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
];
432 struct xfs_trans_res tres
= {0};
437 tp
= libxfs_trans_alloc(mp
, 0);
439 i
= libxfs_trans_reserve(tp
, &tres
, 10, 0);
443 error
= libxfs_trans_iget(mp
, tp
, mp
->m_sb
.sb_rbmino
, 0, 0, &ip
);
446 _("couldn't iget realtime bitmap inode -- error - %d\n"),
450 vers
= xfs_sb_version_hascrc(&mp
->m_sb
) ? 3 : 1;
451 memset(&ip
->i_d
, 0, xfs_icdinode_size(vers
));
453 ip
->i_d
.di_magic
= XFS_DINODE_MAGIC
;
454 ip
->i_d
.di_mode
= S_IFREG
;
455 ip
->i_d
.di_version
= vers
;
456 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
457 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
459 ip
->i_d
.di_nlink
= 1; /* account for sb ptr */
461 times
= XFS_ICHGTIME_CHG
| XFS_ICHGTIME_MOD
;
462 if (ip
->i_d
.di_version
== 3) {
464 ip
->i_d
.di_changecount
= 1;
466 ip
->i_d
.di_flags2
= 0;
467 ip
->i_d
.di_ino
= mp
->m_sb
.sb_rbmino
;
468 memset(&(ip
->i_d
.di_pad2
[0]), 0, sizeof(ip
->i_d
.di_pad2
));
469 platform_uuid_copy(&ip
->i_d
.di_uuid
, &mp
->m_sb
.sb_uuid
);
470 times
|= XFS_ICHGTIME_CREATE
;
472 libxfs_trans_ichgtime(tp
, ip
, times
);
477 ip
->i_df
.if_flags
= XFS_IFEXTENTS
;
478 ip
->i_df
.if_bytes
= ip
->i_df
.if_real_bytes
= 0;
479 ip
->i_df
.if_u1
.if_extents
= NULL
;
481 ip
->i_d
.di_size
= mp
->m_sb
.sb_rbmblocks
* mp
->m_sb
.sb_blocksize
;
486 libxfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
487 libxfs_trans_ihold(tp
, ip
);
488 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
491 * then allocate blocks for file and fill with zeroes (stolen
494 tp
= libxfs_trans_alloc(mp
, 0);
495 error
= libxfs_trans_reserve(tp
, &tres
, mp
->m_sb
.sb_rbmblocks
+
496 (XFS_BM_MAXLEVELS(mp
, XFS_DATA_FORK
) - 1), 0);
500 libxfs_trans_ijoin(tp
, ip
, 0);
502 xfs_bmap_init(&flist
, &first
);
503 while (bno
< mp
->m_sb
.sb_rbmblocks
) {
504 nmap
= XFS_BMAP_MAX_NMAP
;
505 error
= libxfs_bmapi_write(tp
, ip
, bno
,
506 (xfs_extlen_t
)(mp
->m_sb
.sb_rbmblocks
- bno
),
507 0, &first
, mp
->m_sb
.sb_rbmblocks
,
511 _("couldn't allocate realtime bitmap, error = %d\n"),
514 for (i
= 0, ep
= map
; i
< nmap
; i
++, ep
++) {
515 libxfs_device_zero(mp
->m_ddev_targp
,
516 XFS_FSB_TO_DADDR(mp
, ep
->br_startblock
),
517 XFS_FSB_TO_BB(mp
, ep
->br_blockcount
));
518 bno
+= ep
->br_blockcount
;
521 error
= libxfs_bmap_finish(&tp
, &flist
, &committed
);
524 _("allocation of the realtime bitmap failed, error = %d\n"),
527 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
531 fill_rbmino(xfs_mount_t
*mp
)
542 struct xfs_trans_res tres
= {0};
547 tp
= libxfs_trans_alloc(mp
, 0);
549 error
= libxfs_trans_reserve(tp
, &tres
, 10, 0);
553 error
= libxfs_trans_iget(mp
, tp
, mp
->m_sb
.sb_rbmino
, 0, 0, &ip
);
556 _("couldn't iget realtime bitmap inode -- error - %d\n"),
561 while (bno
< mp
->m_sb
.sb_rbmblocks
) {
563 * fill the file one block at a time
566 error
= libxfs_bmapi_write(tp
, ip
, bno
, 1, 0,
567 &first
, 1, &map
, &nmap
, NULL
);
568 if (error
|| nmap
!= 1) {
570 _("couldn't map realtime bitmap block %" PRIu64
", error = %d\n"),
574 ASSERT(map
.br_startblock
!= HOLESTARTBLOCK
);
576 error
= libxfs_trans_read_buf(
578 XFS_FSB_TO_DADDR(mp
, map
.br_startblock
),
579 XFS_FSB_TO_BB(mp
, 1), 1, &bp
, NULL
);
583 _("can't access block %" PRIu64
" (fsbno %" PRIu64
") of realtime bitmap inode %" PRIu64
"\n"),
584 bno
, map
.br_startblock
, mp
->m_sb
.sb_rbmino
);
588 memmove(XFS_BUF_PTR(bp
), bmp
, mp
->m_sb
.sb_blocksize
);
590 libxfs_trans_log_buf(tp
, bp
, 0, mp
->m_sb
.sb_blocksize
- 1);
592 bmp
= (xfs_rtword_t
*)((__psint_t
) bmp
+ mp
->m_sb
.sb_blocksize
);
596 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
601 fill_rsumino(xfs_mount_t
*mp
)
611 xfs_dfiloff_t end_bno
;
613 struct xfs_trans_res tres
= {0};
617 end_bno
= mp
->m_rsumsize
>> mp
->m_sb
.sb_blocklog
;
619 tp
= libxfs_trans_alloc(mp
, 0);
621 error
= libxfs_trans_reserve(tp
, &tres
, 10, 0);
625 error
= libxfs_trans_iget(mp
, tp
, mp
->m_sb
.sb_rsumino
, 0, 0, &ip
);
628 _("couldn't iget realtime summary inode -- error - %d\n"),
633 while (bno
< end_bno
) {
635 * fill the file one block at a time
638 error
= libxfs_bmapi_write(tp
, ip
, bno
, 1, 0,
639 &first
, 1, &map
, &nmap
, NULL
);
640 if (error
|| nmap
!= 1) {
642 _("couldn't map realtime summary inode block %" PRIu64
", error = %d\n"),
646 ASSERT(map
.br_startblock
!= HOLESTARTBLOCK
);
648 error
= libxfs_trans_read_buf(
650 XFS_FSB_TO_DADDR(mp
, map
.br_startblock
),
651 XFS_FSB_TO_BB(mp
, 1), 1, &bp
, NULL
);
655 _("can't access block %" PRIu64
" (fsbno %" PRIu64
") of realtime summary inode %" PRIu64
"\n"),
656 bno
, map
.br_startblock
, mp
->m_sb
.sb_rsumino
);
660 memmove(XFS_BUF_PTR(bp
), smp
, mp
->m_sb
.sb_blocksize
);
662 libxfs_trans_log_buf(tp
, bp
, 0, mp
->m_sb
.sb_blocksize
- 1);
664 smp
= (xfs_suminfo_t
*)((__psint_t
)smp
+ mp
->m_sb
.sb_blocksize
);
668 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
673 mk_rsumino(xfs_mount_t
*mp
)
684 xfs_bmap_free_t flist
;
686 xfs_bmbt_irec_t map
[XFS_BMAP_MAX_NMAP
];
689 struct xfs_trans_res tres
= {0};
694 tp
= libxfs_trans_alloc(mp
, 0);
696 i
= libxfs_trans_reserve(tp
, &M_RES(mp
)->tr_ichange
, 10, 0);
700 error
= libxfs_trans_iget(mp
, tp
, mp
->m_sb
.sb_rsumino
, 0, 0, &ip
);
703 _("couldn't iget realtime summary inode -- error - %d\n"),
707 vers
= xfs_sb_version_hascrc(&mp
->m_sb
) ? 3 : 1;
708 memset(&ip
->i_d
, 0, xfs_icdinode_size(vers
));
710 ip
->i_d
.di_magic
= XFS_DINODE_MAGIC
;
711 ip
->i_d
.di_mode
= S_IFREG
;
712 ip
->i_d
.di_version
= vers
;
713 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
714 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
716 ip
->i_d
.di_nlink
= 1; /* account for sb ptr */
718 times
= XFS_ICHGTIME_CHG
| XFS_ICHGTIME_MOD
;
719 if (ip
->i_d
.di_version
== 3) {
721 ip
->i_d
.di_changecount
= 1;
723 ip
->i_d
.di_flags2
= 0;
724 ip
->i_d
.di_ino
= mp
->m_sb
.sb_rsumino
;
725 memset(&(ip
->i_d
.di_pad2
[0]), 0, sizeof(ip
->i_d
.di_pad2
));
726 platform_uuid_copy(&ip
->i_d
.di_uuid
, &mp
->m_sb
.sb_uuid
);
727 times
|= XFS_ICHGTIME_CREATE
;
729 libxfs_trans_ichgtime(tp
, ip
, times
);
734 ip
->i_df
.if_flags
= XFS_IFEXTENTS
;
735 ip
->i_df
.if_bytes
= ip
->i_df
.if_real_bytes
= 0;
736 ip
->i_df
.if_u1
.if_extents
= NULL
;
738 ip
->i_d
.di_size
= mp
->m_rsumsize
;
743 libxfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
744 libxfs_trans_ihold(tp
, ip
);
745 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
748 * then allocate blocks for file and fill with zeroes (stolen
751 tp
= libxfs_trans_alloc(mp
, 0);
752 xfs_bmap_init(&flist
, &first
);
754 nsumblocks
= mp
->m_rsumsize
>> mp
->m_sb
.sb_blocklog
;
755 tres
.tr_logres
= BBTOB(128);
756 tres
.tr_logcount
= XFS_DEFAULT_PERM_LOG_COUNT
;
757 tres
.tr_logflags
= XFS_TRANS_PERM_LOG_RES
;
758 error
= libxfs_trans_reserve(tp
, &tres
, mp
->m_sb
.sb_rbmblocks
+
759 (XFS_BM_MAXLEVELS(mp
, XFS_DATA_FORK
) - 1), 0);
763 libxfs_trans_ijoin(tp
, ip
, 0);
765 xfs_bmap_init(&flist
, &first
);
766 while (bno
< nsumblocks
) {
767 nmap
= XFS_BMAP_MAX_NMAP
;
768 error
= libxfs_bmapi_write(tp
, ip
, bno
,
769 (xfs_extlen_t
)(nsumblocks
- bno
),
770 0, &first
, nsumblocks
, map
, &nmap
, &flist
);
773 _("couldn't allocate realtime summary inode, error = %d\n"),
776 for (i
= 0, ep
= map
; i
< nmap
; i
++, ep
++) {
777 libxfs_device_zero(mp
->m_ddev_targp
,
778 XFS_FSB_TO_DADDR(mp
, ep
->br_startblock
),
779 XFS_FSB_TO_BB(mp
, ep
->br_blockcount
));
780 bno
+= ep
->br_blockcount
;
783 error
= libxfs_bmap_finish(&tp
, &flist
, &committed
);
786 _("allocation of the realtime summary ino failed, error = %d\n"),
789 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
793 * makes a new root directory.
