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
];
434 tp
= libxfs_trans_alloc(mp
, 0);
436 if ((i
= libxfs_trans_reserve(tp
, 10, 0, 0, 0, 0)))
439 error
= libxfs_trans_iget(mp
, tp
, mp
->m_sb
.sb_rbmino
, 0, 0, &ip
);
442 _("couldn't iget realtime bitmap inode -- error - %d\n"),
446 memset(&ip
->i_d
, 0, sizeof(xfs_icdinode_t
));
448 ip
->i_d
.di_magic
= XFS_DINODE_MAGIC
;
449 ip
->i_d
.di_mode
= S_IFREG
;
450 ip
->i_d
.di_version
= 1;
451 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
452 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
454 ip
->i_d
.di_nlink
= 1; /* account for sb ptr */
459 ip
->i_df
.if_flags
= XFS_IFEXTENTS
;
460 ip
->i_df
.if_bytes
= ip
->i_df
.if_real_bytes
= 0;
461 ip
->i_df
.if_u1
.if_extents
= NULL
;
463 ip
->i_d
.di_size
= mp
->m_sb
.sb_rbmblocks
* mp
->m_sb
.sb_blocksize
;
468 libxfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
469 libxfs_trans_ihold(tp
, ip
);
470 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
473 * then allocate blocks for file and fill with zeroes (stolen
476 tp
= libxfs_trans_alloc(mp
, 0);
477 if ((error
= libxfs_trans_reserve(tp
, mp
->m_sb
.sb_rbmblocks
+
478 (XFS_BM_MAXLEVELS(mp
, XFS_DATA_FORK
) - 1), 0, 0, 0, 0)))
481 libxfs_trans_ijoin(tp
, ip
, 0);
483 xfs_bmap_init(&flist
, &first
);
484 while (bno
< mp
->m_sb
.sb_rbmblocks
) {
485 nmap
= XFS_BMAP_MAX_NMAP
;
486 error
= libxfs_bmapi_write(tp
, ip
, bno
,
487 (xfs_extlen_t
)(mp
->m_sb
.sb_rbmblocks
- bno
),
488 0, &first
, mp
->m_sb
.sb_rbmblocks
,
492 _("couldn't allocate realtime bitmap, error = %d\n"),
495 for (i
= 0, ep
= map
; i
< nmap
; i
++, ep
++) {
496 libxfs_device_zero(mp
->m_dev
,
497 XFS_FSB_TO_DADDR(mp
, ep
->br_startblock
),
498 XFS_FSB_TO_BB(mp
, ep
->br_blockcount
));
499 bno
+= ep
->br_blockcount
;
502 error
= libxfs_bmap_finish(&tp
, &flist
, &committed
);
505 _("allocation of the realtime bitmap failed, error = %d\n"),
508 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
512 fill_rbmino(xfs_mount_t
*mp
)
527 tp
= libxfs_trans_alloc(mp
, 0);
529 if ((error
= libxfs_trans_reserve(tp
, 10, 0, 0, 0, 0)))
532 error
= libxfs_trans_iget(mp
, tp
, mp
->m_sb
.sb_rbmino
, 0, 0, &ip
);
535 _("couldn't iget realtime bitmap inode -- error - %d\n"),
539 while (bno
< mp
->m_sb
.sb_rbmblocks
) {
541 * fill the file one block at a time
544 error
= libxfs_bmapi_write(tp
, ip
, bno
, 1, 0,
545 &first
, 1, &map
, &nmap
, NULL
);
546 if (error
|| nmap
!= 1) {
548 _("couldn't map realtime bitmap block %" PRIu64
", error = %d\n"),
552 ASSERT(map
.br_startblock
!= HOLESTARTBLOCK
);
554 error
= libxfs_trans_read_buf(
556 XFS_FSB_TO_DADDR(mp
, map
.br_startblock
),
557 XFS_FSB_TO_BB(mp
, 1), 1, &bp
, NULL
);
561 _("can't access block %" PRIu64
" (fsbno %" PRIu64
") of realtime bitmap inode %" PRIu64
"\n"),
562 bno
, map
.br_startblock
, mp
->m_sb
.sb_rbmino
);
566 memmove(XFS_BUF_PTR(bp
), bmp
, mp
->m_sb
.sb_blocksize
);
568 libxfs_trans_log_buf(tp
, bp
, 0, mp
->m_sb
.sb_blocksize
- 1);
570 bmp
= (xfs_rtword_t
*)((__psint_t
) bmp
+ mp
->m_sb
.sb_blocksize
);
574 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
579 fill_rsumino(xfs_mount_t
*mp
)
589 xfs_dfiloff_t end_bno
;
594 end_bno
= mp
->m_rsumsize
>> mp
->m_sb
.sb_blocklog
;
596 tp
= libxfs_trans_alloc(mp
, 0);
598 if ((error
= libxfs_trans_reserve(tp
, 10, 0, 0, 0, 0)))
601 error
= libxfs_trans_iget(mp
, tp
, mp
->m_sb
.sb_rsumino
, 0, 0, &ip
);
604 _("couldn't iget realtime summary inode -- error - %d\n"),
608 while (bno
< end_bno
) {
610 * fill the file one block at a time
613 error
= libxfs_bmapi_write(tp
, ip
, bno
, 1, 0,
614 &first
, 1, &map
, &nmap
, NULL
);
615 if (error
|| nmap
!= 1) {
617 _("couldn't map realtime summary inode block %" PRIu64
", error = %d\n"),
621 ASSERT(map
.br_startblock
!= HOLESTARTBLOCK
);
623 error
= libxfs_trans_read_buf(
625 XFS_FSB_TO_DADDR(mp
, map
.br_startblock
),
626 XFS_FSB_TO_BB(mp
, 1), 1, &bp
, NULL
);
630 _("can't access block %" PRIu64
" (fsbno %" PRIu64
") of realtime summary inode %" PRIu64
"\n"),
631 bno
, map
.br_startblock
, mp
->m_sb
.sb_rsumino
);
635 memmove(XFS_BUF_PTR(bp
), smp
, mp
->m_sb
.sb_blocksize
);
637 libxfs_trans_log_buf(tp
, bp
, 0, mp
->m_sb
.sb_blocksize
- 1);
639 smp
= (xfs_suminfo_t
*)((__psint_t
)smp
+ mp
->m_sb
.sb_blocksize
);
643 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
648 mk_rsumino(xfs_mount_t
*mp
)
659 xfs_bmap_free_t flist
;
661 xfs_bmbt_irec_t map
[XFS_BMAP_MAX_NMAP
];
666 tp
= libxfs_trans_alloc(mp
, 0);
668 if ((i
= libxfs_trans_reserve(tp
, 10, XFS_ICHANGE_LOG_RES(mp
), 0,
669 XFS_TRANS_PERM_LOG_RES
, XFS_MKDIR_LOG_COUNT
)))
672 error
= libxfs_trans_iget(mp
, tp
, mp
->m_sb
.sb_rsumino
, 0, 0, &ip
);
675 _("couldn't iget realtime summary inode -- error - %d\n"),
679 memset(&ip
->i_d
, 0, sizeof(xfs_icdinode_t
));
681 ip
->i_d
.di_magic
= XFS_DINODE_MAGIC
;
682 ip
->i_d
.di_mode
= S_IFREG
;
683 ip
->i_d
.di_version
= 1;
684 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
685 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
687 ip
->i_d
.di_nlink
= 1; /* account for sb ptr */
692 ip
->i_df
.if_flags
= XFS_IFEXTENTS
;
693 ip
->i_df
.if_bytes
= ip
->i_df
.if_real_bytes
= 0;
694 ip
->i_df
.if_u1
.if_extents
= NULL
;
696 ip
->i_d
.di_size
= mp
->m_rsumsize
;
701 libxfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
702 libxfs_trans_ihold(tp
, ip
);
703 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
706 * then allocate blocks for file and fill with zeroes (stolen
709 tp
= libxfs_trans_alloc(mp
, 0);
710 xfs_bmap_init(&flist
, &first
);
712 nsumblocks
= mp
->m_rsumsize
>> mp
->m_sb
.sb_blocklog
;
713 if ((error
= libxfs_trans_reserve(tp
,
714 mp
->m_sb
.sb_rbmblocks
+
715 (XFS_BM_MAXLEVELS(mp
, XFS_DATA_FORK
) - 1),
716 BBTOB(128), 0, XFS_TRANS_PERM_LOG_RES
,
717 XFS_DEFAULT_PERM_LOG_COUNT
)))
720 libxfs_trans_ijoin(tp
, ip
, 0);
722 xfs_bmap_init(&flist
, &first
);
723 while (bno
< nsumblocks
) {
724 nmap
= XFS_BMAP_MAX_NMAP
;
725 error
= libxfs_bmapi_write(tp
, ip
, bno
,
726 (xfs_extlen_t
)(nsumblocks
- bno
),
727 0, &first
, nsumblocks
, map
, &nmap
, &flist
);
730 _("couldn't allocate realtime summary inode, error = %d\n"),
733 for (i
= 0, ep
= map
; i
< nmap
; i
++, ep
++) {
734 libxfs_device_zero(mp
->m_dev
,
735 XFS_FSB_TO_DADDR(mp
, ep
->br_startblock
),
736 XFS_FSB_TO_BB(mp
, ep
->br_blockcount
));
737 bno
+= ep
->br_blockcount
;
740 error
= libxfs_bmap_finish(&tp
, &flist
, &committed
);
743 _("allocation of the realtime summary ino failed, error = %d\n"),
746 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
750 * makes a new root directory.
