2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
26 #include "err_protos.h"
33 static struct cred zerocr
;
34 static struct fsxattr zerofsx
;
35 static xfs_ino_t orphanage_ino
;
37 static struct xfs_name xfs_name_dot
= {(unsigned char *)".", 1};
40 * Data structures used to keep track of directories where the ".."
41 * entries are updated. These must be rebuilt after the initial pass
43 typedef struct dotdot_update
{
44 struct dotdot_update
*next
;
45 ino_tree_node_t
*irec
;
50 static dotdot_update_t
*dotdot_update_list
;
51 static int dotdot_update
;
56 ino_tree_node_t
*irec
,
59 dotdot_update_t
*dir
= malloc(sizeof(dotdot_update_t
));
62 do_error(_("malloc failed add_dotdot_update (%zu bytes)\n"),
63 sizeof(dotdot_update_t
));
65 dir
->next
= dotdot_update_list
;
68 dir
->ino_offset
= ino_offset
;
70 dotdot_update_list
= dir
;
74 * Data structures and routines to keep track of directory entries
75 * and whether their leaf entry has been seen. Also used for name
76 * duplicate checking and rebuilding step if required.
78 typedef struct dir_hash_ent
{
79 struct dir_hash_ent
*nextbyaddr
; /* next in addr bucket */
80 struct dir_hash_ent
*nextbyhash
; /* next in name bucket */
81 struct dir_hash_ent
*nextbyorder
; /* next in order added */
82 xfs_dahash_t hashval
; /* hash value of name */
83 __uint32_t address
; /* offset of data entry */
84 xfs_ino_t inum
; /* inode num of entry */
85 short junkit
; /* name starts with / */
86 short seen
; /* have seen leaf entry */
90 typedef struct dir_hash_tab
{
91 int size
; /* size of hash tables */
92 int names_duped
; /* 1 = ent names malloced */
93 dir_hash_ent_t
*first
; /* ptr to first added entry */
94 dir_hash_ent_t
*last
; /* ptr to last added entry */
95 dir_hash_ent_t
**byhash
; /* ptr to name hash buckets */
96 dir_hash_ent_t
**byaddr
; /* ptr to addr hash buckets */
99 #define DIR_HASH_TAB_SIZE(n) \
100 (sizeof(dir_hash_tab_t) + (sizeof(dir_hash_ent_t *) * (n) * 2))
101 #define DIR_HASH_FUNC(t,a) ((a) % (t)->size)
104 * Track the contents of the freespace table in a directory.
106 typedef struct freetab
{
107 int naents
; /* expected number of data blocks */
108 int nents
; /* number of data blocks processed */
110 xfs_dir2_data_off_t v
;
114 #define FREETAB_SIZE(n) \
115 (offsetof(freetab_t, ents) + (sizeof(struct freetab_ent) * (n)))
117 #define DIR_HASH_CK_OK 0
118 #define DIR_HASH_CK_DUPLEAF 1
119 #define DIR_HASH_CK_BADHASH 2
120 #define DIR_HASH_CK_NODATA 3
121 #define DIR_HASH_CK_NOLEAF 4
122 #define DIR_HASH_CK_BADSTALE 5
123 #define DIR_HASH_CK_TOTAL 6
126 * Returns 0 if the name already exists (ie. a duplicate)
131 dir_hash_tab_t
*hashtab
,
137 xfs_dahash_t hash
= 0;
143 struct xfs_name xname
;
145 ASSERT(!hashtab
->names_duped
);
150 junk
= name
[0] == '/';
151 byaddr
= DIR_HASH_FUNC(hashtab
, addr
);
155 hash
= mp
->m_dirnameops
->hashname(&xname
);
156 byhash
= DIR_HASH_FUNC(hashtab
, hash
);
159 * search hash bucket for existing name.
161 for (p
= hashtab
->byhash
[byhash
]; p
; p
= p
->nextbyhash
) {
162 if (p
->hashval
== hash
&& p
->name
.len
== namelen
) {
163 if (memcmp(p
->name
.name
, name
, namelen
) == 0) {
172 if ((p
= malloc(sizeof(*p
))) == NULL
)
173 do_error(_("malloc failed in dir_hash_add (%zu bytes)\n"),
176 p
->nextbyaddr
= hashtab
->byaddr
[byaddr
];
177 hashtab
->byaddr
[byaddr
] = p
;
179 hashtab
->last
->nextbyorder
= p
;
182 p
->nextbyorder
= NULL
;
185 if (!(p
->junkit
= junk
)) {
187 p
->nextbyhash
= hashtab
->byhash
[byhash
];
188 hashtab
->byhash
[byhash
] = p
;
199 * checks to see if any data entries are not in the leaf blocks
203 dir_hash_tab_t
*hashtab
)
208 for (i
= 0; i
< hashtab
->size
; i
++) {
209 for (p
= hashtab
->byaddr
[i
]; p
; p
= p
->nextbyaddr
) {
219 dir_hash_tab_t
*hashtab
,
223 static char *seevalstr
[DIR_HASH_CK_TOTAL
];
227 seevalstr
[DIR_HASH_CK_OK
] = _("ok");
228 seevalstr
[DIR_HASH_CK_DUPLEAF
] = _("duplicate leaf");
229 seevalstr
[DIR_HASH_CK_BADHASH
] = _("hash value mismatch");
230 seevalstr
[DIR_HASH_CK_NODATA
] = _("no data entry");
231 seevalstr
[DIR_HASH_CK_NOLEAF
] = _("no leaf entry");
232 seevalstr
[DIR_HASH_CK_BADSTALE
] = _("bad stale count");
236 if (seeval
== DIR_HASH_CK_OK
&& dir_hash_unseen(hashtab
))
237 seeval
= DIR_HASH_CK_NOLEAF
;
238 if (seeval
== DIR_HASH_CK_OK
)
240 do_warn(_("bad hash table for directory inode %" PRIu64
" (%s): "),
241 ip
->i_ino
, seevalstr
[seeval
]);
243 do_warn(_("rebuilding\n"));
245 do_warn(_("would rebuild\n"));
251 dir_hash_tab_t
*hashtab
)
257 for (i
= 0; i
< hashtab
->size
; i
++) {
258 for (p
= hashtab
->byaddr
[i
]; p
; p
= n
) {
260 if (hashtab
->names_duped
)
261 free((void *)p
->name
.name
);
268 static dir_hash_tab_t
*
272 dir_hash_tab_t
*hashtab
;
275 hsize
= size
/ (16 * 4);
280 if ((hashtab
= calloc(DIR_HASH_TAB_SIZE(hsize
), 1)) == NULL
)
281 do_error(_("calloc failed in dir_hash_init\n"));
282 hashtab
->size
= hsize
;
283 hashtab
->byhash
= (dir_hash_ent_t
**)((char *)hashtab
+
284 sizeof(dir_hash_tab_t
));
285 hashtab
->byaddr
= (dir_hash_ent_t
**)((char *)hashtab
+
286 sizeof(dir_hash_tab_t
) + sizeof(dir_hash_ent_t
*) * hsize
);
292 dir_hash_tab_t
*hashtab
,
294 xfs_dir2_dataptr_t addr
)
299 i
= DIR_HASH_FUNC(hashtab
, addr
);
300 for (p
= hashtab
->byaddr
[i
]; p
; p
= p
->nextbyaddr
) {
301 if (p
->address
!= addr
)
304 return DIR_HASH_CK_DUPLEAF
;
305 if (p
->junkit
== 0 && p
->hashval
!= hash
)
306 return DIR_HASH_CK_BADHASH
;
308 return DIR_HASH_CK_OK
;
310 return DIR_HASH_CK_NODATA
;
314 * checks to make sure leafs match a data entry, and that the stale
319 dir_hash_tab_t
*hashtab
,
320 xfs_dir2_leaf_entry_t
*ents
,
328 for (i
= j
= 0; i
< count
; i
++) {
329 if (be32_to_cpu(ents
[i
].address
) == XFS_DIR2_NULL_DATAPTR
) {
333 rval
= dir_hash_see(hashtab
, be32_to_cpu(ents
[i
].hashval
),
334 be32_to_cpu(ents
[i
].address
));
335 if (rval
!= DIR_HASH_CK_OK
)
338 return j
== stale
? DIR_HASH_CK_OK
: DIR_HASH_CK_BADSTALE
;
342 * Convert name pointers into locally allocated memory.
343 * This must only be done after all the entries have been added.
346 dir_hash_dup_names(dir_hash_tab_t
*hashtab
)
351 if (hashtab
->names_duped
)
354 for (p
= hashtab
->first
; p
; p
= p
->nextbyorder
) {
355 name
= malloc(p
->name
.len
);
356 memcpy(name
, p
->name
.name
, p
->name
.len
);
359 hashtab
->names_duped
= 1;
363 * Given a block number in a fork, return the next valid block number
365 * If this is the last block number then NULLFILEOFF is returned.
367 * This was originally in the kernel, but only used in xfs_repair.