796 mk_root_dir(xfs_mount_t
*mp
)
802 const mode_t mode
= 0755;
803 ino_tree_node_t
*irec
;
807 ASSERT(xfs_sb_version_hasdirv2(&mp
->m_sb
));
809 tp
= libxfs_trans_alloc(mp
, 0);
812 i
= libxfs_trans_reserve(tp
, &M_RES(mp
)->tr_ichange
, 10, 0);
816 error
= libxfs_trans_iget(mp
, tp
, mp
->m_sb
.sb_rootino
, 0, 0, &ip
);
818 do_error(_("could not iget root inode -- error - %d\n"), error
);
822 * take care of the core -- initialization from xfs_ialloc()
824 vers
= xfs_sb_version_hascrc(&mp
->m_sb
) ? 3 : 1;
825 memset(&ip
->i_d
, 0, xfs_icdinode_size(vers
));
827 ip
->i_d
.di_magic
= XFS_DINODE_MAGIC
;
828 ip
->i_d
.di_mode
= (__uint16_t
) mode
|S_IFDIR
;
829 ip
->i_d
.di_version
= vers
;
830 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
831 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
833 ip
->i_d
.di_nlink
= 1; /* account for . */
835 times
= XFS_ICHGTIME_CHG
| XFS_ICHGTIME_MOD
;
836 if (ip
->i_d
.di_version
== 3) {
838 ip
->i_d
.di_changecount
= 1;
840 ip
->i_d
.di_flags2
= 0;
841 ip
->i_d
.di_ino
= mp
->m_sb
.sb_rootino
;
842 memset(&(ip
->i_d
.di_pad2
[0]), 0, sizeof(ip
->i_d
.di_pad2
));
843 platform_uuid_copy(&ip
->i_d
.di_uuid
, &mp
->m_sb
.sb_uuid
);
844 times
|= XFS_ICHGTIME_CREATE
;
846 libxfs_trans_ichgtime(tp
, ip
, times
);
848 libxfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
853 ip
->i_df
.if_flags
= XFS_IFEXTENTS
;
854 ip
->i_df
.if_bytes
= ip
->i_df
.if_real_bytes
= 0;
855 ip
->i_df
.if_u1
.if_extents
= NULL
;
860 * initialize the directory
862 libxfs_dir_init(tp
, ip
, ip
);
864 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
866 irec
= find_inode_rec(mp
, XFS_INO_TO_AGNO(mp
, mp
->m_sb
.sb_rootino
),
867 XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_rootino
));
868 set_inode_isadir(irec
, XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_rootino
) -
873 * orphanage name == lost+found
876 mk_orphanage(xfs_mount_t
*mp
)
883 ino_tree_node_t
*irec
;
888 xfs_bmap_free_t flist
;
889 const int mode
= 0755;
891 struct xfs_name xname
;
893 ASSERT(xfs_sb_version_hasdirv2(&mp
->m_sb
));
896 * check for an existing lost+found first, if it exists, return
897 * its inode. Otherwise, we can create it. Bad lost+found inodes
898 * would have been cleared in phase3 and phase4.
901 if ((i
= libxfs_iget(mp
, NULL
, mp
->m_sb
.sb_rootino
, 0, &pip
, 0)))
902 do_error(_("%d - couldn't iget root inode to obtain %s\n"),
905 xname
.name
= (unsigned char *)ORPHANAGE
;
906 xname
.len
= strlen(ORPHANAGE
);
907 if (libxfs_dir_lookup(NULL
, pip
, &xname
, &ino
, NULL
) == 0)
911 * could not be found, create it
914 tp
= libxfs_trans_alloc(mp
, 0);
915 xfs_bmap_init(&flist
, &first
);
917 nres
= XFS_MKDIR_SPACE_RES(mp
, xname
.len
);
918 i
= libxfs_trans_reserve(tp
, &M_RES(mp
)->tr_mkdir
, nres
, 0);
923 * use iget/ijoin instead of trans_iget because the ialloc
924 * wrapper can commit the transaction and start a new one
926 /* if ((i = libxfs_iget(mp, NULL, mp->m_sb.sb_rootino, 0, &pip, 0)))
927 do_error(_("%d - couldn't iget root inode to make %s\n"),
930 error
= libxfs_inode_alloc(&tp
, pip
, mode
|S_IFDIR
,
931 1, 0, &zerocr
, &zerofsx
, &ip
);
933 do_error(_("%s inode allocation failed %d\n"),
936 ip
->i_d
.di_nlink
++; /* account for . */
939 irec
= find_inode_rec(mp
,
940 XFS_INO_TO_AGNO(mp
, ino
),
941 XFS_INO_TO_AGINO(mp
, ino
));
945 * This inode is allocated from a newly created inode
946 * chunk and therefore did not exist when inode chunks
947 * were processed in phase3. Add this group of inodes to
948 * the entry avl tree as if they were discovered in phase3.
950 irec
= set_inode_free_alloc(mp
, XFS_INO_TO_AGNO(mp
, ino
),
951 XFS_INO_TO_AGINO(mp
, ino
));
954 for (i
= 0; i
< XFS_INODES_PER_CHUNK
; i
++)
955 set_inode_free(irec
, i
);
958 ino_offset
= get_inode_offset(mp
, ino
, irec
);
961 * Mark the inode allocated to lost+found as used in the AVL tree
962 * so it is not skipped in phase 7
964 set_inode_used(irec
, ino_offset
);
965 add_inode_ref(irec
, ino_offset
);
968 * now that we know the transaction will stay around,
969 * add the root inode to it
971 libxfs_trans_ijoin(tp
, pip
, 0);
974 * create the actual entry
976 error
= libxfs_dir_createname(tp
, pip
, &xname
, ip
->i_ino
, &first
,
980 _("can't make %s, createname error %d\n"),
984 * bump up the link count in the root directory to account
985 * for .. in the new directory
988 add_inode_ref(find_inode_rec(mp
,
989 XFS_INO_TO_AGNO(mp
, mp
->m_sb
.sb_rootino
),
990 XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_rootino
)), 0);
994 libxfs_trans_log_inode(tp
, pip
, XFS_ILOG_CORE
);
995 libxfs_dir_init(tp
, ip
, pip
);
996 libxfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
998 error
= libxfs_bmap_finish(&tp
, &flist
, &committed
);
1000 do_error(_("%s directory creation failed -- bmapf error %d\n"),
1005 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
1006 add_inode_reached(irec
,ino_offset
);
1012 * move a file to the orphange.
1017 xfs_ino_t ino
, /* inode # to be moved */
1018 int isa_dir
) /* 1 if inode is a directory */
1020 xfs_inode_t
*orphanage_ip
;
1021 xfs_ino_t entry_ino_num
;
1024 xfs_fsblock_t first
;
1025 xfs_bmap_free_t flist
;
1028 unsigned char fname
[MAXPATHLEN
+ 1];
1031 ino_tree_node_t
*irec
;
1033 struct xfs_name xname
;
1035 ASSERT(xfs_sb_version_hasdirv2(&mp
->m_sb
));
1038 xname
.len
= snprintf((char *)fname
, sizeof(fname
), "%llu",
1039 (unsigned long long)ino
);
1041 err
= libxfs_iget(mp
, NULL
, orphanage_ino
, 0, &orphanage_ip
, 0);
1043 do_error(_("%d - couldn't iget orphanage inode\n"), err
);
1045 * Make sure the filename is unique in the lost+found
1048 while (libxfs_dir_lookup(NULL
, orphanage_ip
, &xname
, &entry_ino_num
,
1050 xname
.len
= snprintf((char *)fname
, sizeof(fname
), "%llu.%d",
1051 (unsigned long long)ino
, ++incr
);
1053 tp
= libxfs_trans_alloc(mp
, 0);
1055 if ((err
= libxfs_iget(mp
, NULL
, ino
, 0, &ino_p
, 0)))
1056 do_error(_("%d - couldn't iget disconnected inode\n"), err
);
1059 irec
= find_inode_rec(mp
, XFS_INO_TO_AGNO(mp
, orphanage_ino
),
1060 XFS_INO_TO_AGINO(mp
, orphanage_ino
));
1062 ino_offset
= XFS_INO_TO_AGINO(mp
, orphanage_ino
) -
1064 nres
= XFS_DIRENTER_SPACE_RES(mp
, fnamelen
) +
1065 XFS_DIRENTER_SPACE_RES(mp
, 2);
1066 err
= libxfs_dir_lookup(tp
, ino_p
, &xfs_name_dotdot
,
1067 &entry_ino_num
, NULL
);
1069 ASSERT(err
== ENOENT
);
1071 err
= libxfs_trans_reserve(tp
, &M_RES(mp
)->tr_rename
,
1075 _("space reservation failed (%d), filesystem may be out of space\n"),
1078 libxfs_trans_ijoin(tp
, orphanage_ip
, 0);
1079 libxfs_trans_ijoin(tp
, ino_p
, 0);
1081 xfs_bmap_init(&flist
, &first
);
1082 err
= libxfs_dir_createname(tp
, orphanage_ip
, &xname
,
1083 ino
, &first
, &flist
, nres
);
1086 _("name create failed in %s (%d), filesystem may be out of space\n"),
1090 add_inode_ref(irec
, ino_offset
);
1092 orphanage_ip
->i_d
.di_nlink
++;
1093 libxfs_trans_log_inode(tp
, orphanage_ip
, XFS_ILOG_CORE
);
1095 err
= libxfs_dir_createname(tp
, ino_p
, &xfs_name_dotdot
,
1096 orphanage_ino
, &first
, &flist
, nres
);
1099 _("creation of .. entry failed (%d), filesystem may be out of space\n"),
1102 ino_p
->i_d
.di_nlink
++;
1103 libxfs_trans_log_inode(tp
, ino_p
, XFS_ILOG_CORE
);
1105 err
= libxfs_bmap_finish(&tp
, &flist
, &committed
);
1108 _("bmap finish failed (err - %d), filesystem may be out of space\n"),
1111 libxfs_trans_commit(tp
,
1112 XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
1114 err
= libxfs_trans_reserve(tp
, &M_RES(mp
)->tr_rename
,
1118 _("space reservation failed (%d), filesystem may be out of space\n"),
1121 libxfs_trans_ijoin(tp
, orphanage_ip
, 0);
1122 libxfs_trans_ijoin(tp
, ino_p
, 0);
1124 xfs_bmap_init(&flist
, &first
);
1126 err
= libxfs_dir_createname(tp
, orphanage_ip
, &xname
,
1127 ino
, &first
, &flist
, nres
);
1130 _("name create failed in %s (%d), filesystem may be out of space\n"),
1134 add_inode_ref(irec
, ino_offset
);
1136 orphanage_ip
->i_d
.di_nlink
++;
1137 libxfs_trans_log_inode(tp
, orphanage_ip
, XFS_ILOG_CORE
);
1140 * don't replace .. value if it already points
1141 * to us. that'll pop a libxfs/kernel ASSERT.