753 mk_root_dir(xfs_mount_t
*mp
)
759 const mode_t mode
= 0755;
760 ino_tree_node_t
*irec
;
762 ASSERT(xfs_sb_version_hasdirv2(&mp
->m_sb
));
764 tp
= libxfs_trans_alloc(mp
, 0);
767 if ((i
= libxfs_trans_reserve(tp
, 10, XFS_ICHANGE_LOG_RES(mp
), 0,
768 XFS_TRANS_PERM_LOG_RES
, XFS_MKDIR_LOG_COUNT
)))
771 error
= libxfs_trans_iget(mp
, tp
, mp
->m_sb
.sb_rootino
, 0, 0, &ip
);
773 do_error(_("could not iget root inode -- error - %d\n"), error
);
777 * take care of the core -- initialization from xfs_ialloc()
779 memset(&ip
->i_d
, 0, sizeof(xfs_icdinode_t
));
781 ip
->i_d
.di_magic
= XFS_DINODE_MAGIC
;
782 ip
->i_d
.di_mode
= (__uint16_t
) mode
|S_IFDIR
;
783 ip
->i_d
.di_version
= 1;
784 ip
->i_d
.di_format
= XFS_DINODE_FMT_EXTENTS
;
785 ip
->i_d
.di_aformat
= XFS_DINODE_FMT_EXTENTS
;
787 ip
->i_d
.di_nlink
= 1; /* account for . */
789 libxfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
794 ip
->i_df
.if_flags
= XFS_IFEXTENTS
;
795 ip
->i_df
.if_bytes
= ip
->i_df
.if_real_bytes
= 0;
796 ip
->i_df
.if_u1
.if_extents
= NULL
;
801 * initialize the directory
803 libxfs_dir_init(tp
, ip
, ip
);
805 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
807 irec
= find_inode_rec(mp
, XFS_INO_TO_AGNO(mp
, mp
->m_sb
.sb_rootino
),
808 XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_rootino
));
809 set_inode_isadir(irec
, XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_rootino
) -
814 * orphanage name == lost+found
817 mk_orphanage(xfs_mount_t
*mp
)
824 ino_tree_node_t
*irec
;
829 xfs_bmap_free_t flist
;
830 const int mode
= 0755;
832 struct xfs_name xname
;
834 ASSERT(xfs_sb_version_hasdirv2(&mp
->m_sb
));
837 * check for an existing lost+found first, if it exists, return
838 * its inode. Otherwise, we can create it. Bad lost+found inodes
839 * would have been cleared in phase3 and phase4.
842 if ((i
= libxfs_iget(mp
, NULL
, mp
->m_sb
.sb_rootino
, 0, &pip
, 0)))
843 do_error(_("%d - couldn't iget root inode to obtain %s\n"),
846 xname
.name
= (unsigned char *)ORPHANAGE
;
847 xname
.len
= strlen(ORPHANAGE
);
848 if (libxfs_dir_lookup(NULL
, pip
, &xname
, &ino
, NULL
) == 0)
852 * could not be found, create it
855 tp
= libxfs_trans_alloc(mp
, 0);
856 xfs_bmap_init(&flist
, &first
);
858 nres
= XFS_MKDIR_SPACE_RES(mp
, xname
.len
);
859 if ((i
= libxfs_trans_reserve(tp
, nres
, XFS_MKDIR_LOG_RES(mp
), 0,
860 XFS_TRANS_PERM_LOG_RES
, XFS_MKDIR_LOG_COUNT
)))
864 * use iget/ijoin instead of trans_iget because the ialloc
865 * wrapper can commit the transaction and start a new one
867 /* if ((i = libxfs_iget(mp, NULL, mp->m_sb.sb_rootino, 0, &pip, 0)))
868 do_error(_("%d - couldn't iget root inode to make %s\n"),
871 error
= libxfs_inode_alloc(&tp
, pip
, mode
|S_IFDIR
,
872 1, 0, &zerocr
, &zerofsx
, &ip
);
874 do_error(_("%s inode allocation failed %d\n"),
877 ip
->i_d
.di_nlink
++; /* account for . */
880 irec
= find_inode_rec(mp
,
881 XFS_INO_TO_AGNO(mp
, ino
),
882 XFS_INO_TO_AGINO(mp
, ino
));
883 ino_offset
= get_inode_offset(mp
, ino
, irec
);
886 * Mark the inode allocated to lost+found as used in the AVL tree
887 * so it is not skipped in phase 7
889 set_inode_used(irec
, ino_offset
);
890 add_inode_ref(irec
, ino_offset
);
893 * now that we know the transaction will stay around,
894 * add the root inode to it
896 libxfs_trans_ijoin(tp
, pip
, 0);
899 * create the actual entry
901 error
= libxfs_dir_createname(tp
, pip
, &xname
, ip
->i_ino
, &first
,
905 _("can't make %s, createname error %d\n"),
909 * bump up the link count in the root directory to account
910 * for .. in the new directory
913 add_inode_ref(find_inode_rec(mp
,
914 XFS_INO_TO_AGNO(mp
, mp
->m_sb
.sb_rootino
),
915 XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_rootino
)), 0);
919 libxfs_trans_log_inode(tp
, pip
, XFS_ILOG_CORE
);
920 libxfs_dir_init(tp
, ip
, pip
);
921 libxfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
923 error
= libxfs_bmap_finish(&tp
, &flist
, &committed
);
925 do_error(_("%s directory creation failed -- bmapf error %d\n"),
930 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
931 add_inode_reached(irec
,ino_offset
);
937 * move a file to the orphange.
942 xfs_ino_t ino
, /* inode # to be moved */
943 int isa_dir
) /* 1 if inode is a directory */
945 xfs_inode_t
*orphanage_ip
;
946 xfs_ino_t entry_ino_num
;
950 xfs_bmap_free_t flist
;
953 unsigned char fname
[MAXPATHLEN
+ 1];
956 ino_tree_node_t
*irec
;
958 struct xfs_name xname
;
960 ASSERT(xfs_sb_version_hasdirv2(&mp
->m_sb
));
963 xname
.len
= snprintf((char *)fname
, sizeof(fname
), "%llu",
964 (unsigned long long)ino
);
966 err
= libxfs_iget(mp
, NULL
, orphanage_ino
, 0, &orphanage_ip
, 0);
968 do_error(_("%d - couldn't iget orphanage inode\n"), err
);
970 * Make sure the filename is unique in the lost+found
973 while (libxfs_dir_lookup(NULL
, orphanage_ip
, &xname
, &entry_ino_num
,
975 xname
.len
= snprintf((char *)fname
, sizeof(fname
), "%llu.%d",
976 (unsigned long long)ino
, ++incr
);
978 tp
= libxfs_trans_alloc(mp
, 0);
980 if ((err
= libxfs_iget(mp
, NULL
, ino
, 0, &ino_p
, 0)))
981 do_error(_("%d - couldn't iget disconnected inode\n"), err
);
984 irec
= find_inode_rec(mp
, XFS_INO_TO_AGNO(mp
, orphanage_ino
),
985 XFS_INO_TO_AGINO(mp
, orphanage_ino
));
987 ino_offset
= XFS_INO_TO_AGINO(mp
, orphanage_ino
) -
989 nres
= XFS_DIRENTER_SPACE_RES(mp
, fnamelen
) +
990 XFS_DIRENTER_SPACE_RES(mp
, 2);
991 err
= libxfs_dir_lookup(tp
, ino_p
, &xfs_name_dotdot
,
992 &entry_ino_num
, NULL
);
994 ASSERT(err
== ENOENT
);
996 if ((err
= libxfs_trans_reserve(tp
, nres
,
997 XFS_RENAME_LOG_RES(mp
), 0,
998 XFS_TRANS_PERM_LOG_RES
,
999 XFS_RENAME_LOG_COUNT
)))
1001 _("space reservation failed (%d), filesystem may be out of space\n"),
1004 libxfs_trans_ijoin(tp
, orphanage_ip
, 0);
1005 libxfs_trans_ijoin(tp
, ino_p
, 0);
1007 xfs_bmap_init(&flist
, &first
);
1008 err
= libxfs_dir_createname(tp
, orphanage_ip
, &xname
,
1009 ino
, &first
, &flist
, nres
);
1012 _("name create failed in %s (%d), filesystem may be out of space\n"),
1016 add_inode_ref(irec
, ino_offset
);
1018 orphanage_ip
->i_d
.di_nlink
++;
1019 libxfs_trans_log_inode(tp
, orphanage_ip
, XFS_ILOG_CORE
);
1021 err
= libxfs_dir_createname(tp
, ino_p
, &xfs_name_dotdot
,
1022 orphanage_ino
, &first
, &flist
, nres
);
1025 _("creation of .. entry failed (%d), filesystem may be out of space\n"),
1028 ino_p
->i_d
.di_nlink
++;
1029 libxfs_trans_log_inode(tp
, ino_p
, XFS_ILOG_CORE
);
1031 err
= libxfs_bmap_finish(&tp
, &flist
, &committed
);
1034 _("bmap finish failed (err - %d), filesystem may be out of space\n"),
1037 libxfs_trans_commit(tp
,
1038 XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
1040 if ((err
= libxfs_trans_reserve(tp
, nres
,
1041 XFS_RENAME_LOG_RES(mp
), 0,
1042 XFS_TRANS_PERM_LOG_RES
,
1043 XFS_RENAME_LOG_COUNT
)))
1045 _("space reservation failed (%d), filesystem may be out of space\n"),
1048 libxfs_trans_ijoin(tp
, orphanage_ip
, 0);
1049 libxfs_trans_ijoin(tp
, ino_p
, 0);
1051 xfs_bmap_init(&flist
, &first
);
1053 err
= libxfs_dir_createname(tp
, orphanage_ip
, &xname
,
1054 ino
, &first
, &flist
, nres
);
1057 _("name create failed in %s (%d), filesystem may be out of space\n"),
1061 add_inode_ref(irec
, ino_offset
);
1063 orphanage_ip
->i_d
.di_nlink
++;
1064 libxfs_trans_log_inode(tp
, orphanage_ip
, XFS_ILOG_CORE
);
1067 * don't replace .. value if it already points
1068 * to us. that'll pop a libxfs/kernel ASSERT.