371 xfs_trans_t
*tp
, /* transaction pointer */
372 xfs_inode_t
*ip
, /* incore inode */
373 xfs_fileoff_t
*bnop
, /* current block */
374 int whichfork
) /* data or attr fork */
376 xfs_fileoff_t bno
; /* current block */
377 int eof
; /* hit end of file */
378 int error
; /* error return value */
379 xfs_bmbt_irec_t got
; /* current extent value */
380 xfs_ifork_t
*ifp
; /* inode fork pointer */
381 xfs_extnum_t lastx
; /* last extent used */
382 xfs_bmbt_irec_t prev
; /* previous extent value */
384 if (XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_BTREE
&&
385 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_EXTENTS
&&
386 XFS_IFORK_FORMAT(ip
, whichfork
) != XFS_DINODE_FMT_LOCAL
)
388 if (XFS_IFORK_FORMAT(ip
, whichfork
) == XFS_DINODE_FMT_LOCAL
) {
392 ifp
= XFS_IFORK_PTR(ip
, whichfork
);
393 if (!(ifp
->if_flags
& XFS_IFEXTENTS
) &&
394 (error
= xfs_iread_extents(tp
, ip
, whichfork
)))
397 xfs_bmap_search_extents(ip
, bno
, whichfork
, &eof
, &lastx
, &got
, &prev
);
401 *bnop
= got
.br_startoff
< bno
? bno
: got
.br_startoff
;
411 do_error(_("ran out of disk space!\n"));
413 do_error(_("xfs_trans_reserve returned %d\n"), err
);
417 mk_rbmino(xfs_mount_t
*mp
)
427 xfs_bmap_free_t flist
;
429 xfs_bmbt_irec_t map
[XFS_BMAP_MAX_NMAP
];
436 tp
= libxfs_trans_alloc(mp
, 0);
438 if ((i
= libxfs_trans_reserve(tp
, 10, 0, 0, 0, 0)))
441 error
= libxfs_trans_iget(mp
, tp
, mp
->m_sb
.sb_rbmino
, 0, 0, &ip
);
444 _("couldn't iget realtime bitmap inode -- error - %d\n"),
448 vers
= xfs_sb_version_hascrc(&mp
->m_sb
) ? 3 : 1;
449 memset(&ip
->i_d
, 0, xfs_icdinode_size(vers
));
451 ip
->i_d
.di_magic
= XFS_DINODE_MAGIC
;
452 ip
->i_d
.di_mode
= S_IFREG
;
453 ip
->i_d
.di_version
= vers
;
454 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
455 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
457 ip
->i_d
.di_nlink
= 1; /* account for sb ptr */
459 times
= XFS_ICHGTIME_CHG
| XFS_ICHGTIME_MOD
;
460 if (ip
->i_d
.di_version
== 3) {
462 ip
->i_d
.di_changecount
= 1;
464 ip
->i_d
.di_flags2
= 0;
465 ip
->i_d
.di_ino
= mp
->m_sb
.sb_rbmino
;
466 memset(&(ip
->i_d
.di_pad2
[0]), 0, sizeof(ip
->i_d
.di_pad2
));
467 platform_uuid_copy(&ip
->i_d
.di_uuid
, &mp
->m_sb
.sb_uuid
);
468 times
|= XFS_ICHGTIME_CREATE
;
470 libxfs_trans_ichgtime(tp
, ip
, times
);
475 ip
->i_df
.if_flags
= XFS_IFEXTENTS
;
476 ip
->i_df
.if_bytes
= ip
->i_df
.if_real_bytes
= 0;
477 ip
->i_df
.if_u1
.if_extents
= NULL
;
479 ip
->i_d
.di_size
= mp
->m_sb
.sb_rbmblocks
* mp
->m_sb
.sb_blocksize
;
484 libxfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
485 libxfs_trans_ihold(tp
, ip
);
486 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
489 * then allocate blocks for file and fill with zeroes (stolen
492 tp
= libxfs_trans_alloc(mp
, 0);
493 if ((error
= libxfs_trans_reserve(tp
, mp
->m_sb
.sb_rbmblocks
+
494 (XFS_BM_MAXLEVELS(mp
, XFS_DATA_FORK
) - 1), 0, 0, 0, 0)))
497 libxfs_trans_ijoin(tp
, ip
, 0);
499 xfs_bmap_init(&flist
, &first
);
500 while (bno
< mp
->m_sb
.sb_rbmblocks
) {
501 nmap
= XFS_BMAP_MAX_NMAP
;
502 error
= libxfs_bmapi_write(tp
, ip
, bno
,
503 (xfs_extlen_t
)(mp
->m_sb
.sb_rbmblocks
- bno
),
504 0, &first
, mp
->m_sb
.sb_rbmblocks
,
508 _("couldn't allocate realtime bitmap, error = %d\n"),
511 for (i
= 0, ep
= map
; i
< nmap
; i
++, ep
++) {
512 libxfs_device_zero(mp
->m_ddev_targp
,
513 XFS_FSB_TO_DADDR(mp
, ep
->br_startblock
),
514 XFS_FSB_TO_BB(mp
, ep
->br_blockcount
));
515 bno
+= ep
->br_blockcount
;
518 error
= libxfs_bmap_finish(&tp
, &flist
, &committed
);
521 _("allocation of the realtime bitmap failed, error = %d\n"),
524 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
528 fill_rbmino(xfs_mount_t
*mp
)
543 tp
= libxfs_trans_alloc(mp
, 0);
545 if ((error
= libxfs_trans_reserve(tp
, 10, 0, 0, 0, 0)))
548 error
= libxfs_trans_iget(mp
, tp
, mp
->m_sb
.sb_rbmino
, 0, 0, &ip
);
551 _("couldn't iget realtime bitmap inode -- error - %d\n"),
555 while (bno
< mp
->m_sb
.sb_rbmblocks
) {
557 * fill the file one block at a time
560 error
= libxfs_bmapi_write(tp
, ip
, bno
, 1, 0,
561 &first
, 1, &map
, &nmap
, NULL
);
562 if (error
|| nmap
!= 1) {
564 _("couldn't map realtime bitmap block %" PRIu64
", error = %d\n"),
568 ASSERT(map
.br_startblock
!= HOLESTARTBLOCK
);
570 error
= libxfs_trans_read_buf(
572 XFS_FSB_TO_DADDR(mp
, map
.br_startblock
),
573 XFS_FSB_TO_BB(mp
, 1), 1, &bp
, NULL
);
577 _("can't access block %" PRIu64
" (fsbno %" PRIu64
") of realtime bitmap inode %" PRIu64
"\n"),
578 bno
, map
.br_startblock
, mp
->m_sb
.sb_rbmino
);
582 memmove(XFS_BUF_PTR(bp
), bmp
, mp
->m_sb
.sb_blocksize
);
584 libxfs_trans_log_buf(tp
, bp
, 0, mp
->m_sb
.sb_blocksize
- 1);
586 bmp
= (xfs_rtword_t
*)((__psint_t
) bmp
+ mp
->m_sb
.sb_blocksize
);
590 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
595 fill_rsumino(xfs_mount_t
*mp
)
605 xfs_dfiloff_t end_bno
;
610 end_bno
= mp
->m_rsumsize
>> mp
->m_sb
.sb_blocklog
;
612 tp
= libxfs_trans_alloc(mp
, 0);
614 if ((error
= libxfs_trans_reserve(tp
, 10, 0, 0, 0, 0)))
617 error
= libxfs_trans_iget(mp
, tp
, mp
->m_sb
.sb_rsumino
, 0, 0, &ip
);
620 _("couldn't iget realtime summary inode -- error - %d\n"),
624 while (bno
< end_bno
) {
626 * fill the file one block at a time
629 error
= libxfs_bmapi_write(tp
, ip
, bno
, 1, 0,
630 &first
, 1, &map
, &nmap
, NULL
);
631 if (error
|| nmap
!= 1) {
633 _("couldn't map realtime summary inode block %" PRIu64
", error = %d\n"),
637 ASSERT(map
.br_startblock
!= HOLESTARTBLOCK
);
639 error
= libxfs_trans_read_buf(
641 XFS_FSB_TO_DADDR(mp
, map
.br_startblock
),
642 XFS_FSB_TO_BB(mp
, 1), 1, &bp
, NULL
);
646 _("can't access block %" PRIu64
" (fsbno %" PRIu64
") of realtime summary inode %" PRIu64
"\n"),
647 bno
, map
.br_startblock
, mp
->m_sb
.sb_rsumino
);
651 memmove(XFS_BUF_PTR(bp
), smp
, mp
->m_sb
.sb_blocksize
);
653 libxfs_trans_log_buf(tp
, bp
, 0, mp
->m_sb
.sb_blocksize
- 1);
655 smp
= (xfs_suminfo_t
*)((__psint_t
)smp
+ mp
->m_sb
.sb_blocksize
);
659 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
664 mk_rsumino(xfs_mount_t
*mp
)
675 xfs_bmap_free_t flist
;
677 xfs_bmbt_irec_t map
[XFS_BMAP_MAX_NMAP
];
684 tp
= libxfs_trans_alloc(mp
, 0);
686 if ((i
= libxfs_trans_reserve(tp
, 10, XFS_ICHANGE_LOG_RES(mp
), 0,
687 XFS_TRANS_PERM_LOG_RES
, XFS_MKDIR_LOG_COUNT
)))
690 error
= libxfs_trans_iget(mp
, tp
, mp
->m_sb
.sb_rsumino
, 0, 0, &ip
);
693 _("couldn't iget realtime summary inode -- error - %d\n"),
697 vers
= xfs_sb_version_hascrc(&mp
->m_sb
) ? 3 : 1;
698 memset(&ip
->i_d
, 0, xfs_icdinode_size(vers
));
700 ip
->i_d
.di_magic
= XFS_DINODE_MAGIC
;
701 ip
->i_d
.di_mode
= S_IFREG
;
702 ip
->i_d
.di_version
= vers
;
703 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
704 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
706 ip
->i_d
.di_nlink
= 1; /* account for sb ptr */
708 times
= XFS_ICHGTIME_CHG
| XFS_ICHGTIME_MOD
;
709 if (ip
->i_d
.di_version
== 3) {
711 ip
->i_d
.di_changecount
= 1;
713 ip
->i_d
.di_flags2
= 0;
714 ip
->i_d
.di_ino
= mp
->m_sb
.sb_rsumino
;
715 memset(&(ip
->i_d
.di_pad2
[0]), 0, sizeof(ip
->i_d
.di_pad2
));
716 platform_uuid_copy(&ip
->i_d
.di_uuid
, &mp
->m_sb
.sb_uuid
);
717 times
|= XFS_ICHGTIME_CREATE
;
719 libxfs_trans_ichgtime(tp
, ip
, times
);
724 ip
->i_df
.if_flags
= XFS_IFEXTENTS
;
725 ip
->i_df
.if_bytes
= ip
->i_df
.if_real_bytes
= 0;
726 ip
->i_df
.if_u1
.if_extents
= NULL
;
728 ip
->i_d
.di_size
= mp
->m_rsumsize
;
733 libxfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
734 libxfs_trans_ihold(tp
, ip
);
735 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
738 * then allocate blocks for file and fill with zeroes (stolen
741 tp
= libxfs_trans_alloc(mp
, 0);
742 xfs_bmap_init(&flist
, &first
);
744 nsumblocks
= mp
->m_rsumsize
>> mp
->m_sb
.sb_blocklog
;
745 if ((error
= libxfs_trans_reserve(tp
,
746 mp
->m_sb
.sb_rbmblocks
+
747 (XFS_BM_MAXLEVELS(mp
, XFS_DATA_FORK
) - 1),
748 BBTOB(128), 0, XFS_TRANS_PERM_LOG_RES
,
749 XFS_DEFAULT_PERM_LOG_COUNT
)))
752 libxfs_trans_ijoin(tp
, ip
, 0);
754 xfs_bmap_init(&flist
, &first
);
755 while (bno
< nsumblocks
) {
756 nmap
= XFS_BMAP_MAX_NMAP
;
757 error
= libxfs_bmapi_write(tp
, ip
, bno
,
758 (xfs_extlen_t
)(nsumblocks
- bno
),
759 0, &first
, nsumblocks
, map
, &nmap
, &flist
);
762 _("couldn't allocate realtime summary inode, error = %d\n"),
765 for (i
= 0, ep
= map
; i
< nmap
; i
++, ep
++) {
766 libxfs_device_zero(mp
->m_ddev_targp
,
767 XFS_FSB_TO_DADDR(mp
, ep
->br_startblock
),
768 XFS_FSB_TO_BB(mp
, ep
->br_blockcount
));
769 bno
+= ep
->br_blockcount
;
772 error
= libxfs_bmap_finish(&tp
, &flist
, &committed
);
775 _("allocation of the realtime summary ino failed, error = %d\n"),
778 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
782 * makes a new root directory.