1143 if (entry_ino_num
!= orphanage_ino
) {
1144 err
= libxfs_dir_replace(tp
, ino_p
,
1145 &xfs_name_dotdot
, orphanage_ino
,
1146 &first
, &flist
, nres
);
1149 _("name replace op failed (%d), filesystem may be out of space\n"),
1153 err
= libxfs_bmap_finish(&tp
, &flist
, &committed
);
1156 _("bmap finish failed (%d), filesystem may be out of space\n"),
1159 libxfs_trans_commit(tp
,
1160 XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
1165 * use the remove log reservation as that's
1166 * more accurate. we're only creating the
1167 * links, we're not doing the inode allocation
1168 * also accounted for in the create
1170 nres
= XFS_DIRENTER_SPACE_RES(mp
, xname
.len
);
1171 err
= libxfs_trans_reserve(tp
, &M_RES(mp
)->tr_remove
,
1175 _("space reservation failed (%d), filesystem may be out of space\n"),
1178 libxfs_trans_ijoin(tp
, orphanage_ip
, 0);
1179 libxfs_trans_ijoin(tp
, ino_p
, 0);
1181 xfs_bmap_init(&flist
, &first
);
1182 err
= libxfs_dir_createname(tp
, orphanage_ip
, &xname
, ino
,
1183 &first
, &flist
, nres
);
1186 _("name create failed in %s (%d), filesystem may be out of space\n"),
1190 ino_p
->i_d
.di_nlink
= 1;
1191 libxfs_trans_log_inode(tp
, ino_p
, XFS_ILOG_CORE
);
1193 err
= libxfs_bmap_finish(&tp
, &flist
, &committed
);
1196 _("bmap finish failed (%d), filesystem may be out of space\n"),
1199 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
1210 do_warn(msg
, iname
, ino1
, ino2
);
1213 do_warn(_(", marking entry to be junked\n"));
1217 do_warn(_(", would junk entry\n"));
1222 * Unexpected failure during the rebuild will leave the entries in
1223 * lost+found on the next run
1227 longform_dir2_rebuild(
1231 ino_tree_node_t
*irec
,
1233 dir_hash_tab_t
*hashtab
)
1238 xfs_fileoff_t lastblock
;
1239 xfs_fsblock_t firstblock
;
1240 xfs_bmap_free_t flist
;
1247 * trash directory completely and rebuild from scratch using the
1248 * name/inode pairs in the hash table
1251 do_warn(_("rebuilding directory inode %" PRIu64
"\n"), ino
);
1254 * first attempt to locate the parent inode, if it can't be
1255 * found, set it to the root inode and it'll be moved to the
1256 * orphanage later (the inode number here needs to be valid
1257 * for the libxfs_dir_init() call).
1259 pip
.i_ino
= get_inode_parent(irec
, ino_offset
);
1260 if (pip
.i_ino
== NULLFSINO
)
1261 pip
.i_ino
= mp
->m_sb
.sb_rootino
;
1263 xfs_bmap_init(&flist
, &firstblock
);
1265 tp
= libxfs_trans_alloc(mp
, 0);
1266 nres
= XFS_REMOVE_SPACE_RES(mp
);
1267 error
= libxfs_trans_reserve(tp
, &M_RES(mp
)->tr_remove
, nres
, 0);
1270 libxfs_trans_ijoin(tp
, ip
, 0);
1271 libxfs_trans_ihold(tp
, ip
);
1273 if ((error
= libxfs_bmap_last_offset(tp
, ip
, &lastblock
,
1275 do_error(_("xfs_bmap_last_offset failed -- error - %d\n"),
1278 /* free all data, leaf, node and freespace blocks */
1279 error
= libxfs_bunmapi(tp
, ip
, 0, lastblock
, XFS_BMAPI_METADATA
, 0,
1280 &firstblock
, &flist
, &done
);
1282 do_warn(_("xfs_bunmapi failed -- error - %d\n"), error
);
1283 libxfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
|
1290 libxfs_dir_init(tp
, ip
, &pip
);
1292 error
= libxfs_bmap_finish(&tp
, &flist
, &committed
);
1294 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
1296 /* go through the hash list and re-add the inodes */
1298 for (p
= hashtab
->first
; p
; p
= p
->nextbyorder
) {
1300 if (p
->name
.name
[0] == '/' || (p
->name
.name
[0] == '.' &&
1301 (p
->name
.len
== 1 || (p
->name
.len
== 2 &&
1302 p
->name
.name
[1] == '.'))))
1305 tp
= libxfs_trans_alloc(mp
, 0);
1306 nres
= XFS_CREATE_SPACE_RES(mp
, p
->name
.len
);
1307 error
= libxfs_trans_reserve(tp
, &M_RES(mp
)->tr_create
,
1311 _("space reservation failed (%d), filesystem may be out of space\n"),
1316 libxfs_trans_ijoin(tp
, ip
, 0);
1317 libxfs_trans_ihold(tp
, ip
);
1319 xfs_bmap_init(&flist
, &firstblock
);
1320 error
= libxfs_dir_createname(tp
, ip
, &p
->name
, p
->inum
,
1321 &firstblock
, &flist
, nres
);
1324 _("name create failed in ino %" PRIu64
" (%d), filesystem may be out of space\n"),
1326 libxfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
|
1331 error
= libxfs_bmap_finish(&tp
, &flist
, &committed
);
1334 _("bmap finish failed (%d), filesystem may be out of space\n"),
1336 libxfs_bmap_cancel(&flist
);
1337 libxfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
|
1342 libxfs_trans_commit(tp
,
1343 XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
1349 * Kill a block in a version 2 inode.
1350 * Makes its own transaction.