1070 if (entry_ino_num
!= orphanage_ino
) {
1071 err
= libxfs_dir_replace(tp
, ino_p
,
1072 &xfs_name_dotdot
, orphanage_ino
,
1073 &first
, &flist
, nres
);
1076 _("name replace op failed (%d), filesystem may be out of space\n"),
1080 err
= libxfs_bmap_finish(&tp
, &flist
, &committed
);
1083 _("bmap finish failed (%d), filesystem may be out of space\n"),
1086 libxfs_trans_commit(tp
,
1087 XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
1092 * use the remove log reservation as that's
1093 * more accurate. we're only creating the
1094 * links, we're not doing the inode allocation
1095 * also accounted for in the create
1097 nres
= XFS_DIRENTER_SPACE_RES(mp
, xname
.len
);
1098 err
= libxfs_trans_reserve(tp
, nres
, XFS_REMOVE_LOG_RES(mp
), 0,
1099 XFS_TRANS_PERM_LOG_RES
, XFS_REMOVE_LOG_COUNT
);
1102 _("space reservation failed (%d), filesystem may be out of space\n"),
1105 libxfs_trans_ijoin(tp
, orphanage_ip
, 0);
1106 libxfs_trans_ijoin(tp
, ino_p
, 0);
1108 xfs_bmap_init(&flist
, &first
);
1109 err
= libxfs_dir_createname(tp
, orphanage_ip
, &xname
, ino
,
1110 &first
, &flist
, nres
);
1113 _("name create failed in %s (%d), filesystem may be out of space\n"),
1117 ino_p
->i_d
.di_nlink
= 1;
1118 libxfs_trans_log_inode(tp
, ino_p
, XFS_ILOG_CORE
);
1120 err
= libxfs_bmap_finish(&tp
, &flist
, &committed
);
1123 _("bmap finish failed (%d), filesystem may be out of space\n"),
1126 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
1137 do_warn(msg
, iname
, ino1
, ino2
);
1140 do_warn(_(", marking entry to be junked\n"));
1144 do_warn(_(", would junk entry\n"));
1149 * Unexpected failure during the rebuild will leave the entries in
1150 * lost+found on the next run
1154 longform_dir2_rebuild(
1158 ino_tree_node_t
*irec
,
1160 dir_hash_tab_t
*hashtab
)
1165 xfs_fileoff_t lastblock
;
1166 xfs_fsblock_t firstblock
;
1167 xfs_bmap_free_t flist
;
1174 * trash directory completely and rebuild from scratch using the
1175 * name/inode pairs in the hash table
1178 do_warn(_("rebuilding directory inode %" PRIu64
"\n"), ino
);
1181 * first attempt to locate the parent inode, if it can't be
1182 * found, set it to the root inode and it'll be moved to the
1183 * orphanage later (the inode number here needs to be valid
1184 * for the libxfs_dir_init() call).
1186 pip
.i_ino
= get_inode_parent(irec
, ino_offset
);
1187 if (pip
.i_ino
== NULLFSINO
)
1188 pip
.i_ino
= mp
->m_sb
.sb_rootino
;
1190 xfs_bmap_init(&flist
, &firstblock
);
1192 tp
= libxfs_trans_alloc(mp
, 0);
1193 nres
= XFS_REMOVE_SPACE_RES(mp
);
1194 error
= libxfs_trans_reserve(tp
, nres
, XFS_REMOVE_LOG_RES(mp
), 0,
1195 XFS_TRANS_PERM_LOG_RES
, XFS_REMOVE_LOG_COUNT
);
1198 libxfs_trans_ijoin(tp
, ip
, 0);
1199 libxfs_trans_ihold(tp
, ip
);
1201 if ((error
= libxfs_bmap_last_offset(tp
, ip
, &lastblock
,
1203 do_error(_("xfs_bmap_last_offset failed -- error - %d\n"),
1206 /* free all data, leaf, node and freespace blocks */
1207 error
= libxfs_bunmapi(tp
, ip
, 0, lastblock
, XFS_BMAPI_METADATA
, 0,
1208 &firstblock
, &flist
, &done
);
1210 do_warn(_("xfs_bunmapi failed -- error - %d\n"), error
);
1211 libxfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
|
1218 libxfs_dir_init(tp
, ip
, &pip
);
1220 error
= libxfs_bmap_finish(&tp
, &flist
, &committed
);
1222 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
1224 /* go through the hash list and re-add the inodes */
1226 for (p
= hashtab
->first
; p
; p
= p
->nextbyorder
) {
1228 if (p
->name
.name
[0] == '/' || (p
->name
.name
[0] == '.' &&
1229 (p
->name
.len
== 1 || (p
->name
.len
== 2 &&
1230 p
->name
.name
[1] == '.'))))
1233 tp
= libxfs_trans_alloc(mp
, 0);
1234 nres
= XFS_CREATE_SPACE_RES(mp
, p
->name
.len
);
1235 error
= libxfs_trans_reserve(tp
, nres
, XFS_CREATE_LOG_RES(mp
),
1236 0, XFS_TRANS_PERM_LOG_RES
, XFS_CREATE_LOG_COUNT
);
1239 _("space reservation failed (%d), filesystem may be out of space\n"),
1244 libxfs_trans_ijoin(tp
, ip
, 0);
1245 libxfs_trans_ihold(tp
, ip
);
1247 xfs_bmap_init(&flist
, &firstblock
);
1248 error
= libxfs_dir_createname(tp
, ip
, &p
->name
, p
->inum
,
1249 &firstblock
, &flist
, nres
);
1252 _("name create failed in ino %" PRIu64
" (%d), filesystem may be out of space\n"),
1254 libxfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
|
1259 error
= libxfs_bmap_finish(&tp
, &flist
, &committed
);
1262 _("bmap finish failed (%d), filesystem may be out of space\n"),
1264 libxfs_bmap_cancel(&flist
);
1265 libxfs_trans_cancel(tp
, XFS_TRANS_RELEASE_LOG_RES
|
1270 libxfs_trans_commit(tp
,
1271 XFS_TRANS_RELEASE_LOG_RES
|XFS_TRANS_SYNC
);
1277 * Kill a block in a version 2 inode.
1278 * Makes its own transaction.
1290 xfs_fsblock_t firstblock
;
1291 xfs_bmap_free_t flist
;
1295 tp
= libxfs_trans_alloc(mp
, 0);
1296 nres
= XFS_REMOVE_SPACE_RES(mp
);
1297 error
= libxfs_trans_reserve(tp
, nres
, XFS_REMOVE_LOG_RES(mp
), 0,
1298 XFS_TRANS_PERM_LOG_RES
, XFS_REMOVE_LOG_COUNT
);
1301 libxfs_trans_ijoin(tp
, ip
, 0);
1302 libxfs_trans_ihold(tp
, ip
);
1303 libxfs_trans_bjoin(tp
, bp
);
1304 memset(&args
, 0, sizeof(args
));
1305 xfs_bmap_init(&flist
, &firstblock
);
1308 args
.firstblock
= &firstblock
;
1309 args
.flist
= &flist
;
1310 args
.whichfork
= XFS_DATA_FORK
;
1311 if (da_bno
>= mp
->m_dirleafblk
&& da_bno
< mp
->m_dirfreeblk
)
1312 error
= libxfs_da_shrink_inode(&args
, da_bno
, bp
);
1314 error
= libxfs_dir2_shrink_inode(&args
,
1315 xfs_dir2_da_to_db(mp
, da_bno
), bp
);
1317 do_error(_("shrink_inode failed inode %" PRIu64
" block %u\n"),
1319 libxfs_bmap_finish(&tp
, &flist
, &committed
);
1320 libxfs_trans_commit(tp
, 0);
1324 * process a data block, also checks for .. entry
1325 * and corrects it to match what we think .. should be
1328 longform_dir2_entry_check_data(
1333 ino_tree_node_t
*current_irec
,
1334 int current_ino_offset
,
1335 struct xfs_buf
**bpp
,
1336 dir_hash_tab_t
*hashtab
,
1337 freetab_t
**freetabp
,
1341 xfs_dir2_dataptr_t addr
;
1342 xfs_dir2_leaf_entry_t
*blp
;
1344 xfs_dir2_block_tail_t
*btp
;
1348 xfs_dir2_data_entry_t
*dep
;
1349 xfs_dir2_data_unused_t
*dup
;
1352 xfs_fsblock_t firstblock
;
1353 xfs_bmap_free_t flist
;
1354 char fname
[MAXNAMELEN
+ 1];
1359 ino_tree_node_t
*irec
;
1375 needscan
= needlog
= 0;
1377 freetab
= *freetabp
;
1379 btp
= xfs_dir2_block_tail_p(mp
, (struct xfs_dir2_data_hdr
*)d
);
1380 blp
= xfs_dir2_block_leaf_p(btp
);
1381 endptr
= (char *)blp
;
1382 if (endptr
> (char *)btp
)
1383 endptr
= (char *)btp
;
1384 wantmagic
= XFS_DIR2_BLOCK_MAGIC
;
1386 endptr
= (char *)d
+ mp
->m_dirblksize
;
1387 wantmagic
= XFS_DIR2_DATA_MAGIC
;
1389 db
= xfs_dir2_da_to_db(mp
, da_bno
);
1391 /* check for data block beyond expected end */
1392 if (freetab
->naents
<= db
) {
1393 struct freetab_ent e
;
1395 *freetabp
= freetab
= realloc(freetab
, FREETAB_SIZE(db
+ 1));
1398 _("realloc failed in longform_dir2_entry_check_data (%zu bytes)\n"),
1399 FREETAB_SIZE(db
+ 1));
1403 for (i
= freetab
->naents
; i
< db
; i
++)
1404 freetab
->ents
[i
] = e
;
1405 freetab
->naents
= db
+ 1;
1408 /* check the data block */
1409 while (ptr
< endptr
) {
1411 /* check for freespace */
1412 dup
= (xfs_dir2_data_unused_t
*)ptr
;
1413 if (XFS_DIR2_DATA_FREE_TAG
== be16_to_cpu(dup