785 mk_root_dir(xfs_mount_t
*mp
)
791 const mode_t mode
= 0755;
792 ino_tree_node_t
*irec
;
796 ASSERT(xfs_sb_version_hasdirv2(&mp
->m_sb
));
798 tp
= libxfs_trans_alloc(mp
, 0);
801 if ((i
= libxfs_trans_reserve(tp
, 10, XFS_ICHANGE_LOG_RES(mp
), 0,
802 XFS_TRANS_PERM_LOG_RES
, XFS_MKDIR_LOG_COUNT
)))
805 error
= libxfs_trans_iget(mp
, tp
, mp
->m_sb
.sb_rootino
, 0, 0, &ip
);
807 do_error(_("could not iget root inode -- error - %d\n"), error
);
811 * take care of the core -- initialization from xfs_ialloc()
813 vers
= xfs_sb_version_hascrc(&mp
->m_sb
) ? 3 : 1;
814 memset(&ip
->i_d
, 0, xfs_icdinode_size(vers
));
816 ip
->i_d
.di_magic
= XFS_DINODE_MAGIC
;
817 ip
->i_d
.di_mode
= (__uint16_t
) mode
|S_IFDIR
;
818 ip
->i_d
.di_version
= vers
;
819 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
820 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
822 ip
->i_d
.di_nlink
= 1; /* account for . */
824 times
= XFS_ICHGTIME_CHG
| XFS_ICHGTIME_MOD
;
825 if (ip
->i_d
.di_version
== 3) {
827 ip
->i_d
.di_changecount
= 1;
829 ip
->i_d
.di_flags2
= 0;
830 ip
->i_d
.di_ino
= mp
->m_sb
.sb_rootino
;
831 memset(&(ip
->i_d
.di_pad2
[0]), 0, sizeof(ip
->i_d
.di_pad2
));
832 platform_uuid_copy(&ip
->i_d
.di_uuid
, &mp
->m_sb
.sb_uuid
);
833 times
|= XFS_ICHGTIME_CREATE
;
835 libxfs_trans_ichgtime(tp
, ip
, times
);
837 libxfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
842 ip
->i_df
.if_flags
= XFS_IFEXTENTS
;
843 ip
->i_df
.if_bytes
= ip
->i_df
.if_real_bytes
= 0;
844 ip
->i_df
.if_u1
.if_extents
= NULL
;
849 * initialize the directory
851 libxfs_dir_init(tp
, ip
, ip
);
853 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
855 irec
= find_inode_rec(mp
, XFS_INO_TO_AGNO(mp
, mp
->m_sb
.sb_rootino
),
856 XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_rootino
));
857 set_inode_isadir(irec
, XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_rootino
) -
862 * orphanage name == lost+found
865 mk_orphanage(xfs_mount_t
*mp
)
872 ino_tree_node_t
*irec
;
877 xfs_bmap_free_t flist
;
878 const int mode
= 0755;
880 struct xfs_name xname
;
882 ASSERT(xfs_sb_version_hasdirv2(&mp
->m_sb
));
885 * check for an existing lost+found first, if it exists, return
886 * its inode. Otherwise, we can create it. Bad lost+found inodes
887 * would have been cleared in phase3 and phase4.
890 if ((i
= libxfs_iget(mp
, NULL
, mp
->m_sb
.sb_rootino
, 0, &pip
, 0)))
891 do_error(_("%d - couldn't iget root inode to obtain %s\n"),
894 xname
.name
= (unsigned char *)ORPHANAGE
;
895 xname
.len
= strlen(ORPHANAGE
);
896 if (libxfs_dir_lookup(NULL
, pip
, &xname
, &ino
, NULL
) == 0)
900 * could not be found, create it
903 tp
= libxfs_trans_alloc(mp
, 0);
904 xfs_bmap_init(&flist
, &first
);
906 nres
= XFS_MKDIR_SPACE_RES(mp
, xname
.len
);
907 if ((i
= libxfs_trans_reserve(tp
, nres
, XFS_MKDIR_LOG_RES(mp
), 0,
908 XFS_TRANS_PERM_LOG_RES
, XFS_MKDIR_LOG_COUNT
)))
912 * use iget/ijoin instead of trans_iget because the ialloc
913 * wrapper can commit the transaction and start a new one
915 /* if ((i = libxfs_iget(mp, NULL, mp->m_sb.sb_rootino, 0, &pip, 0)))
916 do_error(_("%d - couldn't iget root inode to make %s\n"),
919 error
= libxfs_inode_alloc(&tp
, pip
, mode
|S_IFDIR
,
920 1, 0, &zerocr
, &zerofsx
, &ip
);
922 do_error(_("%s inode allocation failed %d\n"),
925 ip
->i_d
.di_nlink
++; /* account for . */
928 irec
= find_inode_rec(mp
,
929 XFS_INO_TO_AGNO(mp
, ino
),
930 XFS_INO_TO_AGINO(mp
, ino
));
931 ino_offset
= get_inode_offset(mp
, ino
, irec
);
934 * Mark the inode allocated to lost+found as used in the AVL tree
935 * so it is not skipped in phase 7
937 set_inode_used(irec
, ino_offset
);
938 add_inode_ref(irec
, ino_offset
);
941 * now that we know the transaction will stay around,
942 * add the root inode to it
944 libxfs_trans_ijoin(tp
, pip
, 0);
947 * create the actual entry
949 error
= libxfs_dir_createname(tp
, pip
, &xname
, ip
->i_ino
, &first
,
953 _("can't make %s, createname error %d\n"),
957 * bump up the link count in the root directory to account
958 * for .. in the new directory
961 add_inode_ref(find_inode_rec(mp
,
962 XFS_INO_TO_AGNO(mp
, mp
->m_sb
.sb_rootino
),
963 XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_rootino
)), 0);
967 libxfs_trans_log_inode(tp
, pip
, XFS_ILOG_CORE
);
968 libxfs_dir_init(tp
, ip
, pip
);
969 libxfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
971 error
= libxfs_bmap_finish(&tp
, &flist
, &committed
);
973 do_error(_("%s directory creation failed -- bmapf error %d\n"),
978 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
979 add_inode_reached(irec
,ino_offset
);
985 * move a file to the orphange.
990 xfs_ino_t ino
, /* inode # to be moved */
991 int isa_dir
) /* 1 if inode is a directory */
993 xfs_inode_t
*orphanage_ip
;
994 xfs_ino_t entry_ino_num
;
998 xfs_bmap_free_t flist
;
1001 unsigned char fname
[MAXPATHLEN
+ 1];
1004 ino_tree_node_t
*irec
;
1006 struct xfs_name xname
;
1008 ASSERT(xfs_sb_version_hasdirv2(&mp
->m_sb
));
1011 xname
.len
= snprintf((char *)fname
, sizeof(fname
), "%llu",
1012 (unsigned long long)ino
);
1014 err
= libxfs_iget(mp
, NULL
, orphanage_ino
, 0, &orphanage_ip
, 0);
1016 do_error(_("%d - couldn't iget orphanage inode\n"), err
);
1018 * Make sure the filename is unique in the lost+found
1021 while (libxfs_dir_lookup(NULL
, orphanage_ip
, &xname
, &entry_ino_num
,
1023 xname
.len
= snprintf((char *)fname
, sizeof(fname
), "%llu.%d",
1024 (unsigned long long)ino
, ++incr
);
1026 tp
= libxfs_trans_alloc(mp
, 0);
1028 if ((err
= libxfs_iget(mp
, NULL
, ino
, 0, &ino_p
, 0)))
1029 do_error(_("%d - couldn't iget disconnected inode\n"), err
);
1032 irec
= find_inode_rec(mp
, XFS_INO_TO_AGNO(mp
, orphanage_ino
),
1033 XFS_INO_TO_AGINO(mp
, orphanage_ino
));
1035 ino_offset
= XFS_INO_TO_AGINO(mp
, orphanage_ino
) -
1037 nres
= XFS_DIRENTER_SPACE_RES(mp
, fnamelen
) +
1038 XFS_DIRENTER_SPACE_RES(mp
, 2);
1039 err
= libxfs_dir_lookup(tp
, ino_p
, &xfs_name_dotdot
,
1040 &entry_ino_num
, NULL
);
1042 ASSERT(err
== ENOENT
);
1044 if ((err
= libxfs_trans_reserve(tp
, nres
,
1045 XFS_RENAME_LOG_RES(mp
), 0,
1046 XFS_TRANS_PERM_LOG_RES
,
1047 XFS_RENAME_LOG_COUNT
)))
1049 _("space reservation failed (%d), filesystem may be out of space\n"),
1052 libxfs_trans_ijoin(tp
, orphanage_ip
, 0);
1053 libxfs_trans_ijoin(tp
, ino_p
, 0);
1055 xfs_bmap_init(&flist
, &first
);
1056 err
= libxfs_dir_createname(tp
, orphanage_ip
, &xname
,
1057 ino
, &first
, &flist
, nres
);
1060 _("name create failed in %s (%d), filesystem may be out of space\n"),
1064 add_inode_ref(irec
, ino_offset
);
1066 orphanage_ip
->i_d
.di_nlink
++;
1067 libxfs_trans_log_inode(tp
, orphanage_ip
, XFS_ILOG_CORE
);
1069 err
= libxfs_dir_createname(tp
, ino_p
, &xfs_name_dotdot
,
1070 orphanage_ino
, &first
, &flist
, nres
);
1073 _("creation of .. entry failed (%d), filesystem may be out of space\n"),
1076 ino_p
->i_d
.di_nlink
++;
1077 libxfs_trans_log_inode(tp
, ino_p
, XFS_ILOG_CORE
);
1079 err
= libxfs_bmap_finish(&tp
, &flist
, &committed
);
1082 _("bmap finish failed (err - %d), filesystem may be out of space\n"),
1085 libxfs_trans_commit(tp
,
1086 XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
1088 if ((err
= libxfs_trans_reserve(tp
, nres
,
1089 XFS_RENAME_LOG_RES(mp
), 0,
1090 XFS_TRANS_PERM_LOG_RES
,
1091 XFS_RENAME_LOG_COUNT
)))
1093 _("space reservation failed (%d), filesystem may be out of space\n"),
1096 libxfs_trans_ijoin(tp
, orphanage_ip
, 0);
1097 libxfs_trans_ijoin(tp
, ino_p
, 0);
1099 xfs_bmap_init(&flist
, &first
);
1101 err
= libxfs_dir_createname(tp
, orphanage_ip
, &xname
,
1102 ino
, &first
, &flist
, nres
);
1105 _("name create failed in %s (%d), filesystem may be out of space\n"),
1109 add_inode_ref(irec
, ino_offset
);
1111 orphanage_ip
->i_d
.di_nlink
++;
1112 libxfs_trans_log_inode(tp
, orphanage_ip
, XFS_ILOG_CORE
);
1115 * don't replace .. value if it already points
1116 * to us. that'll pop a libxfs/kernel ASSERT.
1118 if (entry_ino_num
!= orphanage_ino
) {
1119 err
= libxfs_dir_replace(tp
, ino_p
,
1120 &xfs_name_dotdot
, orphanage_ino
,
1121 &first
, &flist
, nres
);
1124 _("name replace op failed (%d), filesystem may be out of space\n"),
1128 err
= libxfs_bmap_finish(&tp
, &flist
, &committed
);
1131 _("bmap finish failed (%d), filesystem may be out of space\n"),
1134 libxfs_trans_commit(tp
,
1135 XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
1140 * use the remove log reservation as that's
1141 * more accurate. we're only creating the
1142 * links, we're not doing the inode allocation
1143 * also accounted for in the create
1145 nres
= XFS_DIRENTER_SPACE_RES(mp
, xname
.len
);
1146 err
= libxfs_trans_reserve(tp
, nres
, XFS_REMOVE_LOG_RES(mp
), 0,
1147 XFS_TRANS_PERM_LOG_RES
, XFS_REMOVE_LOG_COUNT
);
1150 _("space reservation failed (%d), filesystem may be out of space\n"),
1153 libxfs_trans_ijoin(tp
, orphanage_ip
, 0);
1154 libxfs_trans_ijoin(tp
, ino_p
, 0);
1156 xfs_bmap_init(&flist
, &first
);
1157 err
= libxfs_dir_createname(tp
, orphanage_ip
, &xname
, ino
,
1158 &first
, &flist
, nres
);
1161 _("name create failed in %s (%d), filesystem may be out of space\n"),
1165 ino_p
->i_d
.di_nlink
= 1;
1166 libxfs_trans_log_inode(tp
, ino_p
, XFS_ILOG_CORE
);
1168 err
= libxfs_bmap_finish(&tp
, &flist
, &committed
);
1171 _("bmap finish failed (%d), filesystem may be out of space\n"),
1174 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
1185 do_warn(msg
, iname
, ino1
, ino2
);
1188 do_warn(_(", marking entry to be junked\n"));
1192 do_warn(_(", would junk entry\n"));
1197 * Unexpected failure during the rebuild will leave the entries in
1198 * lost+found on the next run
1202 longform_dir2_rebuild(
1206 ino_tree_node_t
*irec
,
1208 dir_hash_tab_t
*hashtab
)
1213 xfs_fileoff_t lastblock
;
1214 xfs_fsblock_t firstblock
;
1215 xfs_bmap_free_t flist
;
1222 * trash directory completely and rebuild from scratch using the
1223 * name/inode pairs in the hash table
1226 do_warn(_("rebuilding directory inode %" PRIu64
"\n"), ino
);
1229 * first attempt to locate the parent inode, if it can't be
1230 * found, set it to the root inode and it'll be moved to the
1231 * orphanage later (the inode number here needs to be valid
1232 * for the libxfs_dir_init() call).