1362 xfs_fsblock_t firstblock
;
1363 xfs_bmap_free_t flist
;
1367 tp
= libxfs_trans_alloc(mp
, 0);
1368 nres
= XFS_REMOVE_SPACE_RES(mp
);
1369 error
= libxfs_trans_reserve(tp
, &M_RES(mp
)->tr_remove
, nres
, 0);
1372 libxfs_trans_ijoin(tp
, ip
, 0);
1373 libxfs_trans_ihold(tp
, ip
);
1374 libxfs_trans_bjoin(tp
, bp
);
1375 memset(&args
, 0, sizeof(args
));
1376 xfs_bmap_init(&flist
, &firstblock
);
1379 args
.firstblock
= &firstblock
;
1380 args
.flist
= &flist
;
1381 args
.whichfork
= XFS_DATA_FORK
;
1382 if (da_bno
>= mp
->m_dirleafblk
&& da_bno
< mp
->m_dirfreeblk
)
1383 error
= libxfs_da_shrink_inode(&args
, da_bno
, bp
);
1385 error
= libxfs_dir2_shrink_inode(&args
,
1386 xfs_dir2_da_to_db(mp
, da_bno
), bp
);
1388 do_error(_("shrink_inode failed inode %" PRIu64
" block %u\n"),
1390 libxfs_bmap_finish(&tp
, &flist
, &committed
);
1391 libxfs_trans_commit(tp
, 0);
1395 * process a data block, also checks for .. entry
1396 * and corrects it to match what we think .. should be
1399 longform_dir2_entry_check_data(
1404 ino_tree_node_t
*current_irec
,
1405 int current_ino_offset
,
1406 struct xfs_buf
**bpp
,
1407 dir_hash_tab_t
*hashtab
,
1408 freetab_t
**freetabp
,
1412 xfs_dir2_dataptr_t addr
;
1413 xfs_dir2_leaf_entry_t
*blp
;
1415 xfs_dir2_block_tail_t
*btp
;
1417 struct xfs_dir2_data_hdr
*d
;
1419 xfs_dir2_data_entry_t
*dep
;
1420 xfs_dir2_data_unused_t
*dup
;
1421 struct xfs_dir2_data_free
*bf
;
1424 xfs_fsblock_t firstblock
;
1425 xfs_bmap_free_t flist
;
1426 char fname
[MAXNAMELEN
+ 1];
1431 ino_tree_node_t
*irec
;
1445 ptr
= (char *)xfs_dir3_data_entry_p(d
);
1447 needscan
= needlog
= 0;
1449 freetab
= *freetabp
;
1451 btp
= xfs_dir2_block_tail_p(mp
, (struct xfs_dir2_data_hdr
*)d
);
1452 blp
= xfs_dir2_block_leaf_p(btp
);
1453 endptr
= (char *)blp
;
1454 if (endptr
> (char *)btp
)
1455 endptr
= (char *)btp
;
1456 if (xfs_sb_version_hascrc(&mp
->m_sb
))
1457 wantmagic
= XFS_DIR3_BLOCK_MAGIC
;
1459 wantmagic
= XFS_DIR2_BLOCK_MAGIC
;
1461 endptr
= (char *)d
+ mp
->m_dirblksize
;
1462 if (xfs_sb_version_hascrc(&mp
->m_sb
))
1463 wantmagic
= XFS_DIR3_DATA_MAGIC
;
1465 wantmagic
= XFS_DIR2_DATA_MAGIC
;
1467 db
= xfs_dir2_da_to_db(mp
, da_bno
);
1469 /* check for data block beyond expected end */
1470 if (freetab
->naents
<= db
) {
1471 struct freetab_ent e
;
1473 *freetabp
= freetab
= realloc(freetab
, FREETAB_SIZE(db
+ 1));
1476 _("realloc failed in longform_dir2_entry_check_data (%zu bytes)\n"),
1477 FREETAB_SIZE(db
+ 1));
1481 for (i
= freetab
->naents
; i
< db
; i
++)
1482 freetab
->ents
[i
] = e
;
1483 freetab
->naents
= db
+ 1;
1486 /* check the data block */
1487 while (ptr
< endptr
) {
1489 /* check for freespace */
1490 dup
= (xfs_dir2_data_unused_t
*)ptr
;
1491 if (XFS_DIR2_DATA_FREE_TAG
== be16_to_cpu(dup
->freetag
)) {
1493 /* check for invalid freespace length */
1494 if (ptr
+ be16_to_cpu(dup
->length
) > endptr
||
1495 be16_to_cpu(dup
->length
) == 0 ||
1496 (be16_to_cpu(dup
->length
) &
1497 (XFS_DIR2_DATA_ALIGN
- 1)))
1500 /* check for invalid tag */
1501 if (be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup
)) !=
1502 (char *)dup
- (char *)d
)
1505 /* check for block with no data entries */
1506 if ((ptr
== (char *)xfs_dir3_data_entry_p(d
)) &&
1507 (ptr
+ be16_to_cpu(dup
->length
) >= endptr
)) {
1513 /* continue at the end of the freespace */
1514 ptr
+= be16_to_cpu(dup
->length
);
1519 /* validate data entry size */
1520 dep
= (xfs_dir2_data_entry_t
*)ptr
;
1521 if (ptr
+ xfs_dir3_data_entsize(mp
, dep
->namelen
) > endptr
)
1523 if (be16_to_cpu(*xfs_dir3_data_entry_tag_p(mp
, dep
)) !=
1524 (char *)dep
- (char *)d
)
1526 ptr
+= xfs_dir3_data_entsize(mp
, dep
->namelen
);
1529 /* did we find an empty or corrupt block? */
1530 if (ptr
!= endptr
) {
1533 _("empty data block %u in directory inode %" PRIu64
": "),
1537 ("corrupt block %u in directory inode %" PRIu64
": "),
1541 do_warn(_("junking block\n"));
1542 dir2_kill_block(mp
, ip
, da_bno
, bp
);
1544 do_warn(_("would junk block\n"));
1547 freetab
->ents
[db
].v
= NULLDATAOFF
;
1552 /* update number of data blocks processed */
1553 if (freetab
->nents
< db
+ 1)
1554 freetab
->nents
= db
+ 1;
1556 tp
= libxfs_trans_alloc(mp
, 0);
1557 error
= libxfs_trans_reserve(tp
, &M_RES(mp
)->tr_remove
, 0, 0);
1560 libxfs_trans_ijoin(tp
, ip
, 0);
1561 libxfs_trans_ihold(tp
, ip
);
1562 libxfs_trans_bjoin(tp
, bp
);
1563 libxfs_trans_bhold(tp
, bp
);
1564 xfs_bmap_init(&flist
, &firstblock
);
1565 if (be32_to_cpu(d
->magic
) != wantmagic
) {
1567 _("bad directory block magic # %#x for directory inode %" PRIu64
" block %d: "),
1568 be32_to_cpu(d
->magic
), ip
->i_ino
, da_bno
);
1570 do_warn(_("fixing magic # to %#x\n"), wantmagic
);
1571 d
->magic
= cpu_to_be32(wantmagic
);
1574 do_warn(_("would fix magic # to %#x\n"), wantmagic
);
1577 ptr
= (char *)xfs_dir3_data_entry_p(d
);
1579 * look at each entry. reference inode pointed to by each
1580 * entry in the incore inode tree.
1581 * if not a directory, set reached flag, increment link count
1582 * if a directory and reached, mark entry as to be deleted.
1583 * if a directory, check to see if recorded parent
1584 * matches current inode #,
1585 * if so, then set reached flag, increment link count
1586 * of current and child dir inodes, push the child
1587 * directory inode onto the directory stack.
1588 * if current inode != parent, then mark entry to be deleted.
1590 while (ptr
< endptr
) {
1591 dup
= (xfs_dir2_data_unused_t
*)ptr
;
1592 if (be16_to_cpu(dup
->freetag
) == XFS_DIR2_DATA_FREE_TAG
) {
1595 _("directory inode %" PRIu64
" block %u has consecutive free entries: "),
1598 do_warn(_("joining together\n"));
1599 len
= be16_to_cpu(dup
->length
);
1600 libxfs_dir2_data_use_free(tp
, bp
, dup
,
1601 ptr
- (char *)d
, len
, &needlog
,
1603 libxfs_dir2_data_make_free(tp
, bp
,
1604 ptr
- (char *)d
, len
, &needlog
,
1607 do_warn(_("would join together\n"));
1609 ptr
+= be16_to_cpu(dup
->length
);
1613 addr
= xfs_dir2_db_off_to_dataptr(mp
, db
, ptr
- (char *)d
);
1614 dep
= (xfs_dir2_data_entry_t
*)ptr
;
1615 ptr
+= xfs_dir3_data_entsize(mp
, dep
->namelen
);
1616 inum
= be64_to_cpu(dep
->inumber
);
1619 * skip bogus entries (leading '/'). they'll be deleted
1620 * later. must still log it, else we leak references to
1623 if (dep
->name
[0] == '/') {
1626 libxfs_dir2_data_log_entry(tp
, bp
, dep
);
1630 memmove(fname
, dep
->name
, dep
->namelen
);
1631 fname
[dep
->namelen
] = '\0';
1632 ASSERT(inum
!= NULLFSINO
);
1634 irec
= find_inode_rec(mp
, XFS_INO_TO_AGNO(mp
, inum
),
1635 XFS_INO_TO_AGINO(mp
, inum
));
1639 _("entry \"%s\" in directory inode %" PRIu64
" points to non-existent inode %" PRIu64
""),
1640 fname
, ip
->i_ino
, inum
)) {
1642 libxfs_dir2_data_log_entry(tp
, bp
, dep
);
1646 ino_offset
= XFS_INO_TO_AGINO(mp
, inum
) - irec
->ino_startnum
;
1649 * if it's a free inode, blow out the entry.
1650 * by now, any inode that we think is free
1653 if (is_inode_free(irec
, ino_offset
)) {
1656 _("entry \"%s\" in directory inode %" PRIu64
" points to free inode %" PRIu64
),
1657 fname
, ip
->i_ino
, inum
)) {
1659 libxfs_dir2_data_log_entry(tp
, bp
, dep
);
1665 * check if this inode is lost+found dir in the root
1667 if (inum
== mp
->m_sb
.sb_rootino
&& strcmp(fname
, ORPHANAGE
) == 0) {
1669 * if it's not a directory, trash it
1671 if (!inode_isadir(irec
, ino_offset
)) {
1674 _("%s (ino %" PRIu64
") in root (%" PRIu64
") is not a directory"),
1675 ORPHANAGE
, inum
, ip
->i_ino
)) {
1677 libxfs_dir2_data_log_entry(tp
, bp
, dep
);
1682 * if this is a dup, it will be picked up below,
1683 * otherwise, mark it as the orphanage for later.
1686 orphanage_ino
= inum
;
1689 * check for duplicate names in directory.
1691 if (!dir_hash_add(mp
, hashtab
, addr
, inum
, dep
->namelen
,
1695 _("entry \"%s\" (ino %" PRIu64
") in dir %" PRIu64
" is a duplicate name"),
1696 fname
, inum
, ip
->i_ino
)) {
1698 libxfs_dir2_data_log_entry(tp
, bp
, dep
);
1700 if (inum
== orphanage_ino
)
1706 * if just scanning to rebuild a directory due to a ".."
1707 * update, just continue
1713 * skip the '..' entry since it's checked when the
1714 * directory is reached by something else. if it never
1715 * gets reached, it'll be moved to the orphanage and we'll
1716 * take care of it then. If it doesn't exist at all, the
1717 * directory needs to be rebuilt first before being added
1720 if (dep
->namelen
== 2 && dep
->name
[0] == '.' &&
1721 dep
->name
[1] == '.') {
1723 /* ".." should be in the first block */
1726 _("entry \"%s\" (ino %" PRIu64
") in dir %" PRIu64
" is not in the the first block"), fname
,
1729 libxfs_dir2_data_log_entry(tp
, bp
, dep
);
1734 ASSERT(no_modify
|| !verify_inum(mp
, inum
));
1736 * special case the . entry. we know there's only one
1737 * '.' and only '.' points to itself because bogus entries
1738 * got trashed in phase 3 if there were > 1.