->freetag
)) {
1415 /* check for invalid freespace length */
1416 if (ptr
+ be16_to_cpu(dup
->length
) > endptr
||
1417 be16_to_cpu(dup
->length
) == 0 ||
1418 (be16_to_cpu(dup
->length
) &
1419 (XFS_DIR2_DATA_ALIGN
- 1)))
1422 /* check for invalid tag */
1423 if (be16_to_cpu(*xfs_dir2_data_unused_tag_p(dup
)) !=
1424 (char *)dup
- (char *)d
)
1427 /* check for block with no data entries */
1428 if ((ptr
== (char *)d
->u
) && (ptr
+
1429 be16_to_cpu(dup
->length
) >= endptr
)) {
1435 /* continue at the end of the freespace */
1436 ptr
+= be16_to_cpu(dup
->length
);
1441 /* validate data entry size */
1442 dep
= (xfs_dir2_data_entry_t
*)ptr
;
1443 if (ptr
+ xfs_dir2_data_entsize(dep
->namelen
) > endptr
)
1445 if (be16_to_cpu(*xfs_dir2_data_entry_tag_p(dep
)) !=
1446 (char *)dep
- (char *)d
)
1448 ptr
+= xfs_dir2_data_entsize(dep
->namelen
);
1451 /* did we find an empty or corrupt block? */
1452 if (ptr
!= endptr
) {
1455 _("empty data block %u in directory inode %" PRIu64
": "),
1459 ("corrupt block %u in directory inode %" PRIu64
": "),
1463 do_warn(_("junking block\n"));
1464 dir2_kill_block(mp
, ip
, da_bno
, bp
);
1466 do_warn(_("would junk block\n"));
1469 freetab
->ents
[db
].v
= NULLDATAOFF
;
1474 /* update number of data blocks processed */
1475 if (freetab
->nents
< db
+ 1)
1476 freetab
->nents
= db
+ 1;
1478 tp
= libxfs_trans_alloc(mp
, 0);
1479 error
= libxfs_trans_reserve(tp
, 0, XFS_REMOVE_LOG_RES(mp
), 0,
1480 XFS_TRANS_PERM_LOG_RES
, XFS_REMOVE_LOG_COUNT
);
1483 libxfs_trans_ijoin(tp
, ip
, 0);
1484 libxfs_trans_ihold(tp
, ip
);
1485 libxfs_trans_bjoin(tp
, bp
);
1486 libxfs_trans_bhold(tp
, bp
);
1487 xfs_bmap_init(&flist
, &firstblock
);
1488 if (be32_to_cpu(d
->hdr
.magic
) != wantmagic
) {
1490 _("bad directory block magic # %#x for directory inode %" PRIu64
" block %d: "),
1491 be32_to_cpu(d
->hdr
.magic
), ip
->i_ino
, da_bno
);
1493 do_warn(_("fixing magic # to %#x\n"), wantmagic
);
1494 d
->hdr
.magic
= cpu_to_be32(wantmagic
);
1497 do_warn(_("would fix magic # to %#x\n"), wantmagic
);
1502 * look at each entry. reference inode pointed to by each
1503 * entry in the incore inode tree.
1504 * if not a directory, set reached flag, increment link count
1505 * if a directory and reached, mark entry as to be deleted.
1506 * if a directory, check to see if recorded parent
1507 * matches current inode #,
1508 * if so, then set reached flag, increment link count
1509 * of current and child dir inodes, push the child
1510 * directory inode onto the directory stack.
1511 * if current inode != parent, then mark entry to be deleted.
1513 while (ptr
< endptr
) {
1514 dup
= (xfs_dir2_data_unused_t
*)ptr
;
1515 if (be16_to_cpu(dup
->freetag
) == XFS_DIR2_DATA_FREE_TAG
) {
1518 _("directory inode %" PRIu64
" block %u has consecutive free entries: "),
1521 do_warn(_("joining together\n"));
1522 len
= be16_to_cpu(dup
->length
);
1523 libxfs_dir2_data_use_free(tp
, bp
, dup
,
1524 ptr
- (char *)d
, len
, &needlog
,
1526 libxfs_dir2_data_make_free(tp
, bp
,
1527 ptr
- (char *)d
, len
, &needlog
,
1530 do_warn(_("would join together\n"));
1532 ptr
+= be16_to_cpu(dup
->length
);
1536 addr
= xfs_dir2_db_off_to_dataptr(mp
, db
, ptr
- (char *)d
);
1537 dep
= (xfs_dir2_data_entry_t
*)ptr
;
1538 ptr
+= xfs_dir2_data_entsize(dep
->namelen
);
1539 inum
= be64_to_cpu(dep
->inumber
);
1542 * skip bogus entries (leading '/'). they'll be deleted
1543 * later. must still log it, else we leak references to
1546 if (dep
->name
[0] == '/') {
1549 libxfs_dir2_data_log_entry(tp
, bp
, dep
);
1553 memmove(fname
, dep
->name
, dep
->namelen
);
1554 fname
[dep
->namelen
] = '\0';
1555 ASSERT(inum
!= NULLFSINO
);
1557 irec
= find_inode_rec(mp
, XFS_INO_TO_AGNO(mp
, inum
),
1558 XFS_INO_TO_AGINO(mp
, inum
));
1562 _("entry \"%s\" in directory inode %" PRIu64
" points to non-existent inode %" PRIu64
""),
1563 fname
, ip
->i_ino
, inum
)) {
1565 libxfs_dir2_data_log_entry(tp
, bp
, dep
);
1569 ino_offset
= XFS_INO_TO_AGINO(mp
, inum
) - irec
->ino_startnum
;
1572 * if it's a free inode, blow out the entry.
1573 * by now, any inode that we think is free
1576 if (is_inode_free(irec
, ino_offset
)) {
1579 _("entry \"%s\" in directory inode %" PRIu64
" points to free inode %" PRIu64
),
1580 fname
, ip
->i_ino
, inum
)) {
1582 libxfs_dir2_data_log_entry(tp
, bp
, dep
);
1588 * check if this inode is lost+found dir in the root
1590 if (inum
== mp
->m_sb
.sb_rootino
&& strcmp(fname
, ORPHANAGE
) == 0) {
1592 * if it's not a directory, trash it
1594 if (!inode_isadir(irec
, ino_offset
)) {
1597 _("%s (ino %" PRIu64
") in root (%" PRIu64
") is not a directory"),
1598 ORPHANAGE
, inum
, ip
->i_ino
)) {
1600 libxfs_dir2_data_log_entry(tp
, bp
, dep
);
1605 * if this is a dup, it will be picked up below,
1606 * otherwise, mark it as the orphanage for later.
1609 orphanage_ino
= inum
;
1612 * check for duplicate names in directory.
1614 if (!dir_hash_add(mp
, hashtab
, addr
, inum
, dep
->namelen
,
1618 _("entry \"%s\" (ino %" PRIu64
") in dir %" PRIu64
" is a duplicate name"),
1619 fname
, inum
, ip
->i_ino
)) {
1621 libxfs_dir2_data_log_entry(tp
, bp
, dep
);
1623 if (inum
== orphanage_ino
)
1629 * if just scanning to rebuild a directory due to a ".."
1630 * update, just continue
1636 * skip the '..' entry since it's checked when the
1637 * directory is reached by something else. if it never
1638 * gets reached, it'll be moved to the orphanage and we'll
1639 * take care of it then. If it doesn't exist at all, the
1640 * directory needs to be rebuilt first before being added
1643 if (dep
->namelen
== 2 && dep
->name
[0] == '.' &&
1644 dep
->name
[1] == '.') {
1646 /* ".." should be in the first block */
1649 _("entry \"%s\" (ino %" PRIu64
") in dir %" PRIu64
" is not in the the first block"), fname
,
1652 libxfs_dir2_data_log_entry(tp
, bp
, dep
);
1657 ASSERT(no_modify
|| !verify_inum(mp
, inum
));
1659 * special case the . entry. we know there's only one
1660 * '.' and only '.' points to itself because bogus entries
1661 * got trashed in phase 3 if there were > 1.
1662 * bump up link count for '.' but don't set reached
1663 * until we're actually reached by another directory
1664 * '..' is already accounted for or will be taken care
1665 * of when directory is moved to orphanage.
1667 if (ip
->i_ino
== inum
) {
1668 ASSERT(dep
->name
[0] == '.' && dep
->namelen
== 1);
1669 add_inode_ref(current_irec
, current_ino_offset
);
1670 if (da_bno
!= 0 || dep
!= (xfs_dir2_data_entry_t
*)d
->u
) {
1671 /* "." should be the first entry */
1674 _("entry \"%s\" in dir %" PRIu64
" is not the first entry"),
1675 fname
, inum
, ip
->i_ino
)) {
1677 libxfs_dir2_data_log_entry(tp
, bp
, dep
);
1684 * skip entries with bogus inumbers if we're in no modify mode
1686 if (no_modify
&& verify_inum(mp
, inum
))
1689 * check easy case first, regular inode, just bump
1690 * the link count and continue
1692 if (!inode_isadir(irec
, ino_offset
)) {
1693 add_inode_reached(irec
, ino_offset
);
1696 parent
= get_inode_parent(irec
, ino_offset
);
1697 ASSERT(parent
!= 0);
1700 * bump up the link counts in parent and child
1701 * directory but if the link doesn't agree with
1702 * the .. in the child, blow out the entry.