1234 pip
.i_ino
= get_inode_parent(irec
, ino_offset
);
1235 if (pip
.i_ino
== NULLFSINO
)
1236 pip
.i_ino
= mp
->m_sb
.sb_rootino
;
1238 xfs_bmap_init(&flist
, &firstblock
);
1240 tp
= libxfs_trans_alloc(mp
, 0);
1241 nres
= XFS_REMOVE_SPACE_RES(mp
);
1242 error
= libxfs_trans_reserve(tp
, nres
, XFS_REMOVE_LOG_RES(mp
), 0,
1243 XFS_TRANS_PERM_LOG_RES
, XFS_REMOVE_LOG_COUNT
);
1246 libxfs_trans_ijoin(tp
, ip
, 0);
1247 libxfs_trans_ihold(tp
, ip
);
1249 if ((error
= libxfs_bmap_last_offset(tp
, ip
, &lastblock
,
1251 do_error(_("xfs_bmap_last_offset failed -- error - %d\n"),
1254 /* free all data, leaf, node and freespace blocks */
1255 error
= libxfs_bunmapi(tp
, ip
, 0, lastblock
, XFS_BMAPI_METADATA
, 0,
1256 &firstblock
, &flist
, &done
);
1258 do_warn(_("xfs_bunmapi failed -- error - %d\n"), error
);
1259 libxfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
|
1266 libxfs_dir_init(tp
, ip
, &pip
);
1268 error
= libxfs_bmap_finish(&tp
, &flist
, &committed
);
1270 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
1272 /* go through the hash list and re-add the inodes */
1274 for (p
= hashtab
->first
; p
; p
= p
->nextbyorder
) {
1276 if (p
->name
.name
[0] == '/' || (p
->name
.name
[0] == '.' &&
1277 (p
->name
.len
== 1 || (p
->name
.len
== 2 &&
1278 p
->name
.name
[1] == '.'))))
1281 tp
= libxfs_trans_alloc(mp
, 0);
1282 nres
= XFS_CREATE_SPACE_RES(mp
, p
->name
.len
);
1283 error
= libxfs_trans_reserve(tp
, nres
, XFS_CREATE_LOG_RES(mp
),
1284 0, XFS_TRANS_PERM_LOG_RES
, XFS_CREATE_LOG_COUNT
);
1287 _("space reservation failed (%d), filesystem may be out of space\n"),
1292 libxfs_trans_ijoin(tp
, ip
, 0);
1293 libxfs_trans_ihold(tp
, ip
);
1295 xfs_bmap_init(&flist
, &firstblock
);
1296 error
= libxfs_dir_createname(tp
, ip
, &p
->name
, p
->inum
,
1297 &firstblock
, &flist
, nres
);
1300 _("name create failed in ino %" PRIu64
" (%d), filesystem may be out of space\n"),
1302 libxfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
|
1307 error
= libxfs_bmap_finish(&tp
, &flist
, &committed
);
1310 _("bmap finish failed (%d), filesystem may be out of space\n"),
1312 libxfs_bmap_cancel(&flist
);
1313 libxfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
|
1318 libxfs_trans_commit(tp
,
1319 XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
1325 * Kill a block in a version 2 inode.
1326 * Makes its own transaction.
1338 xfs_fsblock_t firstblock
;
1339 xfs_bmap_free_t flist
;
1343 tp
= libxfs_trans_alloc(mp
, 0);
1344 nres
= XFS_REMOVE_SPACE_RES(mp
);
1345 error
= libxfs_trans_reserve(tp
, nres
, XFS_REMOVE_LOG_RES(mp
), 0,
1346 XFS_TRANS_PERM_LOG_RES
, XFS_REMOVE_LOG_COUNT
);
1349 libxfs_trans_ijoin(tp
, ip
, 0);
1350 libxfs_trans_ihold(tp
, ip
);
1351 libxfs_trans_bjoin(tp
, bp
);
1352 memset(&args
, 0, sizeof(args
));
1353 xfs_bmap_init(&flist
, &firstblock
);
1356 args
.firstblock
= &firstblock
;
1357 args
.flist
= &flist
;
1358 args
.whichfork
= XFS_DATA_FORK
;
1359 if (da_bno
>= mp
->m_dirleafblk
&& da_bno
< mp
->m_dirfreeblk
)
1360 error
= libxfs_da_shrink_inode(&args
, da_bno
, bp
);
1362 error
= libxfs_dir2_shrink_inode(&args
,
1363 xfs_dir2_da_to_db(mp
, da_bno
), bp
);
1365 do_error(_("shrink_inode failed inode %" PRIu64
" block %u\n"),
1367 libxfs_bmap_finish(&tp
, &flist
, &committed
);
1368 libxfs_trans_commit(tp
, 0);
1372 * process a data block, also checks for .. entry
1373 * and corrects it to match what we think .. should be
1376 longform_dir2_entry_check_data(
1381 ino_tree_node_t
*current_irec
,
1382 int current_ino_offset
,
1383 struct xfs_buf
**bpp
,
1384 dir_hash_tab_t
*hashtab
,
1385 freetab_t
**freetabp
,
1389 xfs_dir2_dataptr_t addr
;
1390 xfs_dir2_leaf_entry_t
*blp
;
1392 xfs_dir2_block_tail_t
*btp
;
1396 xfs_dir2_data_entry_t
*dep
;
1397 xfs_dir2_data_unused_t
*dup
;
1400 xfs_fsblock_t firstblock
;
1401 xfs_bmap_free_t flist
;
1402 char fname
[MAXNAMELEN
+ 1];
1407 ino_tree_node_t
*irec
;
1421 ptr
= (char *)xfs_dir3_data_entry_p(&d
->hdr
);
1423 needscan
= needlog
= 0;
1425 freetab
= *freetabp
;
1427 btp
= xfs_dir2_block_tail_p(mp
, (struct xfs_dir2_data_hdr
*)d
);
1428 blp
= xfs_dir2_block_leaf_p(btp
);
1429 endptr
= (char *)blp
;
1430 if (endptr
> (char *)btp
)
1431 endptr
= (char *)btp
;
1432 if (xfs_sb_version_hascrc(&mp
->m_sb
))
1433 wantmagic
= XFS_DIR3_BLOCK_MAGIC
;
1435 wantmagic
= XFS_DIR2_BLOCK_MAGIC
;
1437 endptr
= (char *)d
+ mp
->m_dirblksize
;
1438 if (xfs_sb_version_hascrc(&mp
->m_sb
))
1439 wantmagic
= XFS_DIR3_DATA_MAGIC
;
1441 wantmagic
= XFS_DIR2_DATA_MAGIC
;
1443 db
= xfs_dir2_da_to_db(mp
, da_bno
);
1445 /* check for data block beyond expected end */
1446 if (freetab
->naents
<= db
) {
1447 struct freetab_ent e
;
1449 *freetabp
= freetab
= realloc(freetab
, FREETAB_SIZE(db
+ 1));
1452 _("realloc failed in longform_dir2_entry_check_data (%zu bytes)\n"),
1453 FREETAB_SIZE(db
+ 1));
1457 for (i
= freetab
->naents
; i
< db
; i
++)
1458 freetab
->ents
[i
] = e
;
1459 freetab
->naents
= db
+ 1;
1462 /* check the data block */
1463 while (ptr
< endptr
) {
1465 /* check for freespace */
1466 dup
= (xfs_dir2_data_unused_t
*)ptr
;
1467 if (XFS_DIR2_DATA_FREE_TAG
== be16_to_cpu(dup
->freetag
)) {
1469 /* check for invalid freespace length */
1470 if (ptr
+ be16_to_cpu(dup
->length
) > endptr
||
1471 be16_to_cpu(dup
->length
) == 0 ||
1472 (be16_to_cpu(dup
->length
) &
1473 (XFS_DIR2_DATA_ALIGN
- 1)))
1476 /* check for invalid tag */
1477 if (be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup
)) !=
1478 (char *)dup
- (char *)d
)
1481 /* check for block with no data entries */
1482 if ((ptr
== (char *)xfs_dir3_data_entry_p(&d
->hdr
)) &&
1483 (ptr
+ be16_to_cpu(dup
->length
) >= endptr
)) {
1489 /* continue at the end of the freespace */
1490 ptr
+= be16_to_cpu(dup
->length
);
1495 /* validate data entry size */
1496 dep
= (xfs_dir2_data_entry_t
*)ptr
;
1497 if (ptr
+ xfs_dir2_data_entsize(dep
->namelen
) > endptr
)
1499 if (be16_to_cpu(*xfs_dir2_data_entry_tag_p(dep
)) !=
1500 (char *)dep
- (char *)d
)
1502 ptr
+= xfs_dir2_data_entsize(dep
->namelen
);
1505 /* did we find an empty or corrupt block? */
1506 if (ptr
!= endptr
) {
1509 _("empty data block %u in directory inode %" PRIu64
": "),
1513 ("corrupt block %u in directory inode %" PRIu64
": "),
1517 do_warn(_("junking block\n"));
1518 dir2_kill_block(mp
, ip
, da_bno
, bp
);
1520 do_warn(_("would junk block\n"));
1523 freetab
->ents
[db
].v
= NULLDATAOFF
;
1528 /* update number of data blocks processed */
1529 if (freetab
->nents
< db
+ 1)
1530 freetab
->nents
= db
+ 1;
1532 tp
= libxfs_trans_alloc(mp
, 0);
1533 error
= libxfs_trans_reserve(tp
, 0, XFS_REMOVE_LOG_RES(mp
), 0,
1534 XFS_TRANS_PERM_LOG_RES
, XFS_REMOVE_LOG_COUNT
);
1537 libxfs_trans_ijoin(tp
, ip
, 0);
1538 libxfs_trans_ihold(tp
, ip
);
1539 libxfs_trans_bjoin(tp
, bp
);
1540 libxfs_trans_bhold(tp
, bp
);
1541 xfs_bmap_init(&flist
, &firstblock
);
1542 if (be32_to_cpu(d
->hdr
.magic
) != wantmagic
) {
1544 _("bad directory block magic # %#x for directory inode %" PRIu64
" block %d: "),
1545 be32_to_cpu(d
->hdr
.magic
), ip
->i_ino
, da_bno
);
1547 do_warn(_("fixing magic # to %#x\n"), wantmagic
);
1548 d
->hdr
.magic
= cpu_to_be32(wantmagic
);
1551 do_warn(_("would fix magic # to %#x\n"), wantmagic
);
1554 ptr
= (char *)xfs_dir3_data_entry_p(&d
->hdr
);
1556 * look at each entry. reference inode pointed to by each
1557 * entry in the incore inode tree.
