 xfs_trans_log_inode(args
->trans
, dp
, XFS_ILOG_CORE
);
592 * See if the directory is a single-block form directory.
596 struct xfs_da_args
*args
,
597 int *vp
) /* out: 1 is block, 0 is not block */
599 xfs_fileoff_t last
; /* last file offset */
602 if ((rval
= xfs_bmap_last_offset(args
->dp
, &last
, XFS_DATA_FORK
)))
604 rval
= XFS_FSB_TO_B(args
->dp
->i_mount
, last
) == args
->geo
->blksize
;
605 if (rval
!= 0 && args
->dp
->i_d
.di_size
!= args
->geo
->blksize
)
606 return -EFSCORRUPTED
;
612 * See if the directory is a single-leaf form directory.
616 struct xfs_da_args
*args
,
617 int *vp
) /* out: 1 is block, 0 is not block */
619 xfs_fileoff_t last
; /* last file offset */
622 if ((rval
= xfs_bmap_last_offset(args
->dp
, &last
, XFS_DATA_FORK
)))
624 *vp
= last
== args
->geo
->leafblk
+ args
->geo
->fsbcount
;
629 * Remove the given block from the directory.
630 * This routine is used for data and free blocks, leaf/node are done
631 * by xfs_da_shrink_inode.
634 xfs_dir2_shrink_inode(
635 struct xfs_da_args
*args
,
639 xfs_fileoff_t bno
; /* directory file offset */
640 xfs_dablk_t da
; /* directory file offset */
641 int done
; /* bunmap is finished */
642 struct xfs_inode
*dp
;
644 struct xfs_mount
*mp
;
645 struct xfs_trans
*tp
;
647 trace_xfs_dir2_shrink_inode(args
, db
);
652 da
= xfs_dir2_db_to_da(args
->geo
, db
);
654 /* Unmap the fsblock(s). */
655 error
= xfs_bunmapi(tp
, dp
, da
, args
->geo
->fsbcount
, 0, 0, &done
);
658 * ENOSPC actually can happen if we're in a removename with no
659 * space reservation, and the resulting block removal would
660 * cause a bmap btree split or conversion from extents to btree.
661 * This can only happen for un-fragmented directory blocks,
662 * since you need to be punching out the middle of an extent.
663 * In this case we need to leave the block in the file, and not
664 * binval it. So the block has to be in a consistent empty
665 * state and appropriately logged. We don't free up the buffer,
666 * the caller can tell it hasn't happened since it got an error
673 * Invalidate the buffer from the transaction.
675 xfs_trans_binval(tp
, bp
);
677 * If it's not a data block, we're done.
679 if (db
>= xfs_dir2_byte_to_db(args
->geo
, XFS_DIR2_LEAF_OFFSET
))
682 * If the block isn't the last one in the directory, we're done.
684 if (dp
->i_d
.di_size
> xfs_dir2_db_off_to_byte(args
->geo
, db
+ 1, 0))
687 if ((error
= xfs_bmap_last_before(tp
, dp
, &bno
, XFS_DATA_FORK
))) {
689 * This can't really happen unless there's kernel corruption.
693 if (db
== args
->geo
->datablk
)
698 * Set the size to the new last block.
700 dp
->i_d
.di_size
= XFS_FSB_TO_B(mp
, bno
);
701 xfs_trans_log_inode(tp
, dp
, XFS_ILOG_CORE
);
705 /* Returns true if the directory entry name is valid. */
712 * MAXNAMELEN includes the trailing null, but (name/length) leave it
713 * out, so use >= for the length check.
715 if (length
>= MAXNAMELEN
)
718 /* There shouldn't be any slashes or nulls here */
719 return !memchr(name
, '/', length
) && !memchr(name
, 0, length
);