1703 * if the directory has already been reached,
1704 * blow away the entry also.
1706 if (is_inode_reached(irec
, ino_offset
)) {
1709 _("entry \"%s\" in dir %" PRIu64
" points to an already connected directory inode %" PRIu64
"\n"),
1710 fname
, ip
->i_ino
, inum
);
1711 } else if (parent
== ip
->i_ino
) {
1712 add_inode_reached(irec
, ino_offset
);
1713 add_inode_ref(current_irec
, current_ino_offset
);
1714 } else if (parent
== NULLFSINO
) {
1715 /* ".." was missing, but this entry refers to it,
1716 so, set it as the parent and mark for rebuild */
1718 _("entry \"%s\" in dir ino %" PRIu64
" doesn't have a .. entry, will set it in ino %" PRIu64
".\n"),
1719 fname
, ip
->i_ino
, inum
);
1720 set_inode_parent(irec
, ino_offset
, ip
->i_ino
);
1721 add_inode_reached(irec
, ino_offset
);
1722 add_inode_ref(current_irec
, current_ino_offset
);
1723 add_dotdot_update(XFS_INO_TO_AGNO(mp
, inum
), irec
,
1728 _("entry \"%s\" in dir inode %" PRIu64
" inconsistent with .. value (%" PRIu64
") in ino %" PRIu64
"\n"),
1729 fname
, ip
->i_ino
, parent
, inum
);
1732 if (inum
== orphanage_ino
)
1738 libxfs_dir2_data_log_entry(tp
, bp
, dep
);
1741 _("\twill clear entry \"%s\"\n"),
1744 do_warn(_("\twould clear entry \"%s\"\n"),
1749 *num_illegal
+= nbad
;
1751 libxfs_dir2_data_freescan(mp
, &d
->hdr
, &needlog
);
1753 libxfs_dir2_data_log_header(tp
, bp
);
1754 libxfs_bmap_finish(&tp
, &flist
, &committed
);
1755 libxfs_trans_commit(tp
, 0);
1756 freetab
->ents
[db
].v
= be16_to_cpu(d
->hdr
.bestfree
[0].length
);
1757 freetab
->ents
[db
].s
= 0;
1761 * Check contents of leaf-form block.
1764 longform_dir2_check_leaf(
1767 dir_hash_tab_t
*hashtab
,
1775 xfs_dir2_leaf_t
*leaf
;
1776 xfs_dir2_leaf_tail_t
*ltp
;
1779 da_bno
= mp
->m_dirleafblk
;
1780 if (libxfs_da_read_buf(NULL
, ip
, da_bno
, -1, &bp
, XFS_DATA_FORK
, NULL
)) {
1782 _("can't read block %u for directory inode %" PRIu64
"\n"),
1787 ltp
= xfs_dir2_leaf_tail_p(mp
, leaf
);
1788 bestsp
= xfs_dir2_leaf_bests_p(ltp
);
1789 if (be16_to_cpu(leaf
->hdr
.info
.magic
) != XFS_DIR2_LEAF1_MAGIC
||
1790 be32_to_cpu(leaf
->hdr
.info
.forw
) ||
1791 be32_to_cpu(leaf
->hdr
.info
.back
) ||
1792 be16_to_cpu(leaf
->hdr
.count
) <
1793 be16_to_cpu(leaf
->hdr
.stale
) ||
1794 be16_to_cpu(leaf
->hdr
.count
) >
1795 xfs_dir2_max_leaf_ents(mp
) ||
1796 (char *)&leaf
->ents
[be16_to_cpu(
1797 leaf
->hdr
.count
)] > (char *)bestsp
) {
1799 _("leaf block %u for directory inode %" PRIu64
" bad header\n"),
1804 seeval
= dir_hash_see_all(hashtab
, leaf
->ents
,
1805 be16_to_cpu(leaf
->hdr
.count
),
1806 be16_to_cpu(leaf
->hdr
.stale
));
1807 if (dir_hash_check(hashtab
, ip
, seeval
)) {
1811 badtail
= freetab
->nents
!= be32_to_cpu(ltp
->bestcount
);
1812 for (i
= 0; !badtail
&& i
< be32_to_cpu(ltp
->bestcount
); i
++) {
1813 freetab
->ents
[i
].s
= 1;
1814 badtail
= freetab
->ents
[i
].v
!= be16_to_cpu(bestsp
[i
]);
1818 _("leaf block %u for directory inode %" PRIu64
" bad tail\n"),
1828 * Check contents of the node blocks (leaves)
1829 * Looks for matching hash values for the data entries.
1832 longform_dir2_check_node(
1835 dir_hash_tab_t
*hashtab
,
1841 xfs_dir2_free_t
*free
;
1843 xfs_dir2_leaf_t
*leaf
;
1844 xfs_fileoff_t next_da_bno
;
1848 for (da_bno
= mp
->m_dirleafblk
, next_da_bno
= 0;
1849 next_da_bno
!= NULLFILEOFF
&& da_bno
< mp
->m_dirfreeblk
;
1850 da_bno
= (xfs_dablk_t
)next_da_bno
) {
1851 next_da_bno
= da_bno
+ mp
->m_dirblkfsbs
- 1;
1852 if (bmap_next_offset(NULL
, ip
, &next_da_bno
, XFS_DATA_FORK
))
1854 if (libxfs_da_read_buf(NULL
, ip
, da_bno
, -1, &bp
,
1855 XFS_DATA_FORK
, NULL
)) {
1857 _("can't read leaf block %u for directory inode %" PRIu64
"\n"),
1862 if (be16_to_cpu(leaf
->hdr
.info
.magic
) != XFS_DIR2_LEAFN_MAGIC
) {
1863 if (be16_to_cpu(leaf
->hdr
.info
.magic
) ==
1864 XFS_DA_NODE_MAGIC
) {
1869 _("unknown magic number %#x for block %u in directory inode %" PRIu64
"\n"),
1870 be16_to_cpu(leaf
->hdr
.info
.magic
),
1875 if (be16_to_cpu(leaf
->hdr
.count
) > xfs_dir2_max_leaf_ents(mp
) ||
1876 be16_to_cpu(leaf
->hdr
.count
) <
1877 be16_to_cpu(leaf
->hdr
.stale
)) {
1879 _("leaf block %u for directory inode %" PRIu64
" bad header\n"),
1884 seeval
= dir_hash_see_all(hashtab
, leaf
->ents
,
1885 be16_to_cpu(leaf
->hdr
.count
),
1886 be16_to_cpu(leaf
->hdr
.stale
));
1888 if (seeval
!= DIR_HASH_CK_OK
)
1891 if (dir_hash_check(hashtab
, ip
, seeval
))
1894 for (da_bno
= mp
->m_dirfreeblk
, next_da_bno
= 0;
1895 next_da_bno
!= NULLFILEOFF
;
1896 da_bno
= (xfs_dablk_t
)next_da_bno
) {
1897 next_da_bno
= da_bno
+ mp
->m_dirblkfsbs
- 1;
1898 if (bmap_next_offset(NULL
, ip
, &next_da_bno
, XFS_DATA_FORK
))
1900 if (libxfs_da_read_buf(NULL
, ip
, da_bno
, -1, &bp
,
1901 XFS_DATA_FORK
, NULL
)) {
1903 _("can't read freespace block %u for directory inode %" PRIu64
"\n"),
1908 fdb
= xfs_dir2_da_to_db(mp
, da_bno
);
1909 if (be32_to_cpu(free
->hdr
.magic
) != XFS_DIR2_FREE_MAGIC
||
1910 be32_to_cpu(free
->hdr
.firstdb
) !=
1911 (fdb
- XFS_DIR2_FREE_FIRSTDB(mp
)) *
1912 xfs_dir2_free_max_bests(mp
) ||
1913 be32_to_cpu(free
->hdr
.nvalid
) <
1914 be32_to_cpu(free
->hdr
.nused
)) {
1916 _("free block %u for directory inode %" PRIu64
" bad header\n"),
1921 for (i
= used
= 0; i
< be32_to_cpu(free
->hdr
.nvalid
); i
++) {
1922 if (i
+ be32_to_cpu(free
->hdr
.firstdb
) >=
1924 freetab
->ents
[i
+ be32_to_cpu(
1925 free
->hdr
.firstdb
)].v
!=
1926 be16_to_cpu(free
->bests
[i
])) {
1928 _("free block %u entry %i for directory ino %" PRIu64
" bad\n"),
1929 da_bno
, i
, ip
->i_ino
);
1933 used
+= be16_to_cpu(free
->bests
[i
]) != NULLDATAOFF
;
1934 freetab
->ents
[i
+ be32_to_cpu(free
->hdr
.firstdb
)].s
= 1;
1936 if (used
!= be32_to_cpu(free
->hdr
.nused
)) {
1938 _("free block %u for directory inode %" PRIu64
" bad nused\n"),
1945 for (i
= 0; i
< freetab
->nents
; i
++) {
1946 if ((freetab
->ents
[i
].s
== 0) &&
1947 (freetab
->ents
[i
].v
!= NULLDATAOFF
)) {
1949 _("missing freetab entry %u for directory inode %" PRIu64
"\n"),
1958 * If a directory is corrupt, we need to read in as many entries as possible,
1959 * destroy the entry and create a new one with recovered name/inode pairs.