1558 * if not a directory, set reached flag, increment link count
1559 * if a directory and reached, mark entry as to be deleted.
1560 * if a directory, check to see if recorded parent
1561 * matches current inode #,
1562 * if so, then set reached flag, increment link count
1563 * of current and child dir inodes, push the child
1564 * directory inode onto the directory stack.
1565 * if current inode != parent, then mark entry to be deleted.
1567 while (ptr
< endptr
) {
1568 dup
= (xfs_dir2_data_unused_t
*)ptr
;
1569 if (be16_to_cpu(dup
->freetag
) == XFS_DIR2_DATA_FREE_TAG
) {
1572 _("directory inode %" PRIu64
" block %u has consecutive free entries: "),
1575 do_warn(_("joining together\n"));
1576 len
= be16_to_cpu(dup
->length
);
1577 libxfs_dir2_data_use_free(tp
, bp
, dup
,
1578 ptr
- (char *)d
, len
, &needlog
,
1580 libxfs_dir2_data_make_free(tp
, bp
,
1581 ptr
- (char *)d
, len
, &needlog
,
1584 do_warn(_("would join together\n"));
1586 ptr
+= be16_to_cpu(dup
->length
);
1590 addr
= xfs_dir2_db_off_to_dataptr(mp
, db
, ptr
- (char *)d
);
1591 dep
= (xfs_dir2_data_entry_t
*)ptr
;
1592 ptr
+= xfs_dir2_data_entsize(dep
->namelen
);
1593 inum
= be64_to_cpu(dep
->inumber
);
1596 * skip bogus entries (leading '/'). they'll be deleted
1597 * later. must still log it, else we leak references to
1600 if (dep
->name
[0] == '/') {
1603 libxfs_dir2_data_log_entry(tp
, bp
, dep
);
1607 memmove(fname
, dep
->name
, dep
->namelen
);
1608 fname
[dep
->namelen
] = '\0';
1609 ASSERT(inum
!= NULLFSINO
);
1611 irec
= find_inode_rec(mp
, XFS_INO_TO_AGNO(mp
, inum
),
1612 XFS_INO_TO_AGINO(mp
, inum
));
1616 _("entry \"%s\" in directory inode %" PRIu64
" points to non-existent inode %" PRIu64
""),
1617 fname
, ip
->i_ino
, inum
)) {
1619 libxfs_dir2_data_log_entry(tp
, bp
, dep
);
1623 ino_offset
= XFS_INO_TO_AGINO(mp
, inum
) - irec
->ino_startnum
;
1626 * if it's a free inode, blow out the entry.
1627 * by now, any inode that we think is free
1630 if (is_inode_free(irec
, ino_offset
)) {
1633 _("entry \"%s\" in directory inode %" PRIu64
" points to free inode %" PRIu64
),
1634 fname
, ip
->i_ino
, inum
)) {
1636 libxfs_dir2_data_log_entry(tp
, bp
, dep
);
1642 * check if this inode is lost+found dir in the root
1644 if (inum
== mp
->m_sb
.sb_rootino
&& strcmp(fname
, ORPHANAGE
) == 0) {
1646 * if it's not a directory, trash it
1648 if (!inode_isadir(irec
, ino_offset
)) {
1651 _("%s (ino %" PRIu64
") in root (%" PRIu64
") is not a directory"),
1652 ORPHANAGE
, inum
, ip
->i_ino
)) {
1654 libxfs_dir2_data_log_entry(tp
, bp
, dep
);
1659 * if this is a dup, it will be picked up below,
1660 * otherwise, mark it as the orphanage for later.
1663 orphanage_ino
= inum
;
1666 * check for duplicate names in directory.
1668 if (!dir_hash_add(mp
, hashtab
, addr
, inum
, dep
->namelen
,
1672 _("entry \"%s\" (ino %" PRIu64
") in dir %" PRIu64
" is a duplicate name"),
1673 fname
, inum
, ip
->i_ino
)) {
1675 libxfs_dir2_data_log_entry(tp
, bp
, dep
);
1677 if (inum
== orphanage_ino
)
1683 * if just scanning to rebuild a directory due to a ".."
1684 * update, just continue
1690 * skip the '..' entry since it's checked when the
1691 * directory is reached by something else. if it never
1692 * gets reached, it'll be moved to the orphanage and we'll
1693 * take care of it then. If it doesn't exist at all, the
1694 * directory needs to be rebuilt first before being added
1697 if (dep
->namelen
== 2 && dep
->name
[0] == '.' &&
1698 dep
->name
[1] == '.') {
1700 /* ".." should be in the first block */
1703 _("entry \"%s\" (ino %" PRIu64
") in dir %" PRIu64
" is not in the the first block"), fname
,
1706 libxfs_dir2_data_log_entry(tp
, bp
, dep
);
1711 ASSERT(no_modify
|| !verify_inum(mp
, inum
));
1713 * special case the . entry. we know there's only one
1714 * '.' and only '.' points to itself because bogus entries
1715 * got trashed in phase 3 if there were > 1.
1716 * bump up link count for '.' but don't set reached
1717 * until we're actually reached by another directory
1718 * '..' is already accounted for or will be taken care
1719 * of when directory is moved to orphanage.
1721 if (ip
->i_ino
== inum
) {
1722 ASSERT(dep
->name
[0] == '.' && dep
->namelen
== 1);
1723 add_inode_ref(current_irec
, current_ino_offset
);
1725 dep
!= xfs_dir3_data_entry_p(&d
->hdr
)) {
1726 /* "." should be the first entry */
1729 _("entry \"%s\" in dir %" PRIu64
" is not the first entry"),
1730 fname
, inum
, ip
->i_ino
)) {
1732 libxfs_dir2_data_log_entry(tp
, bp
, dep
);
1739 * skip entries with bogus inumbers if we're in no modify mode
1741 if (no_modify
&& verify_inum(mp
, inum
))
1744 * check easy case first, regular inode, just bump
1745 * the link count and continue
1747 if (!inode_isadir(irec
, ino_offset
)) {
1748 add_inode_reached(irec
, ino_offset
);
1751 parent
= get_inode_parent(irec
, ino_offset
);
1752 ASSERT(parent
!= 0);
1755 * bump up the link counts in parent and child
1756 * directory but if the link doesn't agree with
1757 * the .. in the child, blow out the entry.
1758 * if the directory has already been reached,
1759 * blow away the entry also.
1761 if (is_inode_reached(irec
, ino_offset
)) {
1764 _("entry \"%s\" in dir %" PRIu64
" points to an already connected directory inode %" PRIu64
"\n"),
1765 fname
, ip
->i_ino
, inum
);
1766 } else if (parent
== ip
->i_ino
) {
1767 add_inode_reached(irec
, ino_offset
);
1768 add_inode_ref(current_irec
, current_ino_offset
);
1769 } else if (parent
== NULLFSINO
) {
1770 /* ".." was missing, but this entry refers to it,
1771 so, set it as the parent and mark for rebuild */
1773 _("entry \"%s\" in dir ino %" PRIu64
" doesn't have a .. entry, will set it in ino %" PRIu64
".\n"),
1774 fname
, ip
->i_ino
, inum
);
1775 set_inode_parent(irec
, ino_offset
, ip
->i_ino
);
1776 add_inode_reached(irec
, ino_offset
);
1777 add_inode_ref(current_irec
, current_ino_offset
);
1778 add_dotdot_update(XFS_INO_TO_AGNO(mp
, inum
), irec
,
1783 _("entry \"%s\" in dir inode %" PRIu64
" inconsistent with .. value (%" PRIu64
") in ino %" PRIu64
"\n"),
1784 fname
, ip
->i_ino
, parent
, inum
);
1787 if (inum
== orphanage_ino
)
1793 libxfs_dir2_data_log_entry(tp
, bp
, dep
);
1796 _("\twill clear entry \"%s\"\n"),
1799 do_warn(_("\twould clear entry \"%s\"\n"),
1804 *num_illegal
+= nbad
;
1806 libxfs_dir2_data_freescan(mp
, &d
->hdr
, &needlog
);
1808 libxfs_dir2_data_log_header(tp
, bp
);
1809 libxfs_bmap_finish(&tp
, &flist
, &committed
);
1810 libxfs_trans_commit(tp
, 0);
1811 freetab
->ents
[db
].v
= be16_to_cpu(d
->hdr
.bestfree
[0].length
);
1812 freetab
->ents
[db
].s
= 0;
1816 * Check contents of leaf-form block.
1819 longform_dir2_check_leaf(
1822 dir_hash_tab_t
*hashtab
,
1830 xfs_dir2_leaf_t
*leaf
;
1831 xfs_dir2_leaf_tail_t
*ltp
;
1833 struct xfs_dir2_leaf_entry
*ents
;
1834 struct xfs_dir3_icleaf_hdr leafhdr
;
1836 da_bno
= mp
->m_dirleafblk
;
1837 if (libxfs_da_read_buf(NULL
, ip
, da_bno
, -1, &bp
, XFS_DATA_FORK
,
1838 &xfs_dir3_leaf1_buf_ops
)) {
1840 _("can't read block %u for directory inode %" PRIu64
"\n"),
1845 xfs_dir3_leaf_hdr_from_disk(&leafhdr
, leaf
);
1846 ents
= xfs_dir3_leaf_ents_p(leaf
);
1847 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
1848 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
1849 if (!(leafhdr
.magic
== XFS_DIR2_LEAF1_MAGIC
||
1850 leafhdr
.magic
== XFS_DIR3_LEAF1_MAGIC
) ||
1851 leafhdr
.forw
|| leafhdr
.back
||
1852 leafhdr
.count
< leaf
->hdr
.stale
||
1854 xfs_dir3_max_leaf_ents(mp
, leaf
) ||
1855 (char *)&ents
[leafhdr
.count
] > (char *)bestsp
) {
1857 _("leaf block %u for directory inode %" PRIu64
" bad header\n"),
1862 seeval
= dir_hash_see_all(hashtab
, ents
, leafhdr
.count
, leafhdr
.stale
);
1863 if (dir_hash_check(hashtab
, ip
, seeval
)) {
1867 badtail
= freetab
->nents
!= be32_to_cpu(ltp
->bestcount
);
1868 for (i
= 0; !badtail
&& i
< be32_to_cpu(ltp
->bestcount
); i
++) {
1869 freetab
->ents
[i
].s
= 1;
1870 badtail
= freetab
->ents
[i
].v
!= be16_to_cpu(bestsp
[i
]);
1874 _("leaf block %u for directory inode %" PRIu64
" bad tail\n"),
1884 * Check contents of the node blocks (leaves)
1885 * Looks for matching hash values for the data entries.