1739 * bump up link count for '.' but don't set reached
1740 * until we're actually reached by another directory
1741 * '..' is already accounted for or will be taken care
1742 * of when directory is moved to orphanage.
1744 if (ip
->i_ino
== inum
) {
1745 ASSERT(dep
->name
[0] == '.' && dep
->namelen
== 1);
1746 add_inode_ref(current_irec
, current_ino_offset
);
1748 dep
!= xfs_dir3_data_entry_p(d
)) {
1749 /* "." should be the first entry */
1752 _("entry \"%s\" in dir %" PRIu64
" is not the first entry"),
1753 fname
, inum
, ip
->i_ino
)) {
1755 libxfs_dir2_data_log_entry(tp
, bp
, dep
);
1762 * skip entries with bogus inumbers if we're in no modify mode
1764 if (no_modify
&& verify_inum(mp
, inum
))
1767 * check easy case first, regular inode, just bump
1768 * the link count and continue
1770 if (!inode_isadir(irec
, ino_offset
)) {
1771 add_inode_reached(irec
, ino_offset
);
1774 parent
= get_inode_parent(irec
, ino_offset
);
1775 ASSERT(parent
!= 0);
1778 * bump up the link counts in parent and child
1779 * directory but if the link doesn't agree with
1780 * the .. in the child, blow out the entry.
1781 * if the directory has already been reached,
1782 * blow away the entry also.
1784 if (is_inode_reached(irec
, ino_offset
)) {
1787 _("entry \"%s\" in dir %" PRIu64
" points to an already connected directory inode %" PRIu64
"\n"),
1788 fname
, ip
->i_ino
, inum
);
1789 } else if (parent
== ip
->i_ino
) {
1790 add_inode_reached(irec
, ino_offset
);
1791 add_inode_ref(current_irec
, current_ino_offset
);
1792 } else if (parent
== NULLFSINO
) {
1793 /* ".." was missing, but this entry refers to it,
1794 so, set it as the parent and mark for rebuild */
1796 _("entry \"%s\" in dir ino %" PRIu64
" doesn't have a .. entry, will set it in ino %" PRIu64
".\n"),
1797 fname
, ip
->i_ino
, inum
);
1798 set_inode_parent(irec
, ino_offset
, ip
->i_ino
);
1799 add_inode_reached(irec
, ino_offset
);
1800 add_inode_ref(current_irec
, current_ino_offset
);
1801 add_dotdot_update(XFS_INO_TO_AGNO(mp
, inum
), irec
,
1806 _("entry \"%s\" in dir inode %" PRIu64
" inconsistent with .. value (%" PRIu64
") in ino %" PRIu64
"\n"),
1807 fname
, ip
->i_ino
, parent
, inum
);
1810 if (inum
== orphanage_ino
)
1816 libxfs_dir2_data_log_entry(tp
, bp
, dep
);
1819 _("\twill clear entry \"%s\"\n"),
1822 do_warn(_("\twould clear entry \"%s\"\n"),
1827 *num_illegal
+= nbad
;
1829 libxfs_dir2_data_freescan(mp
, d
, &needlog
);
1831 libxfs_dir2_data_log_header(tp
, bp
);
1832 libxfs_bmap_finish(&tp
, &flist
, &committed
);
1833 libxfs_trans_commit(tp
, 0);
1835 /* record the largest free space in the freetab for later checking */
1836 bf
= xfs_dir3_data_bestfree_p(d
);
1837 freetab
->ents
[db
].v
= be16_to_cpu(bf
[0].length
);
1838 freetab
->ents
[db
].s
= 0;
1842 * Check contents of leaf-form block.
1845 longform_dir2_check_leaf(
1848 dir_hash_tab_t
*hashtab
,
1856 xfs_dir2_leaf_t
*leaf
;
1857 xfs_dir2_leaf_tail_t
*ltp
;
1859 struct xfs_dir2_leaf_entry
*ents
;
1860 struct xfs_dir3_icleaf_hdr leafhdr
;
1862 da_bno
= mp
->m_dirleafblk
;
1863 if (libxfs_da_read_buf(NULL
, ip
, da_bno
, -1, &bp
, XFS_DATA_FORK
,
1864 &xfs_dir3_leaf1_buf_ops
)) {
1866 _("can't read block %u for directory inode %" PRIu64
"\n"),
1871 xfs_dir3_leaf_hdr_from_disk(&leafhdr
, leaf
);
1872 ents
= xfs_dir3_leaf_ents_p(leaf
);
1873 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
1874 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
1875 if (!(leafhdr
.magic
== XFS_DIR2_LEAF1_MAGIC
||
1876 leafhdr
.magic
== XFS_DIR3_LEAF1_MAGIC
) ||
1877 leafhdr
.forw
|| leafhdr
.back
||
1878 leafhdr
.count
< leafhdr
.stale
||
1880 xfs_dir3_max_leaf_ents(mp
, leaf
) ||
1881 (char *)&ents
[leafhdr
.count
] > (char *)bestsp
) {
1883 _("leaf block %u for directory inode %" PRIu64
" bad header\n"),
1888 seeval
= dir_hash_see_all(hashtab
, ents
, leafhdr
.count
, leafhdr
.stale
);
1889 if (dir_hash_check(hashtab
, ip
, seeval
)) {
1893 badtail
= freetab
->nents
!= be32_to_cpu(ltp
->bestcount
);
1894 for (i
= 0; !badtail
&& i
< be32_to_cpu(ltp
->bestcount
); i
++) {
1895 freetab
->ents
[i
].s
= 1;
1896 badtail
= freetab
->ents
[i
].v
!= be16_to_cpu(bestsp
[i
]);
1900 _("leaf block %u for directory inode %" PRIu64
" bad tail\n"),
1910 * Check contents of the node blocks (leaves)
1911 * Looks for matching hash values for the data entries.
1914 longform_dir2_check_node(
1917 dir_hash_tab_t
*hashtab
,
1923 xfs_dir2_free_t
*free
;
1925 xfs_dir2_leaf_t
*leaf
;
1926 xfs_fileoff_t next_da_bno
;
1929 struct xfs_dir2_leaf_entry
*ents
;
1930 struct xfs_dir3_icleaf_hdr leafhdr
;
1931 struct xfs_dir3_icfree_hdr freehdr
;
1934 for (da_bno
= mp
->m_dirleafblk
, next_da_bno
= 0;
1935 next_da_bno
!= NULLFILEOFF
&& da_bno
< mp
->m_dirfreeblk
;
1936 da_bno
= (xfs_dablk_t
)next_da_bno
) {
1937 next_da_bno
= da_bno
+ mp
->m_dirblkfsbs
- 1;
1938 if (bmap_next_offset(NULL
, ip
, &next_da_bno
, XFS_DATA_FORK
))
1940 if (libxfs_da_read_buf(NULL
, ip
, da_bno
, -1, &bp
,
1941 XFS_DATA_FORK
, &xfs_dir3_leafn_buf_ops
)) {
1943 _("can't read leaf block %u for directory inode %" PRIu64
"\n"),
1948 xfs_dir3_leaf_hdr_from_disk(&leafhdr
, leaf
);
1949 ents
= xfs_dir3_leaf_ents_p(leaf
);
1950 if (!(leafhdr
.magic
== XFS_DIR2_LEAFN_MAGIC
||
1951 leafhdr
.magic
== XFS_DIR3_LEAFN_MAGIC
)) {
1952 if (leafhdr
.magic
== XFS_DA_NODE_MAGIC
||
1953 leafhdr
.magic
== XFS_DA3_NODE_MAGIC
) {
1958 _("unknown magic number %#x for block %u in directory inode %" PRIu64
"\n"),
1959 leafhdr
.magic
, da_bno
, ip
->i_ino
);
1963 if (leafhdr
.count
> xfs_dir3_max_leaf_ents(mp
, leaf
) ||
1964 leafhdr
.count
< leafhdr
.stale
) {
1966 _("leaf block %u for directory inode %" PRIu64
" bad header\n"),
1971 seeval
= dir_hash_see_all(hashtab
, ents
,
1972 leafhdr
.count
, leafhdr
.stale
);
1974 if (seeval
!= DIR_HASH_CK_OK
)
1977 if (dir_hash_check(hashtab
, ip
, seeval
))
1980 for (da_bno
= mp
->m_dirfreeblk
, next_da_bno
= 0;
1981 next_da_bno
!= NULLFILEOFF
;
1982 da_bno
= (xfs_dablk_t
)next_da_bno
) {
1983 next_da_bno
= da_bno
+ mp
->m_dirblkfsbs
- 1;
1984 if (bmap_next_offset(NULL
, ip
, &next_da_bno
, XFS_DATA_FORK
))
1986 if (libxfs_da_read_buf(NULL
, ip
, da_bno
, -1, &bp
,
1987 XFS_DATA_FORK
, &xfs_dir3_free_buf_ops
)) {
1989 _("can't read freespace block %u for directory inode %" PRIu64
"\n"),
1994 xfs_dir3_free_hdr_from_disk(&freehdr
, free
);
1995 bests
= xfs_dir3_free_bests_p(mp
, free
);
1996 fdb
= xfs_dir2_da_to_db(mp
, da_bno
);
1997 if (!(freehdr
.magic
== XFS_DIR2_FREE_MAGIC
||
1998 freehdr
.magic
== XFS_DIR3_FREE_MAGIC
) ||
2000 (fdb
- XFS_DIR2_FREE_FIRSTDB(mp
)) *
2001 xfs_dir3_free_max_bests(mp
) ||
2002 freehdr
.nvalid
< freehdr
.nused
) {
2004 _("free block %u for directory inode %" PRIu64
" bad header\n"),
2009 for (i
= used
= 0; i
< freehdr
.nvalid
; i
++) {
2010 if (i
+ freehdr
.firstdb
>= freetab
->nents
||
2011 freetab
->ents
[i
+ freehdr
.firstdb
].v
!=
2012 be16_to_cpu(bests
[i
])) {
2014 _("free block %u entry %i for directory ino %" PRIu64
" bad\n"),
2015 da_bno
, i
, ip
->i_ino
);
2019 used
+= be16_to_cpu(bests
[i
]) != NULLDATAOFF
;
2020 freetab
->ents
[i
+ freehdr
.firstdb
].s
= 1;
2022 if (used
!= freehdr
.nused
) {
2024 _("free block %u for directory inode %" PRIu64
" bad nused\n"),
2031 for (i
= 0; i
< freetab
->nents
; i
++) {
2032 if ((freetab
->ents
[i
].s
== 0) &&
2033 (freetab
->ents
[i
].v
!= NULLDATAOFF
)) {
2035 _("missing freetab entry %u for directory inode %" PRIu64
"\n"),
2044 * If a directory is corrupt, we need to read in as many entries as possible,
2045 * destroy the entry and create a new one with recovered name/inode pairs.