1960 * (ie. get libxfs to do all the grunt work)
1963 longform_dir2_entry_check(xfs_mount_t
*mp
,
1968 ino_tree_node_t
*irec
,
1970 dir_hash_tab_t
*hashtab
)
1972 xfs_dir2_block_t
*block
;
1973 struct xfs_buf
**bplist
;
1980 xfs_fileoff_t next_da_bno
;
1986 freetab
= malloc(FREETAB_SIZE(ip
->i_d
.di_size
/ mp
->m_dirblksize
));
1989 _("malloc failed in longform_dir2_entry_check (%" PRId64
" bytes)\n"),
1990 FREETAB_SIZE(ip
->i_d
.di_size
/ mp
->m_dirblksize
));
1993 freetab
->naents
= ip
->i_d
.di_size
/ mp
->m_dirblksize
;
1995 for (i
= 0; i
< freetab
->naents
; i
++) {
1996 freetab
->ents
[i
].v
= NULLDATAOFF
;
1997 freetab
->ents
[i
].s
= 0;
1999 num_bps
= freetab
->naents
;
2000 bplist
= calloc(num_bps
, sizeof(struct xfs_buf
*));
2001 /* is this a block, leaf, or node directory? */
2002 libxfs_dir2_isblock(NULL
, ip
, &isblock
);
2003 libxfs_dir2_isleaf(NULL
, ip
, &isleaf
);
2005 /* check directory "data" blocks (ie. name/inode pairs) */
2006 for (da_bno
= 0, next_da_bno
= 0;
2007 next_da_bno
!= NULLFILEOFF
&& da_bno
< mp
->m_dirleafblk
;
2008 da_bno
= (xfs_dablk_t
)next_da_bno
) {
2009 next_da_bno
= da_bno
+ mp
->m_dirblkfsbs
- 1;
2010 if (bmap_next_offset(NULL
, ip
, &next_da_bno
, XFS_DATA_FORK
))
2012 db
= xfs_dir2_da_to_db(mp
, da_bno
);
2013 if (db
>= num_bps
) {
2014 /* more data blocks than expected */
2016 bplist
= realloc(bplist
, num_bps
* sizeof(struct xfs_buf
*));
2019 _("realloc failed in longform_dir2_entry_check (%zu bytes)\n"),
2020 num_bps
* sizeof(struct xfs_buf
*));
2022 if (libxfs_da_read_buf(NULL
, ip
, da_bno
, -1, &bplist
[db
],
2023 XFS_DATA_FORK
, NULL
)) {
2025 _("can't read data block %u for directory inode %" PRIu64
"\n"),
2028 continue; /* try and read all "data" blocks */
2030 longform_dir2_entry_check_data(mp
, ip
, num_illegal
, need_dot
,
2031 irec
, ino_offset
, &bplist
[db
], hashtab
,
2032 &freetab
, da_bno
, isblock
);
2034 fixit
= (*num_illegal
!= 0) || dir2_is_badino(ino
) || *need_dot
;
2036 if (!dotdot_update
) {
2037 /* check btree and freespace */
2039 xfs_dir2_block_tail_t
*btp
;
2040 xfs_dir2_leaf_entry_t
*blp
;
2042 block
= bplist
[0]->b_addr
;
2043 btp
= xfs_dir2_block_tail_p(mp
, &block
->hdr
);
2044 blp
= xfs_dir2_block_leaf_p(btp
);
2045 seeval
= dir_hash_see_all(hashtab
, blp
,
2046 be32_to_cpu(btp
->count
),
2047 be32_to_cpu(btp
->stale
));
2048 if (dir_hash_check(hashtab
, ip
, seeval
))
2050 } else if (isleaf
) {
2051 fixit
|= longform_dir2_check_leaf(mp
, ip
, hashtab
,
2054 fixit
|= longform_dir2_check_node(mp
, ip
, hashtab
,
2058 if (!no_modify
&& (fixit
|| dotdot_update
)) {
2059 dir_hash_dup_names(hashtab
);
2060 for (i
= 0; i
< freetab
->naents
; i
++)
2062 libxfs_putbuf(bplist
[i
]);
2063 longform_dir2_rebuild(mp
, ino
, ip
, irec
, ino_offset
, hashtab
);
2067 for (i
= 0; i
< freetab
->naents
; i
++)
2069 libxfs_putbuf(bplist
[i
]);
2077 * shortform directory v2 processing routines -- entry verification and
2078 * bad entry deletion (pruning).
2081 shortform_dir2_entry_check(xfs_mount_t
*mp
,
2085 ino_tree_node_t
*current_irec
,
2086 int current_ino_offset
,
2087 dir_hash_tab_t
*hashtab
)
2092 xfs_dir2_sf_entry_t
*sfep
, *next_sfep
, *tmp_sfep
;
2094 ino_tree_node_t
*irec
;
2104 char fname
[MAXNAMELEN
+ 1];
2108 sfp
= (xfs_dir2_sf_t
*) ifp
->if_u1
.if_data
;
2112 max_size
= ifp
->if_bytes
;
2113 ASSERT(ip
->i_d
.di_size
<= ifp
->if_bytes
);
2116 * if just rebuild a directory due to a "..", update and return
2118 if (dotdot_update
) {
2119 parent
= get_inode_parent(current_irec
, current_ino_offset
);
2122 _("would set .. in sf dir inode %" PRIu64
" to %" PRIu64
"\n"),
2126 _("setting .. in sf dir inode %" PRIu64
" to %" PRIu64
"\n"),
2128 xfs_dir2_sf_put_parent_ino(&sfp
->hdr
, parent
);
2135 * no '.' entry in shortform dirs, just bump up ref count by 1
2136 * '..' was already (or will be) accounted for and checked when
2137 * the directory is reached or will be taken care of when the
2138 * directory is moved to orphanage.
2140 add_inode_ref(current_irec
, current_ino_offset
);
2143 * Initialise i8 counter -- the parent inode number counts as well.
2145 i8
= xfs_dir2_sf_get_parent_ino(&sfp
->hdr
) > XFS_DIR2_MAX_SHORT_INUM
;
2148 * now run through entries, stop at first bad entry, don't need
2149 * to skip over '..' since that's encoded in its own field and
2150 * no need to worry about '.' since it doesn't exist.
2152 sfep
= next_sfep
= xfs_dir2_sf_firstentry(&sfp
->hdr
);
2154 for (i
= 0; i
< sfp
->hdr
.count
&& max_size
>
2155 (__psint_t
)next_sfep
- (__psint_t
)sfp
;
2156 sfep
= next_sfep
, i
++) {
2161 lino
= xfs_dir2_sfe_get_ino(&sfp
->hdr
, sfep
);
2163 namelen
= sfep
->namelen
;
2165 ASSERT(no_modify
|| namelen
> 0);
2167 if (no_modify
&& namelen
== 0) {
2169 * if we're really lucky, this is
2170 * the last entry in which case we
2171 * can use the dir size to set the
2172 * namelen value. otherwise, forget
2173 * it because we're not going to be
2174 * able to find the next entry.
2178 if (i
== sfp
->hdr
.count
- 1) {
2179 namelen
= ip
->i_d
.di_size
-
2180 ((__psint_t
) &sfep
->name
[0] -
2184 * don't process the rest of the directory,
2185 * break out of processing loop
2189 } else if (no_modify
&& (__psint_t
) sfep
- (__psint_t
) sfp
+
2190 + xfs_dir2_sf_entsize(&sfp
->hdr
, sfep
->namelen
)
2191 > ip
->i_d
.di_size
) {
2194 if (i
== sfp
->hdr
.count
- 1) {
2195 namelen
= ip
->i_d
.di_size
-
2196 ((__psint_t
) &sfep
->name
[0] -
2200 * don't process the rest of the directory,
2201 * break out of processing loop
2207 memmove(fname
, sfep
->name
, sfep
->namelen
);
2208 fname
[sfep
->namelen
] = '\0';
2210 ASSERT(no_modify
|| (lino
!= NULLFSINO
&& lino
!= 0));
2211 ASSERT(no_modify
|| !verify_inum(mp
, lino
));
2214 * Also skip entries with bogus inode numbers if we're
2215 * in no modify mode.
2218 if (no_modify
&& verify_inum(mp
, lino
)) {
2219 next_sfep
= (xfs_dir2_sf_entry_t
*)((__psint_t
)sfep
+
2220 xfs_dir2_sf_entsize(&sfp
->hdr
, sfep
->namelen
));
2224 irec
= find_inode_rec(mp
, XFS_INO_TO_AGNO(mp
, lino
),
2225 XFS_INO_TO_AGINO(mp
, lino
));
2229 _("entry \"%s\" in shortform directory %" PRIu64
" references non-existent inode %" PRIu64
"\n"),
2234 ino_offset
= XFS_INO_TO_AGINO(mp
, lino
) - irec
->ino_startnum
;
2237 * if it's a free inode, blow out the entry.
2238 * by now, any inode that we think is free
2241 if (is_inode_free(irec
, ino_offset
)) {
2243 _("entry \"%s\" in shortform directory inode %" PRIu64
" points to free inode %" PRIu64
"\n"),
2248 * check if this inode is lost+found dir in the root
2250 if (ino
== mp
->m_sb
.sb_rootino
&& strcmp(fname
, ORPHANAGE
) == 0) {
2252 * if it's not a directory, trash it
2254 if (!inode_isadir(irec
, ino_offset
)) {
2256 _("%s (ino %" PRIu64
") in root (%" PRIu64
") is not a directory"),
2257 ORPHANAGE
, lino
, ino
);
2261 * if this is a dup, it will be picked up below,
2262 * otherwise, mark it as the orphanage for later.
2265 orphanage_ino
= lino
;
2268 * check for duplicate names in directory.
2270 if (!dir_hash_add(mp
, hashtab
, (xfs_dir2_dataptr_t
)
2271 (sfep
- xfs_dir2_sf_firstentry(&sfp
->hdr
)),
2272 lino
, sfep
->namelen
, sfep
->name
)) {
2274 _("entry \"%s\" (ino %" PRIu64
") in dir %" PRIu64
" is a duplicate name"),
2279 if (!inode_isadir(irec
, ino_offset
)) {
2281 * check easy case first, regular inode, just bump
2284 add_inode_reached(irec
, ino_offset
);
2286 parent
= get_inode_parent(irec
, ino_offset
);
2289 * bump up the link counts in parent and child.