1888 longform_dir2_check_node(
1891 dir_hash_tab_t
*hashtab
,
1897 xfs_dir2_free_t
*free
;
1899 xfs_dir2_leaf_t
*leaf
;
1900 xfs_fileoff_t next_da_bno
;
1903 struct xfs_dir2_leaf_entry
*ents
;
1904 struct xfs_dir3_icleaf_hdr leafhdr
;
1905 struct xfs_dir3_icfree_hdr freehdr
;
1908 for (da_bno
= mp
->m_dirleafblk
, next_da_bno
= 0;
1909 next_da_bno
!= NULLFILEOFF
&& da_bno
< mp
->m_dirfreeblk
;
1910 da_bno
= (xfs_dablk_t
)next_da_bno
) {
1911 next_da_bno
= da_bno
+ mp
->m_dirblkfsbs
- 1;
1912 if (bmap_next_offset(NULL
, ip
, &next_da_bno
, XFS_DATA_FORK
))
1914 if (libxfs_da_read_buf(NULL
, ip
, da_bno
, -1, &bp
,
1915 XFS_DATA_FORK
, &xfs_dir3_leafn_buf_ops
)) {
1917 _("can't read leaf block %u for directory inode %" PRIu64
"\n"),
1922 xfs_dir3_leaf_hdr_from_disk(&leafhdr
, leaf
);
1923 ents
= xfs_dir3_leaf_ents_p(leaf
);
1924 if (!(leafhdr
.magic
== XFS_DIR2_LEAFN_MAGIC
||
1925 leafhdr
.magic
== XFS_DIR3_LEAFN_MAGIC
)) {
1926 if (leafhdr
.magic
== XFS_DA_NODE_MAGIC
||
1927 leafhdr
.magic
== XFS_DA3_NODE_MAGIC
) {
1932 _("unknown magic number %#x for block %u in directory inode %" PRIu64
"\n"),
1933 leafhdr
.magic
, da_bno
, ip
->i_ino
);
1937 if (leafhdr
.count
> xfs_dir3_max_leaf_ents(mp
, leaf
) ||
1938 leafhdr
.count
< leafhdr
.stale
) {
1940 _("leaf block %u for directory inode %" PRIu64
" bad header\n"),
1945 seeval
= dir_hash_see_all(hashtab
, ents
,
1946 leafhdr
.count
, leafhdr
.stale
);
1948 if (seeval
!= DIR_HASH_CK_OK
)
1951 if (dir_hash_check(hashtab
, ip
, seeval
))
1954 for (da_bno
= mp
->m_dirfreeblk
, next_da_bno
= 0;
1955 next_da_bno
!= NULLFILEOFF
;
1956 da_bno
= (xfs_dablk_t
)next_da_bno
) {
1957 next_da_bno
= da_bno
+ mp
->m_dirblkfsbs
- 1;
1958 if (bmap_next_offset(NULL
, ip
, &next_da_bno
, XFS_DATA_FORK
))
1960 if (libxfs_da_read_buf(NULL
, ip
, da_bno
, -1, &bp
,
1961 XFS_DATA_FORK
, &xfs_dir3_free_buf_ops
)) {
1963 _("can't read freespace block %u for directory inode %" PRIu64
"\n"),
1968 xfs_dir3_free_hdr_from_disk(&freehdr
, free
);
1969 bests
= xfs_dir3_free_bests_p(mp
, free
);
1970 fdb
= xfs_dir2_da_to_db(mp
, da_bno
);
1971 if (!(freehdr
.magic
== XFS_DIR2_FREE_MAGIC
||
1972 freehdr
.magic
== XFS_DIR3_FREE_MAGIC
) ||
1974 (fdb
- XFS_DIR2_FREE_FIRSTDB(mp
)) *
1975 xfs_dir3_free_max_bests(mp
) ||
1976 freehdr
.nvalid
< freehdr
.nused
) {
1978 _("free block %u for directory inode %" PRIu64
" bad header\n"),
1983 for (i
= used
= 0; i
< freehdr
.nvalid
; i
++) {
1984 if (i
+ freehdr
.firstdb
>= freetab
->nents
||
1985 freetab
->ents
[i
+ freehdr
.firstdb
].v
!=
1986 be16_to_cpu(bests
[i
])) {
1988 _("free block %u entry %i for directory ino %" PRIu64
" bad\n"),
1989 da_bno
, i
, ip
->i_ino
);
1993 used
+= be16_to_cpu(bests
[i
]) != NULLDATAOFF
;
1994 freetab
->ents
[i
+ freehdr
.firstdb
].s
= 1;
1996 if (used
!= freehdr
.nused
) {
1998 _("free block %u for directory inode %" PRIu64
" bad nused\n"),
2005 for (i
= 0; i
< freetab
->nents
; i
++) {
2006 if ((freetab
->ents
[i
].s
== 0) &&
2007 (freetab
->ents
[i
].v
!= NULLDATAOFF
)) {
2009 _("missing freetab entry %u for directory inode %" PRIu64
"\n"),
2018 * If a directory is corrupt, we need to read in as many entries as possible,
2019 * destroy the entry and create a new one with recovered name/inode pairs.
2020 * (ie. get libxfs to do all the grunt work)
2023 longform_dir2_entry_check(xfs_mount_t
*mp
,
2028 ino_tree_node_t
*irec
,
2030 dir_hash_tab_t
*hashtab
)
2032 xfs_dir2_block_t
*block
;
2033 struct xfs_buf
**bplist
;
2040 xfs_fileoff_t next_da_bno
;
2046 freetab
= malloc(FREETAB_SIZE(ip
->i_d
.di_size
/ mp
->m_dirblksize
));
2049 _("malloc failed in longform_dir2_entry_check (%" PRId64
" bytes)\n"),
2050 FREETAB_SIZE(ip
->i_d
.di_size
/ mp
->m_dirblksize
));
2053 freetab
->naents
= ip
->i_d
.di_size
/ mp
->m_dirblksize
;
2055 for (i
= 0; i
< freetab
->naents
; i
++) {
2056 freetab
->ents
[i
].v
= NULLDATAOFF
;
2057 freetab
->ents
[i
].s
= 0;
2059 num_bps
= freetab
->naents
;
2060 bplist
= calloc(num_bps
, sizeof(struct xfs_buf
*));
2061 /* is this a block, leaf, or node directory? */
2062 libxfs_dir2_isblock(NULL
, ip
, &isblock
);
2063 libxfs_dir2_isleaf(NULL
, ip
, &isleaf
);
2065 /* check directory "data" blocks (ie. name/inode pairs) */
2066 for (da_bno
= 0, next_da_bno
= 0;
2067 next_da_bno
!= NULLFILEOFF
&& da_bno
< mp
->m_dirleafblk
;
2068 da_bno
= (xfs_dablk_t
)next_da_bno
) {
2069 next_da_bno
= da_bno
+ mp
->m_dirblkfsbs
- 1;
2070 if (bmap_next_offset(NULL
, ip
, &next_da_bno
, XFS_DATA_FORK
))
2072 db
= xfs_dir2_da_to_db(mp
, da_bno
);
2073 if (db
>= num_bps
) {
2074 /* more data blocks than expected */
2076 bplist
= realloc(bplist
, num_bps
* sizeof(struct xfs_buf
*));
2079 _("realloc failed in longform_dir2_entry_check (%zu bytes)\n"),
2080 num_bps
* sizeof(struct xfs_buf
*));
2082 if (libxfs_da_read_buf(NULL
, ip
, da_bno
, -1, &bplist
[db
],
2083 XFS_DATA_FORK
, &xfs_dir3_data_buf_ops
)) {
2085 _("can't read data block %u for directory inode %" PRIu64
"\n"),
2088 continue; /* try and read all "data" blocks */
2090 longform_dir2_entry_check_data(mp
, ip
, num_illegal
, need_dot
,
2091 irec
, ino_offset
, &bplist
[db
], hashtab
,
2092 &freetab
, da_bno
, isblock
);
2094 fixit
= (*num_illegal
!= 0) || dir2_is_badino(ino
) || *need_dot
;
2096 if (!dotdot_update
) {
2097 /* check btree and freespace */
2099 xfs_dir2_block_tail_t
*btp
;
2100 xfs_dir2_leaf_entry_t
*blp
;
2102 block
= bplist
[0]->b_addr
;
2103 btp
= xfs_dir2_block_tail_p(mp
, &block
->hdr
);
2104 blp
= xfs_dir2_block_leaf_p(btp
);
2105 seeval
= dir_hash_see_all(hashtab
, blp
,
2106 be32_to_cpu(btp
->count
),
2107 be32_to_cpu(btp
->stale
));
2108 if (dir_hash_check(hashtab
, ip
, seeval
))
2110 } else if (isleaf
) {
2111 fixit
|= longform_dir2_check_leaf(mp
, ip
, hashtab
,
2114 fixit
|= longform_dir2_check_node(mp
, ip
, hashtab
,
2118 if (!no_modify
&& (fixit
|| dotdot_update
)) {
2119 dir_hash_dup_names(hashtab
);
2120 for (i
= 0; i
< freetab
->naents
; i
++)
2122 libxfs_putbuf(bplist
[i
]);
2123 longform_dir2_rebuild(mp
, ino
, ip
, irec
, ino_offset
, hashtab
);
2127 for (i
= 0; i
< freetab
->naents
; i
++)
2129 libxfs_putbuf(bplist
[i
]);
2137 * shortform directory v2 processing routines -- entry verification and
2138 * bad entry deletion (pruning).
2141 shortform_dir2_entry_check(xfs_mount_t
*mp
,
2145 ino_tree_node_t
*current_irec
,
2146 int current_ino_offset
,
2147 dir_hash_tab_t
*hashtab
)
2152 xfs_dir2_sf_entry_t
*sfep
, *next_sfep
, *tmp_sfep
;
2154 ino_tree_node_t
*irec
;
2164 char fname
[MAXNAMELEN
+ 1];
2168 sfp
= (xfs_dir2_sf_t
*) ifp
->if_u1
.if_data
;
2172 max_size
= ifp
->if_bytes
;
2173 ASSERT(ip
->i_d
.di_size
<= ifp
->if_bytes
);
2176 * if just rebuild a directory due to a "..", update and return
2178 if (dotdot_update
) {
2179 parent
= get_inode_parent(current_irec
, current_ino_offset
);
2182 _("would set .. in sf dir inode %" PRIu64
" to %" PRIu64
"\n"),
2186 _("setting .. in sf dir inode %" PRIu64
" to %" PRIu64
"\n"),
2188 xfs_dir2_sf_put_parent_ino(&sfp
->hdr
, parent
);
2195 * no '.' entry in shortform dirs, just bump up ref count by 1
2196 * '..' was already (or will be) accounted for and checked when
2197 * the directory is reached or will be taken care of when the
2198 * directory is moved to orphanage.
2200 add_inode_ref(current_irec
, current_ino_offset
);
2203 * Initialise i8 counter -- the parent inode number counts as well.
2205 i8
= xfs_dir2_sf_get_parent_ino(&sfp
->hdr
) > XFS_DIR2_MAX_SHORT_INUM
;
2208 * now run through entries, stop at first bad entry, don't need
2209 * to skip over '..' since that's encoded in its own field and
2210 * no need to worry about '.' since it doesn't exist.