2046 * (ie. get libxfs to do all the grunt work)
2049 longform_dir2_entry_check(xfs_mount_t
*mp
,
2054 ino_tree_node_t
*irec
,
2056 dir_hash_tab_t
*hashtab
)
2058 struct xfs_buf
**bplist
;
2065 xfs_fileoff_t next_da_bno
;
2071 freetab
= malloc(FREETAB_SIZE(ip
->i_d
.di_size
/ mp
->m_dirblksize
));
2074 _("malloc failed in longform_dir2_entry_check (%" PRId64
" bytes)\n"),
2075 FREETAB_SIZE(ip
->i_d
.di_size
/ mp
->m_dirblksize
));
2078 freetab
->naents
= ip
->i_d
.di_size
/ mp
->m_dirblksize
;
2080 for (i
= 0; i
< freetab
->naents
; i
++) {
2081 freetab
->ents
[i
].v
= NULLDATAOFF
;
2082 freetab
->ents
[i
].s
= 0;
2084 num_bps
= freetab
->naents
;
2085 bplist
= calloc(num_bps
, sizeof(struct xfs_buf
*));
2086 /* is this a block, leaf, or node directory? */
2087 libxfs_dir2_isblock(NULL
, ip
, &isblock
);
2088 libxfs_dir2_isleaf(NULL
, ip
, &isleaf
);
2090 /* check directory "data" blocks (ie. name/inode pairs) */
2091 for (da_bno
= 0, next_da_bno
= 0;
2092 next_da_bno
!= NULLFILEOFF
&& da_bno
< mp
->m_dirleafblk
;
2093 da_bno
= (xfs_dablk_t
)next_da_bno
) {
2094 next_da_bno
= da_bno
+ mp
->m_dirblkfsbs
- 1;
2095 if (bmap_next_offset(NULL
, ip
, &next_da_bno
, XFS_DATA_FORK
))
2097 db
= xfs_dir2_da_to_db(mp
, da_bno
);
2098 if (db
>= num_bps
) {
2099 /* more data blocks than expected */
2101 bplist
= realloc(bplist
, num_bps
* sizeof(struct xfs_buf
*));
2104 _("realloc failed in longform_dir2_entry_check (%zu bytes)\n"),
2105 num_bps
* sizeof(struct xfs_buf
*));
2107 if (libxfs_da_read_buf(NULL
, ip
, da_bno
, -1, &bplist
[db
],
2108 XFS_DATA_FORK
, &xfs_dir3_data_buf_ops
)) {
2110 _("can't read data block %u for directory inode %" PRIu64
"\n"),
2113 continue; /* try and read all "data" blocks */
2115 longform_dir2_entry_check_data(mp
, ip
, num_illegal
, need_dot
,
2116 irec
, ino_offset
, &bplist
[db
], hashtab
,
2117 &freetab
, da_bno
, isblock
);
2119 fixit
= (*num_illegal
!= 0) || dir2_is_badino(ino
) || *need_dot
;
2121 if (!dotdot_update
) {
2122 /* check btree and freespace */
2124 struct xfs_dir2_data_hdr
*block
;
2125 xfs_dir2_block_tail_t
*btp
;
2126 xfs_dir2_leaf_entry_t
*blp
;
2128 block
= bplist
[0]->b_addr
;
2129 btp
= xfs_dir2_block_tail_p(mp
, block
);
2130 blp
= xfs_dir2_block_leaf_p(btp
);
2131 seeval
= dir_hash_see_all(hashtab
, blp
,
2132 be32_to_cpu(btp
->count
),
2133 be32_to_cpu(btp
->stale
));
2134 if (dir_hash_check(hashtab
, ip
, seeval
))
2136 } else if (isleaf
) {
2137 fixit
|= longform_dir2_check_leaf(mp
, ip
, hashtab
,
2140 fixit
|= longform_dir2_check_node(mp
, ip
, hashtab
,
2144 if (!no_modify
&& (fixit
|| dotdot_update
)) {
2145 dir_hash_dup_names(hashtab
);
2146 for (i
= 0; i
< freetab
->naents
; i
++)
2148 libxfs_putbuf(bplist
[i
]);
2149 longform_dir2_rebuild(mp
, ino
, ip
, irec
, ino_offset
, hashtab
);
2153 for (i
= 0; i
< freetab
->naents
; i
++)
2155 libxfs_putbuf(bplist
[i
]);
2163 * shortform directory v2 processing routines -- entry verification and
2164 * bad entry deletion (pruning).
2167 shortform_dir2_entry_check(xfs_mount_t
*mp
,
2171 ino_tree_node_t
*current_irec
,
2172 int current_ino_offset
,
2173 dir_hash_tab_t
*hashtab
)
2177 struct xfs_dir2_sf_hdr
*sfp
;
2178 xfs_dir2_sf_entry_t
*sfep
, *next_sfep
, *tmp_sfep
;
2180 ino_tree_node_t
*irec
;
2190 char fname
[MAXNAMELEN
+ 1];
2194 sfp
= (struct xfs_dir2_sf_hdr
*) ifp
->if_u1
.if_data
;
2198 max_size
= ifp
->if_bytes
;
2199 ASSERT(ip
->i_d
.di_size
<= ifp
->if_bytes
);
2202 * if just rebuild a directory due to a "..", update and return
2204 if (dotdot_update
) {
2205 parent
= get_inode_parent(current_irec
, current_ino_offset
);
2208 _("would set .. in sf dir inode %" PRIu64
" to %" PRIu64
"\n"),
2212 _("setting .. in sf dir inode %" PRIu64
" to %" PRIu64
"\n"),
2214 xfs_dir2_sf_put_parent_ino(sfp
, parent
);
2221 * no '.' entry in shortform dirs, just bump up ref count by 1
2222 * '..' was already (or will be) accounted for and checked when
2223 * the directory is reached or will be taken care of when the
2224 * directory is moved to orphanage.
2226 add_inode_ref(current_irec
, current_ino_offset
);
2229 * Initialise i8 counter -- the parent inode number counts as well.
2231 i8
= xfs_dir2_sf_get_parent_ino(sfp
) > XFS_DIR2_MAX_SHORT_INUM
;
2234 * now run through entries, stop at first bad entry, don't need
2235 * to skip over '..' since that's encoded in its own field and
2236 * no need to worry about '.' since it doesn't exist.
2238 sfep
= next_sfep
= xfs_dir2_sf_firstentry(sfp
);
2240 for (i
= 0; i
< sfp
->count
&& max_size
>
2241 (__psint_t
)next_sfep
- (__psint_t
)sfp
;
2242 sfep
= next_sfep
, i
++) {
2247 lino
= xfs_dir3_sfe_get_ino(mp
, sfp
, sfep
);
2249 namelen
= sfep
->namelen
;
2251 ASSERT(no_modify
|| namelen
> 0);
2253 if (no_modify
&& namelen
== 0) {
2255 * if we're really lucky, this is
2256 * the last entry in which case we
2257 * can use the dir size to set the
2258 * namelen value. otherwise, forget
2259 * it because we're not going to be
2260 * able to find the next entry.
2264 if (i
== sfp
->count
- 1) {
2265 namelen
= ip
->i_d
.di_size
-
2266 ((__psint_t
) &sfep
->name
[0] -
2270 * don't process the rest of the directory,
2271 * break out of processing loop
2275 } else if (no_modify
&& (__psint_t
) sfep
- (__psint_t
) sfp
+
2276 + xfs_dir3_sf_entsize(mp
, sfp
, sfep
->namelen
)
2277 > ip
->i_d
.di_size
) {
2280 if (i
== sfp
->count
- 1) {
2281 namelen
= ip
->i_d
.di_size
-
2282 ((__psint_t
) &sfep
->name
[0] -
2286 * don't process the rest of the directory,
2287 * break out of processing loop
2293 memmove(fname
, sfep
->name
, sfep
->namelen
);
2294 fname
[sfep
->namelen
] = '\0';
2296 ASSERT(no_modify
|| (lino
!= NULLFSINO
&& lino
!= 0));
2297 ASSERT(no_modify
|| !verify_inum(mp
, lino
));
2300 * Also skip entries with bogus inode numbers if we're
2301 * in no modify mode.
2304 if (no_modify
&& verify_inum(mp
, lino
)) {
2305 next_sfep
= (xfs_dir2_sf_entry_t
*)((__psint_t
)sfep
+
2306 xfs_dir3_sf_entsize(mp
, sfp
, sfep
->namelen
));
2310 irec
= find_inode_rec(mp
, XFS_INO_TO_AGNO(mp
, lino
),
2311 XFS_INO_TO_AGINO(mp
, lino
));
2315 _("entry \"%s\" in shortform directory %" PRIu64
" references non-existent inode %" PRIu64
"\n"),
2320 ino_offset
= XFS_INO_TO_AGINO(mp
, lino
) - irec
->ino_startnum
;
2323 * if it's a free inode, blow out the entry.
2324 * by now, any inode that we think is free
2327 if (is_inode_free(irec
, ino_offset
)) {
2329 _("entry \"%s\" in shortform directory inode %" PRIu64
" points to free inode %" PRIu64
"\n"),
2334 * check if this inode is lost+found dir in the root
2336 if (ino
== mp
->m_sb
.sb_rootino
&& strcmp(fname
, ORPHANAGE
) == 0) {
2338 * if it's not a directory, trash it
2340 if (!inode_isadir(irec
, ino_offset
)) {
2342 _("%s (ino %" PRIu64
") in root (%" PRIu64
") is not a directory"),
2343 ORPHANAGE
, lino
, ino
);
2347 * if this is a dup, it will be picked up below,
2348 * otherwise, mark it as the orphanage for later.