2290 * directory but if the link doesn't agree with
2291 * the .. in the child, blow out the entry
2293 if (is_inode_reached(irec
, ino_offset
)) {
2296 _("entry \"%s\" in directory inode %" PRIu64
2297 " references already connected inode %" PRIu64
".\n"),
2299 } else if (parent
== ino
) {
2300 add_inode_reached(irec
, ino_offset
);
2301 add_inode_ref(current_irec
, current_ino_offset
);
2302 } else if (parent
== NULLFSINO
) {
2303 /* ".." was missing, but this entry refers to it,
2304 so, set it as the parent and mark for rebuild */
2306 _("entry \"%s\" in dir ino %" PRIu64
" doesn't have a .. entry, will set it in ino %" PRIu64
".\n"),
2308 set_inode_parent(irec
, ino_offset
, ino
);
2309 add_inode_reached(irec
, ino_offset
);
2310 add_inode_ref(current_irec
, current_ino_offset
);
2311 add_dotdot_update(XFS_INO_TO_AGNO(mp
, lino
),
2316 _("entry \"%s\" in directory inode %" PRIu64
2317 " not consistent with .. value (%" PRIu64
2318 ") in inode %" PRIu64
",\n"),
2319 fname
, ino
, parent
, lino
);
2325 if (lino
== orphanage_ino
)
2328 tmp_elen
= xfs_dir2_sf_entsize(&sfp
->hdr
,
2330 tmp_sfep
= (xfs_dir2_sf_entry_t
*)
2331 ((__psint_t
) sfep
+ tmp_elen
);
2332 tmp_len
= max_size
- ((__psint_t
) tmp_sfep
2334 max_size
-= tmp_elen
;
2335 bytes_deleted
+= tmp_elen
;
2337 memmove(sfep
, tmp_sfep
, tmp_len
);
2339 sfp
->hdr
.count
-= 1;
2340 memset((void *)((__psint_t
)sfep
+ tmp_len
), 0,
2344 * set the tmp value to the current
2345 * pointer so we'll process the entry
2351 * WARNING: drop the index i by one
2352 * so it matches the decremented count for
2353 * accurate comparisons in the loop test
2360 do_warn(_("junking entry\n"));
2364 do_warn(_("would junk entry\n"));
2366 } else if (lino
> XFS_DIR2_MAX_SHORT_INUM
)
2370 * go onto next entry unless we've just junked an
2371 * entry in which the current entry pointer points
2372 * to an unprocessed entry. have to take into entries
2373 * with bad namelen into account in no modify mode since we
2374 * calculate size based on next_sfep.
2376 ASSERT(no_modify
|| bad_sfnamelen
== 0);
2378 next_sfep
= (tmp_sfep
== NULL
)
2379 ? (xfs_dir2_sf_entry_t
*) ((__psint_t
) sfep
2381 ? xfs_dir2_sf_entsize(&sfp
->hdr
, sfep
->namelen
)
2382 : xfs_dir2_sf_entsize(&sfp
->hdr
, namelen
)))
2386 if (sfp
->hdr
.i8count
!= i8
) {
2388 do_warn(_("would fix i8count in inode %" PRIu64
"\n"),
2392 tmp_sfep
= next_sfep
;
2393 process_sf_dir2_fixi8(sfp
, &tmp_sfep
);
2395 (__psint_t
)next_sfep
-
2396 (__psint_t
)tmp_sfep
;
2397 next_sfep
= tmp_sfep
;
2399 sfp
->hdr
.i8count
= i8
;
2401 do_warn(_("fixing i8count in inode %" PRIu64
"\n"),
2407 * sync up sizes if required
2410 ASSERT(bytes_deleted
> 0);
2412 libxfs_idata_realloc(ip
, -bytes_deleted
, XFS_DATA_FORK
);
2413 ip
->i_d
.di_size
-= bytes_deleted
;
2416 if (ip
->i_d
.di_size
!= ip
->i_df
.if_bytes
) {
2417 ASSERT(ip
->i_df
.if_bytes
== (xfs_fsize_t
)
2418 ((__psint_t
) next_sfep
- (__psint_t
) sfp
));
2419 ip
->i_d
.di_size
= (xfs_fsize_t
)
2420 ((__psint_t
) next_sfep
- (__psint_t
) sfp
);
2422 _("setting size to %" PRId64
" bytes to reflect junked entries\n"),
2429 * processes all reachable inodes in directories
2434 xfs_agnumber_t agno
,
2435 ino_tree_node_t
*irec
,
2439 xfs_bmap_free_t flist
;
2440 xfs_fsblock_t first
;
2443 dir_hash_tab_t
*hashtab
;
2444 int need_dot
, committed
;
2445 int dirty
, num_illegal
, error
, nres
;
2447 ino
= XFS_AGINO_TO_INO(mp
, agno
, irec
->ino_startnum
+ ino_offset
);
2450 * open up directory inode, check all entries,
2451 * then call prune_dir_entries to remove all
2452 * remaining illegal directory entries.
2455 ASSERT(!is_inode_refchecked(irec
, ino_offset
) || dotdot_update
);
2457 error
= libxfs_iget(mp
, NULL
, ino
, 0, &ip
, 0);
2461 _("couldn't map inode %" PRIu64
", err = %d\n"),
2465 _("couldn't map inode %" PRIu64
", err = %d\n"),
2468 * see below for what we're doing if this
2469 * is root. Why do we need to do this here?
2470 * to ensure that the root doesn't show up
2471 * as being disconnected in the no_modify case.
2473 if (mp
->m_sb
.sb_rootino
== ino
) {
2474 add_inode_reached(irec
, 0);
2475 add_inode_ref(irec
, 0);
2479 add_inode_refchecked(irec
, 0);
2483 need_dot
= dirty
= num_illegal
= 0;
2485 if (mp
->m_sb
.sb_rootino
== ino
) {
2487 * mark root inode reached and bump up
2488 * link count for root inode to account
2489 * for '..' entry since the root inode is
2490 * never reached by a parent. we know
2491 * that root's '..' is always good --
2492 * guaranteed by phase 3 and/or below.
2494 add_inode_reached(irec
, ino_offset
);
2497 add_inode_refchecked(irec
, ino_offset
);
2499 hashtab
= dir_hash_init(ip
->i_d
.di_size
);
2502 * look for bogus entries
2504 switch (ip
->i_d
.di_format
) {
2505 case XFS_DINODE_FMT_EXTENTS
:
2506 case XFS_DINODE_FMT_BTREE
:
2508 * also check for missing '.' in longform dirs.
2509 * missing .. entries are added if required when
2510 * the directory is connected to lost+found. but
2511 * we need to create '.' entries here.
2513 longform_dir2_entry_check(mp
, ino
, ip
,
2514 &num_illegal
, &need_dot
,
2519 case XFS_DINODE_FMT_LOCAL
:
2520 tp
= libxfs_trans_alloc(mp
, 0);
2522 * using the remove reservation is overkill
2523 * since at most we'll only need to log the
2524 * inode but it's easier than wedging a
2525 * new define in ourselves.
2527 nres
= no_modify
? 0 : XFS_REMOVE_SPACE_RES(mp
);
2528 error
= libxfs_trans_reserve(tp
, nres
,
2529 XFS_REMOVE_LOG_RES(mp
), 0,
2530 XFS_TRANS_PERM_LOG_RES
,
2531 XFS_REMOVE_LOG_COUNT
);
2535 libxfs_trans_ijoin(tp
, ip
, 0);
2536 libxfs_trans_ihold(tp
, ip
);
2538 shortform_dir2_entry_check(mp
, ino
, ip
, &dirty
,
2542 ASSERT(dirty
== 0 || (dirty
&& !no_modify
));
2544 libxfs_trans_log_inode(tp
, ip
,
2545 XFS_ILOG_CORE
| XFS_ILOG_DDATA
);
2546 libxfs_trans_commit(tp
,
2547 XFS_TRANS_RELEASE_LOG_RES
|
2550 libxfs_trans_cancel(tp
,
2551 XFS_TRANS_RELEASE_LOG_RES
);
2558 dir_hash_done(hashtab
);
2561 * if we have to create a .. for /, do it now *before*
2562 * we delete the bogus entries, otherwise the directory
2563 * could transform into a shortform dir which would
2564 * probably cause the simulation to choke. Even
2565 * if the illegal entries get shifted around, it's ok
2566 * because the entries are structurally intact and in
2567 * in hash-value order so the simulation won't get confused
2568 * if it has to move them around.
2570 if (!no_modify
&& need_root_dotdot
&& ino
== mp
->m_sb
.sb_rootino
) {
2571 ASSERT(ip
->i_d
.di_format
!= XFS_DINODE_FMT_LOCAL
);
2573 do_warn(_("recreating root directory .. entry\n"));
2575 tp
= libxfs_trans_alloc(mp
, 0);
2578 nres
= XFS_MKDIR_SPACE_RES(mp
, 2);
2579 error
= libxfs_trans_reserve(tp
, nres
, XFS_MKDIR_LOG_RES(mp
),
2580 0, XFS_TRANS_PERM_LOG_RES
, XFS_MKDIR_LOG_COUNT
);
2584 libxfs_trans_ijoin(tp
, ip
, 0);
2585 libxfs_trans_ihold(tp
, ip
);
2587 xfs_bmap_init(&flist
, &first
);
2589 error
= libxfs_dir_createname(tp
, ip
, &xfs_name_dotdot
,
2590 ip
->i_ino
, &first
, &flist
, nres
);
2593 _("can't make \"..\" entry in root inode %" PRIu64
", createname error %d\n"), ino
, error
);
2595 libxfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
2597 error
= libxfs_bmap_finish(&tp
, &flist
, &committed
);
2599 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
|
2602 need_root_dotdot
= 0;
2603 } else if (need_root_dotdot
&& ino
== mp
->m_sb
.sb_rootino
) {
2604 do_warn(_("would recreate root directory .. entry\n"));
2608 * if we need to create the '.' entry, do so only if
2609 * the directory is a longform dir. if it's been
2610 * turned into a shortform dir, then the inode is ok
2611 * since shortform dirs have no '.' entry and the inode
2612 * has already been committed by prune_lf_dir_entry().