2212 sfep
= next_sfep
= xfs_dir2_sf_firstentry(&sfp
->hdr
);
2214 for (i
= 0; i
< sfp
->hdr
.count
&& max_size
>
2215 (__psint_t
)next_sfep
- (__psint_t
)sfp
;
2216 sfep
= next_sfep
, i
++) {
2221 lino
= xfs_dir2_sfe_get_ino(&sfp
->hdr
, sfep
);
2223 namelen
= sfep
->namelen
;
2225 ASSERT(no_modify
|| namelen
> 0);
2227 if (no_modify
&& namelen
== 0) {
2229 * if we're really lucky, this is
2230 * the last entry in which case we
2231 * can use the dir size to set the
2232 * namelen value. otherwise, forget
2233 * it because we're not going to be
2234 * able to find the next entry.
2238 if (i
== sfp
->hdr
.count
- 1) {
2239 namelen
= ip
->i_d
.di_size
-
2240 ((__psint_t
) &sfep
->name
[0] -
2244 * don't process the rest of the directory,
2245 * break out of processing loop
2249 } else if (no_modify
&& (__psint_t
) sfep
- (__psint_t
) sfp
+
2250 + xfs_dir2_sf_entsize(&sfp
->hdr
, sfep
->namelen
)
2251 > ip
->i_d
.di_size
) {
2254 if (i
== sfp
->hdr
.count
- 1) {
2255 namelen
= ip
->i_d
.di_size
-
2256 ((__psint_t
) &sfep
->name
[0] -
2260 * don't process the rest of the directory,
2261 * break out of processing loop
2267 memmove(fname
, sfep
->name
, sfep
->namelen
);
2268 fname
[sfep
->namelen
] = '\0';
2270 ASSERT(no_modify
|| (lino
!= NULLFSINO
&& lino
!= 0));
2271 ASSERT(no_modify
|| !verify_inum(mp
, lino
));
2274 * Also skip entries with bogus inode numbers if we're
2275 * in no modify mode.
2278 if (no_modify
&& verify_inum(mp
, lino
)) {
2279 next_sfep
= (xfs_dir2_sf_entry_t
*)((__psint_t
)sfep
+
2280 xfs_dir2_sf_entsize(&sfp
->hdr
, sfep
->namelen
));
2284 irec
= find_inode_rec(mp
, XFS_INO_TO_AGNO(mp
, lino
),
2285 XFS_INO_TO_AGINO(mp
, lino
));
2289 _("entry \"%s\" in shortform directory %" PRIu64
" references non-existent inode %" PRIu64
"\n"),
2294 ino_offset
= XFS_INO_TO_AGINO(mp
, lino
) - irec
->ino_startnum
;
2297 * if it's a free inode, blow out the entry.
2298 * by now, any inode that we think is free
2301 if (is_inode_free(irec
, ino_offset
)) {
2303 _("entry \"%s\" in shortform directory inode %" PRIu64
" points to free inode %" PRIu64
"\n"),
2308 * check if this inode is lost+found dir in the root
2310 if (ino
== mp
->m_sb
.sb_rootino
&& strcmp(fname
, ORPHANAGE
) == 0) {
2312 * if it's not a directory, trash it
2314 if (!inode_isadir(irec
, ino_offset
)) {
2316 _("%s (ino %" PRIu64
") in root (%" PRIu64
") is not a directory"),
2317 ORPHANAGE
, lino
, ino
);
2321 * if this is a dup, it will be picked up below,
2322 * otherwise, mark it as the orphanage for later.
2325 orphanage_ino
= lino
;
2328 * check for duplicate names in directory.
2330 if (!dir_hash_add(mp
, hashtab
, (xfs_dir2_dataptr_t
)
2331 (sfep
- xfs_dir2_sf_firstentry(&sfp
->hdr
)),
2332 lino
, sfep
->namelen
, sfep
->name
)) {
2334 _("entry \"%s\" (ino %" PRIu64
") in dir %" PRIu64
" is a duplicate name"),
2339 if (!inode_isadir(irec
, ino_offset
)) {
2341 * check easy case first, regular inode, just bump
2344 add_inode_reached(irec
, ino_offset
);
2346 parent
= get_inode_parent(irec
, ino_offset
);
2349 * bump up the link counts in parent and child.
2350 * directory but if the link doesn't agree with
2351 * the .. in the child, blow out the entry
2353 if (is_inode_reached(irec
, ino_offset
)) {
2356 _("entry \"%s\" in directory inode %" PRIu64
2357 " references already connected inode %" PRIu64
".\n"),
2359 } else if (parent
== ino
) {
2360 add_inode_reached(irec
, ino_offset
);
2361 add_inode_ref(current_irec
, current_ino_offset
);
2362 } else if (parent
== NULLFSINO
) {
2363 /* ".." was missing, but this entry refers to it,
2364 so, set it as the parent and mark for rebuild */
2366 _("entry \"%s\" in dir ino %" PRIu64
" doesn't have a .. entry, will set it in ino %" PRIu64
".\n"),
2368 set_inode_parent(irec
, ino_offset
, ino
);
2369 add_inode_reached(irec
, ino_offset
);
2370 add_inode_ref(current_irec
, current_ino_offset
);
2371 add_dotdot_update(XFS_INO_TO_AGNO(mp
, lino
),
2376 _("entry \"%s\" in directory inode %" PRIu64
2377 " not consistent with .. value (%" PRIu64
2378 ") in inode %" PRIu64
",\n"),
2379 fname
, ino
, parent
, lino
);
2385 if (lino
== orphanage_ino
)
2388 tmp_elen
= xfs_dir2_sf_entsize(&sfp
->hdr
,
2390 tmp_sfep
= (xfs_dir2_sf_entry_t
*)
2391 ((__psint_t
) sfep
+ tmp_elen
);
2392 tmp_len
= max_size
- ((__psint_t
) tmp_sfep
2394 max_size
-= tmp_elen
;
2395 bytes_deleted
+= tmp_elen
;
2397 memmove(sfep
, tmp_sfep
, tmp_len
);
2399 sfp
->hdr
.count
-= 1;
2400 memset((void *)((__psint_t
)sfep
+ tmp_len
), 0,
2404 * set the tmp value to the current
2405 * pointer so we'll process the entry
2411 * WARNING: drop the index i by one
2412 * so it matches the decremented count for
2413 * accurate comparisons in the loop test
2420 do_warn(_("junking entry\n"));
2424 do_warn(_("would junk entry\n"));
2426 } else if (lino
> XFS_DIR2_MAX_SHORT_INUM
)
2430 * go onto next entry unless we've just junked an
2431 * entry in which the current entry pointer points
2432 * to an unprocessed entry. have to take into entries
2433 * with bad namelen into account in no modify mode since we
2434 * calculate size based on next_sfep.
2436 ASSERT(no_modify
|| bad_sfnamelen
== 0);
2438 next_sfep
= (tmp_sfep
== NULL
)
2439 ? (xfs_dir2_sf_entry_t
*) ((__psint_t
) sfep
2441 ? xfs_dir2_sf_entsize(&sfp
->hdr
, sfep
->namelen
)
2442 : xfs_dir2_sf_entsize(&sfp
->hdr
, namelen
)))
2446 if (sfp
->hdr
.i8count
!= i8
) {
2448 do_warn(_("would fix i8count in inode %" PRIu64
"\n"),
2452 tmp_sfep
= next_sfep
;
2453 process_sf_dir2_fixi8(sfp
, &tmp_sfep
);
2455 (__psint_t
)next_sfep
-
2456 (__psint_t
)tmp_sfep
;
2457 next_sfep
= tmp_sfep
;
2459 sfp
->hdr
.i8count
= i8
;
2461 do_warn(_("fixing i8count in inode %" PRIu64
"\n"),
2467 * sync up sizes if required
2470 ASSERT(bytes_deleted
> 0);
2472 libxfs_idata_realloc(ip
, -bytes_deleted
, XFS_DATA_FORK
);
2473 ip
->i_d
.di_size
-= bytes_deleted
;
2476 if (ip
->i_d
.di_size
!= ip
->i_df
.if_bytes
) {
2477 ASSERT(ip
->i_df
.if_bytes
== (xfs_fsize_t
)
2478 ((__psint_t
) next_sfep
- (__psint_t
) sfp
));
2479 ip
->i_d
.di_size
= (xfs_fsize_t
)
2480 ((__psint_t
) next_sfep
- (__psint_t
) sfp
);
2482 _("setting size to %" PRId64
" bytes to reflect junked entries\n"),
2489 * processes all reachable inodes in directories
2494 xfs_agnumber_t agno
,
2495 ino_tree_node_t
*irec
,
2499 xfs_bmap_free_t flist
;
2500 xfs_fsblock_t first
;
2503 dir_hash_tab_t
*hashtab
;
2504 int need_dot
, committed
;
2505 int dirty
, num_illegal
, error
, nres
;
2507 ino
= XFS_AGINO_TO_INO(mp
, agno
, irec
->ino_startnum
+ ino_offset
);
2510 * open up directory inode, check all entries,
2511 * then call prune_dir_entries to remove all
2512 * remaining illegal directory entries.
2515 ASSERT(!is_inode_refchecked(irec
, ino_offset
) || dotdot_update
);
2517 error
= libxfs_iget(mp
, NULL
, ino
, 0, &ip
, 0);
2521 _("couldn't map inode %" PRIu64
", err = %d\n"),
2525 _("couldn't map inode %" PRIu64
", err = %d\n"),
2528 * see below for what we're doing if this
2529 * is root. Why do we need to do this here?
2530 * to ensure that the root doesn't show up
2531 * as being disconnected in the no_modify case.
2533 if (mp
->m_sb
.sb_rootino
== ino
) {
2534 add_inode_reached(irec
, 0);
2535 add_inode_ref(irec
, 0);
2539 add_inode_refchecked(irec
, 0);
2543 need_dot
= dirty
= num_illegal
= 0;
2545 if (mp
->m_sb
.sb_rootino
== ino
) {
2547 * mark root inode reached and bump up
2548 * link count for root inode to account
2549 * for '..' entry since the root inode is
2550 * never reached by a parent. we know
2551 * that root's '..' is always good --
2552 * guaranteed by phase 3 and/or below.
2554 add_inode_reached(irec
, ino_offset
);
2557 add_inode_refchecked(irec
, ino_offset
);
2559 hashtab
= dir_hash_init(ip
->i_d
.di_size
);
2562 * look for bogus entries
2564 switch (ip
->i_d
.di_format
) {
2565 case XFS_DINODE_FMT_EXTENTS
:
2566 case XFS_DINODE_FMT_BTREE
:
2568 * also check for missing '.' in longform dirs.
2569 * missing .. entries are added if required when
2570 * the directory is connected to lost+found. but
2571 * we need to create '.' entries here.
2573 longform_dir2_entry_check(mp
, ino
, ip
,
2574 &num_illegal
, &need_dot
,
2579 case XFS_DINODE_FMT_LOCAL
:
2580 tp
= libxfs_trans_alloc(mp
, 0);
2582 * using the remove reservation is overkill
2583 * since at most we'll only need to log the
2584 * inode but it's easier than wedging a
2585 * new define in ourselves.