2351 orphanage_ino
= lino
;
2354 * check for duplicate names in directory.
2356 if (!dir_hash_add(mp
, hashtab
, (xfs_dir2_dataptr_t
)
2357 (sfep
- xfs_dir2_sf_firstentry(sfp
)),
2358 lino
, sfep
->namelen
, sfep
->name
)) {
2360 _("entry \"%s\" (ino %" PRIu64
") in dir %" PRIu64
" is a duplicate name"),
2365 if (!inode_isadir(irec
, ino_offset
)) {
2367 * check easy case first, regular inode, just bump
2370 add_inode_reached(irec
, ino_offset
);
2372 parent
= get_inode_parent(irec
, ino_offset
);
2375 * bump up the link counts in parent and child.
2376 * directory but if the link doesn't agree with
2377 * the .. in the child, blow out the entry
2379 if (is_inode_reached(irec
, ino_offset
)) {
2382 _("entry \"%s\" in directory inode %" PRIu64
2383 " references already connected inode %" PRIu64
".\n"),
2385 } else if (parent
== ino
) {
2386 add_inode_reached(irec
, ino_offset
);
2387 add_inode_ref(current_irec
, current_ino_offset
);
2388 } else if (parent
== NULLFSINO
) {
2389 /* ".." was missing, but this entry refers to it,
2390 so, set it as the parent and mark for rebuild */
2392 _("entry \"%s\" in dir ino %" PRIu64
" doesn't have a .. entry, will set it in ino %" PRIu64
".\n"),
2394 set_inode_parent(irec
, ino_offset
, ino
);
2395 add_inode_reached(irec
, ino_offset
);
2396 add_inode_ref(current_irec
, current_ino_offset
);
2397 add_dotdot_update(XFS_INO_TO_AGNO(mp
, lino
),
2402 _("entry \"%s\" in directory inode %" PRIu64
2403 " not consistent with .. value (%" PRIu64
2404 ") in inode %" PRIu64
",\n"),
2405 fname
, ino
, parent
, lino
);
2411 if (lino
== orphanage_ino
)
2414 tmp_elen
= xfs_dir3_sf_entsize(mp
, sfp
,
2416 tmp_sfep
= (xfs_dir2_sf_entry_t
*)
2417 ((__psint_t
) sfep
+ tmp_elen
);
2418 tmp_len
= max_size
- ((__psint_t
) tmp_sfep
2420 max_size
-= tmp_elen
;
2421 bytes_deleted
+= tmp_elen
;
2423 memmove(sfep
, tmp_sfep
, tmp_len
);
2426 memset((void *)((__psint_t
)sfep
+ tmp_len
), 0,
2430 * set the tmp value to the current
2431 * pointer so we'll process the entry
2437 * WARNING: drop the index i by one
2438 * so it matches the decremented count for
2439 * accurate comparisons in the loop test
2446 do_warn(_("junking entry\n"));
2450 do_warn(_("would junk entry\n"));
2452 } else if (lino
> XFS_DIR2_MAX_SHORT_INUM
)
2456 * go onto next entry unless we've just junked an
2457 * entry in which the current entry pointer points
2458 * to an unprocessed entry. have to take into entries
2459 * with bad namelen into account in no modify mode since we
2460 * calculate size based on next_sfep.
2462 ASSERT(no_modify
|| bad_sfnamelen
== 0);
2464 next_sfep
= (tmp_sfep
== NULL
)
2465 ? (xfs_dir2_sf_entry_t
*) ((__psint_t
) sfep
2467 ? xfs_dir3_sf_entsize(mp
, sfp
, sfep
->namelen
)
2468 : xfs_dir3_sf_entsize(mp
, sfp
, namelen
)))
2472 if (sfp
->i8count
!= i8
) {
2474 do_warn(_("would fix i8count in inode %" PRIu64
"\n"),
2478 tmp_sfep
= next_sfep
;
2479 process_sf_dir2_fixi8(mp
, sfp
, &tmp_sfep
);
2481 (__psint_t
)next_sfep
-
2482 (__psint_t
)tmp_sfep
;
2483 next_sfep
= tmp_sfep
;
2487 do_warn(_("fixing i8count in inode %" PRIu64
"\n"),
2493 * sync up sizes if required
2496 ASSERT(bytes_deleted
> 0);
2498 libxfs_idata_realloc(ip
, -bytes_deleted
, XFS_DATA_FORK
);
2499 ip
->i_d
.di_size
-= bytes_deleted
;
2502 if (ip
->i_d
.di_size
!= ip
->i_df
.if_bytes
) {
2503 ASSERT(ip
->i_df
.if_bytes
== (xfs_fsize_t
)
2504 ((__psint_t
) next_sfep
- (__psint_t
) sfp
));
2505 ip
->i_d
.di_size
= (xfs_fsize_t
)
2506 ((__psint_t
) next_sfep
- (__psint_t
) sfp
);
2508 _("setting size to %" PRId64
" bytes to reflect junked entries\n"),
2515 * processes all reachable inodes in directories
2520 xfs_agnumber_t agno
,
2521 ino_tree_node_t
*irec
,
2525 xfs_bmap_free_t flist
;
2526 xfs_fsblock_t first
;
2529 dir_hash_tab_t
*hashtab
;
2530 int need_dot
, committed
;
2531 int dirty
, num_illegal
, error
, nres
;
2533 ino
= XFS_AGINO_TO_INO(mp
, agno
, irec
->ino_startnum
+ ino_offset
);
2536 * open up directory inode, check all entries,
2537 * then call prune_dir_entries to remove all
2538 * remaining illegal directory entries.
2541 ASSERT(!is_inode_refchecked(irec
, ino_offset
) || dotdot_update
);
2543 error
= libxfs_iget(mp
, NULL
, ino
, 0, &ip
, 0);
2547 _("couldn't map inode %" PRIu64
", err = %d\n"),
2551 _("couldn't map inode %" PRIu64
", err = %d\n"),
2554 * see below for what we're doing if this
2555 * is root. Why do we need to do this here?
2556 * to ensure that the root doesn't show up
2557 * as being disconnected in the no_modify case.
2559 if (mp
->m_sb
.sb_rootino
== ino
) {
2560 add_inode_reached(irec
, 0);
2561 add_inode_ref(irec
, 0);
2565 add_inode_refchecked(irec
, 0);
2569 need_dot
= dirty
= num_illegal
= 0;
2571 if (mp
->m_sb
.sb_rootino
== ino
) {
2573 * mark root inode reached and bump up
2574 * link count for root inode to account
2575 * for '..' entry since the root inode is
2576 * never reached by a parent. we know
2577 * that root's '..' is always good --
2578 * guaranteed by phase 3 and/or below.
2580 add_inode_reached(irec
, ino_offset
);
2583 add_inode_refchecked(irec
, ino_offset
);
2585 hashtab
= dir_hash_init(ip
->i_d
.di_size
);
2588 * look for bogus entries
2590 switch (ip
->i_d
.di_format
) {
2591 case XFS_DINODE_FMT_EXTENTS
:
2592 case XFS_DINODE_FMT_BTREE
:
2594 * also check for missing '.' in longform dirs.
2595 * missing .. entries are added if required when
2596 * the directory is connected to lost+found. but
2597 * we need to create '.' entries here.
2599 longform_dir2_entry_check(mp
, ino
, ip
,
2600 &num_illegal
, &need_dot
,
2605 case XFS_DINODE_FMT_LOCAL
:
2606 tp
= libxfs_trans_alloc(mp
, 0);
2608 * using the remove reservation is overkill
2609 * since at most we'll only need to log the
2610 * inode but it's easier than wedging a
2611 * new define in ourselves.
2613 nres
= no_modify
? 0 : XFS_REMOVE_SPACE_RES(mp
);
2614 error
= libxfs_trans_reserve(tp
, &M_RES(mp
)->tr_remove
,
2619 libxfs_trans_ijoin(tp
, ip
, 0);
2620 libxfs_trans_ihold(tp
, ip
);
2622 shortform_dir2_entry_check(mp
, ino
, ip
, &dirty
,
2626 ASSERT(dirty
== 0 || (dirty
&& !no_modify
));
2628 libxfs_trans_log_inode(tp
, ip
,
2629 XFS_ILOG_CORE
| XFS_ILOG_DDATA
);
2630 libxfs_trans_commit(tp
,
2631 XFS_TRANS_RELEASE_LOG_RES
|
2634 libxfs_trans_cancel(tp
,
2635 XFS_TRANS_RELEASE_LOG_RES
);
2642 dir_hash_done(hashtab
);
2645 * if we have to create a .. for /, do it now *before*
2646 * we delete the bogus entries, otherwise the directory
2647 * could transform into a shortform dir which would
2648 * probably cause the simulation to choke. Even
2649 * if the illegal entries get shifted around, it's ok
2650 * because the entries are structurally intact and in
2651 * in hash-value order so the simulation won't get confused
2652 * if it has to move them around.
2654 if (!no_modify
&& need_root_dotdot
&& ino
== mp
->m_sb
.sb_rootino
) {
2655 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_LOCAL
);
2657 do_warn(_("recreating root directory .. entry\n"));
2659 tp
= libxfs_trans_alloc(mp
, 0);
2662 nres
= XFS_MKDIR_SPACE_RES(mp
, 2);
2663 error
= libxfs_trans_reserve(tp
, &M_RES(mp
)->tr_mkdir
, nres
, 0);
2667 libxfs_trans_ijoin(tp
, ip
, 0);
2668 libxfs_trans_ihold(tp
, ip
);
2670 xfs_bmap_init(&flist
, &first
);
2672 error
= libxfs_dir_createname(tp
, ip
, &xfs_name_dotdot
,
2673 ip
->i_ino
, &first
, &flist
, nres
);
2676 _("can't make \"..\" entry in root inode %" PRIu64
", createname error %d\n"), ino
, error
);
2678 libxfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
2680 error
= libxfs_bmap_finish(&tp
, &flist
, &committed
);
2682 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
|
2685 need_root_dotdot
= 0;
2686 } else if (need_root_dotdot
&& ino
== mp
->m_sb
.sb_rootino
) {
2687 do_warn(_("would recreate root directory .. entry\n"));
2691 * if we need to create the '.' entry, do so only if
2692 * the directory is a longform dir. if it's been
2693 * turned into a shortform dir, then the inode is ok
2694 * since shortform dirs have no '.' entry and the inode
2695 * has already been committed by prune_lf_dir_entry().