2616 * bump up our link count but don't
2617 * bump up the inode link count. chances
2618 * are good that even though we lost '.'
2619 * the inode link counts reflect '.' so
2620 * leave the inode link count alone and if
2621 * it turns out to be wrong, we'll catch
2624 add_inode_ref(irec
, ino_offset
);
2628 _("would create missing \".\" entry in dir ino %" PRIu64
"\n"),
2630 } else if (ip
->i_d
.di_format
!= XFS_DINODE_FMT_LOCAL
) {
2632 * need to create . entry in longform dir.
2635 _("creating missing \".\" entry in dir ino %" PRIu64
"\n"), ino
);
2637 tp
= libxfs_trans_alloc(mp
, 0);
2640 nres
= XFS_MKDIR_SPACE_RES(mp
, 1);
2641 error
= libxfs_trans_reserve(tp
, nres
,
2642 XFS_MKDIR_LOG_RES(mp
),
2644 XFS_TRANS_PERM_LOG_RES
,
2645 XFS_MKDIR_LOG_COUNT
);
2650 libxfs_trans_ijoin(tp
, ip
, 0);
2651 libxfs_trans_ihold(tp
, ip
);
2653 xfs_bmap_init(&flist
, &first
);
2655 error
= libxfs_dir_createname(tp
, ip
, &xfs_name_dot
,
2656 ip
->i_ino
, &first
, &flist
, nres
);
2659 _("can't make \".\" entry in dir ino %" PRIu64
", createname error %d\n"),
2662 libxfs_trans_log_inode(tp
, ip
, XFS_ILOG_CORE
);
2664 error
= libxfs_bmap_finish(&tp
, &flist
, &committed
);
2666 libxfs_trans_commit(tp
, XFS_TRANS_RELEASE_LOG_RES
2674 * mark realtime bitmap and summary inodes as reached.
2675 * quota inode will be marked here as well
2678 mark_standalone_inodes(xfs_mount_t
*mp
)
2680 ino_tree_node_t
*irec
;
2683 irec
= find_inode_rec(mp
, XFS_INO_TO_AGNO(mp
, mp
->m_sb
.sb_rbmino
),
2684 XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_rbmino
));
2686 ASSERT(irec
!= NULL
);
2688 offset
= XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_rbmino
) -
2691 add_inode_reached(irec
, offset
);
2693 irec
= find_inode_rec(mp
, XFS_INO_TO_AGNO(mp
, mp
->m_sb
.sb_rsumino
),
2694 XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_rsumino
));
2696 offset
= XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_rsumino
) -
2699 ASSERT(irec
!= NULL
);
2701 add_inode_reached(irec
, offset
);
2704 if (mp
->m_sb
.sb_uquotino
2705 && mp
->m_sb
.sb_uquotino
!= NULLFSINO
) {
2706 irec
= find_inode_rec(mp
, XFS_INO_TO_AGNO(mp
,
2707 mp
->m_sb
.sb_uquotino
),
2708 XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_uquotino
));
2709 offset
= XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_uquotino
)
2710 - irec
->ino_startnum
;
2711 add_inode_reached(irec
, offset
);
2713 if (mp
->m_sb
.sb_gquotino
2714 && mp
->m_sb
.sb_gquotino
!= NULLFSINO
) {
2715 irec
= find_inode_rec(mp
, XFS_INO_TO_AGNO(mp
,
2716 mp
->m_sb
.sb_gquotino
),
2717 XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_gquotino
));
2718 offset
= XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_gquotino
)
2719 - irec
->ino_startnum
;
2720 add_inode_reached(irec
, offset
);
2726 check_for_orphaned_inodes(
2728 xfs_agnumber_t agno
,
2729 ino_tree_node_t
*irec
)
2734 for (i
= 0; i
< XFS_INODES_PER_CHUNK
; i
++) {
2735 ASSERT(is_inode_confirmed(irec
, i
));
2736 if (is_inode_free(irec
, i
))
2739 if (is_inode_reached(irec
, i
))
2742 ASSERT(inode_isadir(irec
, i
) ||
2743 num_inode_references(irec
, i
) == 0);
2745 ino
= XFS_AGINO_TO_INO(mp
, agno
, i
+ irec
->ino_startnum
);
2746 if (inode_isadir(irec
, i
))
2747 do_warn(_("disconnected dir inode %" PRIu64
", "), ino
);
2749 do_warn(_("disconnected inode %" PRIu64
", "), ino
);
2752 orphanage_ino
= mk_orphanage(mp
);
2753 do_warn(_("moving to %s\n"), ORPHANAGE
);
2754 mv_orphanage(mp
, ino
, inode_isadir(irec
, i
));
2756 do_warn(_("would move to %s\n"), ORPHANAGE
);
2759 * for read-only case, even though the inode isn't
2760 * really reachable, set the flag (and bump our link
2761 * count) anyway to fool phase 7
2763 add_inode_reached(irec
, i
);
2770 xfs_agnumber_t agno
,
2773 ino_tree_node_t
*irec
;
2775 prefetch_args_t
*pf_args
= arg
;
2777 wait_for_inode_prefetch(pf_args
);
2780 do_log(_(" - agno = %d\n"), agno
);
2782 for (irec
= findfirst_inode_rec(agno
); irec
; irec
= next_ino_rec(irec
)) {
2783 if (irec
->ino_isa_dir
== 0)
2787 sem_post(&pf_args
->ra_count
);
2789 for (i
= 0; i
< XFS_INODES_PER_CHUNK
; i
++) {
2790 if (inode_isadir(irec
, i
))
2791 process_dir_inode(wq
->mp
, agno
, irec
, i
);
2794 cleanup_inode_prefetch(pf_args
);
2798 update_missing_dotdot_entries(
2801 dotdot_update_t
*dir
;
2804 * these entries parents were updated, rebuild them again
2805 * set dotdot_update flag so processing routines do not count links
2808 while (dotdot_update_list
) {
2809 dir
= dotdot_update_list
;
2810 dotdot_update_list
= dir
->next
;
2811 process_dir_inode(mp
, dir
->agno
, dir
->irec
, dir
->ino_offset
);
2822 prefetch_args_t
*pf_args
[2];
2825 * we always do prefetch for phase 6 as it will fill in the gaps
2826 * not read during phase 3 prefetch.
2829 pf_args
[0] = start_inode_prefetch(0, 1, NULL
);
2830 for (i
= 0; i
< glob_agcount
; i
++) {
2831 pf_args
[(~i
) & 1] = start_inode_prefetch(i
+ 1, 1,
2833 traverse_function(&queue
, i
, pf_args
[i
& 1]);
2838 phase6(xfs_mount_t
*mp
)
2840 ino_tree_node_t
*irec
;
2843 memset(&zerocr
, 0, sizeof(struct cred
));
2844 memset(&zerofsx
, 0, sizeof(struct fsxattr
));
2847 do_log(_("Phase 6 - check inode connectivity...\n"));
2849 incore_ext_teardown(mp
);
2851 add_ino_ex_data(mp
);
2854 * verify existence of root directory - if we have to
2855 * make one, it's ok for the incore data structs not to
2856 * know about it since everything about it (and the other
2857 * inodes in its chunk if a new chunk was created) are ok
2859 if (need_root_inode
) {
2861 do_warn(_("reinitializing root directory\n"));
2863 need_root_inode
= 0;
2864 need_root_dotdot
= 0;
2866 do_warn(_("would reinitialize root directory\n"));
2872 do_warn(_("reinitializing realtime bitmap inode\n"));
2876 do_warn(_("would reinitialize realtime bitmap inode\n"));
2882 do_warn(_("reinitializing realtime summary inode\n"));
2886 do_warn(_("would reinitialize realtime summary inode\n"));
2892 _(" - resetting contents of realtime bitmap and summary inodes\n"));
2893 if (fill_rbmino(mp
)) {
2895 _("Warning: realtime bitmap may be inconsistent\n"));
2898 if (fill_rsumino(mp
)) {
2900 _("Warning: realtime bitmap may be inconsistent\n"));
2904 mark_standalone_inodes(mp
);
2906 do_log(_(" - traversing filesystem ...\n"));
2908 irec
= find_inode_rec(mp
, XFS_INO_TO_AGNO(mp
, mp
->m_sb
.sb_rootino
),
2909 XFS_INO_TO_AGINO(mp
, mp
->m_sb
.sb_rootino
));
2912 * we always have a root inode, even if it's free...
2913 * if the root is free, forget it, lost+found is already gone
2915 if (is_inode_free(irec
, 0) || !inode_isadir(irec
, 0)) {
2916 need_root_inode
= 1;
2920 * then process all inodes by walking incore inode tree
2925 * any directories that had updated ".." entries, rebuild them now
2927 update_missing_dotdot_entries(mp
);
2929 do_log(_(" - traversal finished ...\n"));
2930 do_log(_(" - moving disconnected inodes to %s ...\n"),
2934 * move all disconnected inodes to the orphanage
2936 for (i
= 0; i
< glob_agcount
; i
++) {
2937 irec
= findfirst_inode_rec(i
);
2938 while (irec
!= NULL
) {
2939 check_for_orphaned_inodes(mp
, i
, irec
);
2940 irec
= next_ino_rec(irec
);