2587 nres
= no_modify
? 0 : XFS_REMOVE_SPACE_RES(mp
);
2588 error
= libxfs_trans_reserve(tp
, nres
,
2589 XFS_REMOVE_LOG_RES(mp
), 0,
2590 XFS_TRANS_PERM_LOG_RES
,
2591 XFS_REMOVE_LOG_COUNT
);
2595 libxfs_trans_ijoin(tp
, ip
, 0);
2596 libxfs_trans_ihold(tp
, ip
);
2598 shortform_dir2_entry_check(mp
, ino
, ip
, &dirty
,
2602 ASSERT(dirty
== 0 || (dirty
&& !no_modify
));
2604 libxfs_trans_log_inode(tp
, ip
,
2605 XFS_ILOG_CORE
| XFS_ILOG_DDATA
);
2606 libxfs_trans_commit(tp
,
2607 XFS_TRANS_RELEASE_LOG_RES
|
2610 libxfs_trans_cancel(tp
,
2611 XFS_TRANS_RELEASE_LOG_RES
);
2618 dir_hash_done(hashtab
);
2621 * if we have to create a .. for /, do it now *before*
2622 * we delete the bogus entries, otherwise the directory
2623 * could transform into a shortform dir which would
2624 * probably cause the simulation to choke. Even
2625 * if the illegal entries get shifted around, it's ok
2626 * because the entries are structurally intact and in
2627 * in hash-value order so the simulation won't get confused
2628 * if it has to move them around.
2630 if (!no_modify
&& need_root_dotdot
&& ino
== mp
->m_sb
.sb_rootino
) {
2631 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_LOCAL
);
2633 do_warn(_("recreating root directory .. entry\n"));
2635 tp
= libxfs_trans_alloc(mp
, 0);
2638 nres
= XFS_MKDIR_SPACE_RES(mp
, 2);
2639 error
= libxfs_trans_reserve(tp
, nres
, XFS_MKDIR_LOG_RES(mp
),
2640 0, XFS_TRANS_PERM_LOG_RES
, XFS_MKDIR_LOG_COUNT
);
2644 libxfs_trans_ijoin(tp
, ip
, 0);
2645 libxfs_trans_ihold(tp
, ip
);
2647 xfs_bmap_init(&flist
, &first
);
2649 error
= libxfs_dir_createname(tp
, ip
, &xfs_name_dotdot
,
2650 ip
->i_ino
, &first
, &flist
, nres
);
2653 _("can't make \"..\" entry in root inode %" PRIu64
", createname error %d\n"), ino
, error
);
2655 libxfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
2657 error
= libxfs_bmap_finish(&tp
, &flist
, &committed
);
2659 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
|
2662 need_root_dotdot
= 0;
2663 } else if (need_root_dotdot
&& ino
== mp
->m_sb
.sb_rootino
) {
2664 do_warn(_("would recreate root directory .. entry\n"));
2668 * if we need to create the '.' entry, do so only if
2669 * the directory is a longform dir. if it's been
2670 * turned into a shortform dir, then the inode is ok
2671 * since shortform dirs have no '.' entry and the inode
2672 * has already been committed by prune_lf_dir_entry().
2676 * bump up our link count but don't
2677 * bump up the inode link count. chances
2678 * are good that even though we lost '.'
2679 * the inode link counts reflect '.' so
2680 * leave the inode link count alone and if
2681 * it turns out to be wrong, we'll catch
2684 add_inode_ref(irec
, ino_offset
);
2688 _("would create missing \".\" entry in dir ino %" PRIu64
"\n"),
2690 } else if (ip
->i_d
.di_format
!= XFS_DINODE_FMT_LOCAL
) {
2692 * need to create . entry in longform dir.
2695 _("creating missing \".\" entry in dir ino %" PRIu64
"\n"), ino
);
2697 tp
= libxfs_trans_alloc(mp
, 0);
2700 nres
= XFS_MKDIR_SPACE_RES(mp
, 1);
2701 error
= libxfs_trans_reserve(tp
, nres
,
2702 XFS_MKDIR_LOG_RES(mp
),
2704 XFS_TRANS_PERM_LOG_RES
,
2705 XFS_MKDIR_LOG_COUNT
);
2710 libxfs_trans_ijoin(tp
, ip
, 0);
2711 libxfs_trans_ihold(tp
, ip
);
2713 xfs_bmap_init(&flist
, &first
);
2715 error
= libxfs_dir_createname(tp
, ip
, &xfs_name_dot
,
2716 ip
->i_ino
, &first
, &flist
, nres
);
2719 _("can't make \".\" entry in dir ino %" PRIu64
", createname error %d\n"),
2722 libxfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
2724 error
= libxfs_bmap_finish(&tp
, &flist
, &committed
);
2726 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
2734 * mark realtime bitmap and summary inodes as reached.
2735 * quota inode will be marked here as well
2738 mark_standalone_inodes(xfs_mount_t
*mp
)
2740 ino_tree_node_t
*irec
;
2743 irec
= find_inode_rec(mp
, XFS_INO_TO_AGNO(mp
, mp
->m_sb
.sb_rbmino
),
2744 XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_rbmino
));
2746 ASSERT(irec
!= NULL
);
2748 offset
= XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_rbmino
) -
2751 add_inode_reached(irec
, offset
);
2753 irec
= find_inode_rec(mp
, XFS_INO_TO_AGNO(mp
, mp
->m_sb
.sb_rsumino
),
2754 XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_rsumino
));
2756 offset
= XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_rsumino
) -
2759 ASSERT(irec
!= NULL
);
2761 add_inode_reached(irec
, offset
);
2764 if (mp
->m_sb
.sb_uquotino
2765 && mp
->m_sb
.sb_uquotino
!= NULLFSINO
) {
2766 irec
= find_inode_rec(mp
, XFS_INO_TO_AGNO(mp
,
2767 mp
->m_sb
.sb_uquotino
),
2768 XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_uquotino
));
2769 offset
= XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_uquotino
)
2770 - irec
->ino_startnum
;
2771 add_inode_reached(irec
, offset
);
2773 if (mp
->m_sb
.sb_gquotino
2774 && mp
->m_sb
.sb_gquotino
!= NULLFSINO
) {
2775 irec
= find_inode_rec(mp
, XFS_INO_TO_AGNO(mp
,
2776 mp
->m_sb
.sb_gquotino
),
2777 XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_gquotino
));
2778 offset
= XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_gquotino
)
2779 - irec
->ino_startnum
;
2780 add_inode_reached(irec
, offset
);
2786 check_for_orphaned_inodes(
2788 xfs_agnumber_t agno
,
2789 ino_tree_node_t
*irec
)
2794 for (i
= 0; i
< XFS_INODES_PER_CHUNK
; i
++) {
2795 ASSERT(is_inode_confirmed(irec
, i
));
2796 if (is_inode_free(irec
, i
))
2799 if (is_inode_reached(irec
, i
))
2802 ASSERT(inode_isadir(irec
, i
) ||
2803 num_inode_references(irec
, i
) == 0);
2805 ino
= XFS_AGINO_TO_INO(mp
, agno
, i
+ irec
->ino_startnum
);
2806 if (inode_isadir(irec
, i
))
2807 do_warn(_("disconnected dir inode %" PRIu64
", "), ino
);
2809 do_warn(_("disconnected inode %" PRIu64
", "), ino
);
2812 orphanage_ino
= mk_orphanage(mp
);
2813 do_warn(_("moving to %s\n"), ORPHANAGE
);
2814 mv_orphanage(mp
, ino
, inode_isadir(irec
, i
));
2816 do_warn(_("would move to %s\n"), ORPHANAGE
);
2819 * for read-only case, even though the inode isn't
2820 * really reachable, set the flag (and bump our link
2821 * count) anyway to fool phase 7
2823 add_inode_reached(irec
, i
);
2830 xfs_agnumber_t agno
,
2833 ino_tree_node_t
*irec
;
2835 prefetch_args_t
*pf_args
= arg
;
2837 wait_for_inode_prefetch(pf_args
);
2840 do_log(_(" - agno = %d\n"), agno
);
2842 for (irec
= findfirst_inode_rec(agno
); irec
; irec
= next_ino_rec(irec
)) {
2843 if (irec
->ino_isa_dir
== 0)
2847 sem_post(&pf_args
->ra_count
);
2849 for (i
= 0; i
< XFS_INODES_PER_CHUNK
; i
++) {
2850 if (inode_isadir(irec
, i
))
2851 process_dir_inode(wq
->mp
, agno
, irec
, i
);
2854 cleanup_inode_prefetch(pf_args
);
2858 update_missing_dotdot_entries(
2861 dotdot_update_t
*dir
;
2864 * these entries parents were updated, rebuild them again
2865 * set dotdot_update flag so processing routines do not count links
2868 while (dotdot_update_list
) {
2869 dir
= dotdot_update_list
;
2870 dotdot_update_list
= dir
->next
;
2871 process_dir_inode(mp
, dir
->agno
, dir
->irec
, dir
->ino_offset
);
2882 prefetch_args_t
*pf_args
[2];
2885 * we always do prefetch for phase 6 as it will fill in the gaps
2886 * not read during phase 3 prefetch.
2889 pf_args
[0] = start_inode_prefetch(0, 1, NULL
);
2890 for (i
= 0; i
< glob_agcount
; i
++) {
2891 pf_args
[(~i
) & 1] = start_inode_prefetch(i
+ 1, 1,
2893 traverse_function(&queue
, i
, pf_args
[i
& 1]);
2898 phase6(xfs_mount_t
*mp
)
2900 ino_tree_node_t
*irec
;
2903 memset(&zerocr
, 0, sizeof(struct cred
));
2904 memset(&zerofsx
, 0, sizeof(struct fsxattr
));
2907 do_log(_("Phase 6 - check inode connectivity...\n"));
2909 incore_ext_teardown(mp
);
2911 add_ino_ex_data(mp
);
2914 * verify existence of root directory - if we have to
2915 * make one, it's ok for the incore data structs not to
2916 * know about it since everything about it (and the other
2917 * inodes in its chunk if a new chunk was created) are ok
2919 if (need_root_inode
) {
2921 do_warn(_("reinitializing root directory\n"));
2923 need_root_inode
= 0;
2924 need_root_dotdot
= 0;
2926 do_warn(_("would reinitialize root directory\n"));
2932 do_warn(_("reinitializing realtime bitmap inode\n"));
2936 do_warn(_("would reinitialize realtime bitmap inode\n"));
2942 do_warn(_("reinitializing realtime summary inode\n"));
2946 do_warn(_("would reinitialize realtime summary inode\n"));
2952 _(" - resetting contents of realtime bitmap and summary inodes\n"));
2953 if (fill_rbmino(mp
)) {
2955 _("Warning: realtime bitmap may be inconsistent\n"));
2958 if (fill_rsumino(mp
)) {
2960 _("Warning: realtime bitmap may be inconsistent\n"));
2964 mark_standalone_inodes(mp
);
2966 do_log(_(" - traversing filesystem ...\n"));
2968 irec
= find_inode_rec(mp
, XFS_INO_TO_AGNO(mp
, mp
->m_sb
.sb_rootino
),
2969 XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_rootino
));
2972 * we always have a root inode, even if it's free...
2973 * if the root is free, forget it, lost+found is already gone
2975 if (is_inode_free(irec
, 0) || !inode_isadir(irec
, 0)) {
2976 need_root_inode
= 1;
2980 * then process all inodes by walking incore inode tree
2985 * any directories that had updated ".." entries, rebuild them now
2987 update_missing_dotdot_entries(mp
);
2989 do_log(_(" - traversal finished ...\n"));
2990 do_log(_(" - moving disconnected inodes to %s ...\n"),
2994 * move all disconnected inodes to the orphanage
2996 for (i
= 0; i
< glob_agcount
; i
++) {
2997 irec
= findfirst_inode_rec(i
);
2998 while (irec
!= NULL
) {
2999 check_for_orphaned_inodes(mp
, i
, irec
);
3000 irec
= next_ino_rec(irec
);