2699 * bump up our link count but don't
2700 * bump up the inode link count. chances
2701 * are good that even though we lost '.'
2702 * the inode link counts reflect '.' so
2703 * leave the inode link count alone and if
2704 * it turns out to be wrong, we'll catch
2707 add_inode_ref(irec
, ino_offset
);
2711 _("would create missing \".\" entry in dir ino %" PRIu64
"\n"),
2713 } else if (ip
->i_d
.di_format
!= XFS_DINODE_FMT_LOCAL
) {
2715 * need to create . entry in longform dir.
2718 _("creating missing \".\" entry in dir ino %" PRIu64
"\n"), ino
);
2720 tp
= libxfs_trans_alloc(mp
, 0);
2723 nres
= XFS_MKDIR_SPACE_RES(mp
, 1);
2724 error
= libxfs_trans_reserve(tp
, &M_RES(mp
)->tr_mkdir
,
2729 libxfs_trans_ijoin(tp
, ip
, 0);
2730 libxfs_trans_ihold(tp
, ip
);
2732 xfs_bmap_init(&flist
, &first
);
2734 error
= libxfs_dir_createname(tp
, ip
, &xfs_name_dot
,
2735 ip
->i_ino
, &first
, &flist
, nres
);
2738 _("can't make \".\" entry in dir ino %" PRIu64
", createname error %d\n"),
2741 libxfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
2743 error
= libxfs_bmap_finish(&tp
, &flist
, &committed
);
2745 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
2753 * mark realtime bitmap and summary inodes as reached.
2754 * quota inode will be marked here as well
2757 mark_standalone_inodes(xfs_mount_t
*mp
)
2759 ino_tree_node_t
*irec
;
2762 irec
= find_inode_rec(mp
, XFS_INO_TO_AGNO(mp
, mp
->m_sb
.sb_rbmino
),
2763 XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_rbmino
));
2765 ASSERT(irec
!= NULL
);
2767 offset
= XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_rbmino
) -
2770 add_inode_reached(irec
, offset
);
2772 irec
= find_inode_rec(mp
, XFS_INO_TO_AGNO(mp
, mp
->m_sb
.sb_rsumino
),
2773 XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_rsumino
));
2775 offset
= XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_rsumino
) -
2778 ASSERT(irec
!= NULL
);
2780 add_inode_reached(irec
, offset
);
2783 if (mp
->m_sb
.sb_uquotino
2784 && mp
->m_sb
.sb_uquotino
!= NULLFSINO
) {
2785 irec
= find_inode_rec(mp
, XFS_INO_TO_AGNO(mp
,
2786 mp
->m_sb
.sb_uquotino
),
2787 XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_uquotino
));
2788 offset
= XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_uquotino
)
2789 - irec
->ino_startnum
;
2790 add_inode_reached(irec
, offset
);
2792 if (mp
->m_sb
.sb_gquotino
2793 && mp
->m_sb
.sb_gquotino
!= NULLFSINO
) {
2794 irec
= find_inode_rec(mp
, XFS_INO_TO_AGNO(mp
,
2795 mp
->m_sb
.sb_gquotino
),
2796 XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_gquotino
));
2797 offset
= XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_gquotino
)
2798 - irec
->ino_startnum
;
2799 add_inode_reached(irec
, offset
);
2801 if (mp
->m_sb
.sb_pquotino
2802 && mp
->m_sb
.sb_pquotino
!= NULLFSINO
) {
2803 irec
= find_inode_rec(mp
, XFS_INO_TO_AGNO(mp
,
2804 mp
->m_sb
.sb_pquotino
),
2805 XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_pquotino
));
2806 offset
= XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_pquotino
)
2807 - irec
->ino_startnum
;
2808 add_inode_reached(irec
, offset
);
2814 check_for_orphaned_inodes(
2816 xfs_agnumber_t agno
,
2817 ino_tree_node_t
*irec
)
2822 for (i
= 0; i
< XFS_INODES_PER_CHUNK
; i
++) {
2823 ASSERT(is_inode_confirmed(irec
, i
));
2824 if (is_inode_free(irec
, i
))
2827 if (is_inode_reached(irec
, i
))
2830 ASSERT(inode_isadir(irec
, i
) ||
2831 num_inode_references(irec
, i
) == 0);
2833 ino
= XFS_AGINO_TO_INO(mp
, agno
, i
+ irec
->ino_startnum
);
2834 if (inode_isadir(irec
, i
))
2835 do_warn(_("disconnected dir inode %" PRIu64
", "), ino
);
2837 do_warn(_("disconnected inode %" PRIu64
", "), ino
);
2840 orphanage_ino
= mk_orphanage(mp
);
2841 do_warn(_("moving to %s\n"), ORPHANAGE
);
2842 mv_orphanage(mp
, ino
, inode_isadir(irec
, i
));
2844 do_warn(_("would move to %s\n"), ORPHANAGE
);
2847 * for read-only case, even though the inode isn't
2848 * really reachable, set the flag (and bump our link
2849 * count) anyway to fool phase 7
2851 add_inode_reached(irec
, i
);
2858 xfs_agnumber_t agno
,
2861 ino_tree_node_t
*irec
;
2863 prefetch_args_t
*pf_args
= arg
;
2865 wait_for_inode_prefetch(pf_args
);
2868 do_log(_(" - agno = %d\n"), agno
);
2870 for (irec
= findfirst_inode_rec(agno
); irec
; irec
= next_ino_rec(irec
)) {
2871 if (irec
->ino_isa_dir
== 0)
2875 sem_post(&pf_args
->ra_count
);
2877 for (i
= 0; i
< XFS_INODES_PER_CHUNK
; i
++) {
2878 if (inode_isadir(irec
, i
))
2879 process_dir_inode(wq
->mp
, agno
, irec
, i
);
2882 cleanup_inode_prefetch(pf_args
);
2886 update_missing_dotdot_entries(
2889 dotdot_update_t
*dir
;
2892 * these entries parents were updated, rebuild them again
2893 * set dotdot_update flag so processing routines do not count links
2896 while (dotdot_update_list
) {
2897 dir
= dotdot_update_list
;
2898 dotdot_update_list
= dir
->next
;
2899 process_dir_inode(mp
, dir
->agno
, dir
->irec
, dir
->ino_offset
);
2910 prefetch_args_t
*pf_args
[2];
2913 * we always do prefetch for phase 6 as it will fill in the gaps
2914 * not read during phase 3 prefetch.
2917 pf_args
[0] = start_inode_prefetch(0, 1, NULL
);
2918 for (i
= 0; i
< glob_agcount
; i
++) {
2919 pf_args
[(~i
) & 1] = start_inode_prefetch(i
+ 1, 1,
2921 traverse_function(&queue
, i
, pf_args
[i
& 1]);
2926 phase6(xfs_mount_t
*mp
)
2928 ino_tree_node_t
*irec
;
2931 memset(&zerocr
, 0, sizeof(struct cred
));
2932 memset(&zerofsx
, 0, sizeof(struct fsxattr
));
2935 do_log(_("Phase 6 - check inode connectivity...\n"));
2937 incore_ext_teardown(mp
);
2939 add_ino_ex_data(mp
);
2942 * verify existence of root directory - if we have to
2943 * make one, it's ok for the incore data structs not to
2944 * know about it since everything about it (and the other
2945 * inodes in its chunk if a new chunk was created) are ok
2947 if (need_root_inode
) {
2949 do_warn(_("reinitializing root directory\n"));
2951 need_root_inode
= 0;
2952 need_root_dotdot
= 0;
2954 do_warn(_("would reinitialize root directory\n"));
2960 do_warn(_("reinitializing realtime bitmap inode\n"));
2964 do_warn(_("would reinitialize realtime bitmap inode\n"));
2970 do_warn(_("reinitializing realtime summary inode\n"));
2974 do_warn(_("would reinitialize realtime summary inode\n"));
2980 _(" - resetting contents of realtime bitmap and summary inodes\n"));
2981 if (fill_rbmino(mp
)) {
2983 _("Warning: realtime bitmap may be inconsistent\n"));
2986 if (fill_rsumino(mp
)) {
2988 _("Warning: realtime bitmap may be inconsistent\n"));
2992 mark_standalone_inodes(mp
);
2994 do_log(_(" - traversing filesystem ...\n"));
2996 irec
= find_inode_rec(mp
, XFS_INO_TO_AGNO(mp
, mp
->m_sb
.sb_rootino
),
2997 XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_rootino
));
3000 * we always have a root inode, even if it's free...
3001 * if the root is free, forget it, lost+found is already gone
3003 if (is_inode_free(irec
, 0) || !inode_isadir(irec
, 0)) {
3004 need_root_inode
= 1;
3008 * then process all inodes by walking incore inode tree
3013 * any directories that had updated ".." entries, rebuild them now
3015 update_missing_dotdot_entries(mp
);
3017 do_log(_(" - traversal finished ...\n"));
3018 do_log(_(" - moving disconnected inodes to %s ...\n"),
3022 * move all disconnected inodes to the orphanage
3024 for (i
= 0; i
< glob_agcount
; i
++) {
3025 irec
= findfirst_inode_rec(i
);
3026 while (irec
!= NULL
) {
3027 check_for_orphaned_inodes(mp
, i
, irec
);
3028 irec
= next_ino_rec(irec
);